1992 PhysRevA.46.R7347
1992 PhysRevA.46.R7347
1992 PhysRevA.46.R7347
A general understanding of the different dynamical re- fact that all oscillators are identical (in contrast to [5,8]).
gimes of dissipative systems with many active degrees of Also, they have simple oscillatory dynamics in the absence
freedom is still lacking despite its importance in a wide of interaction (i.e. , they cannot exhibit chaotic motion in
range of fields. Nevertheless, in recent years, many the absence of coupling like in [3]). Finally, the ampli-
specific models of spatiotemporal chaos have been the sub- tude dynamical mode is not frozen like in pure-phase
ject of intense investigations [1-10]. In particular, models [6,7] and this plays a crucial role in the coherent
Ginzburg-Landau equations have been used to describe a chaotic dynamical regime that we have found.
large range of phenomena [I I]. The complex Ginzburg- Our goal is to determine the asymptotic behavior of (2)
Landau equation (CGLE) in its simplest version reads as in the thermodynamic limit (N» I ) as a function of the
parameters a, P, and p. We first describe our numerical
8&A = (I +iP)V A + pA —(I + ia) ~A ~
A . results. Equation (2) was integrated numerically, using
both a fourth-order Runge-Kutta model and an explicit
It displays a variety of dynamical regimes in different re-
method, with a time step that was generally taken to be
gions of parameter space. In particular, all plane-wave
solutions are unstable for aP+ I (
0 [12]. The complicat-
ed behavior of the CGLE in this Benjamin-Feir unstable
dt =0.01 but, if necessary, reduced to dr =0.001 (see
below). Lattices of 50 to 5000 sites were used but, in gen-
eral, reliable results were already obtained with the small-
regime has been numerically investigated both in one [9]
er numbers. Several different initial conditions were test-
and in two spatial dimensions [10] and different phases
ed, such as random initial conditions in a square or circle
have been numerically characterized. Our aim in this
of variable size.
Rapid Communication is to introduce and study a simpler
Four main qualitatively diA'erent types of behavior were
but related model which can be viewed as a mean-field-
observed. We restrict ourselves here to the description of
like version [3,5,8] of Eq. (I). We choose to discretize the
the two-parameter space a= — P, p which appears to be
CGLE on a fully connected lattice of N sites (i.e. , where
sufficient to describe the qualitatively different phases.
each point is connected to all the other points). Taking
the discrete version of the Laplacian on such a lattice, we
They are delimited in Fig. I as a function of P and p. We
describe the four different types of behavior in order of in-
obtain, instead of Eq. (I),
8, Aq =gI+i
]
(A.
I, Aq)+pA~ —(I+—ia) ~AJ~ Aq
creasing complexity.
Region I is the simplest. All the oscillators are in the
same state [i.e., A~(r) =A(t)]. Therefore, the coupling
term in Eq. (2) vanishes and the oscillators follow the lim-
j=l, . . . , N. (2) it cycle of the uncoupled system
This is a system of N identical globally coupled limit-cycle A(t) = Jpexp( —iapt) .
oscillators. Related models of coupled oscillators and cou-
pled maps have already been studied by a number of au- This region can be found below and to the left of the line
thors, partially motivated by their connection to such denoted BF (for Benjamin-Feir) in Fig. l.
different topics as Josephson-junction arrays [61, oscillato- Region II is also quite simple. There, all the oscillators
ry dynamics of neuronal systems [7, 13, 14), or synchron- are rotating with the same frequency. However, their
ized behavior of assemblies of fireflies [8, 15). The present phases are distributed on the limit circle of radius v'p —I
model [Eq. (2)] difl'ers from these previous studies by the in such a way that the forcing mean amplitude
4.0 The behaviors found in regions I, II, and III have al-
BF ready been observed in a number of previous models. In
particular, the clustering in region III appears similar to
what has been reported for globally coupled logistic or cir-
uency locking
cle maps [3] or homogeneous pure-phase oscillators [7].
In contrast to the first three regions, the behavior in re-
2o ) gion IV appears surprisingly complex and new. DiA'erent
oscillators follow diff'erent erratic motions but in a
sufficiently coherent way so that the mean amplitude,
Homogeneous A = (1/N)+~A~, remains of order 1 in the thermodynamic
limit cycle
3
limit. Moreover, the dynamics of appears to be chaotic.
