Overview of coupled map lattices

Kunihiko Kaneko

Citation: Chaos 2, 279 (1992); doi: 10.1063/1.165869

View online: http://dx.doi.org/10.1063/1.165869
View Table of Contents: http://chaos.aip.org/resource/1/CHAOEH/v2/i3
Published by the American Institute of Physics.

Related Articles
Local accumulation times for source, diffusion, and degradation models in two and three dimensions
J. Chem. Phys. 138, 104121 (2013)
Local accumulation times for source, diffusion, and degradation models in two and three dimensions
JCP: BioChem. Phys. 7, 03B607 (2013)
Conjugate feedback induced suppression and generation of oscillations in the Chua circuit: Experiments and
simulations
Chaos 23, 013130 (2013)
Long-time dynamics for a class of Kirchhoff models with memory
J. Math. Phys. 54, 021505 (2013)
Two-particle circular billiards versus randomly perturbed one-particle circular billiards
Chaos 23, 013123 (2013)

Additional information on Chaos

Journal Homepage: http://chaos.aip.org/
Journal Information: http://chaos.aip.org/about/about_the_journal
Top downloads: http://chaos.aip.org/features/most_downloaded
Information for Authors: http://chaos.aip.org/authors

Downloaded 15 Mar 2013 to 35.8.11.2. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
Overview of coupled map lattices
Kunihiko Kaneko
University 0/ Tokyo. Department of Pure and Applied Sciences, College oj Arts and Sciences. Komaba.
Meguro-ku, Tokyo 153, Japan
(Received 28 July 1992; accepted for publication 31 July 1992)
Studies in coupled map lattices are briefly surveyed in connection with the papers in the
present foclis issue.

I. INTRODUCTION ear dynamics and spatial diffusion, as is .seen, for example,

in chemical reaction-diffusion systems and Benard convec-
In 1981 I started a simulation of a model consisting of tion with a large aspect ratio. For studies of this type of
a chain of logistic maps coupled to their neighbors and spatiotemporal chaos, we can choose local chaos and spa-
with parameters in the regions of chaotic behavior_ Soon r tial diffusion as separate procedures. By adopting nonlinear
realized that the complexity in the resUlting spatiotemporal mapping x(i)-x'(i)=!(x(i» for local chaos and the
pattern was beyond our knowledge at that time_ Such cou- (one-dimensional) discretized diffusion process x' (i) - (1
pled systems of mappings are now known as coupled map -e)X'(i) + (e/2)[x'(i+ I) +x'(i-l)], the following
lattices (CML). A CML is a dynamical system with dis- CML is obtained:
crete time, discrete space, and continuous state vari-
ables. I- 19 It usually consists of dynamical elements on a xn+ 1 (i) = (I-e)!(xn(i»+ (e/2) [f(xn(i+ I))
lattice which interact ("couple") with suitably chosen sets
+!(xn(i-l)], (I)
of other elements. The main motivation for the introduc-
tion of CML is to study spatiotemporal chaos. Over the where n is a discrete time step and i is a lattice point
past 10 years, studies in CML have been expanding not (i= 1,2, ... , N=system size) with a periodic boundary con-
only in the field of spatiotemporal chaos and pattern for- dition.
