First-order Synchronization Transition in Locally Coupled Maps
Extracted source text ยท arXiv cond-mat/0402283
%\documentclass[twocolumn,tightenlines,showpacs]{revtex4}
\documentclass[twocolumn,showpacs]{revtex4}
\usepackage{graphicx}
%\usepackage{epsfig}
%\usepackage{iopams}
%\usepackage{cite}
%\usepackage{setstack}
\def\al{{\it et al}}
\def\eqspace{\hspace{1.0cm}}
\def\PRE{{\it Phys. Rev. E} }
\def\JSP{{\it J. Stat. Phys.} }
\def\half{\frac{1}{2}}
\def\av#1{\left< #1 \right>}
\def\eg{{\it e.g.} }
\def\ie{{\it i.e.} }
\begin{document}
\title{First-order Synchronization Transition in Locally Coupled Maps}
\author{ P. K. Mohanty}
\affiliation {Department of Physics of Complex Systems, Weizmann
Institute of Science, Rehovot, Israel 76100.}
\date{\today}
%%%%%%%%%%%%%%%-------NEW COMMAND-------------------%%%%%%%%%%%%%%%%
\newcommand{\cl}{\centerline}
\newcommand{\be}{\begin{equation}}
\newcommand{\ee}{\end{equation}}
\newcommand{\jpa}{J. Phys. A}
\newcommand{\la}{\langle}
\newcommand{\ra}{\rangle}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{abstract}
We study a family of diffusively coupled chaotic maps
on periodic $d$-dimensional square lattices. Even and odd sub-lattices
are updated alternately, introducing an effective {\it delay}.
As the coupling strength is increased,
the system undergoes a first order phase transition from a multi-stable
to a synchronized phase. Further increase in coupling strength shows
de-synchronization where the phase space splits into two ergodic
regions. We argue that the de-synchronization transition
is discontinuous for piece-wise linear maps, and is continuous for
non-linear maps which are differentiable.
\end{abstract}
\pacs{05.45.-a, 05.45.Xt}
\maketitle
Synchronization is observed in a wide class of complex systems.
Typically, it appears when the range of the correlations inside the
system is of the same order as the system size.
Few examples of mutual synchronization in complex dynamical systems
are flashing fireflies\cite{Buck88}, electronic circuits
\cite{Heagy} and chemical reactions \cite{Kuramoto}.
In recent years, synchronization of spatially extended systems
has drawn considerable interest. In particular coupled map
lattices (CMLs) \cite{Kaneko}, initially introduced as simple models of
spatio-temporal chaos, has received a great deal of attention as a model
of synchronization. It has been realized that two different
replicas of the same CML are coupled directly \cite{Grassberger}
or through a common external random noise \cite{Livi}, can
become synchronized for large enough coupling strengths.
Recently, Ahlers and Pikovsky \cite{Ahlers02} pointed out that
synchronization transitions in one dimensional (1-$d$) CMLs are generically
in the universality class of the Kardar-Parisi-Zhang (KPZ) model,
however strongly nonlinear maps could be in the universality class
of Directed percolation (DP).
From a point of view of general
statistical theory synchronization is a non-equilibrium phase transition
where distinct patches of the CMLs oscillate in phase. The phenomenon
is similar to the roughening of growing interfaces and
can be modeled by multiplicative noise partial differential
equation\cite{Pikovsky}. This picture is further modified
by M$\tilde{\rm u}$noz and Pastor-Satorass\cite{Munoz} to incorporate
first order phase transitions (FOPTs). However, it was not clear if
the FOPTs observed in these models were just transients of DP.
In context of non-equilibrium wetting process FOPTs are found in several
(1+1)-dimensional stochastic models with local interactions\cite{Mukamel},
contrary to equilibrium wetting process where phase transitions are not
possible \cite{Zia} in $1-d$ systems having short-range interactions
between interfaces and substrates.
To the best of our knowledge FOPTs in deterministic,
chaotic, extended systems with short range interactions are still lacking,
although it is known to exist for globally coupled maps \cite{Muller}.
