CHAOS 16, 013134 2006

Using white noise to enhance synchronization of coupled chaotic systems

Wei Lina
Research Center and Key Laboratory of Mathematics for Nonlinear Sciences (Fudan University),
Ministry of Education, School of Mathematical Sciences, Fudan University, Shanghai 200433,
Peoples Republic of China

Guanrong Chenb
Department of Electronic Engineering, City University of Hong Kong, Hong Kong S.A.R.,
Peoples Republic of China

Received 1 December 2005; accepted 13 February 2006; published online 30 March 2006
In the paper, complete synchronization of two chaotic oscillators via unidirectional coupling determined by white noise distribution is investigated. It is analytically proved that chaos synchronization could be achieved with probability one merely via white-noise-based coupling. The established
theoretical result supports the observation of an interesting phenomenon that a certain kind of white
noise could enhance chaos synchronization between two chaotic oscillators. Furthermore, numerical
examples are provided to illustrate some possible applications of the theoretical result. 2006
American Institute of Physics. DOI: 10.1063/1.2183734
Noise, in a common perspective, plays a destructive role
in dynamical evolution, leading regular dynamics to behave irregularly and even completely randomly. However, sometimes noise can have constructive effects such
as inducing useful stochastic resonance. Also, it has been
experimentally observed that noise can enhance chaos
synchronization in some conventional coupled oscillators.
Thus, it becomes interesting to ask such questions as the
following: Can chaos synchronization be achieved
merely via white-noise-based coupling between two chaotic oscillators? Are there any analytical conditions
that can be rigorously proved to support the aforementioned experimental observation? These questions will
be positively answered in this paper. More precisely, rigorous conditions are established for achieving complete
synchronization with probability one merely via whitenoise-based coupling between two chaotic oscillators. A
significant consequence is that the theories and techniques presented in the paper could be further generalized for an analytical investigation of noise influence on
complex dynamical networks.
I. INTRODUCTION

The word synchronization means things occur at the

same time or operate in unison. The most classical phenomenon related to this word is attributed to Huygens observation about the synchrony of pendulum clocks.1 Since this
historical discovery, synchronization as a ubiquitous technical issue has become a focal topic of great importance in
applications ranging from physics to engineering, from computer science to biology, and even from economics to brain
science. In the past 30 some years, the word chaos, which
means a state of total confusion and lack of order, has
become another frequently used term in both theoretical
a

Electronic mail: wlin@fudan.edu.cn

Electronic mail: eechen@cityu.edu.hk

1054-1500/2006/161/013134/10/$23.00

studies and experimental investigations on nonlinear dynamical systems.2 As chaotic dynamics are characterized by
seemingly random evolutions, such as the sensitive dependence on initial states with broadband spectra, chaos synchronization is deemed to have a great potential in, for example, secure communications, pattern recognitions, brain
activity analysis, and the optimization of nonlinear
systems.3,4 As a consequence, research on chaos synchronization has been greatly developed, with the focus so far as on
several types of synchronization, including complete, generalized, phase, and lag synchronization, and on topics of anticipating synchronization.5
Noise, commonly regarded as a random and persistent
disturbance obscuring or reducing the clarity of a signal, is of
omnipresence in nature and in man-made systems. The
evaluation and suppression of the noise destructive influence
on dynamical evolutions and signal transmissions are of
great importance. This topic has been analytically and numerically investigated for complete synchronization of unidirectionally coupled chaotic oscillators, admitting noise perturbations in, e.g., Ref. 6. Some sufficient conditions
ensuring a successful and complete synchronization in the
presence of noise have been rigorously established. On the
other hand, a small amount of noise may enhance the dynamics of a system to becoming well regulated, as is contrary to
intuition, which therefore attracts more attention. A good example of this case is noise-induced stochastic resonance.7
Also, a noise inducing stability, phase transition or patterns
in varieties of noise-perturbed dynamical systems and even
in the spatially extended systems shows its constructive
role.8,9 Furthermore, it is experimentally reported that external noise, either of multiplicative ones or of additive ones,
can somewhat enhance different types of chaos
synchronization.10,11 In particular, additive noise is reported
to play a constructive role in chaos synchronization due to
the noise-induced domination of the contraction direction of
the saddle point embedded in the chaotic attractor.12 There

16, 013134-1

2006 American Institute of Physics

Downloaded 04 Apr 2006 to 202.120.224.18. Redistribution subject to AIP license or copyright, see http://chaos.aip.org/chaos/copyright.jsp