mation but also in the fields of biology, mathematics, and As the local map !(x), we can choose any one-
engineering, although its phenomenology is not yet under- dimensional map (with a chaotic behavior). We often take
stood, applications are being actively pursued. the logistic map
Spatiotemporal chaos is chaotic dynamics in spatially
!(x) = I-ax', (2)
extended systems, where the number of effective degrees of
freedom diverges as the size of the system increases. Ex- since the map has already been investigated in detail. The
amples of spatiotemporal chaos are seen in Benard convec- above model [Eqs. (I) and (2)] has extensively been stud-
tion, convection in liquid crystals, Taylor flow, open flow ied as a prototype model for spatiotemporal chaos.
in fluid systems, chemical reaction-diffusion systems, some
solid-state systems such as Josephson junction arrays, II. QUALITATIVE UNIVERSALITY CLASSES IN CML
charge density wave and spin wave turbulence, and in some PHENOMENOLOGY
biological networks.
Our strategy for modeling dynamical phenomena in One of the merits in CML lies in their predictive power
spatially extended systems by CML models is based on the for novel phenomenology forming qualitative universality
following steps: classes. These classes lead to the introduction of novel no-
tions in the dynamics in spatially extended systems. Most
(a) Choose a (set of) field variables on a lattice. This of these classes have been found in the model (l), but the
set of variables is not at a microscopic, but at a macro- same qualitative behavior has been found in a wide range
scopic level. (Examples for a physico-chemical system of CML, and in other dynamical systems such as some
could be temperature, fluid velocity field, or the local con- chains of coupled differential equations, and in some par-
centration of some chemical substances.) The dimension tial differential equations. Some of the predicted classes
and topology of the lattice should be chosen to correspond have later been discovered in experimental systems.
to the physical space in concern. Examples include (i) spatial bifurcation and frozen
(b) Decompose the processes underlying the phenom- chaos (ii) pattern selection with suppression of chaos,6
ena into independent components (e.g., convection, reac- (iii) selection of zigzag pattern and chaotic diffusion of
tion, diffusion, and so on). defects, (iv) spatiotemporal intermittency",6.11,12 (v) soli-
(c) Replace each component by a simple parallel dy- ton turbulence,7 (vi) global traveling waves generated by
namics on a lattice: The dynamics consists of nonlinear local phase SlipS,14 and (vii) spatial bifurcation to down-
transformation of the field variables on each lattice point flow in open flow systems. 8•9
andlor a coupling term among suitably chosen neighbors. In the pattern selection, chaos in each local map gen-
(d) Carry out each unit dynamics ("procedure") suc- erated is almost completely suppressed by the selection of
cessively. some spatial structure, as is studied in one- and two-
Spatiotemporal chaos is often created by local nonlin- dimensional spatial lattices. The dynamics here is quasip-

