1: \documentclass[twocolumn,
2: floatfix,showpacs,
3: amsmath,amssymb,pre,showpacs]{revtex4}
4:
5: \usepackage{graphicx}
6: \usepackage{amsmath}
7: \usepackage{bm}
8: \usepackage{isolatin1}
9: \newcommand{\de}{\partial}
10: \newcommand{\eps}{\varepsilon}
11: \newcommand{\NOT}[1]{\overline{#1}}
12: \newcommand{\AND}{\wedge}
13: \newcommand{\OR}{\vee}
14: \newcommand{\XOR}{\oplus}
15: \renewcommand{\v}[1]{\boldsymbol{#1}}
16: \renewcommand{\P}{\mathcal{P}}
17:
18: \DeclareGraphicsExtensions{.eps}
19:
20: \begin{document}
21:
22:
23: \title{Phase transitions of extended-range probabilistic cellular
24: automata with two absorbing states}
25:
26: \author{Franco Bagnoli}
27: \altaffiliation{also INFN, sez. Firenze and CSDC}
28: \email{franco.bagnoli@unifi.it}
29: \affiliation{Dipartimento di Energetica, Universit\`a di Firenze,\\
30: Via S. Marta 3, 50139 Firenze, Italy}
31: \author{Fabio Franci}
32: \email{fabio@dma.unifi.it}
33: \affiliation{Centro Interdipartimentale per lo Studio delle Dinamiche
34: Complesse (CSDC), \\Universit\`a di Firenze, \\
35: Via Sansone 1,
36: 50019 Sesto Fiorentino, Italy}
37: \author{Ra\'ul Rechtman}
38: \email{rrs@cie.unam.mx}
39: \affiliation{Centro de Investigaci\'on en Energ\'\i a, Universidad
40: Nacional Aut\'onoma
41: de M\'exico, \\Apdo.\ Postal 34, 62580 Temixco, Mor., Mexico\\
42: }
43:
44: \date{\today}
45:
46: \begin{abstract}
47: We study phase transitions in a long-range one-dimensional cellular
48: automaton with two symmetric absorbing states. It includes and extends
49: several other models, like the Ising and Domany-Kinzel ones.
50: It is characterized by a competing ferromagnetic linear coupling and an
51: antiferromagnetic nonlinear one.
52: Despite its simplicity, this model exhibits an extremely rich phase diagram.
53: We present numerical results and mean-field approximations.
54: \end{abstract}
55:
56: \pacs{68.35.Rh,64.60.-i,05.50.+q}
57:
58: \maketitle
59:
60: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
61:
62: \section{Introduction}
63:
64: Simple discrete models are valuable tools for exploring phase
65: transitions, both in equilibrium and out of equilibrium. A
66: paradigmatic example of the first kind is the kinetic Ising model
67: (heat-bath dynamics), which can be formulated as a Probabilistic
68: Cellular Automaton (PCA) (see Appendix and Ref.~\cite{Georges}).
69: One of the simplest
70: examples of a system exhibiting out of equilibrium phase transitions
71: is the Domany-Kinzel (DK) model~\cite{DP}, which is a natural
72: extension of the directed percolation (DP) problem in the language of
73: cellular automata.
74:
75: Frustrations play a central role in characterizing the phase diagrams of many
76: simple models. One can have geometric frustrations, like in a
77: triangular antiferromagnetic Ising model, or fluctuating coupling,
78: like in spin glasses~\cite{spinglass} or p-spin~\cite{pspin} models.
79: For instance, in the p-spin
80: model, the character (ferro or anti-ferro) of the interaction varies
81: widely with the local magnetization (i.e. even a flip of a single
82: spin may invert the interaction). We discuss here a simpler one dimensional
83: model in which frustrations play a central role.
84: This model is a variation of the droplet model in which the particles
85: tend to repel each other when they are dispersed but in which clusters, once
86: formed, cannot break and only particles near the surface can eventually leave
87: them.
88:
89: The model originated from the schematization of the mechanism of opinion
90: formation in a homogeneous (no leader) society~\cite{acri02}, in which it was
91: assumed that
92: a coherent local community exerts a very strong social pressure on an
93: individual's opinion. However, there is the possibility of disagreement
94: with a weak local majority depending on the individual's education.
95: %As an example, let us consider dressing
96: %habits: there are some rules that are ``almost universally'' accepted in a
97: %given society (from ``do not show sexual zones'', to ``tie required'',
98: %according with the context), while others are often violated (``do not wear
99: %clear shoes in winter''). Since these rules are not codified into DNA, one may
100: %assume that they represent a sort of \emph{absorbing state} in social
101: %behavior, due to the homogeneity of local communities. Cultural shifts (burqa,
102: %t-shirts, etc) indeed correspond to a change of the local communities.
103: In the case of a diffuse non-conformistic attitude, people tend to act in the
104: opposite way of the local majority (antiferromagnetic coupling), introducing
105: frustrations. The presence of absorbing states may cause the
106: formation of large coherent clusters, whose interactions give way to
107: interesting patterns.
108:
109: This model could also be applied to a system of charged,
110: magnetized metallic particles. These repel each other due to their charge, but
111: are attracted due to their magnetic coupling. With a proper choice of
112: parameters,
113: one may have repulsion when particles are scattered. However,
114: once a cluster
115: of particles in contact appears, the mobility of charges may lower the electric
116: field among this cluster and an external particle, and the interaction may
117: become attractive. Other examples can be found in aggregation models.
118:
119: In what follows, we discuss a one dimensional spin model
120: characterized by two
121: coupling factors; one that behaves like the Ising (magnetic)
122: coupling, i.e., for a given spin, grows linearly with the local
123: field, and a \emph{nonlinear term} that is very low for a weak local
124: field, but grows quickly when the local field exceeds a certain
125: threshold.
126:
127: In one
128: dimension, no true phase transition appears for finite-range and
129: finite-strength coupling. On the other hand, models presenting absorbing
130: states, like the DK one, do exhibit nonequilibrium phase transitions. The
131: presence of absorbing states may be related to the divergence of some coupling,
132: that becomes infinite (see Appendix and Ref.~\cite{Georges}). Thus, in our
133: model, we simply assume a two-level nonlinear coupling: zero for local field
134: below a given threshold (corresponding to the parameter $Q$ in the following)
135: and infinite otherwise. Thus, in total we have four parameters: the range
136: $R$, the external magnetic field $H$ and local coupling $J$ like in the Ising
137: model, and the threshold $Q$. We denote these models by $RQ$, indicating
138: the range and the threshold, say $R3Q1$, $R5Q2$, \dots
139:
140: The model can also be discussed as a one dimensional probabilistic
141: cellular automaton. In this guise, it may be considered an extension
142: of the DK model, or, more precisely, of the $R3Q1$ model~\cite{Bagnoli_2001}.