013134-2

Chaos 16, 013134 2006

W. Lin and G. Chen

are also many synchronous phenomena inducted by noise

coupling in nature; a typical example is the blinking synchronization of a swarm of fireflies through random mutual communications couplings. Thus, it becomes interesting to ask
such questions as the following: Besides the numerical evidence, are there any analytical arguments illustrating such
constructive effects induced by a specific type of noise?
Can chaos synchronization be achieved merely by a certain
kind of white-noise-based coupling between two chaotic oscillators with a rigorous theory to support it? These questions will be addressed in the paper. More precisely, unidirectional coupling determined by a white noise distribution,
of the multiplicative type, is applied to two chaotic oscillators. Analytical proof of some sufficient conditions are presented, which ensure complete synchronization with probability one between the two chaotic oscillators. Moreover, it
is shown that these analytical results could be further utilized
to illustrate a fact that intrinsic noise may enhance chaos
synchronization in conventional coupled chaotic oscillators.
The rest of the paper is organized as follows. In Sec. II,
a general chaotic oscillator as a driving system is described.
It is rigorously proved that an identical oscillator could synchronize to this chaotic oscillator with probability one
merely via a specific kind of coupling determined by a white
noise distribution. In Sec. III, the Rssler-like system, as an
illustrative example, is introduced for a possible application
of the general theory established in Sec. II. Furthermore, in
Sec. IV, it is analytically illustrated why a small amount of
intrinsic white noise can accelerate the occurrence of chaos
synchronization in the conventional setting of coupled chaotic oscillators. We conclude this paper with some discussions and comments in Sec. V.
II. COMPLETE SYNCHRONIZATION VIA WHITE
NOISE COUPLING

Consider a driving system described by

x t = Axt + fxt,

= W1t , ..., WmtT is m-dimensional Brownian motion. Ac t = t , ..., tT is an m-dimensional

cordingly, W
1
m
white noise vector, in which every two elements are
statistically independent, i.e., Eit = 0, Eit jt
= ijt t, i , j = 1 , ..., m. In particular, when Hyt xt
= l yt xt and m = 1, it becomes a much simpler case
where the noise coupling strength function is linear and the
t = t, is only one dimensional. For
white noise term, W
this simple case, the coupling strength l t actually becomes white noise with intensity l. This is in accord with
many realistic phenomena, since the coupling strengths of
real oscillators, networks, and biological systems are not
necessarily time invariant. As a matter of fact, the strengths,
always disturbed by the external or intrinsic noise, are time
varying, and even the coupling fashion may be determined
by some kind of noise. Also note that the noise-coupling
term in 3 is of a multiplicative form interpreted in the sense
of It, which implies that the dependence of the process yt
t at the same instant time t could be
on the white noise W
neglected. Even though it seems to be a mathematical limiting procedure, this interpretation of multiplicative noise is
reasonable when the rapidity of the environmental fluctuations is far less than the macroscopic time scale intrinsically
possessed by concrete systems, such as the external fluctuations in a large class of biological systems.8,15 Besides, it is
possible to artificially generate noisy signals within a lower
statistical dependence and instantaneous time interval relative to the long macroscopic time scale of the coupling systems. Therefore, the proposed white-noise-based coupling in
the response system 3 is not totally idealizing, but is realizable in describing many concrete systems in real applications. In this regard, the difference between the It interpretation and the Stratonovich interpretation will be further
elucidated in Sec. V.
Let e = e1 , ..., enT = y x. Then, from system 1 and 3,
one has the error dynamics

where state vector x = x1 , ... , xnT Rn, constant matrix A

and
nonlinear
vector
function
fx
Rnn,
= f 1x , ..., f nxT : Rn Rn satisfies
fx fyTfx fy x yTLx y,
x,y Rn ,

for some positive semidefinite symmetric matrix L. This general form of system contains several well-known chaotic oscillators, such as Chuas circuit,13 the Rssler-like system,14
and so on. Actually, the nonlinearity f, which is in the form
of piecewise linearity, always satisfies condition 2.
Given a driving signal xt generated by system 1, we
design a response system with white-noise-based coupling in
the form of
t,
y t = Ayt + fyt + Hyt xtW

where the nonlinear matrix-valued function H

= h1 , ..., hm : Rn Rnm, called the noise coupling strength
function, is locally Lipschitz continuous and satisfies the linear growth condition with H0 O, and Wt

t.
e t = Aet + fyt fxt + HetW

In light of the theory of stochastic differential equations16

and due to the premises on f and H, it can be easily verified
that the error dynamics 4 possesses a unique global solution for t t0, denoted by et ; t0 , e0 for any initial condition
e0 Rn. Clearly, et ; 0 0 is a trivial solution of the error
dynamics 4. Moreover, the complete synchronization between the driving system 1 and the response system 3
could be achieved, provided that the trivial solution of 4 is
globally asymptotically stable, i.e., for any initial value e0
Rn, et ; t0 , e0 0 as t + . Here, simply stands for
the Euclidean norm.
In what follows, it is shown that complete synchronization could be achieved with probability one, provided that
certain conditions on the noise coupling strength function are
satisfied. In other words, it is shown that white noise makes
sense in chaos synchronization. The arguments performed
below are analogous to the Lyapunov method for deterministic systems but are based on some inherent properties of
stochastic differential equations.

Downloaded 04 Apr 2006 to 202.120.224.18. Redistribution subject to AIP license or copyright, see http://chaos.aip.org/chaos/copyright.jsp

013134-3

Chaos 16, 013134 2006

White noise makes sense

To this end, for system 4, introduce the following function:

1
2

1/2

DVesAes + fys fxs

where the derivatives of the function V are

DVe = eTU/U1/2e2 ,

1
t

eTs

t0

ATU + UA 2
+ U
2
2

i=1 eTsUhies2
m

U es
1/2

t0

Vet0
t

ds +

Mt
t

Mt
l2i
t t0
,
2
+
t
min i=1 max
t

for all t t0, which implies

lim sup

t0

+ 21 traceHTesD2VesHesds + Mt,

Vet0

1
max
es
+ L+
Ki
ds

1/2
2 i=1
2
U es2

where the symmetric matrix U is positive definite. In Appendix A, it is shown that et ; e0 never approaches the zero
vector with probability one when the initial value e0 is a non
zero vector. This verifies the reasoning and correctness of
adopting the logarithm function in 5.
By applying Its formula16 to 5 along with system 4,
it yields that
Vet = Vet0 +

Ve = log e Ue = logU e,
T

Vet

t+

Vet
t

l2i

2
i=1 max

,
min

a.s.,

in which 9 has been used. Here, denote by max and min

the largest and the smallest eigenvalues of U, respectively,
and by the largest eigenvalue of the matrix,
m

ATU + UA 2 1
max
+ U + L+
Ki .
2 i=1
2
2
2

D Ve = U/U e 2Uee U/U e ,

2

1/2

1/2

and the continuous martingale is

Mt =

Consequently, one has

l2i
loget

lim sup
2 +
,
t
min
t+
i=1 max

DVesHesdWs

t0

t0

eTsUHes
dWs,
U1/2es2

with Mt0 = 0, and the quadratic variation is

Mt,Mt =

t0

eTsUHes2
ds.
U1/2es4

To further estimate 6, the following two inequalities are

introduced:
hTi ehie eTKie,
eTUhie lieTe,

where each symmetric matrix Ki is positive semidefinite and

each number li is non-negative i = 1 , 2 , ..., m. By the first
inequality of 8 and an elementary inequality for vectors, it
can be verified that Mt , Mt t t0 for some constant 0 and all t t0. This implies that
lim
t+

Mt
t

=0

holds almost surely a.s., due to the strong law of large

numbers.16,17 Moreover, from 2 and an elementary inequality for vectors, it follows that for any 0,
eTUfx fy eT

11

2 1
U + L e.
2
2

10

Now, substituting inequalities 8 and 10 into 6 gives

a.s.,

12

which means that the error et is almost surely convergent

to the trivial solution of 4 with an exponential damping
m 2
2
li max
/ min. As is well known,
rate, provided that i=1
the term almost surely means that an event occurs with
probability one in the physical sense. Therefore, the aboveperformed argument leads to the following proposition.
Proposition I: Complete synchronization between 1
and 3 via white-noise-based coupling could be achieved
with probability one, provided we have the following.
1 There exist positive semidefinite symmetric matrices
Ki and non-negative numbers li i = 1 , ..., m such that the
noise coupling strength function He satisfies 8.
2 There exist two positive definite symmetric matrices,
m 2
li
U , L, and a positive number , such that i=1
2
max / min, where ,min are the same as those defined
previously.
Furthermore, the convergence rate of the synchronization is exponential, which can be estimated by 12.
Clearly, condition 8 means that each element of the
noise coupling strength function could be estimated by some
forms of linear functions. Moreover, when the noise is absent, the damping rate 12 becomes / min, where becomes the largest eigenvalue of the matrix ATU + UA / 2
+ / 2U2 + 1 / 2L. If this is less than zero, then complete
synchronization is achieved in the noiseless system. This is
consistent with the result derived via the classical Lyapunov
direct method.
Generally, one can select U as an identical matrix I, set
= 1, and set the noise strength function He as in a linear
form, i.e., all hie = lie. Thus, eigenvalues min = max = 1 and

Downloaded 04 Apr 2006 to 202.120.224.18. Redistribution subject to AIP license or copyright, see http://chaos.aip.org/chaos/copyright.jsp

013134-4

Chaos 16, 013134 2006

W. Lin and G. Chen

matrix Ki becomes l2i I. A straightforward calculation thereby

gives a sufficient condition for complete synchronization
with probability one between systems 1 and 3 via whitenoise-based coupling with a linear strength function:
m

l2i 2 ,

i=1

13

in which is the largest eigenvalue of the matrix AT + A + I

+ L / 2. As a matter of fact, each li could be regarded as an
intensity of this white noise coupling. The sufficient condition 13 clearly shows that relatively large intensities surely
guarantee the complete synchronization between systems 1
and 3 in the physical sense. The larger the sum of the
intensity squares, the faster the convergence speed of the
synchronization. Moreover, when m = 1 and l1 = l is relatively
large, it is adequate to utilize one-dimensional white-noisebased coupling to achieve complete synchronization. Also, it
should be pointed out that a necessary coupling strength for
achieving chaos synchronization via conventionally deterministic feedback coupling has the same order of ; however,
it follows from 13 that the corresponding noise intensity is
only the extraction of . In conclusion, white noise may indeed have beneficial significance and impact on chaos synchronization.
III. NOISE-COUPLED RSSLER-LIKE SYSTEMS

In order to further illustrate some potential applications

of the analytical results established in the previous section,
consider a concrete chaotic oscillatorthe Rssler-like system:
x1 = x1 + x2 + x3,
x2 = x1 + x2,

14

FIG. 1. Chaotic attractors, generated by the Rssler-like system 14, are,

respectively, plotted in the phase plane a and in the time-state plane b.
Here, the parameters are taken as = 0.03, = 1.5, = 0.2, = 1.5,
= 0.075, = 0.75, = 21.43. The initial value is 1 , 1 , 1T and the timestep size is t = 0.1.