Although a similar chaotic behavior has been found in [6],
0.0 we were surprised to observe it with identical oscillators.
0.0 2, 0 4.0 In order to illustrate and support these assertions diA'erent
types of numerical evidence are now described. First, in
p
Fig. 2 the position of all the oscillators are shown as dots
FIG. l. The phase diagram for the case a = —P. The line BF at two diA'erent times, together with the position of the
is the line at which the homogeneous limit cycle becomes unsta-
mean (the crosses in Fig. 2). The parameters values are
ble. The line Bo limits the stability of the solution where all the
given in the caption. These two times have been chosen to
oscillators are distributed uniformly on a circle of radius v'p —I.
be large enough to be representative of what is observed in
The line B] is the limit of stability of the nonuniform locked
the long-time limit, at least with our numerical capabili-
states. The analytical equations of these lines are given in the
ties. At a given time instant, different oscillators are lo-
main text. The squares are numerically obtained values for the
limit of existence of the complex state characteristic of region
cated at diff'erent position in the complex plane, along a
IV. The thin line is a guide to the eye. For further explanation shape that has roughly the form of a p. The whole shape
we refer to the text. rotates and also deforms in time, especially in the tail
part. Individual oscillators move along this shape, fast
motion around the loop of the p alternating with slow
motion in the tail part. The time evolution of the real part
=(1/N)+~A, vanishes:
of the mean complex amplitude A(t) is represented in
A, = Jp —1 exp Ii [(a —P —a p ) t + i', ]j, Fig. 3(a) together with the power spectrum for this time
series [Fig. 3(b)]. The power spectrum is peaked around
with —
N)-)
N
g e' '=0. (4) a definite frequency but appears to be broadband. More-
over, its width does not decrease when the number of os-
For phases distributed uniformly on the circle (sometimes cillators is enlarged from 50 to 5000.
called "splay-phase" states in the literature [6]), the sta- In order to get additional evidence on the chaotic nature
bility limit is the line Bo in Fig. 1. However, states distri-
of the dynamics the spectrum of Lyapunov exponents for
buted unevenly on the limit circle (but still with A =0) the whole dynamical system has been computed. The
number of positive Lyapunov exponents has been found to
also exist and have a slightly larger domain of stability.
increase with the number of oscillators. In contrast to this
The configuration of this type which has the larger
domain of stability is formed by two populations of oscil-
high-dimensional dynamics, we believe that the reduced
lators, having opposite phases. The boundary of stability
of this last configuration is denoted Bi in Fig. 1. Howev-
1.5
er, in the parameter region between the lines Bp and B~,
when the oscillator phases were initially distributed uni-
formly on the limit cycle, we never observed restabiliza-
tion to a state with an uneven distribution of phases but
always a transition towards the chaotic state of region IV 0.5
described below. In fact, special care was needed in order
to observe the stability of the uniform state up to the line
8p and we found it necessary to reduce the time step to
dt =0.001. —05 I
X
In region III, i.e. , to the right of the line connecting the 'L
open squares and the line BF, the system breaks into a
small number N, of diA'erent macroscopic clusters. All
the oscillators that belong to a given cluster have the same —1.5
complex amplitude but it diAers from one cluster to the
other. Generally, N, was found to be 2 or 3. Within re-
Re A;
gion III, the dynamics electively reduces to that of a low
number of coupled oscillators (with diA'erent weights) and FIG. 2. The positions of the 500 individual oscillators (dots)
it was found to be periodic or quasiperiodic depending on and their mean (crosses) in the complex plane at two different
the parameter subregion. times. The parameter values are p =2.0 and P =1.5.