CHAOS 2 (3), 1992 1054-1500/92/030279-04$001.00 © 1992 American Institute of Physics 279

Downloaded 15 Mar 2013 to 35.8.11.2. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
280 Kunihiko Kaneko: Overview

eriodic or weakly chaotic, modulated on the quasiperiodic ously switches among different ordered states through dis-
motion. This temporal order is sustained through the spa- ordered states. This switching process is known as chaotic
tial order, that is a long-ranged (quasiperiodic) order in itinerancy over attractor ruins. 2o Fluctuations in the tur-
space. In the present volume, such ordered structure is bulent phase do not obey the law of large numbers. Even if
detected by the method of orthogonal decomposition by the motion is fully chaotic, there remains some subtle co-
Lima. A quasiperiodic dynamics is seen even in a higher herence among the elements. This observation may shed
dimensional CML, as is studied by Chato and Manneville. new light on the origin of EEG in the brain. Globally
The importance of transients in spatiotemporal chaos coupled maps have possible abilities in information pro-
has been recognized. IS ,19 Besides transients which decay in cessing as will be discussed later,
the conventional manner, quasistationary supertransients The mean-field model can be regarded as a CML in the
are discovered. There the transient length increases with infinite dimension. On the other hand, most studies in
the system size exponentially or faster. During the tran- CML use one- or two- dimensional lattices. Chate and
sients, the dynamics is quasistationary; measures quantify- Manneville study the behavior in intermediate dimensions
ing chaos (such as Lyapunov exponents or entropies) fluc- (4,5, ... dimensional lattices). They show that the collec-
tuate around some value, until the transients terminate tive behavior of mean field can appear with quasiperiodic
abruptly. An analytic study for these transients is given by temporal dyuamics. Such quasiperiodic order in space and
a simple CML model given by Bunimovich, Livi, time is already noted in the one-dimensional CML (see
Martinez-Mekler, and Ruffo, in this volume. Sec. II), although the origin of spatial order in a higher-
When there is a stable periodic state (window) in a dimensional lattice is not clarified as yet.
single (local) mapping, the periodic state is linearly stable
in the corresponding CML if it is spatially homogeneous. IV. QUANTITATIVE ANALYSIS
The existence of very long transients often prevents the Studied for quantitative analysis of spatiotemporal
lattice system from falling on this homogeneous periodic chaos, the quantifiers in dynamical systems have been ex-
state, thus leading to the destruction of fine window struc- tended. Such quantifiers include, power spectrum in space
tures originally existing in low-dimensional chaos. By this and time; distribution of patterns and pattern entropy; Ly-
destruction of windows, fully developed spatiotemporal apunov exponents, (localization of) Lyapunov vectors,S
chaos is structurally stable, 19 This stability is important in comoving and sUb-space-time Lyapunov exponents;,,1O,19
the maintenance of ecological systems as discussed by Solo, dimension density; Kolmogorov-Sinai entropy density;
Bascompte, and Valls, and is the basis of homeochaos ex- propagation speed of disturbance; mutual information and
amined by Ikegami and the author. correlation in space-time. 5,6.J8
Traditional power spectra and correlations are useful
III. GLOBALLY COUPLED MAPS to characterize the transition of patterns in space-time.
Power-law behaviors of correlation at the spatio-temporal
An extension of CML to globally coupling is often intermittency is found, although the quantitative universal-
important. Globally coupled dynamical systems are rele- ity of critical exponents is doubtedy-13 The onset of
vant to the studies in physics such as Josephson junction spatio-temporal chaos by the period doubling, on the other
arrays, charge density waves, multimode lasers, arrays of hand, is studied by the renormalization group by
optical elements, and in biological networks such as neural Kuznetsov. 21 Some arguments for the critical exponents
dynamics, ecological systems, evolution models, for the spatial correlations, Lyapunov exponents, and
With this in mind, we have introduced a globally cou- phase fluctuations are given by AIstr~m and Stassinopou-
pled map (GCM) as a mean-field-type extension los although further theoretical studies will be necessary to
(Fushimi-Temperley-type mOdel) of CML, with a cou- confirm their arguments.
pling to all other e1ements. 20 The GCM corresponding to In chaos, theoretical studies have been developed
our diffusively coupled map lattice is written as within the framework of statistical mechanics. In CML, it
N may be possible and important to extend this framework to
Xn+IU)=(l-€)f(xn(i)+(dN) I f(xn(j))· (3) a problem with both space and time. The thermodynamic
j=l approach with the Perron-Frobenius operator is extended
The above model has a remarkably rich behavior, to spatio-temporal chaos, with the use of self-consistent
partly similar but much richer than the Sherrington- approximation. 17 ,19 Extension of periodic orbit summation
Kirkpatrick model for spin glass. The important notion to CML is discussed by Politi and Torcini in connection
here is clustering. The elements split into some clusters; all with the Lyapunov analysis. A related study with periodic-
the elements in a cluster oscilIate in synchronization. De- orbit analysis is given by Houlrik.
pending on the numbers of clusters in the GCM, we have By choosing a simple local dynamics with some special
phase transitions among a coherent phase, an ordered piece-wise linear maps, analytic studies are possible for
phase, a partially ordered phase, and a turbulent phase, as CML. Bunimovich and Sinai 16 have constructed a statisti-
the parameter describing the nonlinearity in f(x) is in- cal mechanical formulation for a CML with complete hy-
creased. In the partially ordered phase, there are many perbolicity. In the present volume Bunimovich, Livi,
attractors with different numbers of clusterings and with a Martinez-Mekler, and Ruffo present analytic studies for
variety of partitions. Dynamically, the system spontane- transients of a CML with a piece-wise linear map.

CHAOS, Vol. 2, No.3, 1992

Downloaded 15 Mar 2013 to 35.8.11.2. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
 Kunihiko Kaneko: Overview 281