,
y 1 = y 1 + y 2 + y 3 + l11e1,l12e1W

x3 = gx1 x3,
where the nonlinearity in the system is a piecewise linear
function:
g =

2.56,
2.56, 2.56,
0,

15

and the corresponding parameters are taken as

= R6 / R5R7C2, = 1 / C1R1, = R5 / R4, = R5R7C2 / R6R12C3,
= 1 / C1R3, = 1 / C1R2, = R10R12 / R8R11. Here, all Ri and
Ci represent the values of resistors and capacitors in a real
electronic circuit designed to simulate the dynamics generated by the original Rssler system.14 Figure 1 shows a chaotic attractor of 14 with the above specified parameters,
chosen as a driving signal denoted by x1 , x2 , x3T.
As mentioned in the previous section, it is adequate to
utilize one-dimensional white noise to achieve chaos synchronization. However, multidimensional white noise is always unavoidable in nature and man-made systems, which
may induce even more interesting phenomena. Given the
driving signal x1 , x2 , x3T of 14, one may set up a unidirectional noise-coupled system as a response system,

,
y 2 = y 1 + y 2 + l21e2 + e1,l22e2 + e3W

16

,
y 3 = gy 1 y 3 + l31e3,l32e3W
where each ei = y i xi i = 1 , 2 , 3 is an error, positive numbers lij i , j = 1 , 2, , and are the noise intensity con = t , tT is a two-dimensional white
stants, and W
1
2
noise vector. This proposed white-noise-based coupling implies that the coupling intensities of each state are influenced
by two dependent white noise signals, so that each l pqit
becomes a white noise with intensity l pq.
Accordingly, the linear and the nonlinear parts of the
noise-coupled system 16 could be, respectively, evaluated
by

A= 1
0

fy =

gy 1

So, it follows from 2 that

Downloaded 04 Apr 2006 to 202.120.224.18. Redistribution subject to AIP license or copyright, see http://chaos.aip.org/chaos/copyright.jsp

013134-5

Chaos 16, 013134 2006

White noise makes sense

2 0 0
L= 0 0 0 ,
0 0 0
where = . Moreover, the elements of the noise coupling
strength function for the noise-coupled system 16 could be
written as

/2 0
/2 l21 0 ,
0
0 l31
l11

l12
0
0

/2 .
/2 l32
l22

To guarantee the positiveness of l1 , l2, the following assumptions on noise intensities are imposed:

h1e = l11e1,l21e2 + e1,l31e3T ,

2 4l11l21,

h2e = l12e1,l22e2 + e3,l32e3 .

2 4l22l32 .

17

For simplicity, set all lij l 0 and = l0 1, so

that 17 is obviously satisfied, l1 = l2 = 1 / 2l, and the
largest eigenvalue of Ki i = 1 , 2 is equal to 2 + 2
+ 42 + 4 / 2l2. Also, take U as the identity matrix and set
= . Then, by virtue of Proposition I, it is concluded that
complete synchronization between systems 14 and 16
could be achieved with probability one, provided that

Thus, straightforward computation gives that

hTi ehie eTKie,

eThie lieTe,

i = 1,2,

where the matrices

K1 =

2
+ 2 l21
l11

l21
0

2
l21

0 ,
2
l31

2
l12

K2 = 0
0

2
l22

l22
2
l32 + 2

l22

and constants l1 , l2 are, respectively, the smallest eigenvalues

of matrices

i
ii

Here 0 , 1 satisfies 4 122 + 16 4 0.

For
the
above-given
,
l l* = 1/21 2
42 + 4 / 21/2, in which is the largest eigenvalue
of the matrix

FIG. 2. Successful chaos synchronization between two noise-coupled Rssler-like systems. The variations of the driving signal with the response state are
shown in a with parameters A and in c with B. Sampling orbits from t = 0 to t = 40 are marked by a dotted line in a and c. The variations of Vet / t
with time t are displayed in b with A and in d with B. The initial value of the response system 16 is 1 , 1 , 1T and the time-step size is t = 0.02. Other
parameters are the same as those specified in Fig. 1.

Downloaded 04 Apr 2006 to 202.120.224.18. Redistribution subject to AIP license or copyright, see http://chaos.aip.org/chaos/copyright.jsp

013134-6

Chaos 16, 013134 2006

W. Lin and G. Chen

1
2

1
2

0
+

With the aid of MATLAB, one can numerically transform

condition i into

0,0.3289
I,
and get 0.9625 for the given parameters specified in
Fig. 1. Consequently, complete synchronization between the
two noise-coupled Rssler-like systems could be achieved
with probability one, if the parameter is determined in a
circumscribed interval I and the white noise intensity l is
larger than the critical value l*.
Figure 2 shows the successful chaos synchronization between systems 14 and 16 with two groups of parameters:
A l = 5.55, = 0.1; B l12 = l31 = 2.5, while all the other lij
= 5, and = = 0.2. Parameters in group A are consistent
with hypotheses i and ii in that I and l l* = 1.1727.
Numerical simulation with parameters in group B shows
that if in each row of the noise-coupled system 16, the
noise intensity lij, corresponding to the error induced by the
state variable of this row, is relatively large, and the noise
intensities , are properly taken, then complete synchronization could be achieved with probability one. In particular,
it is easy to see that condition ii becomes a special case of
13 when = 0. All these clearly illustrate a fact that a certain kind of strong white noise makes sense in chaos
synchronization.
In addition, Fig. 3 shows that synchronization fails simply due to the violation of hypothesis i or ii on noise
intensities. It is worthy to mention that hypotheses i and ii
are sufficient conditions for chaos synchronization but not
necessary by virtue of the analytical argument given in the
previous section. So, a violation of one or even both of them
does not necessarily lead to failure of the intended synchronization.
The well-known Euler-Maruyama numerical scheme18 is
used to present scenes of synchronized and unsynchronized
dynamics between systems 14 and 16 as well as the following systems. This implies that the smaller the time-step
size t is used in simulation, the closer the presented sampling orbit is to the true orbit. Therefore, t = 0.02 is used
instead of t = 0.1 as the time-step size in the simulation of
chaos synchronization, avoiding possibly large deviations induced by rough precisions.

IV. NOISE-ENHANCED CHAOS SYNCHRONIZATION

WITH CONVENTIONAL COUPLING

Given the above-performed analysis, we may now reason why a small amount of intrinsic white noise may enhance chaos synchronization between two conventionally
coupled chaotic oscillators.

FIG. 3. Failed chaos synchronization with two groups of noise intensities:

l = 0.1, = 1.0 a, and all lij = 0.5, = = 2.5 b. The sampling orbit from
t = 0 to t = 40 is marked by a dotted line in a. Here, the initial values and the
time-step size are the same as those specified in Fig. 2.

For simplicity, still take the Rssler-like system as an

illustrative example. Let x1 , x2 , x3T be the driving signal of
14. Design a response system with both traditional unidirectional linear coupling and one-dimensional white-noisebased coupling:
,
y 1 = y 1 + y 2 + y 3 + k1e1 +
e1W
,
y 2 = y 1 + y 2 + k2e2 +
e2W

18

,
y 3 = gy 1 y 3 + k3e3 +
e3W
where each ki is a deterministic coupling strength,
is the
= t stands for one-dimensional
noise coupling intensity, W
white noise, and the values of all the other parameters are the
same as those given in Fig. 1. Accordingly, the term
t
could be regarded as a kind of external noise influencing the
deterministic coupling strength ki, as is a ubiquitous phenomenon in real coupling systems.

Downloaded 04 Apr 2006 to 202.120.224.18. Redistribution subject to AIP license or copyright, see http://chaos.aip.org/chaos/copyright.jsp

013134-7

Chaos 16, 013134 2006

White noise makes sense

FIG. 4. The variations of eVet = e21t + e22t + e23t, with time t are, respectively, shown in a when
= 0, in b when
= 1.0, in c when
= 0.7, and in
d when
= 0.2. The initial values and the time-step size are the same as those given in Fig. 2.

On the one hand, when the noise intensity
= 0, the

response system 18 is a noiseless conventional response
system. In this case, if the strengths are taken as
k1 = 22,

k2 = 0,

k3 = 11,

19

then the largest eigenvalue of the Jacobian matrix of the error

system between 14 and 18 evaluated at the driving signal
is approximately 0.0059, larger than but very close to zero,
and the other two are less than zero. So, the trivial solution
of the error system is not globally asymptotically stable;
therefore complete synchronization cannot be surely realized
by this conventional linear coupling, as shown in Fig. 4a.
On the other hand, when the noise intensity
0, the
noise-perturbed error dynamics between systems 14 and
18 become a special case of the generalized system 4,
in that the conventional linear coupling matrix
diagk1e1 , k2e2 , k3e3 can be regarded as a contributor to matrix A. Therefore, by virtue of Proposition I, it is concluded
that the complete synchronization between systems 14 and
18 could be achieved with probability one if the noise intensity

c. Here, the critical value
c, not very large,
could be numerically obtained as
c 0.9881 based on the
given values 19 and other parameters specified in Fig. 2.
Figure 4b shows a successful chaos synchronization when
the noise intensity
= 1.0.

As mentioned in the preceding section,

c is only a

sufficient condition for complete synchronization. It is thus
expected that a relatively smaller noise intensity
could be
utilized to achieve chaos synchronization. Actually, this is
the case, as shown in Fig. 4c, where
= 0.7. Yet, when
is
too small, for example,
= 0.2, chaos synchronization is always failed, as shown in Fig. 4d, since it is close to the
situation where there is no noise coupling at all.
Next, denote by TFirst the first time instant where
the error dynamics satisfied e2Vet 105 and
e2Vet e2Vet+t 105, which implies an almost successful complete synchronization between systems 14 and 18.
The variation of this time instant TFirst with the noise intensity
changing in the interval 0.6, 3.0, is plotted in Fig. 5,
which shows a tendency that the time period approaching a
complete synchronization is somewhat shortened as the noise
intensity increases. Also, this is consistent with the fact that
the larger the noise intensity is selected to be, the faster the
estimation 12 of the upper damping rate becomes.
Furthermore, when the deterministic coupling strengths
ki i = 1 , 2 , 3 are changing around the given values 19,
the phase diagrams of critical value
c versus different
groups of the coupling strengths ki are plotted in Fig. 6,
where
c is directly settled to zero, provided that the complete synchronization could be realized with the given deter-

Downloaded 04 Apr 2006 to 202.120.224.18. Redistribution subject to AIP license or copyright, see http://chaos.aip.org/chaos/copyright.jsp

013134-8

W. Lin and G. Chen

FIG. 5. The variations of the time instant TFirst with the noise intensity

increasing from 0.6 to 3.0 with step size 0.001. All the other parameters are
the same as those specified in Fig. 2 and Fig. 4.

Chaos 16, 013134 2006

ministic strengths in the noiseless systems. When the value

of the noise intensity
is selected from the region above the
parabola-like curve plotted in Fig. 6a or above the hyperplane depicted in Fig. 6b, the complete synchronization
between systems 14 and 18 could be almost surely
achieved; however, this may not be the case when the value
is picked up from the region located beneath the curve or the
hyperplane. Moreover, it can be seen from Fig. 6b that for
the response system 18, the value of
c is dominated
mostly by the strength k2 when k1 is less than zero; nevertheless, this value distinctly goes up with the increase of k1
when k1 is much bigger than zero. All these lead to a conclusion that the critical value
c is completely dependent of
the deterministic structure of the specifically coupled chaotic
oscillators.
All the above show that the occurrence of chaos synchronization between two conventionally coupled Rsslerlike systems could be surprisingly accelerated by a proper
amount of white noise in coupling when this amount is larger
than some relatively small critical value. The analytical argument and numerical simulation presented above can be
analogously generalized to other conventionally coupled
chaotic oscillators with intrinsic noise perturbations, showing
once again, that to some extent white noise makes sense in
chaos synchronization.
Furthermore, it is worthwhile to further investigate the
influence of multidimensional white noise on the conventional coupling techniques. The corresponding results may be
a little different from those in the previous one-dimensional
case, which could be seen from the illustrative examples
within two-dimensional white-noise-based coupling in the
previous section.
V. DISCUSSIONS AND CONCLUSIONS

FIG. 6. The plot of the critical value
c vs the coupling strength k2

1 , 1 with the fixed strengths k1 = 22, k3 = 11 a; the phase diagram of
c
vs both k2 1 , 1 and k1 = 2k3 = 2k 24, 16 b.

As seen previously, the It interpretation is adopted in

the description of the white-noise-based coupling in the paper; however, the Stratonovich interpretation,8,16 taking into
account a systematic correlation between the systems and
external noise, is always preferred in modeling some concrete physical phenomena. When the Stratonovich interpretation is used in the response system 3, the damping rate
induced by noise intensity disappears for simply linear noise
coupling strength functions. Particularly, the conclusions established for the complete synchronization between systems
14 and 18 will be changed: when linear noise coupling
strength functions are taken into account in the sense of Stratonovich, whether the complete synchronization might be enhanced by noise intensity is not yet certain; nevertheless, the
synchronization is not violated by linear coupling within any
intensity and it is most likely determined by the deterministic
structure of the chaotic oscillators. More interestingly, a special kind of nonlinear noise coupling strength functions in
the Stratonovich interpretation might stabilize the error dynamics with a large probability, which is illustrated by onedimensional error dynamics with multiplicative noise in Appendix B and will be further discussed in our future work.
Based on the above discussions, one may naturally ask the
following: Which interpretation is more preferable? Our
viewpoint is that the white-noise-based coupling in 3 could

Downloaded 04 Apr 2006 to 202.120.224.18. Redistribution subject to AIP license or copyright, see http://chaos.aip.org/chaos/copyright.jsp

013134-9

Chaos 16, 013134 2006

White noise makes sense

be interpreted in the sense of It if the generated noise process with the deterministic oscillators possesses the features
described in Sec. II. More importantly, comparing the analytical results with the experimental data properly can lead to
a more accurate model.
Above all, we have analytically and numerically shown
that it is possible to merely use white noise in an appropriate
way to achieve complete chaos synchronization. Indeed, the
established sufficient conditions on noise coupling intensities
not only ensure successful chaos synchronization via whitenoise-based coupling with probability one but also make it
clear that white noise may enhance chaos synchronization in
conventional coupled chaotic oscillators. This indicates that
noise, if used properly, can be beneficial to some real applications. As a matter of fact, this kind of analytical theories
could be further applied to illustrate the mechanisms of
chaos control, neuron information processing, and complex
network evolution, when some sorts of noise influence are
involved in the processing. Also, noise-induced synchronization of other types, as well as noise-induced desynchronization, could also be analytically investigated, leaving some
interesting topics to the future investigations.
Although some other well-known chaotic systems possessing polynomial vector field, such as the Lorenz-like system, the original Rssler system, and the chemical oscillators
discussed in Refs. 11 and 12 were not included in our discussions, numerical simulation and theoretical analysis could
be applied to confirm the boundedness of the error dynamics
generated by such systems via white-noise-based coupling.
Once this boundedness is guaranteed, the analytical argument given in this paper could be similarly applied to these
systems. Consequently, analogous results of complete synchronization via white-noise-based coupling of such systems
can be established.
ACKNOWLEDGMENT

The authors are grateful to the referees for their helpful

comments and suggestions. This research was supported by
the Hong Kong Research Grants Council under the CERG
Grant No. 114/05E1 and by the National Natural Foundation
of China Grant No. 10501008.

Here, we verify the claim that et ; t0 , e0 never approaches the trivial solution with probability one when the
initial value e0 is a nonzero vector.
First, applying an elementary inequality for vectors, one
has

1
+ traceHTeD2Ve1He
2
eTAe + fy fx
e3

1
+
2 i=1

hTi ehie
3
e

e hie
+3
e5
T

est0

t0

+ Les ds + M0t
es

1
+ M0t,
e0

A1

where M0t is a continuous martingale with expectation

EM0t = 0.
In what follows, verification is performed by contradiction. Assume that the above claim is not true. Then, there
exists some nonzero vector e0 Rn such that the probability
P : , = inft t0:et;e0 = 0 0.
Here, denotes the set of all elementary events. Thus, for
some given e0, it is possible to get constants T t0 and M
0, such that the probability P* 0, in which
* = : T,et;t0,e0 M,t t0,T.
For an arbitrarily given 0 , e0, denote the stopping
time by

= minT,inft t0:et ,M.

Clearly, for any event *, the corresponding T and
e ; t0 , e0 = . This, together with A1, leads to

1
eTt0
et0
,
EI* E
I *

e0
which implies P* eTt0 / e0. Here, I* stands for
an indicator function equal to 1 if * or 0 otherwise.
Since is arbitrarily taken, we get P* = 0, letting 0.
However, this is a contradiction, which completes the verification of the claim.

For simplicity, consider a one-dimensional error dynamics with multiplicative noise in the sense of Stratonovich:
,
e = Ae + fe + he W

B1

where the function f satisfies the Lipschitz condition with

Lipschitz constant L 0, f0 = 0, and h is a noise coupling
strength function with h0 = 0. As is well known, system
B1 could be equivalently transformed into the following
system in the sense of It:

Le
DVe1Ae + fy fx

1
ett0
=
+
et
e0

APPENDIX B: COUPLING IN THE SENSE

OF STRATONOVICH

APPENDIX A: NONZERO SOLUTION

where the existence of constant 0 is simply due to assumptions 2 and 8. Then, by means of Its formula, one
obtains

.
e = Ae + fe + 21 hehe + heW

B2

By using a similar argument given in Sec. II to system B2,

one obtains

Downloaded 04 Apr 2006 to 202.120.224.18. Redistribution subject to AIP license or copyright, see http://chaos.aip.org/chaos/copyright.jsp

013134-10

Chaos 16, 013134 2006

W. Lin and G. Chen

loge loget0 t t0
=
+
t
t
t

Ae + fe + 21 hehe
e

loget0 t t0
+
t
t

A+L+

Mt
h2e
+
2
2e
t

Mt
hehee h2e
.
+
2
2e
t

Obviously, if he is a linear function, i.e., he = le, the exponential rate is almost surely lower than A + L, which implies that the noise intensity l could not destroy the stability
of the trivial solution of system B1. Moreover, when the
following ordinary differential equation,
hehee h2e = 2le2

B3

is satisfied, the exponential rate is almost surely lower than

A + L l. Solving Eq. B3 directly gives
he = e

4l log

1
+ C,
e

B4

where C is a positive constant such that he is well defined,

and he 0 as e 0. Thus, if l A + L, C is sufficiently
large, and the nonlinear coupling strength function is designed to be in the form of B4, then the solution initiating
from the neighborhood of the origin will be suppressed to the
trivial solution in an exponential damping rate with a large
probability.
C. Huygens, Philos. Trans. R. Soc. London 4, 937 1669.
E. Ott, C. Grebogi, and J. A. Yorke, Phys. Rev. Lett. 64, 1196 1990; M.
Benedicks and L. Carleson, Ann. Math. 133, 73 1991; T. Shinbrot et al.,
Nature London 363, 411 1993; C. Robinson, Dynamical Systems: Stability, Symbolic Dynamics, and Chaos, 2nd ed. CRC Press, New York,
1998; J. Stark and K. Hardy, Science 29, 1192 2003.
3
B. Ermentrout, J. Math. Biol. 22, 1 1985; C. S. Skarda and W. J. Freeman, Behav. Brain Sci. 10, 161 1987; C. M. Gray, P. Knig, P. A. K.
Engel, and W. Singer, Nature London 338, 334 1989; M. E. J. New1
2

man, C. Morre, and D. J. Watts, Phys. Rev. Lett. 84, 3201 2000; M. Y.
Choi, H. J. Kim, D. Kim, and H. Hong, Phys. Rev. E 61, 371 2000.
4
G. Chen and X. Dong, From Chaos to Order: Methodologies, Perspectives
and Applications World Scientific, Singapore, 1998; Handbook of Chaos
Control, edited by H. G. Schuster Wiley-VCH, Weinheim, 1999; G.
Chen and X. Yu, Chaos Control: Theory and Applications SpringerVerlag, Berlin, 2003; E. M. Bollt, Int. J. Bifurcation Chaos Appl. Sci.
Eng. 13, 269 2003.
5
H. Fujisaka and T. Yamada, Prog. Theor. Phys. 69, 32 1983; C. Grebogi,
E. Ott, and J. A. Yorke, Phys. Rev. Lett. 50, 935 1983; A. Pikovsky, G.
Osipov, M. Rosenblum, M. Zaks, and J. Kurths, Phys. Rev. Lett. 79, 47
1997; L. Kocarev and U. Parlitz, Phys. Rev. Lett. 76, 1816 1996.
6
W. Lin and Y. B. He, Chaos 15, 023705 2005; W. Lin, preprint 2005.
7
L. Gammaitoni et al., Rev. Mod. Phys. 70, 223 1998; T. Wellens, V.
Shatokhin, and A. Buchleitner, Rep. Prog. Phys. 67, 45 2004.
8
W. Horsthemke and R. Lefever, Noise-Induced Transitions: Theory and
Applications in Physics, Chemistry, and Biology Springer-Verlag, Berlin,
1989.
9
R. M. May, Stability and Complexity in Model Ecosystems Princeton
University Press, Princeton, NJ, 1973; J. Garcia-Ojalvo and J. M. Sancho,
Noise in Spatially Extended Systems Springer-Verlag, Berlin, 1999.
10
L. Yu, E. Ott, and Q. Chen, Phys. Rev. Lett. 65, 2935 1990; Physica D
53, 102 1991; A. S. Pikovsky, Phys. Lett. A 165, 33 1992; M. Wang,
Z. Hou, and H. Xin, J. Phys. A 38, 145 2005.
11
S. Boccaletti, J. Kurths, G. Osipov, D. L. Valladares, and C. S. Zhou,
Phys. Rep. 366, 1 2002; C. S. Zhou, J. Kurths, E. Allaria, S. Boccaletti,
R. Meucci, and F. T. Arecchi, Phys. Rev. E 67, 015205 2003; Phys. Rev.
E 67, 066220 2003; C. S. Zhou et al., Phys. Rev. Lett. 89, 014101
2003.
12
C. S. Zhou and J. Kurths, Phys. Rev. Lett. 88, 230602 2002; Chaos 13,
401 2003.
13
T. Matsumoto, IEEE Trans. Circuits Syst. 31, 1055 1984; Chaos in
Nonlinear Electronic Circuits, edited by L. O. Chua and M. Hasler.
14
O. E. Rssler, Phys. Lett. 57, 397 1976; I. A. Heisler, T. Braun, Y.
Zhang, G. Hu, and H. A. Cerdeira, Chaos 13, 185 2003.
15
A. H. Jazwinski, Stochastic Processes and Filtering Theory Academic,
New York, 1970; N. G. van Kampen, J. Stat. Phys. 24, 175 1981.
16
L. Arnold, Stochastic Differential Equations: Theory and Applications
Wiley, New York, 1972; L. Arnold and H. Crauel, Random Dynamical
System, Lecture Notes in Mathematics, 1991, Vol. 1486, p. 1; X. Mao, J.
Math. Anal. Appl. 268, 125 2002.
17
A. Friedman, Stochastic Differential Equations and Applications, Vol. I
Academic, New York, 1975; R. S. Liptser and A. N. Shiryayev, Theory
of Martingales Kluwer, Dordrecht, 1986.
18
P. E. Kloeden and E. Platen, Numerical Solution of Stochastic Differential
Equations Springer-Verlag, Heidelberg, 1992; P. E. Kloeden, E. Platen,
and H. Schurz, Numerical Solution of Stochastic Differential Equations
Through Computer Experiments Springer-Verlag, Heidelberg, 1993.

Downloaded 04 Apr 2006 to 202.120.224.18. Redistribution subject to AIP license or copyright, see http://chaos.aip.org/chaos/copyright.jsp