bifurcation, amplitude death and oscillation patterns in a ... · of oscillators and nonlinear...
TRANSCRIPT
Bifurcation, amplitude death and oscillation patterns in a system of threecoupled van der Pol oscillators with diffusively delayed velocity coupling
Yongli Song,1,a) Jian Xu,2 and Tonghua Zhang3
1Department of Mathematics, Tongji University, Shanghai 200092, China2School of Aerospace Engineering and Applied Mechanics, Tongji University, Shanghai 200092, China3Faculty of Engineering and Industrial Sciences, Swinburne University of Technology, Mail H38, PO Box 218Hawthorn, Victoria 3122, Australia
(Received 8 February 2011; accepted 22 March 2011; published online 20 April 2011)
In this paper, we study a system of three coupled van der Pol oscillators that are coupled through the
damping terms. Hopf bifurcations and amplitude death induced by the coupling time delay are first
investigated by analyzing the related characteristic equation. Then the oscillation patterns of these
bifurcating periodic oscillations are determined and we find that there are two kinds of critical
values of the coupling time delay: one is related to the synchronous periodic oscillations, the other is
related to eight branches of asynchronous periodic solutions bifurcating simultaneously from the
zero solution. The stability of these bifurcating periodic solutions are also explicitly determined by
calculating the normal forms on center manifolds, and the stable synchronous and stable phase-
locked periodic solutions are found. Finally, some numerical simulations are employed to illustrate
and extend our obtained theoretical results and numerical studies also describe the switches of stable
synchronous and phase-locked periodic oscillations. VC 2011 American Institute of Physics.
[doi:10.1063/1.3578046]
Coupled dynamical system arises in a variety of contexts
in the physical, biological, and social sciences and has
broad relevance to many areas of research. Synchroniza-
tion phenomena in the coupled system is ubiquitous and
of interest to researchers in different research fields.
Generally speaking, time delay is inevitable in the
coupled systems since the information flows between the
individual units are never conveyed instantaneously, but
only after some time delay. A nature question is how
delay coupling affects the synchronization patterns. Here,
we consider a delay-coupled system consisting of three
van der Pol oscillators and attempt to analytically investi-
gate how the coupling time delay and the coupling
strength affect the stability and synchronization patterns.
Hopf bifurcations and amplitude death induced by the
coupling time delay are investigated and the correspond-
ing relation between the synchronization patterns of
Hopf bifurcating periodic oscillations and the critical val-
ues of the coupling time delay is explicitly found.
I. INTRODUCTION
Since synchronization phenomena is frequently encoun-
tered in nature and can help to explain many phenomena in
biology, chemistry, physics and has potential applications in
engineering and communication, it has been considered as
one of the fundamental characteristics of collective dynamics
of coupled systems and has permanently remained an object
of intensive research in nonlinear science. System of coupled
self-oscillators is often regarded as a basic model in the theory
of oscillators and nonlinear dynamics dealing with synchroni-
zation phenomenon. A system of coupled nonlinear oscillators
is capable of displaying a rich dynamics and has applications
in various areas of science and technology, such as physical,
chemical, biological, and other systems.1–4 In particular, the
effect of synchronization in systems of coupled oscillators
nowadays provides a unifying framework for different phe-
nomena observed in nature (for more reviews, see Refs. 5–7).
The van der Pol equation has long been studied as a
quintessential example of a self-excited oscillator and has
been used to represent oscillations in a wide variety of appli-
cations.8,9 It evolves in time according to the following sec-
ond order differential equation:
€yðtÞ þ eðy2ðtÞ � 1Þ _yðtÞ þ xðtÞ ¼ 0; (1)
where y is position coordinate and e is a positive real number
indicating the damping strength. It is well known10 that
Eq. (1) has a unique periodic solution which attracts all other
orbits except the origin, which is the unique equilibrium
point of Eq. (1) and unstable. It has been shown in Refs. 11
and 12 that linear feedback can annihilate limit cycles and
stabilize the origin.
Recently, there are increasing interests on the study of
the synchronization behavior in the coupled van der Pol
oscillators (see, e.g., Refs. 13–17 and references therein). In
Ref. 13, the synchronization of four coupled van der Pol
oscillators has been carried out. It has been shown that, for
the coupling constant below a critical value, there exists a
region in which a diversity of phase-shift attractors is pres-
ent, whereas for values above the critical value an in-phase
attractor is predominant. It is also observed that the presence
a)Author to whom correspondence should be addressed. Electronic mail:
1054-1500/2011/21(2)/023111/15/$30.00 VC 2011 American Institute of Physics21, 023111-1
CHAOS 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
of an antiphase attractor in the subcritical region is associated
with sudden changes in the period of the coupled oscillators.
In Ref. 14, the synchronization in the system of coupled noni-
dentical nonisochronous van der Pol–Duffing oscillators with
inertial and dissipative coupling has been studied. In Ref. 15,
Ookawara and Endo have investigated the effect of element
value deviation on the degenerate modes in a ring of three and
four coupled van der Pol oscillators. By using the averaging
method, they proved that, for a ring of three coupled oscilla-
tors, two frequencies bifurcate from the degenerate mode, syn-
chronize if they are close enough, but lose synchronization
when they are separated to some extent; while for a ring of
four coupled oscillators, the two frequencies generally cannot
be synchronized, even if they are close enough. In Ref. 16,
Woafo and Enjieu Kadji have investigated different states of
synchronization in a ring of mutually van der Pol oscillators.
Using the properties of the variational equations of stability,
they calculated the good coupling parameters leading to com-
plete and partial synchronization or disordered states and
obtained a stability map showing domains of synchronization
to an external excitation locally injected in the ring. In Ref. 17,
the stability of the synchronization manifold in a ring and in an
open-ended chain of nearest neighbor coupled self-sustained
systems, each self-sustained system consisting of multilimit
cycle van der Pol oscillators, has been investigated and it has
been found that synchronization occurs in a system of many
coupled modified van der Pol oscillators, and it is stable even
in the presence of a spread of parameters.
However, the coupling delay is not considered in the lit-
eratures mentioned above. In fact, the coupling is certainly
not instantaneous and might exhibit propagation delays
which are comparable to or larger than the characteristic os-
cillatory time scale of the individual oscillators. It has been
shown that a coupling delay can induce complex behavior in
a network and the nodes organize in different synchroniza-
tion patterns. For example, the coupling delay can induce the
amplitude death even for zero frequency mismatch between
the limit cycle oscillators,18–20 which does not occur in the
case without coupling time delay, can lead to chaos when
two lasers are coupled face to face21 and cause the oscillators
to switch from being in-phase to being out-of-phase in vari-
ous types of delay-coupled oscillators.22–25
In the following, we consider a system of three delay-
coupled van der Pol oscillators where three such oscillators
are diffusively coupled through the damping terms
€yiðtÞ þ eðy2i ðtÞ � 1Þ _yiðtÞ þ yiðtÞ
¼ kð _yiþ1ðt� sÞ þ _yiþ2ðt� sÞ � _yiðtÞÞ; i¼ 1;2;3 ðmod3Þ;(2)
where k � 0 represents the coupling strength and s � 0 is
the coupling time delay. It is well known that the dynamics
of the coupled system depends on the properties of each of
the units and on their interactions. Clearly, in system (2), the
parameters k and s reflect the interactions of three van der
Pol oscillators. Here, we aim at addressing the influence of
the coupling strength and time delay on the stability and
synchronized patterns of system (2). We will show that the
coupling time delay will induce to stable synchronous peri-
odic solutions, stable phase-locked periodic solutions, and
unstable mirror-reflecting and standing waves.
We would like to mention that there are several works on
delayed coupling van der Pol oscillators similar to system (2).
In Ref. 26, Wirkus and Rand have studied the dynamics of a
pair of van der Pol oscillators where the coupling is chosen to
be through the damping terms but not of diffusive type. They
used the method of averaging to reduce the problem to the
study of a slow-flow in three dimensions and investigated the
steady state solutions of this slow-flow, with special attention
given to the bifurcations accompanying their change in num-
ber and stability. Sen an Rand,27 entirely by numerical meth-
ods, investigated the dynamics of a pair of relaxation
oscillators of the van der Pol type with delay position coupling
and have found the various phase-locked motions including
the in-phase and out-of-phase modes. Li et al.28 investigated
the dynamics of a pair of van der Pol oscillators with delayed
position and velocity coupling. Using the method of averaging
together with truncation of Taylor expansions, they have found
that the dynamics of 1:1 internal resonance is more complex
than that of non-1:1 internal resonance. More recently, the sys-
tem of a pair of coupled van der Pol oscillators has been stud-
ied by Atay29 and Song.24 In Ref. 29, Atay has investigated
the synchronization and amplitude death using averaging
theory. The parameter ranges for the coupling strength and
delay are calculated such that the oscillators exhibit amplitude
death or stable in-phase or antiphase synchronized oscillations
with identical amplitudes. In Ref. 24, Song has investigated
the stability and oscillation patterns induced by the coupling
time delay. It was shown that for appropriate damping and
coupling strengths there are stability switches and the
synchronized dynamics changes from in-phase to out-of-phase
oscillations (or vice versa) as the coupling time delay is
increased and the critical values of coupling time delay lead-
ing to different oscillation patterns were explicitly calculated.
This paper is organized as follows. Local stability and
delay-induced Hopf bifurcations are presented in Sec. II.
Spatio-temporal patterns of bifurcating periodic solutions are
investigated in Sec. III. In Sec. IV, we calculate the normal
forms on center manifolds and then determine stability of
bifurcating periodic orbits. Numerical simulations are
employed to support and extend the theoretical analysis in
Sec. V. A summary of the results is presented in Sec. VI.
II. LOCAL STABILITY AND DELAY-INDUCED HOPFBIFURCATIONS
Letting YiðtÞ ¼ ðyi; _yiÞ 2 R2; i ¼ 1; 2; system (2) can
be written as
_YiðtÞ ¼ MYiðtÞ þ N Yiþ1ðt� sÞ þ Yiþ2ðt� sÞð Þ þ gðYiðtÞÞ;i ¼ 1; 2; 3 ðmod3Þ; (3)
where
M ¼0 1
�1 e� k
� �; N ¼
0 0
0 k
� �;
ga
b
� �¼
0
�ea2b
� �; ða; bÞT 2 R2:
(4)
023111-2 Song, Xu, and Zhang Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
The linearization of (3) at the zero solution leads to
_YiðtÞ ¼ MYiðtÞ þ N Yiþ1ðt� sÞ þ Yiþ2ðt� sÞð Þi ¼ 1; 2; 3 ðmod3Þ: (5)
The characteristic matrix of linearized system (5) is given by
Msð0; kÞ ¼kI2 �M �Ne�ks �Ne�ks
�Ne�ks kI2 �M �Ne�ks
�Ne�ks �Ne�ks kI2 �M
0B@
1CA;
where I2 is a 2� 2 identity matrix. Assuming that gj 2 R2
are eigenvectors, respectively, of kI2 �M � v1j Ne�ks
� v2j Ne�ks; j ¼ 0; 1; 2, then it is easy to verify that
Msð0; kÞvj ¼ diag T2; T2; T2f gvj; j ¼ 0; 1; 2; (6)
where T2 ¼ kI2 �M � vjNe�ks � v2j Ne�ks, vj ¼ ðgj; vjgj;
v2j gjÞT , and vj ¼ e2pj=3i: So, we have
detMsð0; kÞ ¼Y2
j¼0
detðkI2 �M � vjNe�ks � v2j Ne�ksÞ;
and then the characteristic equation of the linearization of (3)
at the zero solution is
Dðs; kÞ ¼ detMsð0; kÞ ¼ D1D22 ¼ 0; (7)
where
D1 ¼ ðk2 þ ðk � eÞkþ 1� 2kke�ksÞ;D2 ¼ ðk2 þ ðk � eÞkþ 1þ kke�ksÞ:
Note the fact that the root k of the characteristic equation 7
corresponds to the zero eigenvalue of the matrix
kI2 �M � vjNe�ks � v2j Ne�ks; j ¼ 0; 1; 2, which will be an
important information to calculate the eigenvector of system
(5) for the purely imaginary roots of the characteristic equa-
tion 7 in the following two sections.
In the remaining of this section, we investigate the distri-
bution of roots of Eq. (7), which determines the local stability
of the zero solution of (2). It is convenient to start with a more
general second order transcendental polynomial equation
k2 þ pkþ r þ ske�ks ¼ 0; p; r; s 2 R and s 6¼ 0; (8)
which has been extensively studied (see, e.g., Refs. 30 and
31). From Refs. 30 and 31, we first have the following lemma.
Lemma 2.1. Assuming that r > 0, then we have thefollowing.
(i) If jsj < jpj, then Eq. (8) has no purely imaginaryroot.
(ii) If jsj > jpj, then Eq. (8) has a pair of purely imagi-nary roots 6ixþ (6ix�, respectively) at s ¼ ~sþj(s ¼ ~s�k , respectively) such that ~sþj 6¼ ~s�k for anynon-negative integer numbers j, k, while Eq. (8) hastwo pairs of purely simple imaginary roots 6ixþand 6ix� at s ¼ ~sþj (or s ¼ ~s�k ) for some non-nega-tive integer numbers j and k such that ~sþj ¼ ~s�k .
(iii) If jsj ¼ jpj, then Eq. (8) has a pair of purely imagi-nary roots 6i
ffiffirp
at s ¼ ~sþj .(iv) Letting
kðsÞ ¼ gðsÞ þ ixðsÞ
is a solution of Eq. (8) satisfying
gð~s6j Þ ¼ 0; xð~s6
j Þ ¼ x6;
then we have
dReðkð~sþj ÞÞds
> 0;dReðkð~s�j ÞÞ
ds< 0; for jsj > jpj (9)
and
dReðkð~sþj ÞÞds
¼ 0; for jsj ¼ jpj:Here
x6 ¼ffiffiffi2p
2s2 þ 2r � p26
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðs2 þ 2r � p2Þ2 � 4r2
q� �1=2
and
~sþj ¼ 1xþ
2jpþ p� arccos ps
� �� �;
~s�j ¼ 1x�
2jpþ pþ arccos ps
� �� �;
for s > jpj;
~sþj ¼ 1xþ
2jpþ pþ arccos ps
� �� �;
s�j ¼ 1x�
2jpþ p� arccos ps
� �� �;
for s < �jpj;
~sþj ¼ 1ffiffirp 2jpþ pð Þ; for s ¼ p;
~sþj ¼ 1ffiffirp 2ðjþ 1Þpð Þ; for s ¼ �p;
8>>>>>>>>>>><>>>>>>>>>>>:
with j 2 f0; 1; 2;…g.From Lemma 2.1, we have the following two lemmas
on the distribution of roots of the equations D1 ¼ 0 and
D2 ¼ 0.
Lemma 2.2. For the equation D1 ¼ 0, we have the fol-lowing results:
(i) When s ¼ 0; the equation D1 ¼ 0 has only two rootsand they both have negative real parts for k < �e,positive real parts for k > �e, and zero real partsfor k ¼ �e.
(ii) If �e < k < e=3, then the equation D1 ¼ 0 has nopurely imaginary root.
(iii) If either k < �e or k > e=3, then the equationD1 ¼ 0 has a pair of purely simple imaginary roots6ixþ1 (6ix�1 , respectively) at s ¼ sþ1j (s ¼ s�1k,respectively) such that sþ1j 6¼ s�1k for any non-nega-tive integer numbers j, k, while the equation D1 ¼ 0
has two pairs of purely simple imaginary roots6ixþ1 and 6ix�1 at s ¼ sþ1j (or s ¼ s�1k) for somenon-negative integer numbers such that sþ1j ¼ s�1k.
(iv) If either k ¼ �e or k ¼ e=3, then Eq. (8) has a pairof purely imaginary roots 6i at s ¼ sþ1j.
Here,
x61 ¼
ffiffiffi2p
2
hð3k � eÞðk þ eÞ
þ 26ffiffið
p3k � eÞðk þ eÞ ð3k � eÞðk þ eÞ þ 4½ �
i1=2
(10)
023111-3 Amplitude death and oscillation patterns Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
and
sþ1j ¼ 1xþ
1
2jpþ pþ arccos e�k2k
� �� �;
s�1j ¼ 1x�
1
2jpþ p� arccos e�k2k
� �� �;
for k > e=3;
sþ1j ¼ 1xþ
1
2jpþ p� arccos e�k2k
� �� �;
s�1j ¼ 1x�
1
2jpþ pþ arccos e�k2k
� �� �;
for k < �e;
sþ1j ¼ 2jpþ p; for k ¼ e=3;
sþ1j ¼ 2ðjþ 1Þp; for k ¼ �e:
8>>>>>>>>>>><>>>>>>>>>>>:
(11)
Lemma 2.3. For the equation D2 ¼ 0, we have the fol-lowing results:
(i) When s ¼ 0; the equation D2 ¼ 0 has only two rootsand they both have negative real parts for k > e=2,positive real parts for k < e=2, and zero real partsfor k ¼ e=2.
(ii) If k < e=2, then the equation D2 ¼ 0 has no purelyimaginary root.
(iii) If k > e=2, then the equation D2 ¼ 0 has a pair ofpurely imaginary roots 6ixþ2 (6ix�2 , respectively)at s ¼ sþ2j (s ¼ s�2k, respectively) such that sþ2j 6¼ s�2k
for any non-negative integer numbers j, k, while theequation D2 ¼ 0 has two pairs of purely simpleimaginary roots 6ixþ2 and 6ix�2 at s ¼ sþ2j (ors ¼ s�2k) for any non-negative integer numbers j andk such that sþ2j ¼ s�2k.
(iv) If k ¼ e=2, then the equation D2 ¼ 0 has a pair ofpurely imaginary roots 6i at s ¼ sþ2j.
Here,
x62 ¼
ffiffiffi2p
2eð2k � eÞ þ 26
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffieð2k � eÞ eð2k � eÞ þ 4½ �
ph i1=2
(12)
and
sþ2j ¼ 1xþ
2
2jpþ p� arccos k�ek
� �� �;
s�2j ¼ 1x�
2
2jpþ pþ arccos k�ek
� �� �;
for k > e=2;
sþ2j ¼ 2ðjþ 1Þp; for k ¼ e=2:
8>><>>: (13)
From (11), it is easy to conclude that for k > e3; s�10 < sþ10 if
and only if the following assumption (A1) holds
ðA1Þ arccose� k
2k
� �>
xþ1 � x�1xþ1 þ x�1
p;
and from (13), we have sþ20 < s�20 for k > e=2. In addition,
combining (11) and (13), we obtain that for k > e=2;s�10 < sþ20 if and only if the following assumption (A2) holds
ðA2Þ xþ2 arccose� k
2k
� ��x�1 arccos
k� ek
� �> p xþ2 �x�1� �
:
Rearranging the critical values of the coupling time delay
and setting
sþj 2 fsþ1m; sþ2m;m ¼ 0; 1; 2;…g;
s�j 2 fs�1m; s�2m;m ¼ 0; 1; 2;…g
(14)
such that sþj < sþjþ1; s�j < s�jþ1; j ¼ 0; 1; 2;…. Note the fact
that (A2) is not satisfied provided that k < e since for k < e,
xþ2 arccose� k
2k
� �� x�1 arccos
k � ek
� �<
p2
xþ2 � x�1� �
:
Thus, by Lemmas 2.2 and 2.3, we have the following results
on the characteristic equation 7.
Lemma 2.4. Assuming that s6j are defined by (14), we
have the following:
(i) If either k < e, then Eq. (7) has at least a pair of rootswith positive real parts for all s � 0.
(ii) If k > e and the assumptions (A1) and (A2) hold, thenthere exists a positive integer n such thats�n�1 < sþn�1 < sþn < s�n and all roots of Eq. (7) havenegative real parts for s 2 ðs�0 ; sþ0 Þ [ ðs�1 ; sþ1 Þ [ðs�n�1; s
þn�1Þ and at least a pair of roots with positive
real parts for s 2 ½0; s�0 Þ [ ðsþn�1;þ1Þ.
It is well known from the theory of functional differen-
tial equations FDEs; Ref. 32 that the characteristic equation
determines local stability of the steady state. If all the eigen-
values have negative real parts, the steady state is stable and
if the characteristic equation has at least one root with posi-
tive real part, the steady state is unstable. And if the charac-
teristic equation has a pair of pure imaginary roots at the
critical values and the transversality condition holds, Hopf
bifurcations occur when the parameter crosses its critical val-
ues. Therefore, by Lemmas 2.2–2.4 and the transversality
condition (9), we have the following results on the stability
and bifurcation phenomena of the zero steady state of the
delayed coupling van der Pol system (2).
Theorem 2.1: Assuming that s61j , s6
2j , and s6j are defined,
respectively, by (11), (13), and (14), we have the following:
(i) If either k < e, then the zero equilibrium of system(2) is unstable for all s � 0.
(ii) If k > e and the assumptions (A1) and (A2) hold,then there exists a positive integer n such thats�n�1 < sþn�1 < sþn < s�n and the zero equilibrium ofsystem (2) is asymptotically stable for s 2 ðs�0 ; sþ0 Þ[ðs�1 ; sþ1 Þ [ ðs�n�1; s
þn�1Þ and unstable for s 2 ½0; s�0 Þ [
ðsþn�1;þ1Þ.(iii) Near the zero equilibrium, system (2) undergoes
Hopf bifurcations at the critical values s61j and
undergoes equivalent Hopf bifurcations at the criti-cal values s6
2j .
According to Theorem 2.1, the stability switches region
in the plane of damping and coupling strengths is plotted in
Fig. 1. In the shaded region of Fig. 1, there exist stability
switches for the zero equilibrium of system (2) with increas-
ing the coupling time delay s, and in other regions, the zero
equilibrium of system (2) is always unstable for all s � 0:For fixed damping strength, say e ¼ 0:03, the islands of am-
plitude death in the plane of the coupling time delay and
strength, where all oscillations are quenched, is qualitatively
same as Fig. 2. Throughout the remainder of the present pa-
per, the damping and coupling strengths are always restricted
to the shaded region of Fig. 1.
023111-4 Song, Xu, and Zhang Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
III. SPATIO-TEMPORAL PATTERNS OF BIFURCATINGPERIODIC SOLUTIONS
It follows from Sec. II that due to the symmetry of sys-
tem (2), the purely imaginary eigenvalues ix62 are always
multiple at the critical values s62j . Hence, the standard Bopf
bifurcation theory of functional differential equations (see,
e.g., Hale32 and Hassard et al.)33 cannot be applied. In this
section, we first explore the symmetry of system (3) and then
apply the symmetric Hopf bifurcation theorem for delay dif-
ferential equations established in Ref. 34 to describe the spa-
tio-temporal patterns of bifurcating periodic solutions as the
coupling time delay crosses the critical values s62j . We would
also like to mention that this theory has been successfully
employed to study the spatio-temporal patterns in the delay-
coupled neural networks (see Refs. 25, 35–38 and references
therein).
Let G : C! Rn and C be a compact Lie group. By a
compact Lie group C, we mean a closed subgroup of
GLðRnÞ, the group of all invertible linear transformations of
the vector space Rn into itself. Note that the space of n� nmatrices may be identified with Rn2 , which contains GLðRnÞas an open subset. It follows from Ref. 39 that the system
_uðtÞ ¼ GðutÞ is said to be C -equivariant if GðcutÞ ¼ cGðutÞfor all c 2 C: Let C ¼D3 be the dihedral group of order 6,
which is generated by the cyclic group Z3 order 3 together
with the flip of order 2 (see Ref. 39, for more details). Denote
by q, the generator of the cyclic subgroup Z3 and j the flip.
Define the action of D3 on R6 by
q
Y1
Y2
Y3
0B@
1CA ¼
Y2
Y3
Y1
0B@
1CA; j
Y1
Y2
Y3
0B@
1CA ¼
Y2
Y1
Y3
0B@
1CA
for all Yi 2 R2; i ðmod 3Þ;
(15)
and then it is easy to verify the following lemma.
Lemma 3.1. System (3) is D3 equivariant.Assume that the infinitesimal generator AðsÞ of the C0-
semigroup generated by the linear system (5) has a pair of
purely imaginary eigenvalues 6ix62 at the critical value s6
2j .
From Sec. II, we also get that for the critical value s62j , there
are two independent eigenvectors v1 and v2, where vj
¼ ðgj; vjgj; v2j gjÞT ; j ¼ 1; 2; with vj ¼ eð2pj=3Þi and
g1 ¼ g2 ¼ 1; ixþ2 ;� �
for sþ2 ; and g1 ¼ g2 ¼ 1; ix�2� �
for s�2 :
Denoting the action of C ¼D3 on R2 by
qu1
u2
� �¼
� 12�ffiffi3p
2ffiffi3p
2� 1
2
!u1
u2
� �;
ju1
u2
� �¼
� 12�ffiffi3p
2
�ffiffi3p
212
!u1
u2
� �;
we have the following lemma.
Lemma 3.2. R2 is an absolutely irreducible representa-tion of C, and KerDðs6
2j ; ix62jÞ is isomorphic to R2 �R2.
Here a representation R2 of C is absolutely irreducible ifthe only linear mapping that commutes with the action of Cis a scalar multiple of the identity.
Proof: The proof for R2 to be two-dimensional abso-
lutely irreducible representation of C is straightforward and
can be found in, for example, Ref. 39. Clearly,
KerDðs62j ; ix
62 Þ
¼ a1 þ b1ið Þv1 þ a2 þ b2ið Þv2; a1; a2; b1; b2 2 Rf g:
Define a mapping J from KerDðs6j ; ix
62 Þ to R4 by
J a1 þ b1ið Þv1 þ a2 þ b2ið Þv2ð Þ¼ a1 þ a2; b1 � b2; b1 þ b2; a2 � a1ð ÞT
Clearly, J : KerDðs6j ; ib
62jÞ ffi R4 is a linear isomorphism.
Note that
q a1 þ b1ið Þv1 þ a2 þ b2ið Þv2ð Þ¼ a1 þ b1ið Þq v1ð Þ þ a2 þ b2ið Þq v2ð Þ¼ a1 þ b1ið Þeið2p=3Þv1 þ a2 þ b2ið Þe�ið2p=3Þv2
¼ � 12
a1 �ffiffi3p
2b1
þ i � 1
2b1 þ
ffiffi3p
2a1
v1
þ � 12
a2 þffiffi3p
2b2
þ i � 1
2b2 �
ffiffi3p
2a2
v2
FIG. 1. (Color online) Possible region of amplitude death for the coupling
time delay in the e� k plane. There exist stability switches for the zero equi-
librium of system (2) by increasing the coupling time delay s in the shaded
region and the zero equilibrium of system (2) is always unstable for any
s � 0 otherwise. l1 and l2 are curves determined by the equations
arc cosðe� k=2kÞ � ðxþ1 � x�1 Þ=ðxþ1 þ x�1 Þp ¼ 0 and xþ2 arccosðe� k=2kÞ� x�1 arccosðk � e=kÞ � p
2ðxþ2 �x�1 Þ ¼ 0; respectively.
FIG. 2. (Color online) Islands of amplitude death in the plane of the cou-
pling time delay s and the coupling strength k for fixed e ¼ 003.
023111-5 Amplitude death and oscillation patterns Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
and
j a1 þ b1ið Þv1 þ a2 þ b2ið Þv2ð Þ¼ a1 þ b1ið Þeið2p=3Þv2 þ a2 þ b2ið Þe�ið2p=3Þv1
¼ � 12
a1 �ffiffi3p
2b1
þ i � 1
2b1 þ
ffiffi3p
2a1
v2
þ � 12
a2 þffiffi3p
2b2
þ i � 1
2b2 �
ffiffi3p
2a2
v1:
Therefore,
J q a1 þ b1ið Þv1 þ a2 þ b2ið Þv2ð Þð Þ
¼ � 12
a1 þ a2ð Þ �ffiffi3p
2b1 � b2ð Þ;� 1
2b1 � b2ð Þ
þ
ffiffi3p
2a1 þ a2ð Þ;� 1
2b1 þ b2ð Þ �
ffiffi3p
2a2 � a1ð Þ;
� 12
a2 � a1ð Þ þffiffi3p
2b1 þ b2ð Þ
¼ q J a1 þ b1ið Þv1 þ a2 þ b2ið Þv2ð Þð Þ
and
J j a1 þ b1ið Þv1 þ a2 þ b2ið Þv2ð Þð Þ
¼ � 12
a1 þ a2ð Þ �ffiffi3p
2b1 � b2ð Þ; 1
2b1 � b2ð Þ
�
ffiffi3p
2a1 þ a2ð Þ;� 1
2b1 þ b2ð Þ �
ffiffi3p
2a2 � a1ð Þ;
� 12
a2 � a1ð Þ �ffiffi3p
2b1 þ b2ð ÞÞ
¼ j J a1 þ b1ið Þv1 þ a2 þ b2ið Þv2ð Þð Þ:
This completes the proof. h
Let x ¼ 2p=x6i ; i ¼ 1; 2; and denote by Px the Banach
space of all continuous x�periodic mappings x : R! R6.
Then D3 � S1 acts on Px, where S1 is a circle group, by
ðc; hÞxðtÞ ¼ cxðtþ hÞ; ðc; hÞ 2 D3 � S1:
Denote by SPx the set of all x�periodic solutions of (5)
with s ¼ s62j . Then, for s6
2j ,
SPx ¼ 11�621 þ 12�
622 þ 13�
623 þ 14�
624; 11; 12; 13; 14 2 R
� �;
where
�621 ¼ Re eix62
hv1
� �¼ cos x6
2 h� �
Re v1ð Þ � sin x62 h
� �Im v1ð Þ;
e622 ¼ Im eix6
2hv1
� �¼ sin x6
2 h� �
Re v1ð Þ þ cos x62 h
� �Im v1ð Þ;
�623 ¼ Re eix62
hv2
� �¼ cos x6
2 h� �
Re v2ð Þ � sin x62 h
� �Im v2ð Þ;
�624 ¼ Im eix62
hv2
� �¼ sin x6
2 h� �
Re v2ð Þ þ cos x62 h
� �Im v2ð Þ:
(16)
For SPx, we have the following property.
Lemma 3.3. SPx is an D3�invariant subspace of Px
and
j �621
� �¼ � 1
2�623 �
ffiffiffi3p
2�624; j �622
� �¼
ffiffiffi3p
2�623 �
1
2�624;
j �623
� �¼ � 1
2�621 þ
ffiffiffi3p
2�622; j �624
� �¼ �
ffiffiffi3p
2�621 �
1
2�622;
q �621
� �¼ � 1
2�621 �
ffiffiffi3p
2�622; q �622
� �¼
ffiffiffi3p
2�621 �
1
2�622;
q �623
� �¼ � 1
2�623 þ
ffiffiffi3p
2�624; q �624
� �¼ �
ffiffiffi3p
2�623 �
1
2�624:
Proof: Note that g1 ¼ g2 ¼ ð1; ix62 Þ
T : It is easy to verify
that
j Re v1ð Þð Þ ¼ � 12
Re v2ð Þ �ffiffi3p
2Re v2ð Þ;
j Im v1ð Þð Þ ¼ffiffi3p
2Re v2ð Þ � 1
2Im v2ð Þ;
j Re v2ð Þð Þ ¼ � 12
Re v1ð Þ þffiffi3p
2Re v1ð Þ;
j Im v2ð Þð Þ ¼ �ffiffi3p
2Re v1ð Þ � 1
2Im v1ð Þ;
and
q Re v1ð Þð Þ ¼ � 12
Re v1ð Þ �ffiffi3p
2Re v1ð Þ;
q Im v1ð Þð Þ ¼ffiffi3p
2Re v1ð Þ � 1
2Im v1ð Þ;
q Re v2ð Þð Þ ¼ � 12
Re v2ð Þ þffiffi3p
2Re v2ð Þ;
q Im v2ð Þð Þ ¼ �ffiffi3p
2Re v2ð Þ � 1
2Im v2ð Þ:
By these formulae and (16), the lemma follows from a
straightforward calculation. h
Let Rh ¼ fðc; eið2p=xÞhÞ; h 2 ð0;xÞg and FixðRh;SPxÞbe the fixed point set of Rh under the action D3 � S
1. We
consider
FixðRh;SPxÞ¼fxðtÞ2SPx; ðc;eið2p=xÞhÞxðtÞ¼cxðtþhÞ¼xðtÞfor all ðc;eið2p=xÞhÞ2Rhg:
Setting
R6ð2;3Þ ¼ j;61ð Þh i;R6
q ¼ hðq; e6ið2p=3ÞÞi;
we can prove the following lemma.
Lemma 3.4. FixðR62;3; SPxÞ and FixðR6
q ; SPxÞ are allsubspaces of SPx of dimension 2.
Proof: (1) First, x 2 FixðRþ2;3; SPxÞ if and only if
jðxÞ ¼ x: For x ¼P4
i¼1 1i�62i ; we have
jx¼X4
i¼1
1ij �62i
� �¼� 1
213þ
ffiffi3p
214
�621þ
ffiffi3p
213� 1
214
�622
þ � 1211þ
ffiffi3p
212
�623�
ffiffi3p
211þ 1
212
�624
It follows that x 2 FixðRþ2;3; SPxÞ if and only if
� 12�ffiffi3p
2ffiffi3p
2� 1
2
!13
14
� �¼
11
12
� �:
This implies that FixðRþ2;3; SPxÞ is spanned by ��1 and ��2,
where
��1 ¼ �1
2�621 þ
ffiffiffi3p
2�622 þ �623; ��2 ¼ �
ffiffiffi3p
2�621 �
1
2�622 þ �624:
(2) Second, if and only if It is easy to see that �62iðtþ x2ÞÞ
¼ ��62i and then xðtþ x2Þ ¼ �
P4i¼1 1i�
62i : This implies that
x 2 FixðR�2;3; SPxÞ if and only if
023111-6 Song, Xu, and Zhang Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
� 12�ffiffi3p
2ffiffi3p
2� 1
2
!13
14
� �¼ �
11
12
� �:
Therefore, Fix ðR�2;3; SPxÞ is spanned by ��3 and ��4, where
��3 ¼1
2�621 �
ffiffiffi3p
2�622 þ �623; ��4 ¼
ffiffiffi3p
2�621 þ
1
2�622 þ �624:
(3) Third, for x ¼P4
i¼1 1i�62i ; we have
qðxÞ ¼ � 1211 þ
ffiffi3p
212
�621 �
ffiffi3p
211 þ 1
212
�622
� 1213 þ
ffiffi3p
214
�623 þ
ffiffi3p
213 � 1
214
�624
In addition,
cos x62 t� x
3
� �� �¼ cos x6
2 t� 2p3
� �¼ � 1
2cos x6
2 t� �
þffiffi3p
2sin x6
2 t� �
;
sin x62 t� x
3
� �� �¼ sin x6
2 t� 2p3
� �¼ �
ffiffi3p
2cos x6
2 t� �
� 12
sin x62 t
� �:
This, together with the expression of each �62j and
x ¼P4
j¼1 1j�62j ; leads to
x t� x3
� �¼ � 1
211 þ
ffiffi3p
212
�621 þ
ffiffi3p
211 � 1
212
�622
� 1213 þ
ffiffi3p
214
�623 þ
ffiffi3p
213 � 1
214
�624
So, x 2 FixðRþq ; SPxÞ, i.e., qðxðtÞÞ ¼ x t� x3
� �, if and only if
� 1211 þ
ffiffi3p
212 ¼ � 1
211 �
ffiffi3p
212;
�ffiffi3p
211 � 1
212 ¼
ffiffi3p
211 � 1
212;
� 1213 �
ffiffi3p
214 ¼ � 1
213 �
ffiffi3p
214;ffiffi
3p
213 � 1
214 ¼
ffiffi3p
213 � 1
214:
8>>>>><>>>>>:
That is, x 2 FixðRþq ; SPxÞ if and only if 11 ¼ 12 ¼ 0; and
x 2 FixðR�q ; SPxÞ if and only if 13 ¼ 14 ¼ 0: Therefore,
FixðRþq ; SPxÞ is spanned by �623 and �624, and FixðR�q ; SPxÞ is
spanned by �621 and �622. h
Note that, if Y(t) is a periodic solution of (3), then so is
ðc; eið2p=xÞhÞYðtÞ for every ðc; eið2p=xÞhÞ 2Dn � S1: It fol-
lows from the symmetric bifurcation theory of delay differ-
ential equations due to Wu34 that, under usual nonresonance
and transversality conditions, for every subgroup R Dn
�S1 such that the R�fixed-point subspace of SPx is of
dimension 2, symmetric delay differential equations has a
bifurcation of periodic solutions whose spatial–temporal
symmetries can be completely characterized by R. By
Lemma 3.4, the subgroups R6ð2;3Þ and R6
q can be employed to
describe the symmetry of periodic solutions of (3) which ex-
hibit certain spatial–temporal patterns (see Ref. 39 and 34,
for more details). Thus, Lemmas 3.1–3.4 allow us to apply
the general symmetric Hopf bifurcation theorem due to
Wu34 to describe the spatio-temporal patterns of local Hopf
bifurcating periodic oscillations.
Theorem 3.1: System (3) has eight branches of asyn-
chronous periodic solutions of period ps near 2p=x62 , bifur-
cated simultaneously from the zero solution at the critical
values s ¼ s62j , and these are
(i) two phase-locked oscillations: YiðtÞ ¼ Yiþ1 t6ps=3ð Þ,for i(mod 3) and t 2 R;
(ii) three mirror-reflecting waves: YiðtÞ ¼ YjðtÞ 6¼ YkðtÞ,for t 2 R and for some distinct i; j; k 2 f1; 2; 3g; and
(iii) three standing waves: YiðtÞ ¼ Yj tþ ðps=2Þð Þ, for
t 2 R and some pair of distinct elements i; j 2f1; 2; 3g.
IV. THE CALCULATION OF NORMAL FORMS ONCENTER MANIFOLDS AND STABILITY OFBIFURCATING PERIODIC ORBITS
In this section, we employ the algorithm and notations
of Faria and Magalhaes40,41 to derive the normal forms of
system (3) on center manifolds and then we can determine
the direction and stability of bifurcating periodic orbits.
Rescaling the time by t 7! t=s to normalize the delay so
that system (3) can be written as a FDE in the phase space
C ¼ Cð½�1; 0�;R6Þ; and then separating the linear terms
from the nonlinear terms, (3) becomes
_uðtÞ ¼ LsðutÞ þ Fðut; sÞ (17)
where ut 2 C; utðhÞ ¼ uðtþ hÞ;�1 h 0; and for u¼ ðu1;u2;u3;u4;u5;u6ÞT 2 C, we have
LsðuÞ ¼ sM1u2� 2ð0Þ þ sM2uð�1Þ (18)
and
Fðu; sÞ ¼ sð0; �eu21ð0Þu2ð0Þ; 0; �eu2
3ð0Þu4ð0Þ; 0;
� eu25ð0Þu6ð0ÞÞT ; (19)
where u ¼ ðu1;u2;u3;u4;u5;u6ÞT 2 C and M1, M2 are as
follows:
M1 ¼M O2 O2
O2 M O2
O2 O2 M
0B@
1CA; M2 ¼
O2 N N
N O2 N
N N O2
0B@
1CA;
where M, N are defined by (4) and O2 is a 2� 2 zero matrix.
There exists a function of bounded variation such that
the linear map Ls may be expressed in the following integral
form:
LsðuÞ ¼ð0
�1
½dgsðhÞ�uðhÞ
Letting C� ¼ Cð½0; 1�;R6�Þ with R6� being the six-dimen-
sional space of row vectors, define the adjoint bilinear form
on C*�C as follows:
hwðsÞ;/ðhÞi ¼ wð0Þ/ð0Þ �ð0
�1
ðh
0
wðn� hÞdgs� ðhÞ/ðnÞdn;
w 2 C�;/ 2 C: (20)
It follows from Sec. II that the characteristic equation of (18)
has purely imaginary roots 6ix�s� at the critical values
023111-7 Amplitude death and oscillation patterns Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
s ¼ s� of the coupling time delay, where x� ¼ x61 for
s� ¼ s61j and x� ¼ x6
2 for s� ¼ s62j . Setting K0 ¼ �ix�s�;f
ix�s�g and using the formal adjoint theory for FDEs in Ref.
32, the phase space C can be decomposed by K0 as
C ¼ P� Q; where P is the center space for _uðtÞ ¼ Ls�ut; i.e.,
P is the generalized eigenspace associated with K0: Let Uand W be bases for P and for the space P* associated with
the eigenvalues 6ix�s� of the formal adjoint equation,
respectively, and normalized so that , where Ip is the p� pidentity matrix and p¼ dim P. Assume that B is a p� p real
matrix with the point spectrum rðBÞ ¼ K and satisfies simul-
taneously satisfy _U ¼ UB;� _W ¼ BW:Introducing the new parameter l ¼ s� s� such that
l ¼ 0 is a Hopf bifurcation value of (17), Eq. (17) becomes
_uðtÞ ¼ Ls�ut þ ~Fðut; lÞ;
where
~Fðut; lÞ ¼ Llut þ Fðu; s� þ lÞ:
Enlarging the phase space C by considering the Banach
space BC of functions from [–1, 0] into R4 which are uni-
formly continuous on [–1, 0] and with a jump discontinuity
at 0. Using the decomposition ut ¼ UxðtÞ þ y; with
xðtÞ 2 Cp and y 2 Q1 Q \ C1 (C1 is the subset of C con-
sisting of continuously differentiable functions). Then we
can decompose (17) as
_x ¼ BxþWð0Þ ~FðUxþ y; lÞ;_y ¼ AQ1 yþ ðI � pÞX0
~FðUxþ y; lÞ;
where AQ1 : Q1 ! ker p is such that AQ1/ ¼ _/þ X0½Ls� ð/Þ� _/ð0Þ�. We write the Taylor formula
Wð0Þ ~FðUxþ y; lÞ ¼ 12! f
12 ðx; y; lÞ þ 1
3! f13 ðx; y; lÞ þ � � � ;
ðI � pÞX0~FðUxþ y; lÞ ¼ 1
2! f22 ðx; y; lÞ þ 1
3! f23 ðx; y; lÞ þ � � � ;
(21)
where f 1j ðx; y; lÞ and f 2
j ðx; y; lÞ are homogeneous polyno-
mials in ðx; y; lÞ of degree j with coefficients in Cp, ker p,
respectively, and hot stands for higher order terms. There-
fore, differential equation EDE can be written as
_x ¼ BxþPj�2
1j! f
1j ðx; y; lÞ;
_y ¼ AQ1 yþPj�2
1j! f
2j ðx; y; lÞ:
(22)
From the normal form theory of FDEs due to Faria and Mag-
alhaes,40,41 the normal form of (17) on the center manifold
of the origin at l ¼ 0 is given by
_x ¼ Bxþ 1
2!g1
2ðx; 0; lÞ þ1
3!g1
3ðx; 0; lÞ þ h:o:t:; (23)
where g12ðx; 0; lÞ; g1
3ðx; 0; lÞ are the second and third order
terms in ðx; lÞ; respectively, and hot stands for higher order
terms. In the following part of this section, we will calculate
the normal form (23) at the critical values s61j and s6
2j .
A. For the critical values s61j
From Sec. II, at s ¼ s61j the characteristic equation of
(18) has purely imaginary roots 6ix61 s6
1j which are simple.
So, in this subsection, p¼ 4, x� ¼ xþ1 for s� ¼ sþ1j and
x� ¼ x�1 for s� ¼ s�1j and then we have
B ¼ diagðix�s�;�ix�s�Þ
It follows from (6) that the base of the center space P at
s ¼ s61j can be taken as U ¼ /1;/2ð Þ; where
/1ðhÞ ¼ eix�s�hv0; /2ðhÞ ¼ e�ix�s�h�v0; �1 h 0;
where
v0 ¼ ðg0; g0; g0ÞT ; and g0 ¼ 1; ixþ1� �
for sþ1j
and g0 ¼ 1; ix�1� �
for s�1j:(24)
One can easily show that the adjoint basis satisfying
hWðsÞ;UðhÞi ¼ I2 is
WðsÞ ¼w1ðsÞw2ðsÞ
� �¼ 1
3
d�vT0 e�ix�s�s
�dvT0 eix�s�s
!; 0 s 1;
where
d ¼ 1
1þ x2� þ s�x2
�ðk � eÞ � s�x� 1� x2�
� �i: (25)
Let V3j ðC
2Þ be the homogeneous polynomials of degree j in
three variables, x1; x2; l, with coefficients in C2, and let M1
j
denote the operator from V3j ðC
2Þ into itself defined by
ðM1j hÞðx; lÞ ¼ Dxhðx; lÞBx� Bhðx; lÞ; (26)
where h 2 V3j ðC
2Þ. Then, since B is the 2� 2 diagonal ma-
trix, it follows from Ref. 41 that
V3j ðC
2Þ ¼ ImðM1j Þ � KerðM1
j Þ; g1j ðx; 0; lÞ 2 KerðM1
j Þ
and
KerðM1j Þ ¼ Kerfxq1
1 xq2
2 llek : q1k1 þ q2k2 ¼ kk;
k ¼ 1; 2; q1; q2; l 2 N0; q1 þ q2 þ l ¼ jg;
where k1 ¼ ix�s�; k2 ¼ �ix�s� and fe1; e2g is canonical
basis of R2: Hence,
KerðM12Þ ¼ span
x1l
0
� �;
0
x2l
� � �;
KerðM13Þ ¼ span
x21x2
0
� �;
x1l2
0
� �;
0
x1x22
� �;
0
x2l2
� � �:
Since f 00ð0Þ ¼ 0; by (21), we have
1
2!f 12 ðx; 0; lÞ ¼ Wð0ÞLlðUxÞ
¼13
d�vT0
13
�dvT0
!ðLlð/1Þx1 þ Llð/2Þx2Þ;
023111-8 Song, Xu, and Zhang Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
and the second order terms in ðx; lÞ of the normal form on
the center manifold are given by
1
2g1
2ðx; 0; lÞ ¼1
2ProjKerðM1
2Þf
12 ðx; 0; lÞ
¼ ProjKerðM12Þ
13
d�vT0
13
�dvT0
!ðLlð/1Þx1 þ Llð/2Þx2Þ:
Note that
Llð/1Þ ¼ ilx�v0; Llð/2Þ ¼ �ilx�v0:
So,
1
2g1
2ðx; 0; lÞ ¼A1x1l
A1x2l
� �;
with A1 ¼ 13
dx��vT0 v0i: From (24), we have
A1 ¼ dx�ð1þ x2�Þi: (27)
Next we compute the cubic terms g13ðx; 0; lÞ as described in
(23). We first note that
1
3!g1
3ðx; 0; lÞ ¼ ProjKerðM13Þ
1
3!~f 13 ðx; 0; lÞ
¼ 1
3!ProjS
~f 13 ðx; 0; 0Þ þ Oðjxjl2Þ;
for
S ¼ spanx2
1x2
0
� �;
0
x1x22
� � �;
where 13!
~f 13 ðx; 0; lÞ denotes the third order terms after the com-
putation of the normal form up to the second order terms. It is
sufficient to compute only ProjS~f 13 ðx; 0; 0Þ for the purpose of
determining the generic Hopf bifurcation. Since f 00ð0Þ ¼ 0;implying f 1
2 ðx; y; 0Þ ¼ 0; we can deduce that, after the change
of variables that transformed the quadratic terms f 12 ðx; y; lÞ of
the first equation in (22) into g12ðx; y; lÞ; the coefficients of
third order at y ¼ 0; l ¼ 0 are still given by 13! f
13 ðx; 0; 0Þ, i.e.,
1
3!~f 13 ðx; 0; 0Þ ¼
1
3!f 13 ðx; 0; 0Þ:
Thus, from (19) and (21), we have
1
3!~f 13 ðx; 0; 0Þ ¼
1
3!f 13 ðx; 0; 0Þ
¼ Wð0ÞF3ðUx; s�Þ
¼ s�3
d�vT0
�dvT0
!0
�eðx1v01 þ x2�v01Þ2ðx1v02 þ x2�v02Þ0
�eðx1v03 þ x2�v03Þ2ðx1v04 þ x2�v04Þ0
�eðx1v05 þ x2�v05Þ2ðx1v06 þ x2�v06Þ
0BBBBBBBB@
1CCCCCCCCA;
where v0¼ (v01, v02, v03, v04, v05, v06)T. It follows that
13! g
13ðx; 0; 0Þ ¼ 1
3! ProjS~f 13 ðx; 0; 0Þ
¼A2x2
1x2
A2x1x22
!;
where
A2 ¼ �es�d
3ðv2
01�v202 þ v2
03�v204 þ v2
05�v206 þ 2jv01j2jv02j2
þ 2jv03j2jv04j2 þ 2jv05j2jv06j2Þ:
This, together with (24), yields that
A2 ¼ �es�x2�d: (28)
So, the normal form of (17) on the center manifold reads to
_x ¼ BxþA1x1l
A1x2l
� �þ
A2x21x2
A2x1x22
!þ O jxjl2 þ jx4j
� �;
(29)
where A1 and A2 are defined by (27) and (28), respectively.
Changing (29) to real coordinates by the change of variables
x1¼w1 – iw2, x2¼w1þ iw2 and then using polar coordinates
ðq; nÞ; w1 ¼ q cos n; w2 ¼ q sin n; we obtain
_q ¼ K1lqþ K2q3 þ O l2qþ jðq; lÞj4
;
_n ¼ �x�s� þ Oðjðq; lÞjÞ
with K1 ¼ ReðA1Þ; K2 ¼ ReðA2Þ:It is well known42 that the sign of K1 K2 determines the
direction of the bifurcation (supercritical if K1K2 < 0, sub-
critical if K1K2 > 0), and the sign of K2 determines the sta-
bility of the nontrivial periodic orbits (stable if K2 < 0,
unstable if K2 > 0). From (27) and (28), we have
K1 ¼ ReðA1Þ ¼ �x�ð1þ x2�ÞReðdÞ; (30)
and
K2 ¼ ReðA2Þ ¼ �es�x2ReðdÞ: (31)
From (10) and (25), it is easy to verify that for k > e;
ReðdÞ ¼ 1þ x2� þ s�x2
�ðk � eÞC
> 0;
ImðdÞ ¼s�x� 1� x2
�� �C
< 0; for s� ¼ sþ1j;
> 0; for s� ¼ s�1j;
( (32)
where
C ¼ 1þ x2� þ s�x
2�ðk � eÞ
� �2 þ s2�x
2� 1� x2
�� �2
:
By (30), (31), and (32), we have
K1 ¼>0; for s� ¼ sþ1j;
<0; for s� ¼ s�1j;
(K2 < 0;
which immediately leads to the following theorem.
Theorem 4.1: Assuming that k > e and the assumptions
(A1) and (A2) hold, then near sþ1j ðrespectively; s�1jÞ, there
exists a supercritical (respectively, subcritical) bifurcation of
stable synchronous periodic solutions of period p, near
2p=xþ1 ðrespectively; 2p=x�1 Þ, bifurcated from the zero
023111-9 Amplitude death and oscillation patterns Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
solution of system (2), and those nontrivial periodic orbits
are all stable on the center manifold.
B. For the critical values s62j
From Sec. II, at s ¼ s62j the characteristic equation of
(18) has purely imaginary roots 6ix62 s6
2j which are double.
So, in this subsection, p¼ 4, x� ¼ xþ2 for s� ¼ sþ2j and
x� ¼ x�2 for s� ¼ s�2j and then we have
B ¼ diagðix�s�;�ix�s�; ix�s�;�ix�s�Þ
By (6), the base of the center space P at s ¼ s� can be taken
as U ¼ /1;/2;/3;/4ð Þ; where
/1ðhÞ ¼ eix�s�hv1; /2ðhÞ ¼ e�ix�s�h�v1;
/3ðhÞ ¼ eix�s�hv2; /4ðhÞ ¼ e�ix�s�h�v2;� 1 h 0: (33)
It is easy to check that a basis for the adjoint space P* is
WðsÞ ¼
w1ðsÞ
w2ðsÞ
w3ðsÞ
w4ðsÞ
0BBBBB@
1CCCCCA ¼
1
3
de�ix�s�s�vT1
�deix�s�svT1
de�ix�s�s�vT2
�deix�s�svT2
0BBBBB@
1CCCCCA; 0 s 1: (34)
with hWðsÞ;UðhÞi ¼ I4 for the adjoint bilinear form on
C� � C defined by (20), where d is defined by (25).
Let V5j ðC
4Þ be the homogeneous polynomials of degree
j in 5 variables, , with coefficients in C4, and let M1
j denote
the operator from V5j ðC
4Þ into itself defined by (26) with
h 2 V5j ðC
4Þ. Thus, we have
M1j ðlxqekÞ ¼ ix�s�lðq1 � q2 þ q3 � q4 þ ð�1ÞkÞxqek;
jqj ¼ j� 1;
where j � 2; 1 k 4; and fe1; e2; e3; e4g is the canonical
basis of R4: In particular, if jqj ¼ 1, then
KerðM12Þ \ span lxqek : jqj¼1;k¼1;2;3;4f g
¼ span lx1e1;lx3e1;lx2e2;lx4e2;lx1e3;lx3e3;lx2e4;lx4e4f g:(35)
It follows from (33) and (34) that
Wð0Þ ¼ 1
3
d�vT1
�dvT1
d�vT2
�dvT2
0BBBBB@
1CCCCCA;
Uð0Þx ¼ v1; �v1; v2; �v2ð Þx ¼ x1v1 þ x2�v1 þ x3v2 þ x4�v2
and
Uð�1Þx¼ x1e�ix�s�v1þ x2eix�s��v1þ x3e�ix�s�v2þ x4eix�s��v2:
By (21), we have
12! f
12 ðx; 0; lÞ ¼ Wð0ÞLlðUxÞ
¼ ilx�Wð0Þ x1v1 � x2�v1 þ x3v2 � x4�v2ð Þ
¼ ilx� 1þ x2�
� � d x1 � x4ð Þ�d x3 � x2ð Þd x3 � x2ð Þ�d x1 � x4ð Þ
0BBB@
1CCCA:
(36)
and then by (35) the second order terms in ðx; lÞ of the nor-
mal form on the center manifold are given by
12
g12ðx; 0; lÞ ¼ 1
2ProjKerðM1
2Þf
12 ðx; 0; lÞ
¼
B1x1l
B1x2l
B1x3l
B1x4l
0BBB@
1CCCA;
(37)
where
B1 ¼ ix� 1þ x2�
� �d: (38)
Next we compute the cubic terms g13ðx; 0; lÞ as described in
(23). We first note that
g13ðx; 0; lÞ ¼ ProjKerðM1
3Þ~f
13 ðx; 0; lÞ
¼ ProjKerðM13Þ~f
13 ðx; 0; 0Þ þ Oðjxjl2Þ;
where 13!
~f 13 ðx; 0; lÞ denotes the third order terms after the
computation of the normal form up to the second order
terms. It is easy to see from (36) and (37) that g12ðx; 0; 0Þ
¼ f 12 ðx; 0; 0Þ ¼ 0 So,
1
3!~f 13 ðx; 0; 0Þ ¼ 1
3! f13 ðx; 0; 0Þ
¼ Wð0ÞF3ðUx; s�Þ
¼ s�Wð0Þ
0
�e Uð0Þxð Þ1� �2
Uð0Þxð Þ20
�e Uð0Þxð Þ3� �2
Uð0Þxð Þ40
�e Uð0Þxð Þ5� �2
Uð0Þxð Þ6
0BBBBBBBB@
1CCCCCCCCA;
where
Uð0Þxð Þj ¼ x1v1j þ x2�v1j þ x3v2j þ x4�v2j; j ¼ 1; 2; 3; 4; 5; 6:
Also note that for q¼ 3,
M13ðxqejÞ ¼ 0; if and only if q1 � q2 þ q3 � q4 þ ð�1Þj ¼ 0;
j ¼ 1;2; 3;4:
So,
KerðM13Þ ¼ spanfx2
1x2e1; x21x4e1; x
23x2e1; x
23x4e1; x1x2x3e1;
x1x3x4e1; x22x1e2; x
22x3e2; x
24x1e2; x
24x3e2;
x1x2x4e2; x2x3x4e2; x21x2e3; x
21x4e3; x
23x2e3;
x23x4e3; x1x2x3e3; x1x3x4e3; x
22x1e4; x
22x3e4;
x24x1e4; x
24x3e4; x1x2x4e4; x2x3x4e4g
023111-10 Song, Xu, and Zhang Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
and then
1
3!g1
3ðx; 0; 0Þ ¼ 13! ProjKerðM1
3Þ~f
13 ðx; 0; 0Þ
¼
B2 x21x2 þ 2x1x3x4
� ��B2 x2
2x1 þ 2x2x3x4
� �B2 x2
3x4 þ 2x1x2x3
� ��B2 x2
4x3 þ 2x1x2x4
� �
0BBB@
1CCCA;
where
B2 ¼ �es� x�ð Þ2d: (39)
So, the normal form of (17) on the center manifold reads to
_x ¼ x�s�
i 0 0 0
0 �i 0 0
0 0 i 0
0 0 0 �i
0BBB@
1CCCAxþ
B1x1l�B1x2l
B1x3l�B1x4l
0BBB@
1CCCA
þ
B2 x21x2 þ 2x1x3x4
� ��B2 x2
2x1 þ 2x2x3x4
� �B2 x2
3x4 þ 2x1x2x3
� ��B2 x2
4x3 þ 2x1x2x4
� �
0BBB@
1CCCAþ Oðjxjl2 þ jxj4Þ; (40)
where B1 and B2 are defined by (38) and (39), respectively.
Changing (40) to real coordinates by the change of variables
x ¼
1 �i 0 0
1 i 0 0
0 0 1 �i
0 0 1 i
0BBB@
1CCCAw
and letting
q21 ¼ x1x2 ¼ w2
1 þ w22; q2
2 ¼ x3x4 ¼ w23 þ w2
4;
we obtain
_w1
_w2
� �¼ x�s�
w2
�w1
� �þ l
Re B1ð Þw1 þ Re B1ð Þw2
�Im B1ð Þw1 þ Re B1ð Þw2
� �
þ q21 þ 2q2
2
� � Re B2ð Þw1 þ Re B2ð Þw2ð Þ�Im B2ð Þw1 þ Re B2ð Þw2
� �þ Oðjwjl2 þ jwj4Þ;
_w3
_w4
� �¼ x�s�
w4
�w3
� �þ l
Re B1ð Þw3 þ Re B1ð Þw4
�Im B1ð Þw3 þ Re B1ð Þw4
� �
þ 2q21 þ q2
2
� � Re B2ð Þw3 þ Re B2ð Þw4ð Þ�Im B2ð Þw3 þ Re B2ð Þw4
� �þ Oðjwjl2 þ jwj4Þ:
If we use double polar coordinates
w1 ¼ q1 cos v1; w2 ¼ q1 sin v1;
w3 ¼ q2 cos v2; w4 ¼ q2 sin v2;
then we get
_q1 ¼ ðlReðB1Þ þ ðq21 þ 2q2
2ÞReðB2ÞÞq1
þ Oðjðq1; q2Þjl2 þ jðq1; q2Þj4Þ;_q2 ¼ ðlReðB1Þ þ ðq2
2 þ 2q21ÞReðB2ÞÞq2
þ Oðjðq1; q2Þjl2 þ jðq1; q2Þj4Þ;_v1 ¼ �x�s� � ImðB1Þl� ImðB2Þq2
1 � 2ImðB2Þq22
þ Oðjðq1; q2Þjl2 þ jðq1; q2Þj4Þ;_v2 ¼ �x�s� � ImðB1Þl� 2ImðB2Þq2
1 � ImðB2Þq22
þ Oðjðq1; q2Þjl2 þ jðq1; q2Þj4Þ:
Introducing the periodic-scaling parameter x and letting
z1ðtÞ ¼ w1ðsÞ þ iw2ðsÞ;z2ðtÞ ¼ w3ðsÞ þ iw4ðsÞ
with
s ¼ ð1þ rÞx�s�½ ��1t;
we obtain
ð1þ rÞ _z1 ¼ �iz1ðtÞ þl
x�s�ðReðB1Þ � iReðB1ÞÞz1ðtÞ
þ 1
x�s�ðjz1j2 þ 2jz2j2ÞðReðB2Þ � iReðB2ÞÞz1ðtÞ
þ Oðjzjl2 þ jzj4Þ
¼ �iz1ðtÞ þl
x�s��B1z1ðtÞ þ
1
x�s��B2z1ðtÞ
� ðjz1j2 þ 2jz2j2Þ þ Oðjzjl2 þ jzj4Þ:
Similarly, we get an equation for z2(t). Thus, ignoring the
terms Oðl2jzjÞ and Oðjzj4Þ, we get the normal form
ð1þrÞ _z1¼�iz1ðtÞþl
x�s��B1z1ðtÞþ
1
x�s��B2z1ðtÞðjz1j2þ2jz2j2Þ;
ð1þrÞ _z2¼�iz2ðtÞþl
x�s��B1z2ðtÞþ
1
x�s��B2z2ðtÞð2jz1j2þjz2j2Þ:
(41)
Let g : C�C�R! C�C be given so that �gðz1; z2; lÞis the right-hand side of (41). Then (41) can be written as
ð1þ xÞ _zþ gðz; lÞ ¼ 0
According to Ref. 39 (pp. 296–297, Theorems 6.3 and 6.5),
the bifurcations of small-amplitude periodic solutions of (41)
are completely determined by the zeros of equation,
� ið1þ rÞzþ gðz; lÞ ¼ 0; (42)
and their (orbital) stability is determined by the signs of three
eigenvalues of the following matrix:
Dgðz; 0Þ � ið1þ rÞId
By (41), (42) is equivalent to
irz1 þl
x�s��B1z1 þ
1
x�s��B2ðjz1j2 þ 2jz2j2Þz1 ¼ 0;
irz2 þl
x�s��B1z2 þ
1
x�s��B2ðjz2j2 þ 2jz1j2Þz2 ¼ 0:
(43)
023111-11 Amplitude death and oscillation patterns Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
To apply the results of Ref. 39, we write (43) as
Az1
z2
� �þ B
z21�z1
z22�z2
!¼ 0
with
A ¼ A0 þ ANðjz1j2 þ jz2j2Þ; B ¼ B0
for some complex numbers A0, AN, and B0 given by
A0 ¼ �l
x�s��B1 � ix;
AN ¼ �2
x�s��B2;
B0 ¼1
x�s��B2:
(44)
It follows from (25), (32), (39), and (44) that for k > e;
ReðAN þ B0Þ ¼ �1
x�s�Re �B2ð Þ ¼ ex�ReðdÞ > 0;
Reð2AN þ B0Þ ¼ �3
x�s�Re �B2ð Þ3ex�ReðdÞ > 0;
and
ReB0 ¼1
x�s�Re �B2ð Þ ¼ �ex�ReðdÞ < 0:
In addition, from (12), (25), and (38), we obtain that
Re�B1
x�s�
� �¼ �
1þ x2�
� �s�
ReðdÞ
¼x� 1þ x2
�� �
x2� � 1
� �C
> 0; for s� ¼ sþ2j;
< 0; for s� ¼ s�2j:
(
By the results of Ref. 39 (pp. 383, Theorem 3.1), we have
the following theorem on the properties of bifurcation at the
critical values s62j :
Theorem 4.2: Assuming that k > e and the assumptions
(A1) and (A2) hold, then the bifurcations are supercritical
(respectively, subcritical) at the critical values sþ2j (respec-
tively, s�2j), and then on the center manifold the phase-locked
oscillations are orbitally asymptotically stable and the mir-
ror-reflecting waves and standing waves are unstable.
V. NUMERICAL CONTINUATION
In this section, we perform some numerical simulations
to illustrate and expand the results obtained above. In the fol-
lowing numerical simulations, we take e ¼ 0:03; k ¼ 0:1such that ðe; kÞ belongs to the shaded region of Fig. 1, where
stability switches occur with increasing the coupling time
delay. The following numerical simulations were performed
by Simulink of MATLAB and the initial functions are chosen as
default of Simulink for delay differential equations with
y1ð0Þ ¼ 0:2; _y1ð0Þ ¼ 0:1; y2ð0Þ ¼ �0:3; _y2ð0Þ ¼ �0:1; y3ð0Þ¼ 0:5; _y3ð0Þ ¼ �0:4.
From (11) and (13), we have
s�10¼:
1:3322; sþ10¼:
4:6172; s�11¼:
8:2315; sþ11¼:
10:3393;
s�12¼:
15:1307; sþ12¼:
16:0615; sþ13¼:
21:7836; s�13¼:
22:0300;
and
sþ20 ¼:
2:2639; s�20 ¼:
4:0801; sþ21 ¼:
8:3267; s�21 ¼:
10:5916;
sþ22 ¼:
14:3896; s�22 ¼:
17:1032; sþ23 ¼:
20:4524;
s�23 ¼:
23:6147; sþ24 ¼:
265153
So, the critical values of the coupling time delay can be
arranged as
s�10 < sþ20 < s�20 < sþ10 < s�11 < sþ21 < sþ11 < s�21 < sþ22
< s�12 < sþ12 < s�22 < sþ23 < sþ13 < s�13 < s�23 < sþ24:
According to Theorem 2.1, the zero equilibrium of system
(2) is asymptotically stable for s 2 s�10; sþ20
� �[ s�20; s
þ10
� �[
s�11; sþ21
� �. Figure 3 illustrate this result for s ¼ 2 2
s�10; sþ20
� �:
According to Theorems 3.1, 4.1, and 4.2, when s is less
than and close to s�10, stable synchronous periodic solutions
bifurcate from the zero solution. When s is greater than and
close to sþ20 or s is less than and close to s�20, stable phase-
locked periodic solutions bifurcate from the zero solution.
The existence of these stable phase-locked periodic solutions
are illustrated in Figs. 4(a) and 4(b) for s ¼ 2:3 greater than
and close to sþ20 and in Figs. 4(e) and 4(f) for s ¼ 2:4 less
than and close to s�20. When s is greater than and close to sþ10
or s is less than and close to s�11, stable synchronous periodic
solutions again bifurcate from the zero solution. The exis-
tence of these stable synchronous periodic solutions is illus-
trated in Figs. 5(a) and 5(b) for s ¼ 4:7 greater than and
close to sþ10 and in Figs. 5(e) and 5(f) for s ¼ 8:1 less than
and close to s�11. These numerical simulations are exactly
consistent with the results obtained in Secs. II–IV. When s is
far away from these critical values, our theoretical analysis
above is not available. However, numerical simulations
show that these stable phase-locked periodic solutions
always exist for all s 2 sþ20; s�20
� �and stable synchronous
periodic solutions always exist for all s 2 sþ10; s�11
� �and that
further the delay is far away from the critical values, the big-
ger the amplitude of the periodic solution is. Figures 5(c)
and 5(d) illustrate the trajectories of these periodic
FIG. 3. (Color online) Trajectories of the oscillators y1 (solid line), y2
(dashed line), and y3 (dotted line): the zero equilibrium is asymptotically sta-
ble for s 2 s�10; sþ20
� �[ s�20; s
þ10
� �[ s�11; s
þ21
� �. Here s ¼ 2 2 s�10; s
þ20
� �.
023111-12 Song, Xu, and Zhang Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
FIG. 5. (Color online) Trajectories of the oscillators y1 (solid line), y2 (dashed line), and y3 (dotted line) with stable synchronous periodic motion at s ¼ 4:7(a and b), s ¼ 6:5 (c and d), and s ¼ 8:1 (e and f), where s 2 sþ10; s
�11
� �:
FIG. 4. (Color online) Trajectories of the oscillators y1 (solid line), y2 (dashed line), and y3 (dotted line) with stable phase-locked periodic motion at s ¼ 2:3(a and b), s ¼ 3:1 (c and d), and s ¼ 4 (e and f), where s 2 sþ20; s
�20
� �:
023111-13 Amplitude death and oscillation patterns Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
oscillations for s ¼ 3:1 far away from the critical values sþ20
and s�20, and Figs. 6(c) and 6(d) illustrate the trajectories of
these periodic oscillations for s ¼ 6:5 far away from the crit-
ical values sþ10 and s�11. The results of these numerical contin-
uation are described as shown in Fig. 7.
When s > sþ21; the zero equilibrium of system (2) loses
its stability permanently. When s is greater than and close to
sþ21, Theorems 3.1, 4.1, and 4.2 show that there are stable
phase-locked periodic orbits bifurcating from the zero equi-
librium. Numerical simulations show that system (2) always
possesses for all s > sþ21. With increasing the coupling time
delay, oscillation patterns of these stable periodic orbits
switch from being phase-locked to synchronous and back to
phase-locked several times and finally becoming synchro-
nous. Numerical simulations also show that these stable peri-
odic orbit changes its oscillation patterns only when s is less
than and very close to the critical values s�2m and s�1n with
m¼ 1, 2, 3, n¼ 2, 3. More precisely, there exist sufficiently
small positive number l1; l2; l3; l4; l5 such that stable
phase-locked periodic orbits exist for
ðsþ21; s�21 � l1Þ [ ðs�12 � l2; s
�22 � l3Þ [ ðs�13 � l4; s
�23 � l5Þ
and stable synchronous periodic orbits exist for
ðs�21 � l1; s�12 � l1Þ [ ðs�22 � l3; s
�13 � l4Þ [ ðs�23 � l5;þ1Þ:
Figure 7 shows the switch from being phase-locked to syn-
chronous when s is less than and close to s�21: Figures 6(a)
and 6(b) show the existence of stable phase-locked periodic
oscillation for s ¼ 10:4 < s�21 � l1 and Figs. 7(c) and 7(d)
show the existence of stable synchronous periodic oscillation
for s ¼ 10:5 > s�21 � l1:
VI. CONCLUSIONS
We have studied the bifurcation, amplitude death, and os-
cillation patterns induced by the coupling time delay in a sys-
tem of three coupled van der Pol, which are coupled through
the damping terms. We have shown that the coupling time
delay can lead to rich dynamics. The stability of zero solution
is determined in the parameter space. The existence of
FIG. 6. (Color online) Regions for stable synchronous
and phase-locked periodic solutions in the plane of the
coupling delay and strength which are separated by
islands of amplitude death.
FIG. 7. (Color online) Trajectories of the oscillators y1 (solid line), y2 (dashed line), and y3 (dotted line) with stable phase-locked periodic motion at s ¼ 10:4(a and b) and stable synchronous periodic motion at s ¼ 10:5 (c and d), where s is less than and close to s�21:
023111-14 Song, Xu, and Zhang Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions
stability switches induced by the coupling time delay is found
in the region of the parameter plane of the damping strength
and the coupling strength. The islands of amplitude death are
also described in the parameter plane of the coupling time
delay and strength for appropriate fixed damping strength.
In the region where stability switches occur, there are
two kinds of critical values s61j and s6
2j of the coupling time
delay leading to Hopf bifurcating periodic solutions. By
using the symmetric bifurcation theory of delay differential
equations due to Wu, we have shown that at the critical val-
ues s61j , there exist a synchronous periodic solution bifurca-
tion and at the critical values s62j , there are eight branches of
asynchronous periodic solutions bifurcating simultaneously
from the zero solution: two phase-locked oscillations, mir-
ror-reflecting waves and three standing waves. The direction
and stability of these bifurcations is based on the normal
form calculations. We have shown that the bifurcation is
subcritical at the critical values s�kj and supercritical at the
critical values sþkj , k¼ 1, 2, j¼ 0, 1, 2, .... We have also
shown that on the center manifold the synchronous periodic
and phase-locked periodic solutions are stable and other mir-
ror-reflecting and standing waves are unstable.
Numerical studies have been employed to support and
extend the obtained theoretical results. For fixed damping
strength and coupling strength, the numerical simulations
illustrate the existence of stable synchronous and phase-
locked periodic solutions near the critical values of the cou-
pling time delay which is in good agreement with our theoreti-
cal analysis results. For the coupling time delay far away from
the critical values, our theoretical analysis is not available.
However, numerical simulations show that with increasing the
coupling time delay from the zero, the dynamics of the system
switches from being synchronous to stability, to being phase-
locked to stability, and so on. When the zero solution ulti-
mately loses its stability, the oscillation patterns of these
stable periodic solutions switch from being phase-locked to
being synchronous (or vice versa) and finally becoming syn-
chronous with increasing the coupling time delay.
ACKNOWLEDGMENTS
The first two authors (Y. Song and J. Xu) would like to
acknowledge the support from the Shanghai Committee of
Science and Technology (No. 10ZR1431900), the National
Nature Science Foundation of China (No. 10971155), and
the State Key Program of National Natural Science of China
(No. 11032009). The third author (T. Zhang) would like to
acknowledge the support from the Australian Research
Council (No. DP0770420) and the National Nature Science
Foundation of China (No. 11001212).
1J. Grasman, Asymptotic Methods for Relaxation Oscillations and Applica-tions (Springer-Verlag, New York, 1987).
2A. Hohl, A. Gavrielides, T. Erneux, and V. Kovanis, Phys. Rev. Lett. 78,
4745 (1997).3K. Radparvar and B. Z. Kaplan, IEEE Trans. Circuits Syst. 32, 1072
(1985).4T. Saito, IEEE Trans. Circuits Syst. 35, 1147 (1988).5S. Boccaletti, J. Kurths, D. L. Valladares, G. Osipov, and C. S. Zhou,
Phys. Rep. 366(1), 1 (2002).6S. C. Manrubia, A. S. Mikhailov, and A. H. Zanette, Emergence of Dy-namical Order (World Scientific, Singapore, 2004).
7A. Pikovsky, M. G. Rosenblum, and J. Kurths, Synchronization, A Univer-sal Concept in Nonlinear Sciences (Cambridge University Press, Cam-
bridge, 2001).8B. van der Pol, Radio Rev. 1, 701, 754 (1920).9B. van der Pol and J. van der Mark, Philos. Mag. 6(38), 763 (1928).
10F. Verhulst, Nonlinear Differential Equations and Dynamical Systems(Springer-Verlag, Berlin, 1990).
11F. M. Atay, J. Sound Vib. 218 (2), 333 (1998).12J. Wei and W. Jiang, J. Sound Vib. 283, 801 (2005).13M. A. Barron and M. Sen, Nonlinear Dyn. 56, 357 (2009).14A. P. Kuznetsova, N. V. Stankevichb, and L. V. Turukina, Physica D 238,
1203 (2009).15T. Ookawara and T. Endo, IEEE Trans. Circuits Syst., I: Fundam. Theory
Appl. 46, 827 (1999).16P. Woafo and H. G. Enjieu Kadji, Phys. Rev. E 69, 046206 (2006).17R. Yamapi, H. G. Enjieu Kadji, and G. Filatrella, Nonlinear Dyn. 61, 275
(2010).18D. V. Ramana Reddy, A. Sen, and G. L. Johnston, Phys. Rev. Lett. 80,
5109 (1998).19D. V. Ramana Reddy, A. Sen, and G. L. Johnston, Physica D 129, 335
(1999).20Y. Song, J. Wei, and Y. Yuan, J. Nonlinear Sci. 17, 145 (2007).21Y. Aviad, I. Reidler, W. Kinzel, I. Kanter, and M. Rosenbluh, Phys. Rev.
E 78, 025204 (2008).22A. Prasad, J. Kurths, S. K. Dana, and R. Ramaswamy, Phys. Rev. E 74,
035204 (2006).23A. Prasad, S. K. Dana, R. Karnatak, J. Kurths, B. Blasius, and R. Ramaswamy,
Chaos 18, 023111 (2008).24Y. Song, Nonlinear Dyn. 63, 223 (2011).25Y. Song, M. O. Tade, and T. Zhang, Nonlinearity 22, 975 (2009).26S. Wirkus and R. H. Rand, Nonlinear Dyn. 30, 205 (2002).27A. K. Sen and R. H. Rand, Commun. Pure Appl. Anal. 2(4), 567 (2003).28X. Li, J. Ji, and C. H. Hansen, Mech. Res. Commun. 33, 614 (2006).29F. M. Atay, Lect. Notes Control Inf. Sci. 388, 383 (2009).30S. Ruan, Quart. Appl. Math. 59, 159 (2001).31Y. Song, M. Han, and Y. Peng, Chaos, Solitons Fractals 22, 1139 (2004).32J. K. Haleand S. M. Verduyn Lunel, Introduction to Functional Differen-
tial Equations (Springer, New York, 1993).33B. D. Hassard, N. D. Kazarinoff, and Y.-H. Wan, Theory and Applications
of Hopf Bifurcation (Cambridge University Press, Cambridge, 1981).34J. Wu, Trans. Am. Math. Soc. 350, 4799 (1998).35S. A. Campbell, Y. Yuan, and S. Bungay, Nonlinearity 18, 2827 (2005).36S. Guo and L. Huang, Physica D 183, 19 (2003).37J. Wu, T. Faria, and Y. S. Huang, Math. Comput. Modell. 30, 117
(1999).38Y. Yuan and S. A. Campbell, J. Dyn. Differ. Equ. 16(3), 709 (2004).39M. Golubitsky, I. Stewart, and D. Schaeffer, Singularities and Groups in
Bifurcation Theory (Springer-Verlag, New York, 1988), Vol. II.40T. Faria and L. T. Magalhaes, J. Differ. Equ. 122, 181 (1995).41T. Faria and L. T. Magalhaes, J. Differ. Equ. 122, 201 (1995).42S. N. Chow and J. K. Hale, Methods of Bifurcation Theory (Springer, New
York, 1982).
023111-15 Amplitude death and oscillation patterns Chaos 21, 023111 (2011)
Downloaded 02 Aug 2011 to 136.186.80.71. Redistribution subject to AIP license or copyright; see http://chaos.aip.org/about/rights_and_permissions