In the time series analysis of chaotic dynamics, the use biological systems. Among possible applications, we dis-
of orthogonal decomposition provides a powerful tool to cuss some features related with papers included in this
find some low-dimensional dynamics, Lima applies the volume.
method to detect a coherent structure often emerging in Searching with chaos for a stored memory (as an at-
CMLs as a quasiperiodic pattern, tractor) has been discussed in relation to the observation of
chaos in neural systems. 25 With the use of the stochastic
nature of chaos, searching for different regions is possible,
V. APPLICATIONS TO PATTERN DYNAMICS while deterministic dynamics restricts the space of search,
which may eliminate unnecessary wandering. Spatia-
Our strategy of studying dynamical phenomena in spa- temporal chaos can keep some order both in space and
tially extended systems by CML is based on the separation time. Spatial and temporal structures are partially pre-
of parallel procedures and successive operations of them. served in the course of the search with chaos, which must
Most phenomena in pattern dynamics are described by the be relevant to the optimization problem in some spatiotem-
combination of some elementary local dynamics, efficiently poral structures. In this volume, Nozawa has studied a
modeled by CML. globally coupled map combined with the Hopfield's neural
Oono and Puri have proposed a CML for spinodal network coding. His model performs remarkably well for
decomposition,22 which is the first, and most popular, ap- the optimization of the traveling salesman problem. This
plication. When a system is quenched from a disordered search may make effective use of the chaotic itinerancy
state (at high temperature) to an ordered state (at low dynamics in globally coupled maps.
temperature), spatial pattern formation proceeds. Tradi- By relating each procedure in a CML with some infor-
tionally this problem has been studied by the time- mation processing, it may be possible to reach a compro-
dependent Ginzburg-Landau equation or by the kinetic mise among multiple different requests for processing. In
Ising model using Monte Carlo methods. By taking a the paper by Price, Wanbacq, and Oosterlinck, applica-
coarse grained approach, local ordering dynamics can be tions of CMLs to image data processing are given, where
represented by a map with two stable fixed points, e.g., preservation of relevant structure and elimination of noise
f(x) =tanh(,8x). (4) by smoothing are compromised.
A computing machine with a digital state has limits for
The same diffusive coupling form as in Sec. IV can be used information processing. Blum, Shub, and Smale27 have
since the phase transition dynamics includes a term which demonstrated that an analog computer (with real numbers
tends to make two neighboring regions order. It is possible of infinite precision) can decide a problem that a conven-
to extend the model.to include some conservation laws. 22 tional digital machine cannot do. Coupled map lattices
Similar applications are possible to crystal growth, spiral may provide an analog parallel processor. This problem of
wave and traveling wave in excitable media, and so computing with CML is discussed by Holden, Tucker, and
on as demonstrated in the paper written by Levine and Zhang.
Reynolds. Domain growth processes for a bistable system
on a fractal lattice are discussed by Cosenza and Kapral in
this volume. VII. APPLICATIONS TO BIOLOGICAL SYSTEMS
Boiling process can be modeled with a CML, combin-
ing the change from liquid to gas, buoyancy, and heat Coupled chaotic oscillations are seen in a wide range of
diffusion. The model reproduces the nucleus-film transi- biological phenomena. The importance of nonlinear oscil-
tion, observed in experiments of boiling phenomena, as is lations in neural activities has recently been appreci-
shown in the paper written by Yanagita. Similar CML ated,25.28 a globally coupled map may provide an efficient
modeling is possible for thermal convection. The model model of this process. The immune response, for example,
reproduces formation process of convective rolls, onset of consists of the population dynamics of many antibodies
chaos, spatiotemporal intermittency, and transition be- formed in a network. Oscillations of some antibodies are
tween soft and hard turbulence. Indeed, the temperature observed, which may be chaotic in time. In population
distribution changes its form from Gaussian to exponen- dynamics, many species interact through space and/or a
tial, as is found in experiments in the soft-hard turbulence food web. A CML approach is thus useful in the study of
transition. Dependence on the Prandtl and Rayleigh num- ecology and evolution. In an ecological model with hosts
bers is correctly reproduced?3 and parasites on a two-dimensional lattice, Sole, Bas-
compte, and Valls study a spiral pattern and stability of a
chaotic state here.
VI. APPLICATIONS TO INFORMATION PROCESSING When mutation among species is included, the popu-
lation dynamics is described by a coupled map on a gene
The relevance of chaos to biological information pro- space. In this volume, Ikegami and the author study the
cessing has been pointed out and investigated. 3.24-26 High- evolution of ecological networks, by further taking into
dimensional chaos has the potential for information pro- account of mutation rates. The stability of ecological net-
cessing. Since most biological networks include many work is dynamically sustained by a weak and high-
interacting (chaotic) dynamical elements, we expect that dimensional chaos, termed as "homeochaos." The notion
high-dimensional chaos may also play an important role in of homeochaos will be relevant to most biological net-