DYNAMICS OF THE GLOBALLY COUPLED COMPLEX. .. R7349
P4(X) =k[X+2(p —I )] [A. +2(p —2)1+ 3 rotating frame depends only on the two eff'ective parame-
ters II and ~FI. In the parameter range corresponding to
—P +2p(P —1)l . (9) the complex chaotic state of region IV the phase space of
The domain of stability of the 3, = 1 state is thus the re- Eq. (10) seems to be close to a bifurcation between two
gion p ~ 2, P ~ 4(2p —3)/(2p —1) whose boundary is topologies. For example, simulation of the full system
denoted by 8 in Fig. . 1 [Eq. (2)] with P = —a =1.5 and p =2 gives for the main
frequency v=041 and (~A(t)~)=1. 1. The corresponding
~
(p I
We are grateful to Y. Pomeau for very instructive dis-
A rescaled time r =(p —1)t has also been introduced in cussions. We would also like to thank R. Conte and J.
order to show explicitly that the dynamics of B(t) in the Kadtke for their numerical advice.
[1] Y. Kuramoto, Chemical Oscillators, Waves and Tur [13] H. Sompolinsky, A. Crisanti, and H. J. Sommers, Phys.
bulence (Springer, New York, 1984). Rev. Lett. 61, 259 (1988).
[2] H. Chate and P. Manneville, Prog. Theor. Phys. (to be [14] E. Niebur, H. G. Schuster, and D. M. Kummen, Phys.
published). Rev. Lett. 67, 2753 (1991).
[3] K. Kaneko, Phys. Rev. Lett. 63, 219 (1989); 65, 1391 [15] J. Buck and E. Buck, Sci. Am. 234, 74 (1976).
(1990); Physica D 55, 368 (1992), and references therein. [16] P. Grassberger and I. Procaccia, Physica D 9, 189 (1983).
[4] Y. Pomeau, Physica D 23, 3 (1986); H. Chate and P. [17] J. P. Eckmann and D. Ruelle, Rev. Mod. Phys. 57, 617
Manneville, Europhys. Lett. 6, 59 (1988); Physica D 32, (1985), see especially the discussion on p. 648.
409 (1988); P. Manneville, in Propagation in Systems Far [18] Related stability calculations for a chain of coupled oscil-
from Equilibrium, edited by J. E. Weisfreid et al. lators are given in G. B. Ermentrout and N. Kopell, SIAM
(Springer, New York, 1988). J. Math. Anal. 15, 215 (1984). Similar ideas are used for
[5] Y. Kuramoto and Y. Nishikawa, J. Stat. Phys. 49, 569 neural-like oscillators by A. Sherman and J. Rinzel, Proc.
(1987); H. Daido, Phys. Rev. Lett. 68, 1073 (1992). Natl. Acad. Sci. U. S.A. 89, 2471 (1992).
[6] K. Wiesenfeld and P. Hadley, Phys. Rev. Lett. 62, 1335 [19] N —2 zero eigenvalues are also found when the stability of
(1989); J. W. Swift, S. H. Strogatz, and K. Wiesenfeld, "splay-phase states" is investigated for a series arrays of 1V
Physica D 55, 239 (1992). Josephson junctions [S. Nichols and K. Wiesenfeld, Phys.
[7] D. Golomb et al. , Phys. Rev. A 45, 3516 (1992). Rev. 45, 8430 (1992), and references therein]. It would
[8] P. C. Matthews and S. H. Strogatz, Phys. Rev. Lett. 65, be interesting to see whether this is due to the existence of
15 (1990); P. C. Matthews, R. E. Mirallo, and S. H. Stro- more general locked states with nonuniform time delays as
gatz, Physica D 52, 293 (1991). in the present case.
[9] B. Shraiman et al. , Physica D 57, 241 (1992). [20] The critical value F, is given as a function of the two pa-
[10] P. Coullet, L. Gil, and J. Lega, Phys. Rev. Lett. 62, 1619 rameters of the reduced system Q and a [Eq. (10)] by
(1989).
F = [2(1+ Qa) /27(l+a2) ]
[11] P. Coullet and G. Iooss, Phys. Rev. Lett. 64, 866 (1990).
[12] T. B. Benjamin and J. E. Feir, J. Fluid Mech. 27, 417 x (1+9[(Q —a)/(1+ Qa)]
(1967); A. C. Newell, Lect. Appl. Math. 15, 157 (1974);
J. T. Stuart and R. C. DiPrima, Proc. R. Soc. London, —{1 —3[(Q —a)/(1+ Qa)] ] )
Ser. A 362, 27 ( 1 978).