143: This model exhibits a richer
144: phase diagram than the DK one, and we shall show that for
145: $R\to\infty$, the $RQ$ model presents novel features, like a
146: disorder-chaotic transition.
147:
148: It is not surprising that frustrations gives rise to disordered
149: behavior, which may be sometimes considered chaotic from a
150: microscopic point of view. We investigate this aspect using an
151: original chaotic observable, corresponding to a finite-distance
152: Lyapunov exponent. However, when renormalized (or locally averaged),
153: a ``disordered'' behavior simply corresponds to a stable almost-zero
154: magnetization. In the mean-field approximation, it corresponds to a
155: fixed point of observables. In our model also a coherently-chaotic
156: phase appears, in which large patches of sites oscillate widely,
157: corresponding to a chaotic map in the mean-field approximation.
158:
159: In summary, we observe that the \emph{usual} transitions from
160: ordered into active phases becomes much richer, being preceded by a
161: transition from coherently chaotic to simply chaotic, then to more
162: irregular states, the appearance of a parameter-insensitive
163: disordered phase and finally an ``escape'' to ordered (quiescent)
164: states with a discontinuous (first-order) character. The origin of
165: this behavior is the competition between ``linear''
166: antiferromagnetic and ``nonlinear'' ferromagnetic couplings.
167:
168: The paper is organized as follows. In Sec.~\ref{sec:model} we present
169: the $RQ$ model both as a spin system and a PCA. In
170: Sec.~\ref{sec:results} we present numerical results of the model
171: beginning with a review of the simplest non trivial case $R=3$,
172: $Q=1$ and Sec.~\ref{sec:approximations} is devoted to a mean field
173: discussion of the model. The paper ends with some conclusions.
174:
175: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
176:
177: \section{The model}
178: \label{sec:model}
179:
180: Both the kinetic Ising model (see Appendix) and the DK one are
181: Boolean, one dimensional, $R=2$ PCA in which the state of a given
182: cell depends probabilistically on the sum $S$ of the states of its
183: two nearest neighbors at the previous time step. That is, there are
184: three different conditional probabilities, $\tau(1|S)$, $S=0,1,2$ that the
185: future state of the cell is 1 given $S$ neighbors in state 1. In the
186: DK model $\tau(1|0)=0$ which means that there is an absorbing phase
187: corresponding to the configuration where the state on every cell is
188: zero. The presence of more than one absorbing state can induce
189: different behaviors and trigger the appearance of universality
190: classes different from the usual directed
191: percolation (DP) one~\cite{Hinrichsen_1997,Janssen_1981,Grassberger_1982}.
192:
193: We denote by $s_i^t\in\{0,1\}$ the state at site $i=0,\dots,L-1$ and
194: time $t=0,1\dots$. All operations on spatial indices are assumed to
195: be modulo $L$ (periodic boundary conditions). The range of
196: interactions is denoted by $R$, that is, $s_i^{t+1}$ depends on
197: the states at time $t$ in a neighborhood ${\cal N}_i$ that contains
198: the $R$ nearest neighbors of site $i$. It is
199: convenient to introduce also the spin notation: $\sigma_i^t = 2
200: s_i^t-1$, $\sigma_i^t \in \{-1,1\}$.
201:
202: The spin $\sigma_i^t$ ``feels'' the local field $V(m_i^t) =
203: H+[J +K(m_i^t)] m_i^t$, where $H$ is an external field, $J$ a coupling
204: constant and the local magnetization is defined as
205: \[
206: m_i^t=\sum_{j\in{\cal N}_i} \sigma_j^t=2S_i^t-R,
207: \]
208: where
209: \[
210: S_i^t=\sum_{j\in{\cal N}_i} s_j^t.
211: \]
212: The presence of absorbing states is due to the nonlinear coupling $K$
213: given by
214: \[
215: K(m)=K(m(S,R),Q) = \begin{cases}
216: -k & \text{if $S<Q$,}\\
217: k & \text{if $S > R-Q$,}\\
218: 0 & \text{otherwise.}\\
219: \end{cases}
220: \]
221: with $k$ a constant.
222: This term is relevant only if it is dominant with
223: respect to the linear one. We thus choose the limit $k\rightarrow \infty$, so
224: that the transition probabilities $\tau(1|S)$ are given by
225: \begin{equation}\label{Eq:tau_s}
226: \tau(1|S) = \begin{cases}
227: 0 & \text{if $S<Q$,}\\
228: 1 & \text{if $S > R-Q$,}\\
229: \dfrac{1}{1+\exp\left\{-2\left[H+Jm(S,R)\right]\right\}}
230: & \text{otherwise.}\\
231: \end{cases}
232: \end{equation}
233:
234: For $Q=0$ we recover the usual heat-bath
235: dynamics for an Ising-like
236: model with reduced Hamiltonian $\mathcal{H} = \sum_i V(m_i) \sigma_i$.
237: For $R=3$, $Q=1$ we have the model of Ref.~\cite{Bagnoli_2001} with
238: two absorbing states.
239: The quantities $H$ and $J$ range from $-\infty$ to $\infty$.
240: For easy plotting, we use
241: $j = [1+\exp(-2J)]^{-1}$ and $h = [1+\exp(-2H)]^{-1}$
242: as control parameters, mapping the real axis ($-\infty, \infty$) to
243: the interval $[0,1]$.
244:
245: The fraction $c$ of ones in a
246: configuration and the concentration of clusters $\rho$ are defined by
247: \[
248: c=\dfrac{1}{L}\sum_i s_i\quad\text{and}\quad%
249: \rho=\dfrac{1}{L}\sum_i |s_i - s_{i+1}|.
250: \]
251: Both the uniform zero-state and
252: one-state correspond to $\rho\rightarrow 0$ in the thermodynamic
253: limit, while the active state corresponds to $\rho > 0$.
254:
255:
256:
257:
258: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
259:
260: \section{Numerical Results}
261: \label{sec:results}
262:
263: \begin{figure}
264: \includegraphics[width=8cm]{figuraR3.eps}
265: \caption{\label{figuraR3Q1}
266: Phase diagram of $R3Q1$ model. The upper--left region
267: corresponds to the absorbing state 0, the upper--right region to the
268: absorbing state 1 and the lower region to the disordered phase. In the
269: inset we show the asymptotic value of $\rho$ as a function of $j$
270: for $H=0$ ($h=0.5$).}
271: \end{figure}
272:
273: \begin{figure}
274: \begin{tabular}{cc}
275: \includegraphics[width=4cm]{ordineR3.eps} &
276: \includegraphics[width=4cm]{disordine.eps} \\
277: \includegraphics[width=4cm]{percolaz.eps} & \\
278: \end{tabular}
279: \caption{\label{patternsR3}
280: Typical space time patterns of the $R3Q1$ model for $H=0$ ($h=0.5$)
281: with $L=256$.
282: (A~--~top left) Chaotic deterministic pattern $j=0.03125$.
283: (B~--~top right) Disordered phase $j=0.5$.
284: (C~--~bottom left) competition between quiescent phases $j=0.54688$
285: (first-order transition).
286: }
287: \end{figure}
288:
289: Let us first review the $R3Q1$ case, whose phase diagram is reported in
290: Fig.~\ref{figuraR3Q1}. A first order phase transition occurs on the vertical
291: line that ends at a bicritical point where it meets two second order phase
292: transition curves. In the neighborhood of the first order phase transition line
293: a typical pattern of the system is composed by large patches of zeroes and
294: ones, separated by disordered zones (walls) whose width does not grow in time.
295: These walls perform a sort of random motion and annihilate in pairs (see
296: Fig.~\ref{patternsR3}-C). By lowering $J$ an active, disordered phase appears
297: (see Fig.~\ref{patternsR3}-B). The transition between the ordered (absorbing)
298: and disordered (active) phases occurs by destabilization of the width of the
299: walls, that percolate in the whole system. In the limit $J=-\infty$ ($j=0$) we
300: have typical ``chaotic'' class-3 cellular automata
301: patterns~\cite{wolframclasses}, as shown in
302: Fig.~\ref{patternsR3}-A. Damage spreading analysis~\cite{Bagnoli_2001}
303: show that inside the active phase there is a region in which a
304: variation in the initial configuration can influence the asymptotic
305: configuration. This region may be called chaotic. However, in the mean
306: field approximation, this region simply corresponds to a stable fixed
307: point of the average density. Indeed, the fluctuations of the density
308: remains quite small and a spatial coarse-graining would generate a simple
309: random pattern.
310:
311: This underlining ``disorder'' is revealed also by
312: the asymptotic value of $\rho$ for $H=0$ ($h=0.5$)
313: as a function of $j$, shown in the inset of Fig.~\ref{figuraR3Q1}.
314: It exhibits a monotonic behavior, with
315: a maximum for $j=0$ (corresponding to the smallest average length of
316: homogeneous clusters) and a continuous phase transition at the bicritical
317: point, that separates the active ($\rho>0$) and quiescent ($\rho=0$)
318: phases.
319:
320:
321: %
322: \begin{figure}
323: \begin{tabular}{cc}
324: \includegraphics[width=4cm]{H0.500r05Q01N09-056250.eps} &
325: \includegraphics[width=4cm]{H0.500r05Q01N09-421875.eps} \\
326: \includegraphics[width=4cm]{H0.500r05Q01N09-478125.eps} &
327: \includegraphics[width=4cm]{H0.500r05Q01N09-562500.eps}\\
328: \end{tabular}
329: \caption{\label{fig:patterns}
330: Typical space time patterns ($L=256$) for model $R11Q1$ and $H=0$,
331: starting from a disordered initial condition with $\rho_0=0.5$.
332: (A) [top left, $j=0.056250$]: ``coherently-chaotic'' phase.
333: One can observe rare
334: ``triangles'' with ``base'' larger that $R$,
335: corresponding to the unstable absorbing states, and other metastable
336: patterns, corresponding to additional absorbing states for
337: $J\rightarrow -\infty$.
338: (B) [top right, $j=0.421875$]: At the boundary between the chaotic and
339: the irregular phase,
340: the only locally absorbing states are (rare) patches of 0 and 1. Most
341: of the pattern looks random, with some ``triangles'' reminiscent of
342: chaotic CA.
343: (C) [bottom left, $j=0.478125$]: Disordered phase. the pattern looks
344: random and the difference between patterns obtained with different
345: values of the parameters (and the same random numbers) is vanishing.
346: (D) [bottom right, $j=0.562500$]:
347: Quiescent phase. In this phase the only stable states are the absorbing
348: ones. The boundaries separating the phases move randomly until coalescence.}
349: \end{figure}
350: %
351: \begin{figure}
352: \includegraphics[width=8cm]{inset2.eps}
353: \caption{\label{fig:rho}Behavior of $\rho$ for $H=0$. $R3Q1$ (long dashed),
354: $R11Q1$ (solid), $R41Q1$ (dashed)
355: and $R81Q1$ (dotted). Letters correspond to patterns in
356: Fig~\protect\ref{fig:patterns}.}
357: \end{figure}
358:
359: By increasing $R$, new features appear. Typical patterns are
360: shown in Fig.~\ref{fig:patterns} for different values of $j$.
361: As illustrated in Fig.~\ref{fig:rho}, $\rho$ is no longer a monotonic
362: function of $j$, and a new, \emph{less disordered} region appears
363: inside the active one for small values of $j$. The transition between
364: the active and the quiescent phases become sharper with increasing
365: $R$, as shown in Fig.~\ref{fig:rho}.
366:
367: The pattern shown in Fig.~\ref{fig:patterns}-a is reminiscent of
368: ``chaotic'' deterministic cellular automata. We refer to
369: this region as \emph{coherently-chaotic}, since it corresponds to
370: ``irregular'' coherent oscillations of large patches of sites. This
371: region of the parameters is dominated by an almost-deterministic
372: behavior, and the presence of many metastable ``absorbing'' states,
373: revealed by transient regular patterns in Fig.~\ref{fig:patterns}-a.
374: As we discuss in
375: Sec.~\ref{sec:approximations}, the mean field analysis gives a
376: chaotic map for large antiferromagnetic values of $J$,
377: but we were unable to find a clear order parameter that numerically
378: distinguishes this region from the broader ``chaotic'' one in
379: which the asymptotic configuration exhibits a dependence on variations in the
380: initial configuration. In Sec.~\ref{sec:lyap}, we analyze the
381: chaotic region by means of (finite size) Lyapunov exponents.
382:
383: By increasing $J$, the coherent
384: patches shrink and the Lyapunov exponent decreases and finally
385: becomes negative. The system asymptotically loses its dependence on
386: the initial
387: conditions and is dominated by the stochastic components.
388: This phase is termed \emph{irregular}, and appears completely random.
389: Fig.~\ref{fig:patterns}-b shows a typical pattern at the boundary
390: between the chaotic and the irregular phase.
391:
392: Our simulations were carried out using the fragment
393: method~\cite{fragment}, in which
394: a set of configurations (replicas) are
395: updated with different values of parameters using the same random numbers,
396: and the same initial configuration.
397:
398: By observing in rapid sequence the patterns generated with increasing
399: values of $J$, one observes a sudden ``freezing'' of the (random)
400: image, just before the transition between the active to quiescent
401: phases, corresponding to Fig.~\ref{fig:patterns}-c. This
402: effect is due to the insensitivity of the patterns not only to the
403: initial condition, but also to the parameters, see
404: Sec~\ref{Sec:irregular-disorder}. We term this phase
405: \emph{disordered}.
406:
407: Finally, the quiescent phase is
408: asymptotically dominated by the two absorbing states, with the usual
409: annihilating walls dynamics, Fig.~\ref{fig:patterns}-d.
410:
411: Fig.~\ref{fig:rho} shows that the active-quiescent transition becomes
412: sharper when $R$ is increased, and that the cluster density
413: exhibits a cusp at the transition. By enlarging this region one sees
414: that for $H=0$ (Fig.~\ref{fig:H0}) the cluster density exhibits two
415: sharp bends, while for $H =0.42$, (Fig.~\ref{fig:H042}), only one
416: bend is present. This is reminiscent of the universality class
417: change (parity conservation, DP) for $H=0$, $H \ne 0$, respectively, in the
418: R3Q1 model~\cite{Bagnoli_2001}. Notice also that the
419: irregular--disorder transition at $j \simeq 0.494$ is rather
420: peculiar, since $\rho$ first decreases and then suddenly jumps to
421: high values.
422:
423: The bends are not finite size or time effects. As shown in
424: Fig.~\ref{fig:patterns}-c,
425: in this range of parameters the probability of observing
426: a local absorbing configuration (i.e.\ patches of zeroes or ones) is
427: vanishing. All other local configurations have finite probability of
428: originating zeroes or ones in the next time step. The observed
429: transitions are essentially equivalent to those of an equilibrium system,
430: that in one dimension and for short-range interactions cannot exhibit
431: a true phase transition. The bends thus become real salient points only
432: in the limit $R\rightarrow \infty$.
433: %
434: \begin{figure}
435: \includegraphics[width=8cm]{H05.eps}
436: \caption{\label{fig:H0}
437: Comparisons between numerical (thin line) and mean-field (thick, dotted line)
438: results for $R81Q1$ and $H=0$ ($h=0.5$). The estimated critical values are
439: $j^*_{\text{I}} \simeq 0.493827$ and $j^*_{\text{II}} \simeq 0.51384$.
440: This figure has been obtained with much larger simulations of the corresponding line in Fig.~\protect\ref{fig:rho}, and more details emerge. The patterns shown in Fig.~\protect\ref{fig:patterns} (which however refer to $R=11$) illustrate the typical behavior of the system to the left (B), in the middle (C) and to the right (D) of the plateau.}
441: \end{figure}
442: %
443: \begin{figure}
444: \includegraphics[width=8cm]{all.eps}
445: \caption{\label{fig:H042}
446: Comparisons between numerical (thin line) and mean-field (thick, dotted line)
447: results for $R201Q5$ and $H=0.42$ ($h=0.7$). The two set of pluses
448: mark the period-two region.
449: The estimated critical values are
450: $j^*_{\text{I}} \simeq 0.49740$ and $j^*_{\text{II}} \simeq 0.50369$.}
451: \end{figure}
452:
453: \subsection{Chaos and Lyapunov exponent in cellular automata}
454: \label{sec:lyap}
455:
456: State variables in cellular automata are discrete, and thus the usual linear
457: stability analysis classifies them as stable systems. The occurrence of
458: disordered patterns and their propagation in stable dynamical systems can be
459: classified into two main groups: \emph{transient chaos} and \emph{stable
460: chaos}.
461:
462: %
463: \begin{figure}
464: \includegraphics[width=8cm]{Simulazioni.eps}
465: \caption{\label{fig:Sim}
466: Contour plot of the maximum Lyapunov $\lambda$ exponents for different values
467: of $R$ and $j$, for $H=0$. The solid line represent
468: the boundary between the $\lambda\ge 0$ phase and the $\lambda=-\infty$ one.}
469: \end{figure}
470:
471: Transient chaos is an irregular behavior of finite lifetime characterized by
472: the coexistence in the phase space of stable attractors and chaotic non
473: attracting sets, namely chaotic saddles or repellers~\cite{Tel}. After a
474: transient irregular behavior, the system generally collapses abruptly onto a
475: non-chaotic attractor. This kind of behavior is reminiscent of some
476: {\em class-4}
477: (nonchaotic) deterministic cellular automata
478: (DCA)~\cite{wolframclasses}, and can
479: be present in continuous systems having a discrete dynamics as a limiting
480: case~\cite{stablechaos}.
481:
482: Stable chaos constitutes a different kind of transient irregular
483: behavior \cite{CK88,PLOK} which cannot be ascribed to the presence of
484: chaotic saddles and therefore to divergence of nearby trajectories.
485: Moreover, the lifetime of this transient regime may scale
486: exponentially with the system size
487: (supertransients as defined in Refs.~\cite{CK88,PLOK}),
488: and the final stable attractor is
489: practically never reached for large enough systems. One is thus
490: allowed to assume that such transients may be of substantial
491: experimental interest and become the only physically relevant states
492: in the thermodynamic limit.
493:
494: The emergence of this ``chaoticity'' in {\em class-3} (chaotic)
495: DCA dynamics, is effectively illustrated by the damage spreading
496: analysis~\cite{Damage1,Damage2}, which measures the sensitivity to initial
497: conditions and for this reason is considered as the natural extension of the
498: Lyapunov technique to discrete systems. In this method, indeed, one monitors
499: the behavior of the distance between two replicas of the system evolving from
500: slightly different initial conditions, or the speed of propagation of a
501: disturbance~\cite{wolfram}. The dynamics is considered unstable and
502: the DCA is called chaotic, whenever a small initial difference between replicas
503: spreads through the whole system. On the contrary, if the initial difference
504: eventually freezes or disappears, the DCA is considered non chaotic.
505:
506: Due to the limited number of states of the automaton, damage spreading
507: does not account for the maximal ``production of uncertainty'', since the two
508: replicas may synchronize locally just by chance (self-annihilation of the
509: damage). Moreover, there are different definitions of damage spreading for the
510: same rule~\cite{DomanyHinrichsen}.
511:
512: To better understand the nature of the active phase, and up to what extent it
513: can be denoted \emph{chaotic}, we extend the finite-distance Lyapunov exponent
514: definition~\cite{LyapunovCA} to probabilistic cellular automata. A similar
515: approach has been used in Ref.~\cite{luque}, calculating the Lyapunov exponents
516: of a Kauffman random Boolean network in the annealed approximation. As shown in
517: this latter paper, this computation gives the value of the (classic) Lyapunov
518: exponent obtained by the analysis of time-series data using the Wolf
519: algorithm~\cite{wolf}.
520:
521: Given a Boolean function $f(x,y,\dots)$,
522: we define the Boolean derivative $\de f/\de x$ with respect to $x$ by
523: \begin{equation}
524: \label{eq:boolean}
525: \dfrac{\de f}{\de x} = \begin{cases}
526: 1 & \text{if $f(|x-1|, y, \dots) \neq f(x,y,\dots)$,}\\
527: 0 & \text{otherwise,}
528: \end{cases}
529: \end{equation}
530: which represents a measure of sensitivity of a function with respect to $x$.
531: The evolution rule of a probabilistic cellular automaton
532: can be written as
533: \[
534: x_i^{t+1}=\begin{cases}
535: 1 & r<\tau(1|S_i^t),\\
536: 0 & \text{otherwise}
537: \end{cases}
538: \]
539: where $r$ is a random number uniformly distributed in $[0,1]$. The Boolean
540: derivative with respect to $x_j$ is evaluated by using the same $r$
541: in the comparison implied in Eq.~(\ref{eq:boolean}).
542:
543: For a PCA, we can thus build the Jacobian matrix $J_{ij} = \de
544: x_i^{t+1}/\de x_j^t$, $i,j=0,\dots,L-1$.
545: The maximum Lyapunov exponent $\lambda$ can now be defined
546: in the usual way as the expansion rate of a tangent
547: vector $\v{v}(t)$, whose time evolution is given by
548: \[
549: \v{v}(t+1) = \v{J} \v{v}(t).
550: \]
551: When all the components of $\v{v}$ become zero,
552: $\lambda=-\infty$ and no information about the initial configuration may
553: ``percolate'' as $t\to\infty$~\cite{LyapunovCA}. This maximum Lyapunov
554: exponent is also related to the synchronization properties of
555: cellular automata~\cite{synca}.
556:
557: A preliminary numerical computation of $\lambda$ for our model is reported in
558: Fig.~\ref{fig:Sim}. It can be noticed that the boundary $j_c(R)$ of
559: $\lambda\ge0$ is not monotonic with $R$, reaching a maximum value for $R\simeq
560: 11$. By comparisons with Fig.~\ref{fig:rho}, one can see that the chaotic phase
561: is included in the irregular one.
562:
563: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
564: \begin{figure}
565: \includegraphics[width=8cm]{diff.eps}
566: \caption{\label{fig:diff}
567: Susceptibility $\chi$ (continuous line) and average
568: number of clusters $\rho$ (dashed line) for $R11Q1$, $H=0$,
569: $\Delta j = 0.01$}
570:
571: \end{figure}
572:
573: \subsection{Parameter dependence}
574: \label{Sec:irregular-disorder}
575:
576: We can numerically characterize the irregular-disorder transition by looking at the
577: sensitivity of space-time patterns with respect to the variations of parameters.
578: Using the fragment method~\cite{fragment} a set of configurations (replicas) are
579: updated with different values of parameters using the same random numbers,
580: i.e.\ with the same disordered field and initial conditions.
581: The quantity
582: \[
583: \chi(J) = \lim_{\Delta j \rightarrow 0} \lim_{t\rightarrow
584: \infty} \sum_i\left| s_i^t(J+\Delta J) - s_i^t(J)\right|
585: \]
586: is a measure of the susceptibility. For large correlations one expects
587: very small differences among replicas. As shown in
588: Fig.~\ref{fig:diff}, this susceptibility is strongly correlated to
589: the average number of clusters $\rho$.
590: %
591:
592: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
593:
594: \section{Mean field approximation}
595: %\section{Approximations}
596: \label{sec:approximations}
597:
598: In the simplest mean field approximation, the
599: density $c$ evolves in time according to
600: \begin{equation}\label{rmf}
601: c(t+1) = \sum_{S=0}^{R} W^{(R,S)}(c(t))\tau(1|S),
602: \end{equation}
603: with
604: \[
605: W^{(R,S)}(c) = \binom{R}{S} c^S (1-c)^{R-S}.
606: \]
607: An example of a mean-field phase diagram is reported in
608: Fig.~\ref{fig:meanfieldphase}.
609:
610: For large $R$, $W^{(R,S)}$ can be approximated by
611: \begin{equation}\label{gaussian}
612: W^{(R,S)}(c) \simeq \dfrac{1}{\sqrt{2\pi
613: c(1-c)R}}\exp\left[-\dfrac{R(S/R -c)^2}{2c(1-c)}\right],
614: \end{equation}
615: and the summation can be replaced by an integral.
616: The parameters of the resulting equation may be rescaled
617: by using $\tilde J = JR$ and $\tilde Q=Q/R$.
618: In the limit $R \rightarrow \infty$ ($\tilde J$ and $\tilde Q$ fixed),
619: $c(t+1) = f(c(t); H, \tilde{J}) $
620: with
621: \begin{equation}\label{mf}
622: f = \begin{cases}
623: 0 & \text{if $c<\tilde{Q}$,}\\
624: 1 & \text{if $c > 1-\tilde{Q}$,}\\
625: \dfrac{1}{1+\exp(-2(H+\tilde{J}(2c-1))} & \text{otherwise.}\\
626: \end{cases}
627: \end{equation}
628: The requirement of constant rescaled variables means that
629: the variations of $J$ for large $R$ triggers large variations of
630: $\tilde J$, and consequently the transition region becomes sharper.
631:
632: This approximation is supposed to be particularly good in the
633: disordered phase, i.e.\ for $J, H\simeq 0$, as shown in
634: Fig.~\ref{fig:H0}. In the finite-dimensional
635: case, one has to add a noise term, representing the influence of
636: neighboring sites, whose amplitude is of the order of the width of
637: the Gaussian in Eq.~(\ref{gaussian}), $\sqrt{c(1-c)/R}$.
638:
639: In this mean-field approximation, the order parameters $c$ and the
640: numerical density of clusters $\rho$
641: are related by $\rho=2c(1-c)$.
642:
643: For $J, H\simeq 0$ the mean field approximation gives a stable fixed
644: point $c^*(H, \tilde J) \ne 0, 1$ for the density, which assumes the
645: maximum value $c^* =0.5$ for $J=H=0$.
646: %
647: \begin{figure}
648: \includegraphics[width=8cm]{figuraanalitic.eps}
649: \caption{\label{fig:meanfieldphase}
650: Mean-field phase diagram for $R=81$. The upper curves correspond to the
651: transition II from the quiescent to the disordered phases, $R81Q5$ (dashed)
652: and $R81Q1$ (dotted). The solid curve corresponds to the transition I
653: from the disordered to the irregular phase.
654: }
655: \end{figure}
656:
657: \begin{figure}
658: \includegraphics[width=8cm]{campomedio.eps}
659: \caption{ \label{fig:meanfield}
660: Mean field map
661: Eq.~(\ref{rmf}) in the neighborhood of transition I ($\tilde
662: J\simeq -1$, dashed line) and II
663: ($\tilde
664: J\simeq 1$, solid line), for $H=0$ and $\tilde Q=0.02$}
665: \end{figure}
666: %
667:
668: The origin of the irregular-disordered and
669: disorder--quiescent phase transitions marked by (I) and (II) respectively in
670: Figs.~\ref{fig:H0} and \ref{fig:H042}
671: is due to the loss of stability of the mean-field fixed point $c^* \ne 0, 1$
672: (see Eq.~(\ref{mf}) and Fig.~\ref{fig:meanfield}).
673:
674:
675: The transition I is given by
676: \begin{equation}\label{transI}
677: c^*_{\text{I}} = f(c^*_{\text{I}}; H, \tilde{J}); \qquad
678: \left.\dfrac{ \text{d} f(c; H, \tilde{J})}{\text{d}
679: c}\right|_{\text{I}} = -1.
680: \end{equation}
681: which would induce a period-doubling cascade in the mean-field
682: approximation (disregarding the presence of absorbing states).
683: The solution of Eq.~(\ref{transI})
684: %%for $H=H^*_{\text{I}}$ obtaining
685: is
686: %\begin{equation*}
687: %\begin{cases}
688: \begin{align*}
689: c^*_{\text{I}}&= \dfrac{1}{2}\left(1+\sqrt{1+\tilde{J}^{-1}}\right),\\
690: H^*_{\text{I}}& = - \tilde{J}\sqrt{1+\tilde{J}^{-1}} -
691: \dfrac{1}{2}\log\left(\dfrac{1-\sqrt{1+\tilde{J}^{-1}}}
692: {1+\sqrt{1+\tilde{J}^{-1}}}\right).
693: \end{align*}
694: %\end{cases}
695: %\end{equation*}
696: The critical value $J^*_{\text{I}}$ can be
697: obtained numerically once given the value of $H^*=H^*(J^*)$, and
698: corresponds to the left bend in Fig.~\ref{fig:H0} and
699: \ref{fig:H042}.
700:
701: For $H=0$ (Fig.~\ref{fig:H0}) the period-doubling instability
702: brings the local
703: configuration into an absorbing state, and the lattice dynamics is
704: therefore driven by interactions among patches which are locally
705: absorbing. This essentially corresponds to the dynamics of a
706: Deterministic Cellular Automata (DCA) of \emph{chaotic} type, i.e.\
707: a system which is insensitive to infinitesimal perturbations but
708: reacts in an unpredictable way to \emph{finite} perturbations
709: \cite{stablechaos}. In other words, in this region the original stochastic
710: model behaves like a ``chaotic'' deterministic one after a coarse-graining of
711: patches. We expect that this correspondence will become more and more exact
712: with growing $R$. From a theoretic-field point of view this means that the
713: renormalization flux tends towards a ``chaotic'' model instead of the usual
714: fixed-point dynamics.
715:
716: For $H=0.42$, as shown in Fig.~\ref{fig:H042}, the period-two phase has a
717: finite amplitude, before falling into the DCA-like dynamics by reducing $J$.
718:
719:
720: For $J>0$ (transition II) we have
721: \[
722: c^*_{\text{II}} = f(c^*_{\text{II}}; H, \tilde{J}) =
723: \begin{cases}\tilde Q \\ 1-\tilde
724: Q\\\end{cases}
725: \]
726: and thus the critical value $J^*_{\text{II}}$ is
727: \begin{equation*}
728: J^*_{\text{II}} = \dfrac{1}{2\tilde Q -1} \left[\pm H + \dfrac{1}{2}
729: \log \left(\dfrac{\tilde Q}{1-\tilde Q}\right)\right].
730: \end{equation*}
731: Once that a portion of the lattice has been
732: attracted to an absorbing state, it pulls the neighboring regions to
733: this same state, due to the ferromagnetic ($J>0$) coupling.
734:
735: This approximation, disregarding fluctuations, overestimates the
736: critical value $J^*_{\text{II}}$, as shown in Figs.~\ref{fig:H0} and
737: \ref{fig:H042}.
738:
739:
740: \begin{figure}[t]
741: \includegraphics[width=6cm]{fasecm.eps}
742: \caption{\label{fasecm} Case $H=0$, $J=-\infty$: the
743: mean-field $R-Q$ phase diagram of the mean-field approximation.
744: Points mark parameter values for which the
745: absorbing states are the only stable attractors.
746: A plus sign denotes period-2 temporal
747: oscillations between absorbing states,
748: a star denotes the presence of a stable
749: point at $c=0.5$, a cross (circle) denotes period-two (four)
750: oscillations between two non-zero and non-one densities, triangles
751: denote chaotic oscillations.}
752: \end{figure}
753:
754: We analyze extensively the mean-field behavior of the system for $H=0$
755: and $J=-\infty$. As shown in
756: Fig.~\ref{fasecm}, for a given value of $R$,
757: there is always a critical $Q_c$ value of $Q$ for
758: which the active phase disappears, with an approximate correspondence
759: $Q_c \simeq 2/5 R$. One can also observe that the chaotic
760: oscillations of the mean-field map, that roughly corresponds to the
761: coherent-chaotic phase of the model, appear only for large values of
762: $R$. Since the absorbing states are always present, there chaotic
763: oscillations may bring the system into the quiescent phase, as shown
764: by the ``hole'' to the right of triangles in Fig.~\ref{fasecm}.
765:
766: The correspondence between the chaotic behavior of the mean-field
767: approximation and the actual behavior of the system will be the subject
768: of a future work.
769:
770:
771: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
772:
773:
774: \section{Conclusions}
775: \label{sec:conclusions}
776:
777: We have investigated a general one dimensional model with
778: extended-range interactions and symmetric absorbing states. The model
779: is characterized by a competing ferromagnetic linear coupling and an
780: antiferromagnetic nonlinear one.
781:
782: By means of
783: numerical simulations and mean field approximations we have shown
784: that a chaotic phase is present
785: for strong antiferromagnetic coupling.
786: This phase may be identified as a stable-chaotic region, in which the
787: behavior of the (originally stochastic) system is essentially deterministic
788: and its behavior is highly irregular and essentially unpredictable.
789: The mean field map exhibit a chaotic behavior for a large interaction
790: range $R$, and this behavior is reflected in the appearance of many
791: metastable states in the system, for extremely strong
792: antiferromagnetic coupling.
793:
794: A disordered phase, insensitive of parameter variations,
795: appears at the boundary between the active and the quiescent ones, and the
796: transitions appear to be of equilibrium type, i.e.\ truly
797: salient points only in the limit $R\rightarrow\infty$.
798:
799:
800: We expect that in higher dimensions one can recover some aspects of
801: these phase transitions without imposing the presence of absorbing
802: states, i.e. using finite couplings.
803:
804: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
805:
806: \section*{Acknowledgements}
807: Partial economic support from project IN109602 DGAPA--UNAM and the
808: Coordinaci\'on de la Investigaci\'on Cient\'\i fica UNAM is
809: acknowledged.
810:
811: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
812:
813: \appendix\section{Equivalence
814: between dynamic Ising Model and Cellular Automata}
815:
816: As an illustration,
817: we present here a derivation of the equivalence of the kinetic Ising model
818: (here we choose heat bath) with a cellular
819: automaton (the Domany-Kinzel model) in order to elucidate the role of
820: infinite coupling parameters and absorbing states. This derivation is
821: similar to that of Ref.~\cite{Georges} but more general.
822:
823: The Ising model is defined by the couplings among spins. The configuration at
824: time $t$ is denoted as $\boldsymbol{\sigma}= \sigma_1, \sigma_2,\dots$ and the
825: configuration at time $t+1$ as $\boldsymbol{\sigma}'= \sigma'_1,
826: \sigma'_2,\dots$.
827: Let us write the temperature-dependent Hamiltonian as
828: \[
829: \mathcal{H}(\boldsymbol{\sigma}) = \sum_i H(\sigma_{i-1},\sigma_{i}
830: \sigma_{i+1}),
831: \]
832: with
833: \begin{equation}\label{H}
834: H(x,w,y) = J^{(0)} w + J^{(1)} x w + J^{(2)} w y + J^{(3)} xwy.
835: \end{equation}
836:
837: The transition probabilities $\tau$ must obey the detailed balance condition
838: \[
839: \dfrac{\tau(\boldsymbol{\sigma}'|\boldsymbol{\sigma})}
840: {\tau(\boldsymbol{\sigma}|\boldsymbol{\sigma}')} =
841: \exp\bigl(\mathcal{H}(\boldsymbol{\sigma}')-\mathcal{H}(\boldsymbol{\sigma})\bigr).
842: \]
843: In each step we can update in parallel all even or odd sites, obtaining
844: \[
845: \dfrac{\tau(\boldsymbol{\sigma}'|\boldsymbol{\sigma})}
846: {\tau(\boldsymbol{\sigma}|\boldsymbol{\sigma}')} =
847: \prod_{i}'
848: \dfrac{\tau(\sigma_{i-1},\sigma'_i, \sigma_{i+1}|\sigma_{i-1},\sigma_i,
849: \sigma_{i+1})}
850: {\tau(\sigma_{i-1},\sigma_i ,\sigma_{i+1}|\sigma_{i-1},\sigma'_i,
851: \sigma_{i+1})} ,
852: \]
853: where the product is restricted to either even or odd sites.
854: The detailed balance condition can thus be satisfied locally.
855: Choosing the Heat Bath dynamics
856: \[
857: \tau(x,1,y|x,w,y) = \dfrac{\exp\bigl(-H(x,1,y)\bigr)}
858: {\exp\bigl(-H(x,-1,y)\bigr)+\exp\bigl(-H(x,1,y)\bigr)}
859: \]
860: the
861: transition probabilities does not depend on the present value of the
862: spin $w=\sigma_i^t$ so that the lattice (with even or infinite lattice
863: sites) may be decoupled into two noninteracting sublattices with the
864: same geometry of the DK model. From now on,
865: to be coherent with the usual cellular
866: automaton notation, we shall express the transition probabilities in
867: terms of the Boolean variables $s_i=(\sigma_i+1)/2$,
868: and we shall denote the local field
869: as $h(a,b)\equiv H(2a-1,1,2b-1)$.
870:
871: Let us denote the DK transition probabilities $\tau(s'_i|s_{i-1},
872: s_{i+1})$ as
873: \[
874: \tau(1|00)=\varepsilon, \qquad \tau(1|01)=\tau(1|10)=p, \qquad \tau(1|11)=q.
875: \]
876:
877: We get for the heat bath dynamics
878: \[
879: \tau(1|ab) = \dfrac{1}{1+\exp\bigl(-2h(a,b)\bigr)}
880: \]
881: and thus
882: \[
883: h(a,b) = \dfrac{1}{2}\ln \dfrac{\tau(1|ab)}{1-\tau(1|ab)}.
884: \]
885:
886: Substituting into Eq.~\eqref{H} one obtains a linear system
887: \[
888: \begin{split}
889: J^{(0)} - J^{(1)} - J^{(2)} + J^{(3)} &= \frac{1}{2}\ln
890: \frac{\varepsilon}{1-\varepsilon},\\
891: J^{(0)} + J^{(1)} - J^{(2)} - J^{(3)} &= \frac{1}{2}\ln \frac{p}{1-p},\\
892: J^{(0)} - J^{(1)} + J^{(2)} - J^{(3)} &= \frac{1}{2}\ln \frac{p}{1-p},\\
893: J^{(0)} + J^{(1)} + J^{(2)} + J^{(3)} &= \frac{1}{2}\ln \frac{q}{1-q}.
894: \end{split}
895: \]
896: \vspace{3cm}
897:
898: Finally, we have
899: \[
900: \begin{split}
901: J^{(0)} &= \frac{1}{8}\ln \frac{\varepsilon}{1-\varepsilon} +
902: \frac{1}{8}\ln \frac{q}{1-q} + \frac{1}{4}\ln \frac{p}{1-p}, \\
903: J^{(1)} =J^{(2)} &= - \frac{1}{8}\ln \frac{\varepsilon}{1-\varepsilon} +
904: \frac{1}{8}\ln \frac{q}{1-q} , \\
905: J^{(3)} &= \frac{1}{8}\ln \frac{\varepsilon}{1-\varepsilon} +
906: \frac{1}{8}\ln \frac{q}{1-q} - \frac{1}{4}\ln \frac{p}{1-p}. \\
907: \end{split}
908: \]
909:
910: In the limit $\varepsilon \rightarrow 0$ all couplings become infinite.
911: \vfill
912:
913: \begin{thebibliography}{99}
914:
915: \bibitem{Georges} A.~Georges and P.~Le Doussal, J. Stat. Phys.
916: \textbf{54}, 1011 (1989).
917:
918: \bibitem{DP} E.~Kinzel and W.~Domany, Phys. Rev. Lett. {\bf 53} (1984);
919: W.~Kinzel, Z. Phys. B {\bf 58} (1985);
920: W. Kinzel, in \textit{Percolation Structures and Processes},
921: G. Deutsch, R. Zallen and J. Adler, editors (Adam Hilger, Bristol, 1983).
922:
923:
924: \bibitem{spinglass} M. Mézard, G. Parisi, and M. A. Virasoro,
925: \emph{Spin Glass Theory and
926: Beyond} (World Scientific, Singapore, 1987); K.H. Fischer and
927: J.A. Hertz, \emph{Spin Glasses} (Cambridge University Press, Cambridge,
928: England, 1991).
929:
930: \bibitem{pspin} D.J. Gross and M. Mézard, Nucl. Phys. \textbf{B240},
931: 431 (1984); E.
932: Gardner, ibid. \textbf{B257}, 747 (1985).
933:
934: \bibitem{acri02} F. Bagnoli, F. Franci and R. Rechtman,
935: in \emph{Cellular Automata}, S. Bandini, B. Chopard and M. Tomassini (editors),
936: (Springer-Verlag, Berlin 2002) p. 249.
937:
938:
939: \bibitem{Bagnoli_2001} F.~Bagnoli, N.~Boccara and R.~Rechtman,
940: Phys. Rev. E \textbf{63}, 46116 (2001).
941:
942:
943: \bibitem{Hinrichsen_1997}
944: H.~Hinrichsen,
945: Phys. Rev. E {\bf 55}, 219 (1997).
946: \bibitem{Janssen_1981}
947: H.~K. Janssen,
948: Z. Phys. B {\bf 42}, 152 (1981).
949:
950: \bibitem{Grassberger_1982}
951: P.~Grassberger,
952: Z. Phys. B {\bf 47}, 365 (1982).
953:
954: \bibitem{wolframclasses} S. Wolfram: Physica D \textbf{10}, 1 (1984).
955:
956: \bibitem{fragment} F.~Bagnoli, P.~Palmerini and R.~Rechtman,
957: Phys. Rev. E \textbf{55}, 3970 (1997).
958:
959:
960: \bibitem{Tel} T.~Tel, \emph{Proceedings of the 19th IUPAP International
961: Conference on Statistical Physics}, edited by Hao Bai-lin (World Scientific
962: Publishing: Singapore 1996).
963:
964:
965: \bibitem{stablechaos} F.~Bagnoli and R.~Rechtman,
966: Phys. Rev. E \textbf{59}, R1307 (1999);
967: F.~Bagnoli and F.~Cecconi, Phys. Lett. A \textbf{260}, 9-17
968: (2001) and references therein.
969:
970: \bibitem{CK88} J.P. Crutchfield and K. Kaneko, Phys. Rev. Lett. {\bf 60},
971: 2715 (1988), K. Kaneko, Phys. Lett. {\bf 149A}, 105 (1990).
972: \bibitem{PLOK} A. Politi, R. Livi, G.-L. Oppo, and R. Kapral, Europhys. Lett.
973: {\bf 22}, 571 (1993).
974:
975: \bibitem{Damage1} P.~Grassberger, J.~Stat.~Phys. {\bf 79}, 13 (1995).
976: \bibitem{Damage2} F.~Bagnoli, J.~Stat.~Phys. {\bf 79}, 151 (1996).
977:
978:
979: \bibitem{wolfram} S.Wolfram (Ed.)
980: \emph{Theory and Application of Cellular Automata}
981: (Addison-Wesley, Reading, MA, 1986).
982:
983: \bibitem{DomanyHinrichsen}
984: H. Hinrichsen, J. S. Weitz and E. Domany
985: J. Stat. Phys. \textbf{88}, 617-636 (1997).
986:
987: \bibitem{LyapunovCA} F.~Bagnoli, R.~Rechtman and S.~Ruffo, Phys. Lett. A
988: \textbf{172}, 34 (1992).
989:
990: \bibitem{luque} B. Luque and R.V. Sol\'e, Physica A 284 (2000) 33--45
991:
992: \bibitem{wolf} A.Wolf, JB Swift, HL Swinney, and JA Vastano,
993: Physica D, \textbf{16}, 285 (1985).
994:
995: \bibitem{synca}
996: F.~Bagnoli and R.~Rechtman,
997: Phys. Rev. E \textbf{59}, R1307 (1999).
998:
999: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1000: % non-referenced bibitems
1001: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1002: %\bibitem{Dickman}
1003: %A.~P.~F. Atman, R.~Dickman and J.~G.~Moreira
1004: %Phys. Rev. E \textbf{67}, 016107 (2003).
1005:
1006:
1007: %\bibitem{rigid} D. J. Jacobs and M.F. Thorpe, Phys. Rev. E \textbf{53}
1008: %3682 (1996); see also C.F. Moukarzel, Phys. Rev. E 68, 056104 (2003)
1009: %and references therein.
1010:
1011: \end{thebibliography}
1012:
1013: \end{document}
1014:
1015:
1016:
1017: