ARTICLE

Received 28 Jan 2015 | Accepted 1 Jun 2015 | Published 15 Jul 2015 DOI: 10.1038/ncomms8709 OPEN

Restoration of rhythmicity in diffusively coupled

dynamical networks
Wei Zou1,2,3, D.V. Senthilkumar3,4, Raphael Nagao5, István Z. Kiss5, Yang Tang3,6, Aneta Koseska7,8,
Jinqiao Duan1,2,9 & Jürgen Kurths3,10,11,12

Oscillatory behaviour is essential for proper functioning of various physical and biological
processes. However, diffusive coupling is capable of suppressing intrinsic oscillations
due to the manifestation of the phenomena of amplitude and oscillation deaths. Here we
present a scheme to revoke these quenching states in diffusively coupled dynamical
networks, and demonstrate the approach in experiments with an oscillatory chemical
reaction. By introducing a simple feedback factor in the diffusive coupling, we show that the
stable (in)homogeneous steady states can be effectively destabilized to restore dynamic
behaviours of coupled systems. Even a feeble deviation from the normal diffusive coupling
drastically shrinks the death regions in the parameter space. The generality of our method is
corroborated in diverse non-linear systems of diffusively coupled paradigmatic models
with various death scenarios. Our study provides a general framework to strengthen the
robustness of dynamic activity in diffusively coupled dynamical networks.

1 School of Mathematics and Statistics, Huazhong University of Science and Technology, Wuhan 430074, China. 2 Center for Mathematical Sciences,

Huazhong University of Science and Technology, Wuhan 430074, China. 3 Potsdam Institute for Climate Impact Research, Telegraphenberg, D-14415
Potsdam, Germany. 4 Centre for Nonlinear Science and Engineering, School of Electrical and Electronics Engineering, SASTRA University, Thanjavur 613401,
India. 5 Department of Chemistry, Saint Louis University, 3501 Laclede Avenue, St Louis, Missouri 63103, USA. 6 The Key Laboratory of Advanced Control and
Optimization for Chemical Processes, Ministry of Education, East China University of Science and Technology, Shanghai 200237, China. 7 Department of
Systemic Cell Biology, Max Planck Institute of Molecular Physiology, Dortmund D-44227, Germany. 8 Research Centre for Computer Science and Information
Technologies, Macedonian Academy of Sciences and Arts, Skopje, Macedonia. 9 Department of Applied Mathematics, Illinois Institute of Technology,
Chicago, Illinois 60616, USA. 10 Institute of Physics, Humboldt University Berlin, D-12489 Berlin, Germany. 11 Institute for Complex Systems and Mathematical
Biology, University of Aberdeen, Aberdeen AB24 3FX, UK. 12 Department of Control Theory, Nizhny Novgorod State University, Gagarin Avenue 23, 606950
Nizhny Novgorod, Russia. Correspondence and requests for materials should be addressed to W.Z. (email: [email protected]) or to
I.Z.K. (email: [email protected]).

NATURE COMMUNICATIONS | 6:7709 | DOI: 10.1038/ncomms8709 | www.nature.com/naturecommunications 1

& 2015 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8709

M
odelling coupled oscillators offer an alternative and paradigmatic non-linear oscillators. Astonishingly, even a minute
effective approach aimed at understanding a plethora of deviation from the normal diffusive coupling can reduce
intriguing self-organizing phenomena in diverse fields of drastically the stable regimes of AD and OD in the parameter
non-linear sciences ranging from physics, chemistry, biology and space. Both the scheme and the phenomenon are clearly verified
neuroscience to engineering1–4. The phenomenon of oscillation in experiments using electrochemical oscillators: the stability of
quenching (oscillation suppression) is a fascinating emerging the HSS of current generating chemical reactions emanating from
behaviour, whereby individual oscillatory systems cease to oscillate dissolution of nickel wires is revoked by a simple adjustment of
when coupled, which results into the emergence of stationarity of the nature of the interactions. The proposed method is highly
the whole-coupled system5,6. One of the early observations of efficient and quite robust in regaining sustained oscillations in
oscillation quenching dates back to the 19th century by Lord networks of diffusively coupled non-linear oscillators, where the
Rayleigh, who found that two organ pipes of the same pitch dynamic activity has been damaged and lost as a consequence of
standing side by side mutually suppressed their vibration2. Later, the diffusive interaction. Our study leads to a significant
this phenomenon was observed in chemical reactions7,8, lasers9, improvement of our understanding of the role of diffusive
electronic circuits10, neural oscillators11 and so on. Models based coupling in controlling oscillatory activity of coupled complex
on coupled non-linear oscillators have been extensively used to non-linear systems.
understand the intricacies of oscillation quenching for more than
two decades5,6. Very recently, two structurally distinct oscillation Results
quenching processes have been clearly distinguished as: Two coupled Stuart–Landau oscillators. In the following, we will
(i) amplitude death (AD) and (ii) oscillation death (OD)6,12. demonstrate our approach using the paradigmatic model of
Generally, AD refers to stabilization of an already existing coupled Stuart–Landau oscillators, describing dynamics near a
homogeneous steady state (HSS) to which the coupled units are supercritical Hopf bifurcation (further examples of oscillatory
entrained. In contrast, OD is manifested as a novel inhomo- systems are discussed in the Supplementary Notes 1–5). Systems
geneous steady state (IHSS), which occurs due to symmetry of coupled Stuart–Landau oscillators have served as an ideal
breaking of the coupled systems, and the distinct units then model for exploring the mechanisms of AD by analytical
populate different branches of the same IHSS6,12. means for more than a couple of decades20–28. So far, numerous
The phenomena of AD and OD can be responsible for a loss of scenarios such as frequency mismatch20,21, time-delayed
intrinsic dynamics, which may lead to a large degree of coupling22–26, and conjugate and dynamic couplings27,28 and
degradation in the dynamic performance of physical systems. so on have been reported to induce AD in coupled Stuart–Landau
For example, the onset of AD and OD could seriously weaken oscillators. On the other hand, a burst of recent research activity
and even completely destroy the transmission of electricity in a has revealed that OD can even occur in coupled identical Stuart–
power grid, where the energy flows along transmission lines arise Landau oscillators when the coupling destroys the rotational
from periodic oscillations of the rotational phase of generators at symmetry of the system12,31–34. Here we will establish that
the frequency of 50 Hz (60 Hz)13–15. On the other hand, sustained the onset of both AD and OD under all the above-mentioned
oscillations represent one of the most striking and ubiquitous scenarios can be effectively revoked to retain self-sustained
manifestations of dynamic evolution in distinct real-world oscillations by a very simple coupling scheme.
systems16,17, such as spiking neural networks, cardiac and Let us begin with two coupled Stuart–Landau oscillators1
respiratory systems, and El Niño/Southern Oscillation in Earth’s   2 
ocean and atmosphere and so on. Moreover, rhythmic behaviours Z_ j ðt Þ ¼ 1 þ iwj  Zj ðt Þ Zj ðt Þ
play an essential function at all levels of physiological processes,   ð1Þ
which make it possible to organize and control the activity of þ K Zk ðt  tÞ  aZj ðt Þ ;
populations of cells18. For instance, it has been proposed that
highly synchronized theta frequency oscillations serve as a where j, k ¼ 1, 2 (jak), Zj ¼ xj þ iyj and wj are the complex
temporal organizer for cortex in the hippocampus19. These amplitude and the natural frequency of the jth oscillator,
examples furnish the robustness of dynamic activity as a main respectively, K is the coupling strength, and t is the
prerequisite for proper functioning of multitude natural and propagation delay. In contrast to the studies on the traditional
man-made networks. Thus, oscillation quenching in such real diffusive interaction, here we introduce a new factor a in the
systems is destructive and fatal to robustness of rhythmic coupling. The new ingredient offers the opportunity to better
behaviours, which is undesirable and should be circumvented. characterize the diffusion process at the interface of a wide range
For a couple of decades, the major research focus was centred of natural systems. In addition, the new feedback factor a acts as a
at identifying various scenarios that facilitate the emergence of control parameter: a ¼ 1 corresponds to a normal diffusive
AD and OD in coupled oscillatory systems20–34. Only in the coupling, whereas a ¼ 0 refers to a direct coupling scheme. Thus,
recent past, a significant attention has been devoted to explore the new diffusive coupling may naturally occur in more realistic
possibilities of restoring rhythmic activity after the dynamic circumstances as a bridge-linking direct coupling and normal
activity of the coupled networks irreversibly fails, as a means diffusive interaction. We will mainly probe its influence on
to understand sustained oscillatory mechanisms of natural the collective dynamics of coupled non-linear systems, and
systems35–39. However, a general technique to efficiently reveal that an extremely small detuning of a from the normal
overcome the phenomenon of oscillation quenching in coupled diffusion has a significant impact on the oscillatory activity of the
oscillatory systems remains as an open challenge as highlighted in system.
a recent review5. For zero-coupling strength (K ¼ 0), the jth Stuart–Landau
In this paper, we propose a rather simple and efficient oscillator exhibits a stable limit-cycle oscillation Zj ¼ eiwj t and has
approach for revoking oscillation suppression: a limiting factor in an unstable focus at Zj ¼ 0. The occurrence of AD in the coupled
the diffusive coupling effectively recovers rhythmic behaviour of system (Equation 1) for a ¼ 1 has been well studied in ref. 20 for
coupled networks, whose dynamic activity has been destroyed by t ¼ 0 and in ref. 22 for t40. In this respect, Aronson et al.20
the diffusive interaction of the elements. We demonstrate that asserted that AD occurs for sufficiently disparate frequencies,
this simple limiting factor can destabilize the stable HSS and IHSS whereas Reddy et al.22 showed that AD occurs even for coupled
to wipe out the onset of both AD and OD in diffusively coupled identical oscillators if t40. We next show that the coupling with

2 NATURE COMMUNICATIONS | 6:7709 | DOI: 10.1038/ncomms8709 | www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8709 ARTICLE

ao1 overrides the quenching effects by revoking the stability where W is the Lambert function defined as the inverse map of
of AD to regain oscillations in the respective AD parameter G(L) ¼ LeL (ref. 40). Moreover, the boundaries of stable HSS
space. (AD islands) on the (t, K) plane for w1 ¼ w2 ¼ w can be explicitly
When the coupled system (Equation 1) experiences AD, the derived as
HSS at the origin is stabilized. The condition for the onset of AD 
can be determined from a linear stability analysis around cos  1 aKK 1
t1 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
Z1 ¼ Z2 ¼ 0. Assuming that the linear perturbations vary as elt, w  K 2  ð1  aK Þ2
the characteristic equation can be written as  ð5Þ
p  cos  1 aKK 1
ð1 þ iw1  aK  lÞð1 þ iw2  aK  lÞ  K 2 e  2lt ¼ 0: ð2Þ t2 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi :
w þ K 2  ð1  aK Þ2
Thus, for t ¼ 0, the stability condition of AD is:
where g(a) ¼ (1 þ D2/4)/2 if a ¼ 1 and
1/aoKog(a),qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi AD islands in the (t, K) plane for a ¼ 1, 0.99, 0.97 and 0.9 are
gðaÞ ¼ ð  a þ 1 þ ð1  a2 ÞD2 =4Þ=ð1  a2 Þ if ao1, where D ¼ illustrated in Fig. 1c for w ¼ 10. Surprisingly, we observe that the
|w1  w2|. We find that the spread of the stable HSS (AD) in AD island shrinks greatly as a is slightly decreased, and
the (K, a) space narrows down on decreasing a for D42 (Fig. 1a completely disappears if aoaminE0.748. Now, the relation
for D ¼ 5) and eventually vanishes if aoamin. The value of amin between amin and w can be deduced as
can be analytically deduced as amin ¼ 2/D, and exactly agrees with 8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 9
<p K 2  ð1  aK Þ2 1=
the simulation results in Fig. 1a. The exact fit between the aK  1 : K4
w ¼ min ; ð6Þ
simulation amin (black squares) and the analytical amin (red line) : p  2 cos  1 K a;
is depicted in Fig. 1b as a function of a and D. Thus, it is clear that
the feedback factor a plays a crucial role in determining the which is depicted in Fig. 1d using the red line. This line
stability of HSS, which restores oscillations in stable HSS (AD) demarcates the unstable HSS on its left side for any combinations
regimes by revoking its stability for aoamin in coupled non- of (t, K) and stable HSS on its right side for a certain set of (t, K).
identical oscillators. A minimal amin necessary for revoking the stability of HSS for a
The characteristic equation (2) can be further simplified as given w on the (t, K) plane and vice versa can also be determined
l ¼ 1 þ iw  aK  Kelt ; ð3Þ from the above relation. Hence, it is strongly confirmed that the
feedback factor a successfully induces oscillations in stable HSS
for w1 ¼ w2 ¼ w, whose roots can be analytically resolved as (AD) regimes of delay-coupled identical oscillators for aoamin.
1   Atay25 reported that distributed delays favour a stabilization of
l ¼ W  tKe  ð1 þ iw  aK Þt þ 1 þ iw  aK; ð4Þ HSS in a larger parameter space compared with discrete delays. In
t what follows, we will explore the impact of the feedback factor a
on the two coupled Stuart–Landau oscillators with distributed
delays
a 1.0 b 1.0
  2 
Z_ j ðt Þ ¼ 1 þ iwj  Zj ðt Þ Zj ðt Þ
0.8 0.8 21 3
min
Z ð7Þ
0.6 0.6 =2/Δ þ K 4 f ðt ÞZk ðt  t Þdt  aZj ðt Þ5;
0 0 0



0
0.4 0.4
where f represents a distributed delay kernel. If f is the delta
0.2 0.2 function f(t0 ) ¼ d(t0  t), one can recover the coupled system
Δ=5 (Equation 1). Here we will concentrate on w1 ¼ w2 ¼ w, and a
0.0
0 1 2 3 4 5
0.0
2 6 10 14 18 22 26 30
uniformly distributed delay kernel f(t0 ) ¼ 1/(2b) if |t0  t|ob and
K Δ
zero elsewhere as in ref. 25. The kernel f approaches d(t0  t) as
b-0. For a ¼ 1, Atay25 showed that the AD region becomes
c 40 d 100 unbounded along t when b40.008 for w ¼ 30 as illustrated in
=1
Fig. 2a with b ¼ 0.02. To our surprise, on a very small decrease of
80
30 a from 1, we find that the spread of the AD regime shrinks
0.99
60
sharply as shown in Fig. 2b,c for a ¼ 0.99 and 0.98, respectively.
Moreover, the stable AD region breaks into three disjoint and


20
K

0.97
min bounded islands for a ¼ 0.97 (Fig. 2d). Decreasing a further wipes
40
0.9 off the AD region completely from the whole parameter space
10
20 (aoamin ¼ 0.424), thereby restoring stable oscillations by
=10 min switching the stability of stable HSS (AD). Hence, the feedback
0 0 factor a circumvents AD in coupled oscillators even for time
0.0 0.1 0.2 0.0 0.2 0.4 0.6 0.8 1.0
 (s) 
delays distributed over an interval, thereby corroborating the
robust nature of a in invoking oscillations by revoking the
Figure 1 | AD with frequency mismatch and discrete time delay. (a) The stability of AD.
stable AD interval versus a with frequency mismatch D ¼ |w1  w2| ¼ 5 and Conjugate and dynamic couplings are two forms of novel and
t ¼ 0. AD is impossible for any coupling strength K with aoamin ¼ 0.4. distinct interactions facilitating AD in coupled identical oscilla-
(b) The critical amin versus frequency mismatch D for t ¼ 0. (c) AD islands tors without any propagation delays27,28. Could the presence of
in the parameter (t, K) space for different values of a with w1 ¼ w2 ¼ w ¼ 10. the feedback factor a be capable of eradicating the onset of AD in
(d) The minimal value amin versus w. If aoamin AD is completely revoked the coupled systems with both coupling configurations? To clarify
for all t and K. it, we thus consider the following systems of Stuart–Landau

NATURE COMMUNICATIONS | 6:7709 | DOI: 10.1038/ncomms8709 | www.nature.com/naturecommunications 3

& 2015 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8709

a b a 1.0 b 1.0
1,000 1,000
0.8 0.8 min
100 100 0.6 0.6



K

0.4 0.4
10 10
0.2 0.2
=5
=1 =0.99 0.0 0.0
1 1 0 5 10 15 20 25 1 5 10 15 20 25 30
0.02 0.2 0.4 0.6 0.8 1.0 0.02 0.2 0.4 0.6 0.8 1.0 K ω
 (s)  (s)
c d c 1.0 d 1.0
1,000 1,000
0.8 0.8 min
100 100 0.6 0.6


K

0.4 0.4
10 10 0.2 0.2
=5
=0.98 =0.97 0.0 0.0
1 1 0 5 10 15 20 25 2 5 10 15 20 25 30
0.02 0.2 0.4 0.6 0.8 1.0 0.02 0.2 0.4 0.6 0.8 1.0
K 
 (s)  (s)
Figure 3 | AD with conjugate and dynamic couplings. (a,c) The stable AD
Figure 2 | AD with distributed time delays. (a–d) The stability region of
interval K versus a for conjugate and dynamic couplings, respectively. w ¼ 5
AD with time delays uniformly distributed over t±0.02 for a ¼ 1, 0.99,
is fixed. (b,d) The minimal amin versus w for conjugate and dynamic
0.98 and 0.97, respectively. The stable AD region breaks into three
couplings, respectively. The black squares denote numerical results, and the
disconnected and bounded islands for a ¼ 0.97. A feeble decrease in the
red line represents theoretical predictions. The scheme for restoration of
value of the limiting factor a from unity drastically shrinks the stable regions
rhythmicity is valid for coupled systems with distinctly different
of AD in the parameter space, which implies that dynamic activity is
interactions.
efficiently restored. AD is impossible for any values of t and K if
aoamin ¼ 0.424. The intrinsic frequencies are fixed as w1 ¼ w2 ¼ w ¼ 30.
agreement with the numerical results (black squares)
(Fig. 3b). For dynamic coupling
pffiffiffiffiffiffiffiffiffiffiffiffiffiEquation 9, the stable interval
pffiffiffiffiffiffiffiffiffiffiffiffiffi
oscillators with conjugate coupling of HSS (AD) is ðw2  w w2  4Þ=2oKoðw2 þ w w2  4Þ=2
 when a ¼ 1, which requires w42 for the coupled system
x_ j ¼ pj xj  wyj þ K yk  axj ;
 ð8Þ to experience AD. On decreasing a, the stable interval of
y_ j ¼ wxj þ pj yj þ K xk  ayj ; HSS monotonically reduces (Fig. ffi 3c), which ceases to exist if
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where pj ¼ 1  |Zj|2 ¼ 1  x2j  y2j , j, k ¼ 1,2 (jak), and with aoamin ¼ ½w2 ðw2 þ 9Þ  w
2
ðw2  3Þ3 =½2ðw2 þ 1Þ . This theoretical
dynamic coupling relation (red line) agrees very well with the simulation results
  2    (black squares) as depicted in Fig. 3d. Thus, the effect of the
Z_ j ¼ 1 þ iw  Zj  Zj þ K mj  aZj ; feedback factor a in revoking AD to restore the natural harmonics
ð9Þ
m_ j ¼  mj þ Zk : of the coupled systems can be generally applied to distinct
coupling configurations for which a stable AD is possible.
The underlying mechanisms of AD induced by conjugate and Next, we explore the capability of the feedback factor a in
dynamic couplings are completely different compared to that of effectively retrieving rhythmic oscillations in systems where OD
the previous cases. In conjugate coupling, the individual systems has been observed. Consider the following system of two Stuart–
are coupled via dissimilar variables27, whereas additional Landau oscillators via the x component12,
variables are involved to describe the dynamic coupling28. For 
both coupling scenarios, AD can be identified by performing a x_ j ¼ pj xj  wyj þ K xk  axj ;
ð12Þ
linear stability analysis around the origin of the coupled systems y_ j ¼ wxj þ pj yj ;
(Equations 8 and 9). The corresponding characteristic equations
can be obtained as where pj ¼ 1  |Zj|2 ¼ 1  x2j  y2j , j, k ¼ 1,2 (jak). Here the one-
dimensional diffusive coupling involves only the real parts. This
ð1  aK  lÞ2 þ w2  K 2 ¼ 0 ð10Þ
coupling form breaks the rotational symmetry of the coupled
and system, which is a necessary condition for the emergence
of OD. Besides the HSS at the origin Z1 ¼ Z2 ¼ 0 in the
l2  ðiw  aK Þl  1  iw  ð1  aÞK ¼ 0; ð11Þ system (Equation 12), the system also displays an IHSS
respectively. For conjugate coupling (Equation 8), AD occurs in (x*, y*,  x*,  y*) that appears at K p¼ (1 þ w2)/(1 þ a)ffi
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 
withffi x ¼ p ð1  pÞ=ðw2 þ p2 Þ
the interval
 1/aoKog(a), where ffi g(a) ¼ (1 þ w )/2 if a ¼ 1 and
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi via a pitchforkpbifurcation, ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
gðaÞ ¼  a þ 1 þ ð1  a2 Þw2 =ð1  a2 Þ if ao1. Stable HSS and 
y ¼  w ð1  pÞ=ðw þ p2 Þ with p ¼ ð1 þ aÞK=2 
2
(AD) is achieved only for w41, but monotonically decreases with qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a, which is wiped off for aoamin ¼ 1/w. Figure 3a depicts the ð1 þ aÞ2 K 2  4w2 =2. For the normal diffusive coupling, a ¼ 1,
spread of the AD interval in the (K, a) space for w ¼ 5, which it was reported that the origin is unstable for all K40, but the
strictly decreases with decreasing a and vanishes at amin ¼ 0.2. IHSS (OD) is stabilized for K4w2 þ 1/4 (refs 12,31); this is
The critical values of amin below which the coupled oscillators demonstrated by the bifurcation diagram plotted in Fig. 4a with
retain their oscillations for all K40 are shown in Fig. 3b as a w ¼ 10. The bold red lines denote the stable branches of the
function of w. The theoretical prediction (red line) is in good IHSS and the thin lines refer to the unstable ones. We further

4 NATURE COMMUNICATIONS | 6:7709 | DOI: 10.1038/ncomms8709 | www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8709 ARTICLE

a b c
0.04 0.04 0.04

0.02 0.02 0.02

X1,X2
0.00 0.00 0.00

–0.02 –0.02 –0.02

=1 =0.999 =0.998

–0.04 –0.04 –0.04
1 10 100 1,000 1 10 100 1,000 1 10 100 1,000

d e f
0.04 0.04 0.04

0.02 0.02 0.02

X1,X2

0.00 0.00 0.00

–0.02 –0.02 –0.02

=0.997 =0.996 =0.995

–0.04 –0.04 –0.04
1 10 100 1,000 1 10 100 1,000 1 10 100 1,000
K K K

Figure 4 | OD with one-dimensional diffusive coupling via the x component. (a-f) The bifurcation diagrams of the steady-state solutions for a ¼ 1, 0.999,
0.998, 0.997, 0.996 and 0.995, respectively. The bold red lines represent stable IHSS (OD), and the thin black lines denote unstable steady states. Even an
infinitesimal change of a from unity drastically shrinks the stable OD interval, which shows the high efficiency of the method in restoring rhythmic activity.
The frequency is used as w ¼ 10.

depict the bifurcation diagrams of the steady-state solutions for where j ¼ 1, 2,..., N and f is the uniformly distributed delayed
a ¼ 0.999, 0.998, 0.997, 0.996 and 0.995 in Fig. 4b–e, respectively. kernel as in equation 7. The parameter gjk determines the
Astonishingly, we observe that even a very tiny detuning in a topology of the network, that is, gjk ¼ gkj ¼ 1 if jth P
and kth nodes
from 1 shrinks the coupling intervals of the stable IHSS (OD) are linked otherwise gjk ¼ gkj ¼ 0, gjj ¼ 0; and dj ¼ Nk¼1 gjk is the
(Fig. 4b–e). Moreover, no stable IHSS can be found for any jth node degree. The coupled system (Equation 13) (a ¼ 1) was
coupling strength when aoaminE0.995 (Fig. 4f), which attributes studied by Atay for all-to-all networks25, that is, gjk ¼ 1 for all
that OD is successfully revoked and thus oscillations are jak. Performing a linear stability analysis around the origin, the
provoked in the OD regimes. The effect of the limiting factor a characteristic equations determining the stability of the HSS in
in restoration of rhythmic activity in this case is even more coupled system (Equation 13) with wj ¼ w are
pronounced, since infinitesimal changes of a from unity lead to a
sharp shrinkage of the stable IHSS regime, demonstrating the sinhðlbÞ
l ¼ 1 þ iw  aK þ Krj e  lt ; ð14Þ
high efficiency of the proposed method in restoring oscillatory lb
activity. g
By studying diverse systems of two coupled Stuart–Landau where rj’s are the eigenvalues of G ¼ ð djkj ÞNN and can be
oscillators, it has been demonstrated that incorporating the ordered as 1:0 ¼r1  r2  :::   N 1 1  rN   1:0 (ref. 41).
limiting feedback factor a in the normal diffusive coupling is a The network (Equation 13) experiences AD when the largest real
rather simple and high-efficient approach in revoking not only part of roots of equation (14) is negative for all rj. In fact, only
AD but also OD under various distinct death scenarios. Even a two bounding eigenvalues r1 ¼ 1 and  1/(N  1)ZrNZ  1
very minute deviation of a from unity drastically shrinks both AD contribute to the stability regions of AD. The degree of the spread
and OD regions in the parameter space. The new coupler with a of the stable HSS is determined by rN; the spread of AD is larger
strengthens the tolerance to different dynamic deteriorations of for larger values of rN. The stable HSS regime (AD) experienced
coupled non-linear oscillators. by the coupled system (Equation 13) for a ¼ 1 as rN-0 is plotted
in Fig. 5a for uniformly distributed delays with b ¼ 0.07 and
w ¼ 10. The AD regime is connected and unbounded along t. We
Networked coupled Stuart–Landau oscillators. The role of the
are indeed surprised to observe that the AD regime becomes
diffusive feedback factor a in revoking AD persists for an arbi-
disconnected even for a negligible decrease in a; see Fig. 5b
trary network of coupled Stuart–Landau oscillators. To illustrate
for a ¼ 0.999. Decreasing a further, the number of AD
it, let us consider a connected network of N Stuart–Landau
islands decreases as shown in Fig. 5c,d for a ¼ 0.998 and 0.99,
oscillators with an arbitrary topology
respectively. The AD region completely disappears in the
  2  parameter space for aoaminE0.189, thereby revoking the
Z_ j ¼ 1 þ iwj  Zj  Zj stability of HSS (AD) in recovering oscillations in an arbitrary
21 3 network with distributed time delays.
Z ð13Þ
K X N
þ gjk 4 f ðt0 ÞZk ðt  t0 Þdt0  aZj 5;
dj Experiments with chemical oscillations. The efficiency of the
k¼1 0 new coupling approach in revoking deaths to restore oscillations

NATURE COMMUNICATIONS | 6:7709 | DOI: 10.1038/ncomms8709 | www.nature.com/naturecommunications 5

& 2015 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8709

a b a d
50 50 =1
1.2
40 40  = 0.8
V1 1.1

– K (V mA–1)
30 30
K

RE WE1 WE2 CE 1.0

20 20

10 10 V2 0.9
=1 =0.999
0 0 0.8
WE1 WE2
0.07 1 2 3 4 5 0.07 1 2 3 4 5
 (s)  (s) 0.9 1.0 1.1 1.2 1.3
c d  (s)
50 50

40 40 b 150 Control off Control on ( = 1) Control on (= 0)

30 30

Current (μΑ)
K

20 20
100
10 10
=0.998 =0.99
0 0
0.07 1 2 3 4 5 0.07 1 2 3 4 5 50
 (s)  (s)
0 50 100 150
Figure 5 | AD in an arbitrary network with distributed time delays. Time (s)
(a–d) The upper limiting stability region of AD as rN-0 with time delays
uniformly distributed over t±0.07 for different values of a. AD regime c
150 =0  = 0.1  = 0.2  = 0.3  = 0.4 =0
becomes disconnected even for a tiny decrease of a from 1 to 0.999. Only
three AD islands survive for a ¼ 0.998. Decreasing a further, the number of
Current (μΑ)

AD islands decreases. The AD region completely disappears in the 100

parameter space if aoamin ¼ 0.189. The intrinsic frequency is fixed as
w ¼ 10.
50

0 100 200 300 400 500 600 700

has been experimentally verified in an oscillatory chemical
Time (s)
reaction system with electrochemical dissolution of nickel in
sulfuric acidic media. Figure 6a depicts a schematic view of the Figure 6 | Experiments with coupled electrochemical reactions.
experimental setup. The reactions take place on the surface of two (a) Schematic of the experimental setup containing the reference (RE),
nickel wires with diameter of 0.69 mm, spaced by 2.0 mm and counter (CE) and two Ni working electrodes (WE) in an electrochemical
embedded in epoxy resin. When an individual resistance of cell with 50 ml 3 mol L  1 sulfuric acid at a temperature of 10 oC.
2.50 kO are attached to each wire, and the potential V0 of each (b) Establishing AD with a ¼ 1 and regaining oscillations with a ¼ 0.
wire are set with a bipotentiostat (Bank Instruments) to 1.150 V Without control, the oscillations phase drift (to40.13 s, K ¼ 0). With direct
versus Hg/Hg2SO4/sat. K2SO4 reference electrode, oscillatory feedback control AD is observed (40.13 soto127.1 s, K ¼  1.0 V mA  1,
current can be recorded with the built-in ammeters of the t ¼ 1.09 s and a ¼ 1). With a ¼ 0, the oscillations are anti-phase
bipotentiostat between the wires and the counter electrode synchronized. (t4127.1 s, K ¼  1.0 VmA  1, t ¼ 1.09 s and a ¼ 0).
(1.6 mm diameter Pt coated Ti wire). This potential value is just (c) Current time series with several values of a ¼ 0, 0.1, 0.2, 0.3 and 0.4.
above (by 25 mV) a Hopf bifurcation. The current oscillations of (d) AD domains for a ¼ 1 and 0.8 in the parameter space for K versus t.
the two electrodes (I1(t) and I2(t)), shown in Fig. 6b, occur due to The set of discrete data points are approximated by a cubic spline
the non-linear kinetics of nickel dissolution in the presence of interpolation, which indicates the border between AD and oscillatory states.
nickel-oxide films that provide hidden negative differential
resistance characteristics to the electrochemical process42,43.
The interactions between the chemical reactions on the two a ¼ 1, the amplitude quickly decreases, and the system establishes
metal surfaces are introduced by polarizing the two electrodes a stable HSS: the delayed interaction thus causes the phenomenon
(indexed as j ¼ 1 and 2) according to the following equations: of AD. The oscillations are reborn in an anti-phase configuration
  when a is adjusted to 0 for t4127.1 s. Corroborating with this
Vj ðt Þ ¼ V0 þ K oI ðt  tÞ4  aIj ðt Þ ð15Þ
scenario, the gradual increase of a, shown in Fig. 6c, clearly

where Ij ðt Þ are the offset corrected currents of the electrodes indicates a promoting effect on AD. For ao0.4, the oscillatory
Ij ðt Þ ¼ Ij ðt Þ  o , o is the average current overtime determined behaviour can be reinstated. Figure 6d depicts the AD domains
before the experiments, and oI ðt Þ4 ¼ ðI1 ðt Þ þ I2 ðt ÞÞ=2 is the for two values of a ¼ 1 and 0.8 in K versus t diagram. As
mean current of oscillations at time t. The potential perturba- predicted by our theory, a small decrease of a markedly
tions, with values of a and K previously chosen with a Labview suppresses the AD regions.
Real Time software, are applied at a rate of 1,000 Hz, which is The experimental realization of our method with chemical
much faster than the typical timescale of the oscillations (2.5 s). oscillations was successful, despite the fact that while the Stuart–
Without control, the chemical reactions on the two electrodes Landau system is a simple, weakly non-linear, two variable model,
take place in an oscillatory manner with slightly different periods the chemical system represents an ensemble of complex physical
(2.504s and 2.524 s), as shown in Fig. 6b. Because of the and chemical processes with many variables and uncertainties.
lack of coupling, the oscillations exhibit phase drift behaviour The oscillations in the chemical reactions generally originate
(Supplementary Fig. 1)43. When the control is turned on with from a Hopf bifurcation, which can be well represented by the

6 NATURE COMMUNICATIONS | 6:7709 | DOI: 10.1038/ncomms8709 | www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8709 ARTICLE

Stuart–Landau equations. Therefore, it is expected that the maintaining the oscillations. Assuming that the intrinsic
phenomenon of retrieving oscillations with our scheme will work dynamics of a cell is oscillatory, this means that it is important
for a large number of other chemical and biological systems, to develop isolated reference states inside the cells where the
whose oscillatory behaviour arises as a result of a Hopf periodic variations of the cell life cycle are less affected. In neural
bifurcation44. networks, two major types of coupling forms are: gap junctions
and chemical synapses50. Gap junctions are based on the
potential difference between the cells (a ¼ 1), while chemical
Discussion synapses are through substances generated by action potential
We have proposed a rather simple and efficient coupling scheme and thus do not directly depend on the potential difference
to revoke both AD and OD to retrieve rhythmic behaviours in (a ¼ 0). Our prediction indicates that chemical synapses play a
diffusively coupled dynamical networks of non-linear oscillators. crucial role in maintaining the oscillations, and by changing the
By introducing a limiting feedback factor a in the diffusive ratio of gap junction and electrical coupling could be a versatile
coupling, we have revealed that this approach is generically robust way to tune the extent of oscillations in the networks. Our study
in revoking AD in coupled Stuart–Landau oscillators under also has important implications for the design of coupling
distinct scenarios such as frequency mismatch, delayed coupling schemes for synchronization engineering, that is, for tuning the
with both discrete and distributed time delays, and conjugate and phase relationship between oscillatory units. Feedback-based
dynamic couplings. We have shown that the OD phenomenon algorithms can be applied directly or through difference
can also be revoked in coupled Stuart–Landau oscillators. schemes51. In general, to avoid AD, direct feedback methods
Intriguingly, a minimal decrease of a from unity drastically could provide wider parameter ranges at strong feedback gains
reduces the size of the stable regions of both AD and OD in the where synchronization structures are to be attained without
parameter space. The effect of a minute deviation of a from unity amplitude suppression52.
manifests in switching the stability of the stable HSS and IHSS. A population of coupled non-linear oscillators are deemed to
Furthermore, we have experimentally confirmed the efficient role serve as an efficient tool to enhance our understanding of
of the diffusive control factor a in restoring oscillations by collective dynamics spontaneously emerging in real-life systems.
destabilizing AD in a delay-coupled chemical reaction system The coupling in an ensemble of non-linear oscillators could be
with Ni electrodissolution. designed in different ways. In particular, diffusive coupling is a
The surprising effect of a minute deviation of the parameter a natural and fundamental type of interaction, which has attracted
from unity could be interpreted in the following intuitive way. In great interests in different context in a wide range of disciplines
the death scenarios, the normal diffusive coupling (a ¼ 1) brings a including physics, chemistry and biology for many decades.
strong additional dissipation into the coupled oscillatory systems Communication between cells and most of biological process
to suppress the oscillations. In our proposed diffusive coupling, occur through diffusion of ions. In chemical reactions, mixing of
the value of a describes the relation between the incoming different reactants generally involves diffusive interactions.
and the outgoing flows. Specifically, the level of incoming flow is Therefore, in this work, we have considered the diffusive coupling
measured by K, whereas the outgoing one by Ka. Clearly, the rate as a starting point of our study in revealing restoration of
of the incoming flow is greater than that of the outgoing one if oscillations of coupled networks. By introducing a new parameter
0oao1, which leads to asymmetric flows in the whole-coupled modifying the nature of the normal diffusive coupling, we have
systems once a deviates from unity. The surplus incoming flow elaborated that even a very tiny deviation from the traditional
supplies an internal energy source to compensate the dissipation diffusive coupling causes a drastic influence on the dynamic
of coupled systems, which serves as a driving force to induce activity of coupled dynamical systems. Our results illuminate the
oscillatory activity. role of diffusive coupling in the generation of sustained rhythmic
It is to be noted that we have corroborated our results by activity in real-world systems. It was reported that the
employing the paradigmatic Stuart–Landau oscillator, which phenomena of AD and OD are also possible for coupled non-
represents a normal form describing dynamics near a supercritical linear systems with non-diffusive couplings53–57. How to revoke
Hopf bifurcation. Our findings are expected to hold true for a AD and OD in non-diffusively coupled systems constitutes our
broad class of coupled non-linear systems near a Hopf bifurcation. next task to be studied.
In fact, we have confirmed that dynamic activity can be recovered Finally, our proposed coupling scheme can be easily realized
effectively in many other diffusively coupled non-linear oscillators not only in different model studies but also for diverse
experiencing AD or OD, such as Brusselators (Supplementary experimental setups. We firmly speculate that our approach is
Note 1), chaotic Lorenz oscillators (Supplementary Note 2), feasible in other distinct experiments, where AD or OD has been
Pikovsky-Rabinovich circuit models (Supplementary Note 3), reported58–61. The non-trivial influence of the new diffusive
synthetic genetic relaxation oscillators and membrane models coupling scheme on the qualitative properties of dynamical
(Supplementary Note 4). The generality of our method has systems indicates a new path for future studies of other collective
also been successfully validated in diffusively coupled dynamical behaviours in diffusively coupled non-linear oscillators such as
networks experiencing distinctly different deteriorations of chimera states62,63, explosive synchronization64,65 and glass
dynamic activity such as partial deaths45 (Supplementary states66–68. The framework of our study sheds significantly new
Note 5) and aging transition46,47 (Supplementary Note 6). Thus, insights on the diffusive coupling in manipulating oscillatory
the new ingredient a in the coupling serves as a very general dynamics of coupled complex non-linear systems, which will have
framework to strengthen the robustness of dynamic activity in a strong impact and invoke wide interests in the field of complex
diffusively coupled networks. systems science as well as in various applications from biology via
Our proposed coupling scheme has important consequences in engineering to social sciences.
particular in biology. Recent studies of synthetic biology have
made significant progress in the construction of populations of
intracellular oscillators that mimic naturally occurring clocks as a Methods
Numerical simulations. All numerical integrations of coupled dynamical equa-
means to gain insight about the functionality of biological tions are performed by employing a standard explicit fourth-order Runge–Kutta
networks48,49. In this regard, it would be beneficial to develop method with integration steps h ¼ 0.01 and 0.001, respectively. There are no
sensing compartments, which are non-difference (a ¼ 0) for qualitative differences between two different steps. Random initial conditions can

NATURE COMMUNICATIONS | 6:7709 | DOI: 10.1038/ncomms8709 | www.nature.com/naturecommunications 7

& 2015 Macmillan Publishers Limited. All rights reserved.
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8709

be adopted for AD in coupled Stuart–Landau oscillators, which means that 17. Tchumatchenko, T. & Clopath, C. Oscillations emerging from noise-driven
AD is not only linearly but also globally stable in the phase space. In contrast, as steady state in networks with electrical synapses and subthreshold resonance.
OD always coexists with stable limit-cycle motions, initial conditions near the IHSS Nat. Commun. 5, 5512 (2014).
are used for OD, or many different random initial conditions are tried until 18. Koukkari, W. L. & Sothern, R. B. Introducing Biological Rhythms (Springer,
reaching OD. 2006).
19. Lisman, J. & Buzsaki, G. A neural coding scheme formed by the combined
Steady-state solutions for AD and OD. The onset of AD in coupled Stuart– function of gamma and theta oscillations. Schizophr. Bull. 34, 974–980 (2008).
Landau oscillators implies that the unstable origin is stabilized by the coupling. 20. Aronson, D. G., Ermentrout, G. B. & Kopell, N. Amplitude response of coupled
Thus the steady-state solution for AD in coupled Stuart–Landau oscillators oscillators. Physica. D 41, 403–449 (1990).
is HSS at Zj ¼ 0. In contrast, OD manifests due to the appearance of stable IHSS. 21. Mirollo, R. E. & Strogatz, S. H. Amplitude death in an array of limit-cycle
In general, for OD in coupled non-linear oscillators, the Newton–Raphson oscillators. J. Stat. Phys. 60, 245–262 (1990).
algorithm69 is used to locate the IHSS solution. In coupled Stuart–Landau 22. Reddy, D. V. R., Sen, A. & Johnston, G. L. Time delay induced death in coupled
oscillators in equation (12), the IHSS solution for OD is analytically derived, which limit cycle oscillators. Phys. Rev. Lett. 80, 5109–5112 (1998).
is also confirmed by the Newton–Raphson algorithm. 23. Reddy, D. V. R., Sen, A. & Johnston, G. L. Time delay effects on coupled limit
cycle oscillators at Hopf bifurcation. Physica. D 129, 13–34 (1999).
24. Strogatz, S. H. Death by delay. Nature 394, 316–317 (1998).
Linear stability analysis. The occurrences of AD and OD are verified from a
25. Atay, F. M. Distributed delays facilitate amplitude death of coupled oscillators.
standard linear stability analysis after obtaining steady-state solutions of AD and
OD in coupled Stuart–Landau oscillators. AD (OD) is stable, if and only if all the Phys. Rev. Lett. 91, 094101 (2003).
real parts of resulted characteristic eigenvalue equations are negative. Once a time 26. Zou, W. & Zhan, M. Partial time-delay coupling enlarges death island of
delay is involved, the corresponding characteristic eigenvalue equations are coupled oscillators. Phys. Rev. E 80, 065204 (2009).
transcendental, which have infinitely many (both complex and real) roots with 27. Karnatak, R., Ramaswamy, R. & Prasad, A. Amplitude death in the absence
negative real parts, but only a finite number of roots with positive real parts of time delays in identical coupled oscillators. Phys. Rev. E 76, 035201(R)
attributed to instability. Generally, the characteristic roots of transcendental (2007).
characteristic equations cannot be solved analytically. We numerically compute the 28. Konishi, K. Amplitude death induced by dynamic coupling. Phys. Rev. E 68,
largest real part of characteristic roots with the pseudospectral differentiation 067202 (2003).
techniques70. For one special case, the characteristic roots can be expressed 29. Prigogine, I. & Lefever, R. Symmetry breaking instabilities in dissipative
explicitly using the Lambert function W as indicated in equation (4). systems. J. Chem. Phys. 48, 1695–1700 (1968).
30. Bar-Eli, K. On the stability of coupled chemical oscillators. Physica. D 14,
242–252 (1985).
AD boundaries. The AD boundaries in equation (5) are analytically derived. The 31. Zakharova, A. et al. Time delay control of symmetry-breaking primary and
analysis is based on the fact that as tuning t or K, the stability of HSS in coupled
secondary oscillation death. Europhys. Lett. 104, 50004 (2013).
Stuart–Landau oscillators may be switched only if a characteristic eigenvalue root l
32. Zakharova, A., Kapeller, M. & Schöll, E. Chimera death: symmetry breaking in
in equation (3) crosses the imaginary axis. For the critical situation, we have l ¼ iO.
dynamical networks. Phys. Rev. Lett. 112, 154101 (2014).
Inserting into equation (3) and separating into real and imaginary part, the AD
boundary curves can be obtained after some straightforward algebraic 33. Zou, W., Senthilkumar, D. V., Koseska, A. & Kurths, J. Generalizing the
manipulations. transition from amplitude to oscillation death in coupled oscillators. Phys. Rev.
E 88, 050901 (2013).
34. Zou, W., Senthilkumar, D. V., Duan, J. & Kurths, J. Emergence of amplitude
The critical value of amin. The values of amin theoretically predicted in and oscillation death in identical coupled oscillators. Phys. Rev. E 90, 032906
equation (6) are from the intersection condition of AD boundaries in equation (5). (2014).
Generally, the threshold value of amin in all cases is calculated by meticulously 35. Konishi, K. Limitation of time-delay induced amplitude death. Phys. Lett. A
decreasing a from 1 until the stable regime of HSS (AD) or IHSS (OD) is extinct in 341, 401–409 (2005).
the whole parameter space. 36. Morino, K., Tanaka, G. & Aihara, K. Efficient recovery of dynamic behavior in
coupled oscillator networks. Phys. Rev. E 88, 032909 (2013).
References 37. Majdandzic, A. et al. Spontaneous recovery in dynamical networks. Nat. Phys.
1. Kuramoto, Y. Chemical Oscillations. Waves, and Turbulence (Springer, 1984). 10, 34–38 (2013).
2. Pikovsky, A., Rosenblum, M. & Kurths, J. Synchronization: A Universal Concept 38. Zou, W., Yao, C. & Zhan, M. Eliminating delay-induced oscillation death by
in Nonlinear Sciences (Cambridge Univ. Press, 2001). gradient coupling. Phys. Rev. E 82, 056203 (2010).
3. Strogatz, S. H. From Kuramoto to Crawford: exploring the onset of 39. Zou, W., Senthilkumar, D. V., Zhan, M. & Kurths, J. Reviving oscillations in
synchronization in populations of coupled oscillators. Physica. D 143, 1–20 coupled nonlinear oscillators. Phys. Rev. Lett. 111, 014101 (2013).
(2000). 40. Hale, J. K. Functional Differential Equations (Springer, 1971).
4. Acebrón, J. A., Bonilla, L. L., Vicente, C. J. P., Ritort, F. & Spigler, R. The 41. Atay, F. M. Oscillator death in coupled functional differential equations near
Kuramoto model: a simple paradigm for synchronization phenomena. Rev. Hopf bifurcation. J. Diff. Eqn. 221, 190–209 (2006).
Mod. Phys. 77, 137 (2005). 42. Koper, M. Nonlinear phenomena in electrochemical systems. J. Chem. Soc.
5. Saxena, G., Prasad, A. & Ramaswamy, R. Amplitude death: the emergence of Faraday Trans. 94, 1369 (1998).
stationarity in coupled nonlinear systems. Phys. Rep. 521, 205–228 (2012). 43. Kiss, I. Z., Zhai, Y. & Hudson, J. L. Emerging coherence in a population of
6. Koseska, A., Volkov, E. & Kurths, J. Oscillation quenching mechanisms: chemical oscillators. Science 296, 1676–1678 (2002).
amplitude vs oscillation death. Phys. Rep. 531, 173–199 (2013). 44. Epstein, I. R. & Pojman, J. A. Introduction to nonlinear chemical dynamics.
7. Bar-Eli, K. & Reuveni, S. Stable stationary states of coupled chemical oscillators. Oscillations, Waves, Patterns and Chaos (Oxford Univ. Press, 1998).
experimental evidence. J. Phys. Chem. 89, 1329–1330 (1985). 45. Atay, F. M. Total and partial amplitude death in networks of diffusively coupled
8. Crowley, M. F. & Epstein, I. R. Experimental and theoretical studies of a oscillators. Physica. D 183, 1–18 (2003).
coupled chemical oscillator: phase death, multlstability, and in-phase and out- 46. Daido, H. & Nakanishi, H. Aging transition and universal scaling in oscillator
of-phase entrainment. J. Phys. Chem. 93, 2496–2502 (1989). networks. Phys. Rev. Lett. 93, 104101 (2004).
9. Kuntsevich, B. F. & Pisarchik, A. N. Synchronization effects in a dual- 47. Tanaka, G., Morino, K., Daido, H. & Aihara, K. Dynamical robustness of
wavelength class-B laser with modulated losses. Phys. Rev. E 64, 046221 (2001). coupled heterogeneous oscillators. Phys. Rev. E 89, 052906 (2014).
10. Reddy, D. V. R., Sen, A. & Johnston, G. L. Experimental evidence of time-delay- 48. Danino, T., Mondragón-Palomino, O., Tsimring, L. & Hasty, J. A synchronized
induced death in coupled limit-cycle oscillators. Phys. Rev. Lett. 85, 3381–3384 quorum of genetic clocks. Nature 463, 326–330 (2010).
(2000). 49. Mondragón-Palomino, O., Danino, T., Selimkhanov, J., Tsimring, L. & Hasty, J.
11. Ermentrout, G. B. & Kopell, N. Oscillator death in systems of coupled neural Entrainment of a population of synthetic genetic oscillators. Science 333,
oscillators. SIAM J. Appl. Math. 50, 125–146 (1990). 1315–1319 (2011).
12. Koseska, A., Volkov, E. & Kurths, J. Transition from amplitude to oscillation 50. Kopell, N. & Ermentrout, G. B. Chemical and electrical synapses perform
death via Turing bifurcation. Phys. Rev. Lett. 111, 024103 (2013). complementary roles in the synchronization of interneuronal networks. Proc.
13. Döfler, F., Chertkov, M. & Bullo, F. Synchronization in complex oscillator Natl Acad. Sci. USA 101, 15482–15487 (2004).
networks and smart grids. Proc. Natl Acad. Sci. USA 110, 2005–2010 (2013). 51. Rosenblum, M. & Pikovsky, A. Delayed feedback control of collective
14. Motter, A. E., Myers, S. A., Anghel, M. & Nishikawa, T. Spontaneous synchrony synchrony: An approach to suppression of pathological brain rhythms. Phys.
in power-grid networks. Nat. Phys. 9, 191–197 (2013). Rev. E 70, 041904 (2004).
15. Menck, P. J., Heitzig, J., Kurths, J. & Schellnhuber, H. J. How dead ends 52. Kiss, I. Z., Rusin, C. G., Kori, H. & Hudson, J. L. Engineering complex
undermine power grid stability. Nat. Commun. 5, 3969 (2014). dynamical structures: Sequential patterns and desynchronization. Science 316,
16. Jenkins, A. Self-oscillation. Phys. Rep. 525, 167–222 (2013). 1886–1889 (2007).

8 NATURE COMMUNICATIONS | 6:7709 | DOI: 10.1038/ncomms8709 | www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8709 ARTICLE

53. Pisarchik, A. N. Oscillation death in coupled nonautonomous systems with 69. Parker, T. S. & Chua, L. O. Practical Numerical Algorithms for Chaotic Systems
parametrical modulation. Phys. Lett. A 318, 65–70 (2003). (Springer, 1989).
54. Wei, M. D. & Lun, J. C. Amplitude death in coupled chaotic solid-state lasers 70. Breda, D. Solution operator approximation for characteristic roots of delay
with cavity-configuration-dependent instabilities. Appl. Phys. Lett. 91, 061121 differential equations. Appl. Numer. Math. 56, 305–317 (2006).
(2007).
55. Bora, M. P. & Sarmah, D. Oscillation death in a coupled van der Pol-Mathieu
system. Pramana 81, 677–690 (2013).
Acknowledgements
We acknowledge financial support from the National Natural Science Foundation of
56. Palazzi, M. J. & Cosenza, M. G. Amplitude death in coupled robust-chaos
China (No. 11202082, No. 61203235, No. 11371367 and No. 11271290), the Funda-
oscillators. Eur. Phys. J. Special Topics 223, 2831–2836 (2014).
mental Research Funds for the Central Universities of China under Grant No.
57. Chaudhuri, U. & Prasad, A. Complicated basins and the phenomenon of
2014QT005, IRTG1740(DFG-FAPESP), and SERB-DST Fast Track scheme for young
amplitude death in coupled hidden attractors. Phys. Lett. A 378, 713–718
scientist under Grant No. ST/FTP/PS-119/2013, NSF CHE-0955555 and Grant No.
(2014).
229171/2013-3 (CNPq).
58. Herrero, R. et al. Experimental observation of the amplitude death effect in two
coupled nonlinear oscillators. Phys. Rev. Lett. 84, 5312–5315 (2000).
59. Neufeld, Z., Kiss, I. Z., Zhou, C. S. & Kurths, J. Synchronization and oscillator Author contributions
death in oscillatory media with stirring. Phys. Rev. Lett. 91, 084101 (2003). W.Z. conducted theoretical and numerical studies. W.Z. and D.V.S. drafted the manu-
60. Ozden, I. et al. Strong coupling of nonlinear electronic and biological script. R.N. and I.Z.K. designed and performed the experiment. Y.T. and A.K. analysed
oscillators: Reaching the ‘amplitude death’ regime. Phys. Rev. Lett. 93, 158102 the models and data. J.D. and J.K. directed and supervised the study. All the authors
(2004). discussed the results, drew conclusions and edited the manuscript.
61. Banerjee, T. & Ghosh, D. Experimental observation of a transition from
amplitude to oscillation death in coupled oscillators. Phys. Rev. E 89, 062902
(2014). Additional information
62. Hagerstrom, A. M. et al. Experimental observation of chimeras in coupled-map Supplementary Information accompanies this paper at http://www.nature.com/
lattices. Nat. Phys. 8, 658–661 (2012). naturecommunications
63. Tinsley, M. R., Nkomo, S. & Showalter, K. Chimera and phase-cluster states in Competing financial interests: The authors declare no competing financial interests.
populations of coupled chemical oscillators. Nat. Phys. 8, 662–665 (2012).
64. Gómez-Gardeñes, J., Gómez, S., Arenas, A. & Moreno, Y. Explosive Reprints and permission information is available online at http://npg.nature.com/
synchronization transitions in scale-free networks. Phys. Rev. Lett. 106, 128701 reprintsandpermissions/
(2011).
65. Leyva, L. et al. Explosive first-order transition to synchrony in networked How to cite this article: Zou, W. et al. Restoration of rhythmicity in diffusively coupled
chaotic oscillators. Phys. Rev. Lett. 108, 168702 (2012). dynamical networks. Nat. Commun. 6:7709 doi: 10.1038/ncomms8709 (2015).
66. Daido, H. Population dynamics of randomly interacting self-oscillators. I
Tractable models without frustration. Prog. Theor. Phys. 77, 622–634 (1987). This work is licensed under a Creative Commons Attribution 4.0
67. Daido, H. Quasientrainment and slow relaxation in a population of oscillators International License. The images or other third party material in this
with random and frustrated interactions. Phys. Rev. Lett. 68, 1073–1076 (1992). article are included in the article’s Creative Commons license, unless indicated otherwise
68. Iatsenko, D., McClintock, P. V. E. & Stefanovska, A. Q. Glassy states and super- in the credit line; if the material is not included under the Creative Commons license,
relaxation in populations of coupled phase oscillators. Nat. Commun. 5, 4118 users will need to obtain permission from the license holder to reproduce the material.
(2014). To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

NATURE COMMUNICATIONS | 6:7709 | DOI: 10.1038/ncomms8709 | www.nature.com/naturecommunications 9

& 2015 Macmillan Publishers Limited. All rights reserved.