In this Letter we introduce and study a single parameter
family of piece-wise linear chaotic maps which are diffusively
coupled on a $d$-dimensional square lattice. A {\it delay} is
introduced dynamically between sub-lattices by updating them
alternately.
One of our interests would be to find if, starting from a random
initial condition, these sub-lattices synchronize at later times.
The answer turns out to be 'no', for both very high and low diffusion
strengths $\epsilon$. However for intermediate $\epsilon$ synchronization
occurs with the suppression of spatio-temporal chaos. This
synchronized phase is an {\it unique absorbing state} of the system and
for piece-wise linear maps (PLMs) the phase boundary is identical with the boundary
for a stable fixed point.
For $\epsilon=0$, the system visits the whole
phase-space in time. The phase-space volume decreases as
$\epsilon$ is increased and at a critical strength
$\epsilon_A$, the phase-space suddenly shrinks to a point
which corresponds to the fixed point of the primitive map.
This fixed point is stable until $\epsilon=\epsilon_B>\epsilon_A$.
Further increase in $\epsilon$ results in
de-synchronization where the phase space splits into
{\em two} ergodic regions about the {\it collective bifurcation points}.
It may be argued that for PLMs the width of the bifurcation
is independent of $\epsilon$ and thus synchronization error is
discontinuous at the critical point. However for non-linear differentiable
maps the width of the bifurcation vanishes as $\sqrt{\epsilon-\epsilon_{c}}$
which results in a continuous transition. We argue that this continuous
transition is in a different universality class than that of KPZ and DP
with critical exponents $\beta=1/2$, $\beta/\nu=0$ and $\gamma=1$ in
$1$-dimension.
{\it The Model :} Consider a $d$-dimensional hyper-cubic lattice
${\cal L}$ of coupled identical maps $f(m,z_{\vec i })$, where
$z_{\vec i }$ is a real variable at site $\vec{i}\equiv (i_1,i_2\dots i_d)$
with $i_k$ varying from $1$ to $L$. We define the {\it even} and {\it odd}
sub-lattices (${\cal L}^e$ and ${\cal L}^o$ respectively) as
${\cal L}^{e,o}= \{\vec i : \sum_k i_k =$ {\it even,odd}$\}$,
and denote $x_{\vec i}$ ($y_{\vec i}$) as the variable of ${\cal L}^e$
(${\cal L}^o$). Starting from a random initial configuration,
$\{x_{\vec i}\}$ and $\{y_{\vec i}\}$ are updated alternately as
\begin{eqnarray}
x_ {\vec i}^{t+1} &=& (1-\epsilon) f( x_ {\vec i}^{t})
+\frac{\epsilon}{2d} \sum_{{\vec j}\in {\cal N}_{\vec i} }
f( y_ {\vec j}^{t}),\cr
y_ {\vec i}^{t+1} &=& (1-\epsilon) f( y_ {\vec i}^{t})
+\frac{\epsilon}{2d} \sum_{{\vec j}\in {\cal N}_{\vec i} }
f( x_ {\vec j}^{t+1}),
\label{eq:XY}
\end{eqnarray}
where ${\cal N}_{\vec i}$ is a set of $2d$ nearest neighbors of
$\vec i$ and $\epsilon$ is the coupling strength, can be seen as a
diffusion constant. Equivalently, in the first half unit of time
$\{x_{\vec i}\}$ are updated while $\{y_{\vec i}\}$ are at rest
and the reverse happens in second half. We will see later that the
{\it delay } which is introduced dynamically between sub-lattices
is responsible for a complete synchronization of the system.
Note that periodic boundary configuration in all $d$ dimensions
are used throughout.
Synchronization occurs when the difference between $z_{\vec i}$ and its
neighbors vanish at all sites as $t \to \infty$. Thus, the order
parameter can be defined as $\phi=\langle\phi^t \rangle$ where
\begin{equation}
\phi^t = \frac{1}{2dL^d} \sum_{\vec i \in {\cal L}}
\sum_{\vec j\in {\cal N}_{\vec i}}
|z_{\vec i}^t-z_{\vec j}^t|,
\end{equation}
and steady state average is taken over time and
realizations. Obviously $\phi$ vanishes in the synchronized phase and
in the unsynchronized phase $\phi>0$. A trivial synchronized phase would
correspond to the stable fixed point of the CLM, $i~e$,
$\{z_{\vec i}= z^*\}$. It is
easy to see from Eq. (\ref{eq:XY}) that $z^*= f(m, z^*).$
For chaotic CLMs without {\it delay} the largest Lyapunov exponent is
independent of $\epsilon$ and is the same as the Lyapunov exponent of
primitive map which is positive. Hence, a fixed point solution
$\{z_{\vec i}= z^*\}$ is unstable for any $\epsilon$.
With a {\it delay}, however, it can become negative in a region
$\epsilon_B \le \epsilon \le \epsilon_A$. For
PLMs $\epsilon_{A,B}$ corresponds to the boundary of linearly
stable region which can be calculated as follows.
Let us take the initial state to be close to the fixed-point, $i.e.,$
$x_{\vec i}= z^*+\delta x_{\vec i}$ and $y_{\vec i}= z^*+\delta y_{\vec i}.$
In Fourier-space the evolution of $\delta x_{\vec i}$ and
$\delta y_{\vec i}$ reads as,
\begin{equation}
\left( \begin{array}{c} \delta x_{\vec k} \\
\delta y_{\vec k} \end{array}
\right)^{t+1} =
\left(\begin{array}{cc}\tilde\mu& R_{\vec k}\\
\tilde\mu R_{\vec k} &
\tilde\mu+ R_{\vec k}^2\end{array}\right)
\left(\begin{array}{c} \delta x_{\vec k} \\ \delta y_{\vec k}
\end{array} \right)^t,\label{eq:mat}\end{equation}
$$
{\rm where~} R_{\vec k} = \frac{\epsilon\mu}{2d}
\sum_{\vec r \in {\cal N}_{\vec 0}}
e^{i{\vec k}.{\vec r}}, ~\mu=f^\prime(m, z^*)
$$
and $\tilde\mu=\mu(1-\epsilon).$ Let $E_\pm$ denote the eigenvalues
of the matrix defined in Eq. (\ref{eq:mat}). From the stability
requirements $|Max({\cal R}(E_+), {\cal R}(E_-))| <1$,
we find $\epsilon_A= 1-1/\mu$ and $\epsilon_B= (\mu+1)/(2\mu)$ which
are drawn in the inset of Fig. \ref{fig:phase} as a phase boundary for
the synchronized phase. For simplicity, it is assumed here that primitive
maps have only single nonzero fixed point $z^*$. One can further
generalize it to maps with more fixed points.
%----------------------------------------------
\begin{figure}
\includegraphics[width=7.0 cm]{fig1.eps}
\caption{\label{fig:phase}
Numerically obtained $\phi$ is plotted against $\epsilon$ for
tent map ($\mu=-2$).
The inset shows the phase diagram in
$\epsilon$-$\mu$ plane.}
%The fixed point $z_i=z^*$ is stable in the
%shaded region, bounded by $\mu=(1-\epsilon)^{-1}$ and
%$\mu=(2\epsilon-1)^{-1}.$ For piece-wise liner maps these phase boundaries are
%identical with that of synchronized phase.}
\end{figure}
%----------------------------------------------
To find out the behavior of $\phi$ close to these transitions we
first restrict ourselves to one dimension and study a specific
single parameter family of maps :
\begin{equation}
f(m,z) = \left \{ \begin{array}{ll}
m z/(m-1) & z<a\\
m(1-z) & z \ge a
\end{array}\right. ,
\label{eq:map}
\end{equation}
where $a = 1-1/m$. This piece-wise linear mapping of $[0,1]$ onto itself
is everywhere expanding for $m>1$, and thus chaotic,
with an invariant density uniform on $[0,1]$.
A particular example of this family with $m=2$ is known as {\it tent map}.
Note that the fixed point is $z^*=m/(m+1)$.
\begin{figure}
\includegraphics[width=7.0 cm]{fig2.eps}
\caption{\label{fig:multi} This figure shows how the phase space
changes in $z_1$-$z_2$ plane as $\epsilon\to \epsilon_A$;
equivalently $\delta\to 0$. For large $\delta$, the phase space is
identical for two different initial configurations $S_1$
and $S_2$. However, dynamically different shapes
are generated as $\delta\to 0$. The
symbol '$*$' represents the fixed point $z^*=.6$. }
\end{figure}
{\it Synchronization :}
Let us first discuss the transition from the unsynchronized
phase $A$ (inset of Fig. \ref{fig:phase}) to the synchronized phase.
Close to the transition we take $\epsilon= \epsilon_A-\delta$
and find that the system become multi-stable as $\delta \to 0$, $i.e.,$
there are large number steady states and one out of them is chosen by the
system depending on the initial configuration. Multi-stability has been
discussed earlier in the context of CMLs \cite{Chate99}, and
systems of delayed differential equations\cite{Ikeda}.
It may be argued that the multi-stability is extensive, $i.e.,$ the
number of attractors grow exponentially with the system size.
Thus any statistical average has to be taken over large number of
independent realizations, which resticts us to simmulate large
systems. We carried out numerical simmulations for $L=1024$ and $m=2$
({\it tent map}) and find that $\phi$ vanishes discontinuously at
$\epsilon_A=0.5$ (see Fig. \ref{fig:phase}). To confirm that it is
a true first order transition, not just a transient
effect, we monitor the phase space of every neighboring pair of
co-ordinates as $\delta \to 0$. For example in Fig. \ref{fig:multi}
we demonstrate how the phase space changes in $z_1$-$z_2$ plane.
Every other pair of neighboring co-ordinates show similar changes.
In practice, no noticable change is observed in the phase space when
$\delta<10^{-4}$ and then suddenly the fixed point $z_1=z^*=z_2$ appears
at $\delta=0$. In other words, when $\delta\approx 0$,
we have $|z_i-z_{i+1}|>0$ for every realization and thus $\phi$ has
a jump at the critical point $\epsilon_A$.
{\it De-synchronization :} The synchronized state persists
up to $\epsilon=\epsilon_B$ where phase space splits into
{\it two} disconnected ergodic regions. In the new phase
$B$ shown in the inset of Fig. \ref{fig:phase}, the system fluctuates
about its {\it collective bi-bifurcation points}
$x_{i}= x^*$ and $y_i= y^*$. Using Eq. (\ref{eq:XY}) we get
\begin{eqnarray}
x^*&=& (1-\epsilon) f(x^*)+\epsilon f(y^*)\cr
y^*&=& (1-\epsilon) f(y^*)+\epsilon f(x^*),
\label{eq:bi}
\end{eqnarray}
which can be solved for the family of maps defined in Eq. (\ref{eq:map})
as $x^*=\alpha_\pm$ and corresponding $y^*= \alpha_\mp$, with
$\alpha_\pm=\frac{m^2(2\delta+1)\pm (2\delta m+1)}{m^2(2\delta+1)
+m(4\delta +1)}$ and $\delta=\epsilon-\epsilon_B$.
Depending on the initial configuration, different parts of the sub-lattices
are then attracted to $\alpha_+$ or $\alpha_-$ with kink-like interfaces
(inset (a) of Fig. \ref{fig:kink})
separating them. It will be shown later that the width of such a kink $w$,
diverges as $1/\sqrt\delta$ as $\delta\to 0$. Thus, stable kinks can
not be generated when $\delta$ is ${\cal O}(L^{-2})$
and we have $\phi= \alpha_+-\alpha_-$. Clearly the jump in $\phi$
at the critical point $\epsilon_B$ is $\Delta=2/(m^2 +m)$. Note that in the
other limit, $i.e.$, when $L \to \infty$ before $\delta \to 0$, we have a
slightly different $\phi$. In this case the large systems
will generate an average density of kinks, say $\rho$. Immediately,
we have $\phi= \Delta( 1- \frac{\rho{\cal A}}{w\Delta})$, where
${\cal A}$ is the area bounded by an even and an odd kink.
Nevertheless the transition is discontinuous.
The profile of a kink can be calculated as follows.
Let us assume that it starts at site $k$ with $z_{k}=\alpha_-$ and
$z_{k+1}=\alpha_+$. In steady state $z_{k+2}$ has two solutions :
$\alpha_-$ and $\alpha = \frac{m^2(2\delta+1) +(2\delta m-1)}
{m^2(2\delta+1) + m(4\delta +1)}$.
It is obvious that the first solution, $z_{k+2}=\alpha_-$, does not
generate a kink. Thus the steady state profile of a kink can be obtained as
\begin{equation}
\left( \begin{array}{c} z_{k+i+1}\\ z_{k+i+2}\end{array}\right) =
\left( \begin{array}{cc}0&1\\-1 &-2\cos(\theta) \end{array}\right)^i
\left( \begin{array}{c} \alpha_+\\ \alpha\end{array}\right),
\label{eq:kink}
\end{equation}
where $\theta=2\tan^{-1}(\sqrt{\delta/\epsilon_B})$.
Explicit form of $z_{k +i}$ is lengthy to present here,
rather we compare numerically obtained $z_{k+i}$ with Eq. (\ref{eq:kink})
in the inset (b) of Fig. \ref{fig:kink}. When site $i$ is far from
$k$ and $\delta\approx 0$ one can write an approximate
solution as $z_{k+2i}=\alpha_- + \Delta_i$
and $z_{k+2i+1}=\alpha_- -\Delta_i$, where $\Delta_i=
\frac{i^2 \delta}{2m(m+1)}$. The width of the kink is thus
$w=2n$, such that $\Delta_n = \alpha_+ - \alpha_-$.
Clearly, $w$ diverges as $1/\sqrt\delta$. In Fig.\ref{fig:kink} we
have plotted $w$ obtained from numerical simulations
versus $\delta$ in $\log$-scale, which confirms this power-law.
%-------------------------------------------------
\begin{figure}
\includegraphics[width=7.0 cm]{fig3.eps}
\caption{\label{fig:kink} A typical steady state profile of odd (solid-line)
and even (dashed line) sub-lattices are shown in inset (a).
Inset (b) compares the kink profile obtained from simulations (points) with
Eq. (\ref{eq:kink}). In the main figure, width of the
kink $w$ is plotted against $\delta$ in log scale for $m=2$
and $2.5$. The slope of the solid line is set to $-1/2$.}
\end{figure}
%-------------------------------------------------
{\it Higher Dimensions :} The phase diagram for coupled PLMs
in higher dimension is the same as shown in the inset of
Fig.\ref{fig:phase}. Close to the transition point $\epsilon_A$, phase
$A$ is mutistable and shows sudden shrink in phase space volume. The
desynchronization transition occurs at $\epsilon_B$. But unlike one
dimension, the kink-type solutions in phase $B$ are no more stable
and hence jump in the orderparameter is $\Delta=2/(m^2 +m)$.
Numerical simulations in $2$ and $3$ dimensions confirms this finding.
{\it Other Maps :} How do non-linear maps behave when a
{\it delay} is introduced between sub-lattices ? Clearly the
phase boundaries for synchronized phase strongly depends on
corresponding primitive maps. First let us consider
the {\it logistic map} : $f(x)=4x(1-x).$ The fixed point
of this map $z_i = z^*=3/4$ is linearly stable for $1/2<\epsilon<3/4$.
However numerical simulation shows that the synchronization
occurs at $\epsilon_A> 1/2$. Close to the transition point phase $A$ is
multistable and hence we have a first order transition similar
to that of the {\it tent map}. The de-synchronization is found
to occur exactly at $\epsilon_B$, where the linear-stability of the
fixed point breaks down. Unlike the tent maps here we do not have
fluctuations about the bifurcation points. But the kinks are
unavoidable. Using Eqs. (\ref{eq:bi}) and (\ref{eq:kink}) one can
obtain the bifurcation points as $x_i=\alpha_\pm$ and $y_i= \alpha_\mp$,
where $\alpha_\pm= \frac{8\delta+3\pm2\sqrt\delta(8\delta+3)}{4(1+4\delta)}$.
Clearly $\Delta = \alpha_+ -\alpha_- $ vanishes as $\sqrt{\delta}$
indicating that the transition is continuous. Critical exponents of
this continuous transition can be deduced using following arguments.
First, note that fluctuation of
order-parameter comes about from the variation in density of kinks
which fluctuates about a mean density, say $\rho $. Density distribution
about $\rho$ can be assumed to be {\it normal}, which gives
$\phi \sim \sqrt\delta (1-\frac{{\cal A}\rho}{w\Delta})$, where
${\cal A}$ is the area bounded by an odd and an even kink, and
$w$ is the width of the kink. Since $\phi$ is independent of $L$ and
proportional to $\sqrt\delta$, we have critical exponents
$\beta=1/2$ and $\beta/\nu =0$. One can further define ``susceptibility''
as $\chi= (\langle\phi^2\rangle - \langle\phi\rangle^2)/L$.
Now $\chi \sim \delta r^2/L$, and hence $\gamma=1$. It is evident that
the critical dimension $d_c=1$.
We have also done numerical simulations on {\it sine} ($f(x)=\sin(\pi x)$)
and {\it cubic} ($f(x)=\sqrt{27}x(1-x^2)/2)$ maps in one dimension,
which show that the synchronization transition is discontinuous, whereas
de-synchronization occurs continuously with the same critical exponents
as that of the {\it logistic } map.
To get an insight
why de-synchronization is continuous for certain class of
maps we investigate coupled {\it power-law} maps: $f(z) =
1- |2z-1|^b$, where $b=1$ and $2$ corresponds to the {\it tent} and
the {\it logistic} maps respectively. Thus, by tuning $b$ it is possible
to study how the nature of transition changes from being first to
second order. Our numerical simulations suggests that for $b>1$ the
de-synchronization transition is continuous and belongs to the same
universality class as that of the {\it logistic } map, whereas for
$1/2<b\le 1$ it is discontinuous. Note that there are {\it two} non
zero fixed points for $b<1/2$ and present analysis need to be modified
for their discussion .
Basically, synchronization error is the average difference between
bifurcation points $x^*$ and $y^*$, which vanishes
at $\epsilon_B$ where $x^*=1/2=y^*$.
Just above $\epsilon_B$ this difference is proportional
to the difference between the left and right slope of $f(z)$
at $z=1/2$. For $b<1$, since these maps are not differentiable at $z=1/2$
the de-syncronization occurs discontinuously, whereas for $b>1$ the
maps are differentiable and thus we have a continuous transition.
%{\it The nature of transition in delayed coupled maps are thus
%dictated by the differentiability properties of the primitive maps}.
%-------------------------------------------------
\begin{figure}
\includegraphics[width=7.0 cm]{fig4.eps}
\caption{\label{fig:log} Phase diagram of de-synchronization transition is
shown in $\epsilon$-$b$ plane. The nature of transition changes
from being discontinuous (dashed line) to continuous (solid line) at $b=1$.
In the inset, $\phi$ and $\chi$ versus $\delta$ is plotted in
log-scale and the solid lines are drawn with slope
$\beta=1/2$ and $\gamma=1$ respectively for comparision.
}
\end{figure}
%-------------------------------------------------
A few comments are in order. First, the synchronized phase
discussed here is different from earlier studies, where CMLs
remain chaotic in the synchronized phase for both, when they are
coupled directly \cite{Grassberger} or through noise \cite{Livi}.
In our model the spatio-temporal chaos are suppressed
and thus the synchronized phase is a completely ordered phase.
The spatial correlations even persist in de-synchronized phase $B$.
Second, one can
define a different order parameter called ``magnetization''
$M=\langle m^t\rangle$, where $m^t = \sum_i sign(z_i^t-z^*)$.
Then the desynchronyzation transition can be thought as a
transition from fully ferromagnetic to fully antiferrmagnetic
phase, which can be either first or second order. Note that the
continuous transition obtained here is in a different universality
class than that of zero temperature Ising transition in
one dimension.
{\it Conclusion :} In conclusion, we study a single parameter
family of coupled chaotic maps by introducing a {\it delay}
between sub-lattices. We show that these systems undergo a first order
transition to a synchronized phase when the coupling parameter
is varied. The transition occurs as the system enters from a
multiply stable region to a single `collective fixed point' in phase
space. A second transition occurs as this fixed point become unstable
and the phase space breaks up into two ergodic regions about the
collective bifurcation points.
Our analytical results show that
the de-synchronization transition is discontinuous in one and higher
dimensions when the primitive maps are not differentiable.
Differentiable maps also show a first-order synchronization transition,
whereas the de-synchronization occurs continuously with critical exponents
$\beta=1/2$, $\gamma=1$, and $\nu^{-1}=0$. We argue that the
de-synchronization
transition discussed here can also be thought as a
transition from a fully ferromagnetic to a fully anti-ferromagnetic
phase. These results establish a true non-equilibrium first order transition
in one dimensional dynamical systems with short range interactions.
{\it Acknowledgments :} We thank D. Mukamel and E. Levin for
fruitful comments and discussions.
\begin{thebibliography}{0}
\bibitem{Buck88} J. Buck, Quart. Rev. Biol. {\bf 63}, 265(1988).
\bibitem{Heagy} J. F. Heagy, T. L. Carroll, and L. M. Pecora, \pre {\bf 50,}
1874 (1994).
\bibitem{Kuramoto}Y. Kuramoto, {\it Chemical oscilations,
Waves and Turbulence}, Springer, Berlin (1984).
\bibitem{Kaneko}{\it Theory and Application of Coupled map Lattices}, ed. by
K. Kaneko, J. Wiley and Sons, Chichester, (1993).
\bibitem{Grassberger} J. F. Heagy, T. L. Carroll, and L. M. Pecora,
\pre {\bf 50}, 1874(1994); P. Grassberger, \pre {\bf 59}, R2520.
\bibitem{Livi} F. Cecconi, R. Livi and A. Politi, \pre {\bf 57}, 2703 (1998).
L. Baroni, R. Livi, and A. Torcini, \pre {\bf 63},
036226 (2001).
\bibitem{Ahlers02} V. Ahlers, and A. Pikovsky, \prl {\bf 88}, 254101 (2002).
\bibitem{Pikovsky} A. S. Pikovsky and J. Kurths, \pre{\bf 49}, 898 (1994).
\bibitem{Munoz} M. M$\tilde{\rm u}$noj and R. Pastor-Satorras, \prl{\bf 91},
190403 (2003).
\bibitem{Mukamel} H. Hinrichsen, R. Livi, D. Mukamel and A. Politi,
\pre {\bf 61}, R1032 (2000); L. Giada, and M. Marsili, \pre {\bf 62},
6015 (2000).
\bibitem{Zia} R. K. P. Zia, R. Lipowsky, and D. M. Kroll, Am. J. Phys.
{\bf 56}, 160 1(1998).
\bibitem{Muller} R. M$\ddot{\rm u}$ller, K. Lippert,
A. K$\ddot {\rm u}$hnel, and U. Behn, \pre {\bf 56}, 2658 (1997);
S. Kim, S. H. Park and C. S. Ryn \prl, {\bf 78}, 1616 (1997).
\bibitem{Chate99}A. Lema$\hat{\rm i}$tre, and H. Chat${\rm e}^\prime$,
\prl {\bf 82}, 1140 (1999) and references there in.
\bibitem{Ikeda} K. Ikeda, and K. Kondo, \prl {\bf 49}, 1467 (1982);
J. Foss, A. Longtin, B. Mensour, and J. Milton, {\it ibid} {\bf 76},
708 (1996).
\end{thebibliography}
\end{document}
\end{document}