CHAOS, Vol. 2, No.3. 1992

Downloaded 15 Mar 2013 to 35.8.11.2. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
282 Kunihiko Kaneko: Overview

works, where the hqmeochaos may replace the conven- 16L. A. Bunimovich and Ya. G. Sinai, Nonlinearity 1, 491 (1989).
tional notion of homeostasis. 17K. Kaneko, Phys, Lett, A 139, 47 (1989); J, M. Houlrik, I. Webman,
and M, H Jensen, Phys, Rev, A 41, 4210 (1990).
18K. Kaneko, "Stimulating Physics with Coupled Map Lattices," in For~
I Here only partial references are given for eMLs. See the papers in this
mation, Dynamics, and Statistics of Patterns, Vol. 1, edited by K. Ka~
volume and the references cited therein and in Ref. 18. .
'K. Kaneko, Prog. Theor. Phy& 72, 480 (1984); 74,1033 (1985). wasaki, A. Onuki, and M. Suzuki (World Scientific, Singapore, 1990),
3 K. Kaneko, Ph. D. thesis, Collapse of Tori and Genesis of Chaos in 19K. Kaneko, Prog. Theor, Phys. Suppl. 99, 263 (1989); Phys, Lett. A
Dissipative Systems, 1983 (enlarged version is published by World Sci- 149, 105 (1990).
entific, Singapore, 1986). 10K. Kaneko, Phys. Rev. Lett. 63, 219 (1989); 65, 1391 (1990); Physica
4R. J. Deissler, Phys. Lett. A 120, 334 (1984); I. Waller and R. Kapral, D 41, 137 (1990); P. Hadley and K. Wiesenfeld, Phys. Rev, Lett. 62,
Phys. Rev. A 30,2047 (1984). 1335 (1989).
sK. Kaneko, Physica D 23, 436 (1986). 21 S. P. Kuznetsov, Radiophysics 29, 888 (1986) (in Russian).
6K. Kaneko, Physica D 34, I (1989); 37, 60 (1989). 22 Y. Oono and S. Puri, Phys. Rev. Lett. 58, 836 (1986); Phys. Rev. A 38,
7 J. P. Crutchfield and K. Kaneko, "Phenomenology of Spatiotemporal 1542 (1988); Y. Oono and A. Shinozaki, Forma 4,75 (1989).
Chaos," in Directions in Chaos (World Scientific, Singapore, 1987), p. 23T. Yanagita and K. Kaneko, Coupled Map Lattice for Convection,
272. preprint, 199 L
'K. Kaneko, Phys. Lett. A 111, 321 (1985).
241. Tsuda, Prog. Theor. Phys. Suppl. 79, 241 (1984); I. Tsuda, E.
91. S. Aronson, A. V. Gaponov-Grekhov, and M. 1. Rabinovich, Physica
Koerner, and H. Shimizu, Prog. Theor. Phys. 78, 5} (1987); I. Tsuda,
D 33, I (1988).
Chaos~teki~Noukan (Science-sha, Tokyo, 1990, in Japanese).
lOR. J. Deissler and K. Kaneko, Phys. Lett. A 119, 397 (1987).
IIJ. D. Keeler and J. D. Farmer, Physica D 23, 413 (1986). 25W, Freeman and C. A. Skarda, Brain Res. Rev. 10, 147 (1985).
12H. Chate and P. Manneville, Physica D 32, 409 (1988). 26p. Davis, Jap. J. Appl. Phys. Lett. 29, 1238 (1990).
13p. Grassberger and T. Schreiber, Physica D 50, 177 (1991). 27L. Blum, M. Shub, and S. Smale, Bull. Am. Math. Soc. 21, 1 (1989).
14K. Kaneko, Phys. Rev. Lett. 69, 905 (1992). 28C, M. Gray, P. Koenig, A. K. Engel, and W. Singer, Nature 338, 334
15J. P. Crutchfield and K. Kaneko, Phys. Rev. Lett. 60, 2715 (1988). (1989); R. Eckhorn el al., BioI. Cybernetics 60, 121 (1988).

CHAOS, Vol. 2, No.3, 1992

Downloaded 15 Mar 2013 to 35.8.11.2. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions