Вы находитесь на странице: 1из 131

UNIVERSITY OF CALGARY

Chimera States in Nonlocally Coupled Oscillatory

and Complex Oscillatory Systems

by

Chad Gu

A THESIS

SUBMITTED TO THE FACULTY OF GRADUATE STUDIES

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE

DEGREE OF MASTER OF SCIENCE

DEPARTMENT OF PHYSICS AND ASTRONOMY

CALGARY, ALBERTA

December, 2012

c Chad Gu 2012

Abstract

Populations of nonlocally coupled, identical limit cycle oscillators can lead to the peculiar

chimera state, in which subpopulations of synchronized and unsynchronized oscillators coex-

ist in localized spatial domains. We study the existence and robustness of chimera states in

populations of simple, complex and chaotic oscillators, which are different forms of oscilla-

tion realizable in experiments. In the case of simple oscillation, we investigate the existence

and robustness of chimera states under physically relevant constraints in our model such as

time delay and frequency heterogeneity. We show for the first time that chimera states can

be extended to the general setting of complex and chaotic oscillators, in particular, we find

spiral wave chimera together with synchronization defect lines. We also introduce several

methods for selecting parameters in general models that lead to the formation of chimera

states in 1D or 2D systems, which can provide important theoretical guidance to future

experimental studies.

ii
Acknowledgements

The completion of this thesis would not have been possible if not for the support of my

supervisor, Joern Davidsen, and my valuable colleague in the Complexity Science Group,

Ghislain St.-Yves. I thank Joern for his mentorship, in particular his persistent workmanship

has influenced me to strive to do the same, and be critical to my own understanding of a

topic, or my own work. Being able to collaborate with Ghislain was a pleasure, and he

was always available whenever I needed to understand a concept or get feedback on new

ideas. Being able to formulate my own ideas and be part of the creative process of solving

a puzzle has been something that I looked forward to in the academic life, and I am glad I

was able to to find this at the Complexity Science Group. I also thank some of the past and

present members of the Complexity Science Group, in no particular order: Orion Penner,

Elliot Martin, Hon Wai Lau, Seung-Woo Soo, Arsalan Sattari, Javad Moradpour, Aicko Yves

Schumann, Golnoosh Bizhani, Claire Christensen, Fernando Girotto and Vishal Sood, many

of whom have given me advice on my career as a graduate student, motivated me to work

harder, or just shared a few laughters during breaks.

I also thank the new friends I have made during my stay in Calgary, many of whom have

exposed me to the wonders of the Canadian Rockies, which is a gift I will take with me

forever. Most importantly, I thank my friends and family back in Toronto for their support,

in particular my parents, who have always worked tirelessly for me and for the family. Their

achievement will always be greater than what I can only hope to accomplish in my own

lifetime.

iii
Table of Contents

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Synchronization in Globally Coupled Systems . . . . . . . . . . . . . . 5
1.2 Synchronization in Locally Coupled Systems . . . . . . . . . . . . . . . 6
1.3 Synchrony Breaking in Population of Identical Oscillators . . . . . . . . 9
1.3.1 Nonlocal Coupling . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Coexistence of Synchrony and Asynchrony in the Brain . . . . . . . . . 11
1.4.1 Unihemispheric Sleep . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.2 Temporal Correlation Hypothesis . . . . . . . . . . . . . . . . . 12
1.5 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2 Mathematical Background and Literature Review . . . . . . . . . . . . . . . 18
2.1 Definitions and Basic Terminologies . . . . . . . . . . . . . . . . . . . . 18
2.1.1 Notions of Synchronization . . . . . . . . . . . . . . . . . . . . 18
2.1.2 Limit Cycle Oscillations . . . . . . . . . . . . . . . . . . . . . . 19
2.1.3 Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.4 Complex Ginzburg-Landau Equation . . . . . . . . . . . . . . . 22
2.2 Chimera States in 1D and 2D . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.1 Derivation of Nonlocal Coupling . . . . . . . . . . . . . . . . . . 30
2.3 Phase Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3.1 Kuramoto Model . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3.2 Characterization of Synchronization in the Kuramoto Model . . 37
2.4 The Phase Equation for Nonlocally Coupled Systems . . . . . . . . . . 39
2.4.1 Derivation of the Phase Equation . . . . . . . . . . . . . . . . . 39
2.4.2 Analytical Treatment of the Phase Equation . . . . . . . . . . . 42
2.4.3 Simulation of the Phase Equation . . . . . . . . . . . . . . . . . 44
2.5 Results on the 1D Phase Equation . . . . . . . . . . . . . . . . . . . . . 45
2.5.1 Chimera State Solutions in |α| < π . . . . . . . . . . . . . . . . 46
2.5.2 Bistability of Solutions . . . . . . . . . . . . . . . . . . . . . . . 47
2.5.3 Robustness With Respect to Frequency Heterogeneity . . . . . 49
2.5.4 Chimera States in Systems with Time Delay . . . . . . . . . . . 50
2.5.5 Other Coupling Topologies . . . . . . . . . . . . . . . . . . . . . 51
2.6 Complex Oscillatory Systems . . . . . . . . . . . . . . . . . . . . . . . 52
2.6.1 Rössler Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.6.2 Diffusively Coupled Rössler Model and Synchronization Defect
Lines in 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3 1D Ring of Nonlocally Coupled Oscillators . . . . . . . . . . . . . . . . . . . 60
3.1 Electrochemical Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.2 Time-delayed, Nonlocal Complex Ginzburg-Landau Equation . . . . . . 62

iv
3.3 Effects of Time Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.3.1 Variation of Parameters: Relation to the Phase Equation . . . . 66
3.4 Effects of Frequency Heterogeneity . . . . . . . . . . . . . . . . . . . . 69
3.4.1 Test For Bistability . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4 Nonlocally Coupled Complex Oscillatory System in 2D . . . . . . . . . . . . 80
4.1 Nonlocal Rössler Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.1.1 Evaluation of Parameters . . . . . . . . . . . . . . . . . . . . . 81
4.2 Chimera State in the Nonlocal Rössler Model . . . . . . . . . . . . . . . 82
4.2.1 Disappearance of the Chimera State . . . . . . . . . . . . . . . 89
4.3 Synchronization Defect Lines in the Period-4 and Chaotic Regimes . . . 92
4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.1 Future Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.1.1 Analytical Development . . . . . . . . . . . . . . . . . . . . . . 106
5.1.2 Improved Parameter Space Search . . . . . . . . . . . . . . . . 106
5.1.3 Candidate Experimental Systems for Chimera Spiral Waves . . 107
A Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
A.1 Integration of Nonlocally Coupled Models . . . . . . . . . . . . . . . . . 108
A.1.1 No-flux Boundary Condition . . . . . . . . . . . . . . . . . . . . 109
A.2 Computation of the Phase Response Curve . . . . . . . . . . . . . . . . 110
A.3 Detection of Synchronization Defect Lines . . . . . . . . . . . . . . . . 114
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

v
List of Figures and Illustrations

1.1 Example of φf iref ly (t) versus φsignal (t) = Ωt, the latter indicated by the black
curve. Scenarios of synchronization are illustrated, when |Ω − ω| = 0 (green
curve) and |Ω − ω| ≪ 1 (blue curve). It can be seen that although the slope
(instantaneous frequency) of the blue curve is changing, on average the phase
difference is bounded. On the other hand, as indicated by the red curve,
if φf iref ly and φsignal evolve with different frequencies, their difference will
become unbounded. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.1 Simulation of nonlocal CGLE, Eq. (2.3), in a population of N = 128 oscillators


arranged on the interval [−1/2, 1/2] with periodic boundary condition on a
2D domain. Subpopulations of synchronized and unsynchronized oscillators
can be clearly seen from the snapshot of phase φ(x) (black) and the long-time
averaged frequencies hωi (red). Parameters used in the simulation are (a, b,
c) = (-1.0, 0.88, 2.0) and K = 0.1. . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Snapshot of X(r, t) = Re A(r, t) from a simulation of the nonlocal CGLE,
Eq. (2.3), with no-flux boundary condition. Parameters used were (a, b, c) =
(0, 0.4, 2.0) and K = 0.4. A spatial domain of L2 = 50 × 50 is discretized into
a lattice of N 2 points, where N = 512. . . . . . . . . . . . . . . . . . . . . . 28
2.3 Long-time averaged frequencies of the nonlocal CGLE model, at a cross-
section y = 28.8, using same parameters as Figure 2.2. Units on the horizontal
axis given in terms of the real length of the system. . . . . . . . . . . . . . 29
2.4 Simulation of the phase equation, Eq. (2.37) over the range of |α| < π/2.
Blue diamonds (X’s) indicate chimera states in which the synchronized sub-
population is faster (slower) than the desynchronized subpopulation, while
red diamonds (X’s) indicate modulated drift states. . . . . . . . . . . . . . . 47
2.5 Plots of the phase snapshots and long-time averaged frequencies, computed
from simulations of the phase equation on a 1D ring, using N = 256 oscillators,
for α = 1.51 (left column) and α = −1.51 (right column), showing different
chimera state solutions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.6 Time series of oscillatory behaviour showing simple (black), complex (red and
green) and chaotic (blue) behaviours. . . . . . . . . . . . . . . . . . . . . . 53
2.7 Plots show the components of (X, Y, Z) with a = 0.2, b = 0.2 and c = 2.0 (top
left), 3.0 (top right), 4.0 (bottom left) and 5.0 (bottom right). Increase of the
parameter c shows successive period-doubling of the trajectory. . . . . . . . 56
2.8 Projections onto the X-Y plane of the plots in Figure 2.7, shown in the same
order for the same parameters. The twisting in 3D space can be seen as self-
intersection in the plane, which increases the winding number of the curve.
The regimes shown correspond to simple oscillatory (top left), period-2 (top
right), period-4 (bottom left) and chaotic (bottom right) behaviours. . . . . 57

3.1 Schematic representation of the electrochemical experiment setup. . . . . . 61

vi
3.2 The time-averaged order parameter hr(t)i as a function of τ is measured over a
period of roughly 100 oscillations, in a simulation of the time-delayed, nonlocal
CGLE, Eq. (3.2) using (a, b, c) = (0, 0.2, 2.0), K = 0.1 and ǫ = 0. . . . . . . 64
3.3 Snapshots of phase φ and time-averaged frequencies hωi taken over roughly
170 oscillations in a simulation of the time-delayed, nonlocal CGLE, Eq. (3.2).
Parameters are the same as in Figure 3.2. At time delays τ1 /T ∼ 0.224 (left
panels) and τ2 /T ∼ 0.695 (right panels). The order parameter of the states
lie in the range 0.65 < hri < 0.75. . . . . . . . . . . . . . . . . . . . . . . . 65
3.4 Comparison between time-averaged order parameter hr(t)i as a function of τ ,
taken over a period of roughly 100 oscillations, using b = 0.2 or b = 0.8. The
parameters used throughout in the simulations of the time-delayed, nonlocal
CGLE, Eq. (3.2) are a = 0, c = 2.0, K = 0.1 and ǫ = 0. . . . . . . . . . . . . 67
3.5 The map α(τ ) from Eq. (3.11) is plotted for parameter values (a, b) identical
to those in Figure 3.4. The red dashed curves indicate boundaries at −π/2
and π/2. In the vicinity of the intersection of α(τ ) with these boundaries we
expect chimera states to occur (compare with Fig. 3.4). . . . . . . . . . . . 70
3.6 Using the same values of (a, b, c) and K as in Figure 3.2 for the simulation
of the time-delayed, nonlocal CGLE, Eq. (3.2), the frequency heterogeneity is
varied. The plot of hr(t)i as a function of τ , showing the effect of frequency het-
erogeneity on the synchronization order parameter. Frequency heterogeneity
appears to diminish both the completely synchronized states and the chimera
states. At the same control parameter τ , we see that hriǫ>0 < hriǫ=0. . . . . 71
3.7 Snapshots of phase φ and time-averaged frequencies taken over roughly 170
oscillations, hωi with heterogeneity in the frequencies. Parameters used for
the simulation of the time-delayed, nonlocal CGLE, Eq. (3.2)are same as in
Figure 3.2, except ǫ = 0.01, at time delays τ1 /T ∼ 0.201 (left panels) and
τ2 /T ∼ 0.702 (right panels). . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.8 Snapshots of phase φ and time-averaged frequencies taken over roughly 170
oscillations, hωi with heterogeneity in the frequencies. Parameters used for
the simulation of the time-delayed, nonlocal CGLE, Eq. (3.2)are same as in
Figure 3.2, except ǫ = 0.05, at time delays τ1 /T ∼ 0.189 (left panels) and
τ2 /T ∼ 0.731 (right panels). . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.9 Plot of hr(t)i as a function of τ using (a, b, c) = (0.0, 0.2, 2.0) and K = 0.1 for
the simulations of the time-delayed, nonlocal CGLE, Eq. (3.2). Comparisons
are made between the cases of identical and non-identical oscillators, with
varying amount of heterogeneity ǫ = 0, 0.01 and 0.05, using two different
initial conditions (explained in text). . . . . . . . . . . . . . . . . . . . . . . 75

4.1 Top row: Equivalent values of α computed by matching the numerically ob-
tained phase response curves for the Rössler model to the sinusoidal phase
response of Eq. (2.37), for parameters a = 0.2, c = 5.7 and varying b. Several
representative phase response curves are shown. Bottom row: Same as the
top, except in a different parameter regime a = 0.2, b = 0.2 and varying c. . . 83

vii
4.2 Snapshot of the scalar field X(r, t) for the nonlocal Rössler model, exhibiting
a spiral wave chimera. Parameters used are (a, b, c) = (0.2, 1.7, 5.7), in the
period-1 (simple oscillatory) regime of the system. . . . . . . . . . . . . . . . 84
4.3 Snapshot of the scalar field X(r, t) for the nonlocal Rössler model, exhibiting
a spiral wave chimera. Parameters used are (a, b, c) = (0.2, 1.25, 5.7), in the
period-2 regime of the system. . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.4 Processed image showing the presence of a period-1 SDL together with a spiral
wave chimera, corresponding to Figure 4.3. The image processing algorithm
is discussed in Appendix A.3. . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5 Time-averaged frequencies over ∼ 1700 oscillations recorded at the cross sec-
tion y = 27.05, in the corresponding Figure 4.2. Units on the horizontal axis
given in terms of the real length of the system. . . . . . . . . . . . . . . . . 87
4.6 Time-averaged frequencies over ∼ 1700 oscillations recorded at the cross sec-
tion y = 28.125, in the corresponding Figure 4.3. Units on the horizontal axis
given in terms of the real length of the system. . . . . . . . . . . . . . . . . 88
4.7 Snapshot of the scalar field X(r, t) for the nonlocal Rössler model, exhibiting
a spiral wave chimera. Parameters used are (a, b, c) = (0.2, 1.25, 6.7), in the
period-4 regime of the system. . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.8 Snapshot of the scalar field X(r, t) for the nonlocal Rössler model, exhibiting
a spiral wave chimera. Parameters used are (a, b, c) = (0.2, 1.25, 7.0), in the
chaotic regime of the system. . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.9 Snapshots of observed number of maxima n(r, t) corresponding to Figure 4.8,
shown at two different times t = 1000 (left) and t = 7500 (right). . . . . . . 92
4.10 Snapshot of the scalar field X(r, t) for the nonlocal Rössler model. Parameters
used are (a, b, c) = (0.2, 0.9, 5.7), in the period-4 regime of the system. . . . . 93
4.11 Snapshot of the scalar field X(r, t) for the nonlocal Rössler model. Parameters
used are (a, b, c) = (0.2, 0.75, 5.7), in the chaotic regime of the system. . . . . 94
4.12 Snapshots of observed number of maxima n(r, t) corresponding to Figure 4.11,
shown at two different times t = 1000 (left) and t = 7500 (right). . . . . . . 95
4.13 Processed image showing the presence of a period-1 (grey) and period-2 SDLs
(black) in the period-4 system, corresponding to Figure 4.7 in the presence
of a spiral wave chimera. The image processing algorithm is discussed in
Appendix A.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.14 Processed image showing the presence of a period-1 (grey) and period-2 SDLs
(black) in the period-4 system corresponding to Figure 4.10. The image pro-
cessing algorithm is discussed in Appendix A.3. . . . . . . . . . . . . . . . . 97
4.15 Processed image showing the presence of a period-1 (grey) and period-2 SDLs
(black) in the chaotic system corresponding to Figure 4.8 in the presence
of a spiral wave chimera. The image processing algorithm is discussed in
Appendix A.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.16 Processed image showing the presence of a period-1 (grey) and period-2 SDLs
(black) in the chaotic system corresponding to Figure 4.11. The image pro-
cessing algorithm is discussed in Appendix A.3. . . . . . . . . . . . . . . . . 99

viii
4.17 Schematic illustrations of reattachment events in the chaotic regimes. The
circle represents the centre of the spiral wave to which the SDLs are attached.
An SDL can be detached from its connection and reattached at a closed loop
emanating from the centre (A). In another scenario, an SDL that is not con-
nected to the centre initially can be attached to it, but first detaching the
other connected SDLs (B). . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

ix
Chapter 1

Introduction

In nature, periodic motions and processes can be observed in abundance, from planetary

orbits to biochemical cycles that maintain a living multicellular organism. In the latter

context, the robustness of such self-sustained cycles is of paramount importance. These cycles

are not generated due to external periodic forcing, nor do they decay to a steady state due to

friction or other external dissipative processes. The cells in which these biochemical cycles

take place are constantly subjected to changes in the external environment, but the cycles are

preserved in spite of these circumstances. Furthermore, different cells in a living organism

are often required to act in unison - for instance, the simultaneous firing of a population of

neurons. In order for this to occur, interaction, or coupling must take place amongst the

population. We are interested in the study of synchronization, the process through which

many self-maintained oscillatory entities, or oscillators, adjust their rhythmicity through

interaction [Pikovsky et al. , 2001].

In order to get a better physical understanding of what we mean by adjustment of

rhythmicity, we discuss an example from biological observation and experiment. In Southeast

Asia, populations of male fireflies have been observed to gather in trees along the banks of

a river, and flash in synchrony as part of their mating ritual [Buck, 1988]. The synchronous

flashing by no means occurs instantaneously - an individual firefly observes the flashing of

others, and slow down or speeds up in order to compensate. This process begins with only

a few individual fireflies flashing at different rates. Over a long period of time, the whole

population begin to participate in the flashing, which subsequently leads to the behaviour

of synchronous flashing being spread throughout population [Buck, 1976].

To see how an individual insect is influenced by a photo-stimulus, Hanson conducted an

1
experiment where an isolated firefly was presented with a controlled, periodically flashing

signal [Hanson, 1978]. The signal, acting as a driving force, flashes at a frequency Ω, which

may be different from that of the firefly’s internal flashing frequency ω. If the two frequencies

are identical, Ω = ω, the firefly will become locked to the signal and flash simultaneously.

If the discrepancy |Ω − ω| > 0, but is sufficiently small, the firefly will flash at a different

instantaneous frequency, but still following the signal with only a small lag in time, so that

its time-averaged frequency is equal to Ω. If we observe the experiment for a length of time

equal to τ , and count the number of times Nf iref ly that the firefly has flashed and the number

of times Nsignal that the signal has flashed during τ , the difference |Nf iref ly −Nsignal | does not

increase monotonically with length of observation τ . Analogously, imagine the firefly and

the signal as runners on a circular track, neither gets lapped by the other because the firefly

tries to slow down or speed up according to the signal. On the other hand, if |Ω − ω| > 0

is sufficiently large, it becomes impossible for the firefly to adjust to the signal, and the

difference |Nf iref ly − Nsignal | will grow unbounded with time. In the first two cases when

|Ω − ω| vanishes or is sufficiently small, it is intuitive to speak of synchronization of the

firefly’s flashing with that of the signal, when the creature alters its own internal rhythm to

follow a periodic stimulus, even if its flashing is not exactly simultaneous with the signal.

Following this intuition, we then say that there is a lack of synchronization between the

firefly’s flashing and the signal, when |Ω − ω| is sufficiently large.

The observations in the previous paragraph can be stated differently using the concept of

phase. A point in the oscillation at time t can be identified unambiguously with the value

of a continuous phase variable φ(t), which should evolve in proportion to the instantaneous

frequency of the oscillator. In the firefly experiment, for example, let φsignal (t) and φf iref ly (t)

denote the phase of the signal and the phase of the firefly. We know that φsignal (t) = Ωt,

since the signal simply flashes at a constant frequency, but φf iref ly (t) would correspond to

the observed flashing of the firefly in the presence of the stimulus, and evolves with an

2
instantaneous frequency different from ω, as mentioned in the previous paragraph. A model

of the firefly experiment is actually described in [Strogatz, 2000b], but instead of going into

the mathematical details, we illustrate the evolution of the phase φsignal (t) and φf iref ly (t)

corresponding to the scenarios described in the previous paragraph, in Figure 1.1. Given

that φsignal (t) evolves linearly, we can depict the corresponding behaviours of φf iref ly (t) in

the case when the firefly is synchronized with the signal, and when it is not.

Phase

Time

Figure 1.1: Example of φf iref ly (t) versus φsignal (t) = Ωt, the latter indicated by the black
curve. Scenarios of synchronization are illustrated, when |Ω − ω| = 0 (green curve) and
|Ω − ω| ≪ 1 (blue curve). It can be seen that although the slope (instantaneous frequency)
of the blue curve is changing, on average the phase difference is bounded. On the other
hand, as indicated by the red curve, if φf iref ly and φsignal evolve with different frequencies,
their difference will become unbounded.

3
Note that the phase increases by 2π radian per oscillation, so we can glean from the

value of φ(t) both the position in the oscillation as well as the number of full cycles that

has occurred. Phase variables defined this way are called unwrapped, and we observe the

number of oscillations from the unwrapped phase by noting the number of times φ exceeds

an integer multiple of 2π radian. As Figure 1.1 shows, we can instead use the difference

between the unwrapped phases |φf iref ly − φsignal | to characterize synchronization. We shall

have more to say about oscillations and phase in Section 2.1. In this chapter, we refer to the

unwrapped phase whenever we speak of the concept of phase.

Going back to the case of a large population of fireflies, the simplistic scenario in the

experiment no longer applies. There no longer exists a single, external periodic stimulus,

but rather stimuli generated internally from an entire population, acting on a single firefly.

The frequency of each individual becomes adjusted through mutual interaction. The coupling

topology present in the population of fireflies is called global coupling, in which the coupling

is “all-to-all” - each individual is identically influenced by the sum of all others. In contrast,

local coupling, or coupling between only nearest-neighbours, arises in contexts in which

the stimulus that affects the interaction has a limited ranged of influence. Note that local

coupling requires a notion of distance defined between members of the population, which is

not needed in the case of global coupling.

An important question regarding systems of interacting oscillators is the emergence of

synchrony under different conditions. In a large population, conceptualization of this problem

is necessarily more difficult in comparison to the example of the firefly experiment. In the

following sections, we describe the emergence of synchronization in globally and locally

coupled systems.

4
1.1 Synchronization in Globally Coupled Systems

The mutual interaction in a population of fireflies exemplifies the concept of global coupling:

each individual firefly in the population must take into consideration the flashing stimuli

from all others in the population. Even though some fireflies located on the edges of a tree

might be obstructed from others, in order to yield simplification to a physical problem one

can often consider a large enough population, such that the interaction scheme is assumed to

be true on average and describes the bulk of the population. Unavoidably, each individual is

somehow slightly different from the others in both genotype and phenotype, and so intrinsic

differences in the flashing rhythm exist in the population.

This leads us to an important problem in the study of synchronization, which is how

oscillators can spontaneously become locked to a common frequency despite the differences

in their intrinsic, natural frequencies. Said differently, we are interested in knowing if the

phenomenon of synchronization is robust, unaffected by small differences in the natural

rhythms between oscillators. Arthur Winfree studied this problem in the context of a pop-

ulation of oscillators under all-to-all, or global interaction. He noted that the theoretical

problem can be simplified in two ways. If the oscillators interact weakly, their amplitudes

adjust on fast timescales, so that only a description of the evolution of the phases on longer

timescales is needed. Furthermore, oscillators in the population interact in an identical

manner [Winfree, 1967]

N
!
dφi X
= ωi + X(φj ) Z(φi ). (1.1)
dt j=1

Here, ωi denotes the intrinsic frequency of each oscillator. The evolution of φi depends on

the phase-dependent influences X(φj ), summed over the entire population, and its own re-

sponse function Z(φi ). Coupling in the manner of Eq. (1.1) resembles in spirit a mean-field

approximation in physics [Strogatz, 2000a]. Winfree analyzed the simple model Eq. (1.1)

using numerical methods to reveal that when the spread of natural frequencies ωi is suffi-

5
ciently small, a cluster of oscillators in the population will begin to synchronize. Otherwise,

if the spread in ωi is too large, then the system is dominated by unsynchronized behaviour.

These observations obey exactly the intuition drawn from the firefly experiment conducted

by Hanson.

After Winfree, Yoshiki Kuramoto made an attempt to create an analytically tractable

model based on the ideas behind Eq. (1.1) [Kuramoto, 1984]. This resulted in what is

now widely known as the Kuramoto model. As we shall see, the Kuramoto model is a

paradigmic model of collective synchronization in population of oscillators. Synchronization

behaviours in the Kuramoto model are consistent with the intuition put forth by Winfree,

and these results can be derived analytically. The Kuramoto model has also application

in physics and engineering, for example, it can be shown that the equations governing a

coupled array of superconducting Josephson junctions can be reduced to the Kuramoto

model [Wiesenfeld et al. , 1996, Wiesenfeld et al. , 1998].

1.2 Synchronization in Locally Coupled Systems

Unlike in the population of fireflies, in certain systems the mutual interactions between oscil-

lators have a limited influence in space. An example of this being local, or nearest-neighbour

coupling. We show in this section how oscillatory chemical reaction can be an example of

synchronization in locally coupled system, how it is robust with respect to heterogeneity, in

the same way as the globally coupled system we discussed in the previous section.

The first evidence of oscillatory chemical reaction, or a chemical oscillator, was provided

by the Soviet chemist Boris Belousov. Near equilibrium in a chemical reaction, oscillation

is impossible since every elementary step in the kinetics is balanced out by its reverse pro-

cess [Nicolis & Prigogine, 1977]. Belousov’s finding of prolonged chemical oscillations in the

reduction and oxidation of cerium ions, Ce4+ and Ce3+ was indicative of the fact that the

dynamics of the chemical system takes place far from equilibrium. Eventually, equilibrium

6
would be reached in Belousov’s experiment as the reactants run out, since it takes place in

a closed environment. Nonetheless, the system approaches equilibrium in a non-monotonic

manner, as indicated by the presence of oscillation. Belousov’s graduate students, Ana-

tol Zhabotinsky rediscovered Belousov’s reaction sequence for cerium and finally published

this result in 1964, producing the classic recipe for the Belousov-Zhabotinsky (BZ) reaction

[Zhabotinsky, 1964].

A simple chemical oscillator can be achieved by stirring a BZ reaction. The content of

a stirred BZ reaction will change in colours continuously from purple to blue, in a periodic

manner, indicating the oscillation of concentrations of Ce4+ and Ce3+ in the system. We may

imagine that stirring induces a homogeneous mixture of chemicals. The spatially uniformity

of the oscillation indicates that the local oscillation of the concentration of the cerium ions

is at the same phase everywhere, so it can be said that the medium is synchronized. Let

us consider a toy model of chemical oscillators of discrete, tiny reaction beakers prepared

(nearly) identically, stirred, and contain all necessary BZ reactants, connected to each other

in a ring configuration. Local concentration of each species inside each beaker changes

either by reacting with other species inside a single beaker, or by exchange, diffusing into

neighbouring beakers. Without diffusion or exchanges, the concentrations would oscillate

inside each beaker due to the reaction. Diffusion tends to smooth out neighbouring differences

in the phase gradient of the oscillations, leading to a dynamical, oscillatory state in our simple

toy model in which the chemical oscillations inside the beakers become synchronized.

We can mimic the toy model by writing down a set of equations governing the evolution

of u = (u1 , ...., un ), the concentration fields of the n chemical species involved in the reaction

∂t u(xi , t) = F(u) + D[u(xi−1 , t) + u(xi+1 , t) − 2u(xi , t)], (1.2)

where xi denote the ith beaker in the ring. The functions F = (F1 (u), ..., Fn (u)) represent

the reactions taking place between the chemical species. The matrix D = diag(D1, ..., Dn )

7
are the diffusion constants of each species, and describe the extent of diffusion occurring

between the ith beaker and its nearest-neighbours, as is evident in the coupling term inside

the square bracket. In the continuum limit, the coupling term in Eq. (1.2) is exactly the

diffusion operator, or the Laplacian. The corresponding class of models in the continuum

limit is called reaction-diffusion equations, first introduced by Alan Turing in 1952 as a

paradigm for animal embryogensis and spontaneous formation of spatially ordered static or

oscillatory states, or patterns from an initially spatially uniform state [Turing, 1952]. Note

that the same discussion in the toy model above applies to the case of a lattice configuration

without loss of generality. The only change occurs on the right hand side of Eq. (1.2), which

will contain more terms in the local coupling due to a change in topology - for instance,

there are four nearest neighbours in the 2D square lattice. Variants of the BZ reaction can

be performed in petri dish with thin layers of reactants, suspended in a gel, or over some cat-

alytic surface. Since the geometry realized in these examples are essentially a 2D surface, we

imagine our toy model in 2D, in the limit of continuous space - then the intuition of reaction

and diffusion driving the process of mutual synchronization apply perfectly in a chemical

context. In an unstirred BZ reaction medium, spatially uniform oscillation first takes place

in the medium, so that the phase of oscillation at each point is identical. However, non-zero

phase differences at points in the medium can be created, due to imperfections in the prepa-

ration of the catalytic reaction surface, or impurities in the reactants. At these points the

local oscillations become faster or slower. It is thought that the initial spontaneous genera-

tion of target waves, or concentric wavefronts in the BZ reaction, is precisely due to various

kinds of imperfection [Cross & Greenside, 2009]. Just as in the case of globally coupled os-

cillators, synchronization is robust with respect to tiny, intrinsic frequency differences in a

system of locally, or diffusively coupled chemical oscillators, leading to stable wave patterns.

The appearance of waves, which are spatially and temporally periodic structures, implies

that the phase differences between local oscillations at different points in the system are

8
bounded, indicating complete synchronization in system.

If we now disturb the medium by breaking up the target wavefronts (with a laser, or by

hand, depending on the experimental setup), the wavefronts curl around the centre of the

target waves, and spiral wave patterns emerge in the system. This indicates that synchroniza-

tion persists in the system even under perturbation to the wave propagation. Spiral waves

have been observed in the heart [Davidenko et al. , 1992] and the brain [Huang et al. , 2004],

amongst many other examples in biology. These observations are just a few that show how

synchronization phenomena occur generally in naturally-occurring oscillatory systems.

1.3 Synchrony Breaking in Population of Identical Oscillators

If a population of oscillators are identical, such that the free (uncoupled) motion of each

oscillator is governed by a common frequency, and the form of coupling is also identical for

different oscillators, we would expect uniform behaviour in the population due to perfect

symmetry in the set of equations describing the system [Motter, 2010]. This means that

the only outcome through interaction is complete, mutual synchronization. Global and local

coupling, as we have described in the preceeding sections, fit the description of identical

coupling.

However, as we discuss below, the intuition presented in [Motter, 2010] is not complete:

synchrony can be broken in a coupled system of identical oscillators using an identical cou-

pling scheme, as was first observed by Kuramoto [Kuramoto & Battogtokh, 2002]. This new

discovery has drawn the interests of many physicists and applied mathematicians, and is the

main focus of this thesis.

1.3.1 Nonlocal Coupling

In the past ten years, a different interaction scheme called nonlocal coupling has been

the subject of intense research. Nonlocal coupling is the same as global coupling, in that

9
the interaction is also “all-to-all”, but also different in that a pair of oscillators situated

further apart influences each other weakly, and a pair of oscillators that are closer influ-

ences each other strongly: in nonlocal coupling, the strength of interaction decays with

distance. Therefore, nonlocal coupling also requires a notion of distance as in the case of

local coupling. Many studies have demonstrated that nonlocally coupled systems of iden-

tical oscillators come to behave in an asymmetric manner [Kuramoto & Battogtokh, 2002,

Kuramoto & Shima, 2003, Shima & Kuramoto, 2004]. These systems evolve to states in

which subpopulations of mutually synchronized and desynchronized oscillators coexist: in

one part of the system we can observe a common effective frequency, indicating synchroniza-

tion, and in the remainder we observe a smooth distribution of frequencies. Additionally,

these subpopulations are localized in adjacent spatial domains. This is contrary to our in-

tuition drawn from locally and globally coupled systems, which do not exhibit this kind of

coexistence in a population of identical oscillators. Furthermore, nonlocal coupling is also

an identical interaction scheme in the same way as local and global coupling, and thus it is

surprising how such a symmetric interaction topology in a system of identical oscillators can

evolve to an asymmetric state that breaks perfect synchrony in the system [Motter, 2010].

We follow Abrams and Strogatz in calling such states of strange coexistence chimera states

[Abrams & Strogatz, 2004]. Even stranger is the fact that the nonlocally coupled system of

oscillators has been shown to exhibit bistability: for particular parameter values, complete

synchronization and chimera states can both be stable solutions [Laing, 2009a].

In the previous sections, we have described real-life examples of oscillatory systems where

the mutual interactions between oscillators are global or local. The study of nonlocal coupling

can also be motivated by a physiological example, namely the coexistence of synchrony

and asynchrony in the mammalian brain. Oscillatory models have been employed to study

neuronal networks, and there is evidence that neurons do interact more strongly with its

neighbours and weakly with those further away [Singer & Gray, 1995], which is precisely

10
required by nonlocal coupling. While we do not wish to imply that neuronal systems have

a direct relationship to chimera states in the nonlocally coupled models studied in this

thesis, the fact that coexistence of synchrony and asynchrony can occur in highly complex

and dynamically adaptive neuronal networks indicates that the phenomenology of chimera

state, or chimera-like behaviours is general. This points to neuronal system as or a physical

example of nonlocal coupling, as well as a candidate for the observation of chimera states

in a biological setting. In the next section we discuss two experimental studies on the brain

that provide evidence for observation of coexistence of synchrony and asynchrony.

1.4 Coexistence of Synchrony and Asynchrony in the Brain

Experimental studies of the brain usually involve measuring electrical activities non-invasively,

via a technique known as electroencephalography (EEG). Electrodes are attached to the scalp

at different positions of the head, in order to measure the electric potential created by the

waves of ion exchange that propagates in the neuronal networks. The measure EEG signals

reflect the simultaneous firing of a group of neurons with similar spatial orientations, i.e.

groups of neurons in the vicinity of an EEG electrode. EEG activities exhibit oscillations

at different frequency ranges are recorded at different electrode locations attached at var-

ious points on the subject’s head. These identifiable structures in an EEG recording are

associated with different brain functions or states (for example, awake or different stages of

sleep).

1.4.1 Unihemispheric Sleep

Unihemispheric sleep is a behaviour observed in various aquatic mammals, birds and rep-

tiles where the creature “sleeps with half of the brain” [Rattenborg et al. , 2000]. Recorded

EEG results indicate drastically different signals on two halves of the brain. The “sleep-

ing side” shows predominantly high-amplitude, low-frequency oscillations characteristic of

11
slow-wave sleep stages, whereas the “awake side” exhibits low-amplitude, high-frequency

oscillations. The EEG readings from different electrodes attached to the brain shows that

signals from EEG electrodes attached to the awake side of the brain oscillate in unison, indi-

cating synchronization, while the signals from the other half exhibit the opposite behaviour

[Rattenborg et al. , 2000].

Early behavioural and EEG studies of dolphins [Mukhametov et al. , 1977] indicate that

they sleep alternatingly with one eye open during episodes of unihemispheric sleep, to allow

episodes of rest for each hemisphere of the brain. This is complemented by constant slow-

swimming or hovering at the surface of the water, which allows the creatures to breath

without breaking out of sleep. Packs of dolphins have also been observed to swim in a

circular direction during unihemispheric sleep, which allows an individual to monitor the

group as well as external predation risks, depending on its position in the circular pack

and which eye remains open. These are the several pieces of evidence of the evolutionary

function posed by unihemispheric sleep, which is characterized on the neurophysiological

level by coexistence of synchronous and asynchronous brain activities.

1.4.2 Temporal Correlation Hypothesis

In the processing of information gained from the environment, the brain must take on the

task of distinguishing an unimaginably large number of possibilities of external stimuli, and

encode the appropriate physiological response in all cases. The problem of distinguishing

visual stimuli and eliciting the appropriate physiological response is called visual feature

integration [Singer & Gray, 1995]. For example, if one is presented with a delicious meal, he

or she may begin to salivate, feel hungry, or perform different kinds of memory association.

These tasks correspond to different senses, and the populations of neurons responsible for

these responses are located in different parts of the brain. Over a finite time interval, these

neurons will fire synchronously, which can be measured as oscillatory patterns in the electric

potential. On the other hand, neurons that are located in close vicinity of these active

12
populations do not become recruited in this synchronous firing. Hence, the firing patterns of

these neurons are asynchronous with the active populations. This framework of measurement

is called temporal correlation hypothesis, which seeks to explain the encoding of responses to

external stimuli by correlated, synchronous activities of populations of neurons over different

spatial locations in the brain [Singer & Gray, 1995].

Different experiments on visual feature integration have been performed to show be-

haviours synchronization across different cortical columns (defined as the intersection of

neurons in different cortical layers with a probe inserted perpendicularly to the cortical sur-

face) [Ts’o et al. , 1986, Ts’o & Gilbert, 1988] to different hemispheres [Engel et al. , 1991].

Even within the neuronal population of a single cortical column, subpopulations of neurons

can be recruited to synchronized, while the rest take on different firing frequencies. Further-

more, such synchronization can be relinquished, so that in the next episodes of synchronized

activities different numbers and/or different members of the cortical column can be recruited.

1.5 Motivation

Through the use of examples, we have introduced the concept of synchronization in a system

of mutually interacting oscillators. We find that the phenomenon of synchronization is

robust with respect to different coupling schemes as well as small inherent differences in the

oscillators, as can be seen in examples of naturally-occurring systems, such as populations

of fireflies, or laboratory experiments of oscillatory chemical reactions. In the ideal scenario

where each oscillator is identical, only complete synchronization can appear in these systems,

which are coupled identically via a local or global coupling scheme. The recent discovery of

chimera states in nonlocally coupled systems of identical oscillators has been a surprise and

challenges the validity of this statement. Chimera states behave asymmetrically, in which

subpopulations of synchronized and desynchronized oscillators coexist, despite the fact that

nonlocal coupling also has an identical coupling topology.

13
This thesis is concerned with the study of chimera states in populations of nonlocally

coupled oscillators. The research conducted is in part motivated by a candidate experi-

mental system, an array of nonlocally coupled electrochemical oscillators, which can mimic

different forms of simple, complex and chaotic oscillations [Blaha, 2012]. Recent works have

experimentally demonstrated chimera states in very restricted settings, either exhibiting only

spatial asynchrony (but complete synchronization in the phase) [Hagerstrom et al. , 2012], or

employs an interaction topology that only mimics nonlocal coupling [Tinsley et al. , 2012].

In particular, the experiment in [Tinsley et al. , 2012] is inspired by previous theoretical

works, and only demonstrates chimera states in a restrictive sense, since their results do

not depend on spatial dimensions, unlike the most general form of nonlocal coupling. Ad-

ditionally, new phenomena that are not predicted by current theories have been reported in

the experimental system of [Tinsley et al. , 2012]. This suggests that the current theoretical

understanding of nonlocally coupled systems and of chimera states is incomplete.

Although some current analytical theories exist for the chimera state, they are based

on a highly idealized model in terms of only the dynamics of the phase variable containing

only a single control parameter, and thus are not suitable to direct comparison with any

particular experimental setups. Moreover, the idealized phase model does not yield as wide

a variety of dynamics as oscillator models with both amplitude and phase degrees of freedom.

Realistically, oscillators in experiments are described by both amplitude and phase, and their

behaviours have dependence on multiple control parameters. Because of this, we lack the

knowledge of parameter regimes in which chimera states can be observed, and may resort

to searching blindly. To resolve this problem, we propose a method to evaluate whether a

given set of parameters leads to the formation of chimera states.

Based on the motivations outlined in the previous paragraphs, in this thesis we study a

model of nonlocally coupled oscillators with both amplitude and phase degrees of freedom.

The model describes our candidate experimental system in the simple oscillatory regime,

14
but we also expect it to also be a suitable description for other general experimental setups.

We introduce constraints that appear in the candidate experimental setup, including effects

such as time-delay in the interaction as well as small frequency heterogeneity. The addition

of frequency heterogeneity in our model should give an indication whether chimera states

are robust with respect to systematic errors encountered in experimental scenarios.

Furthermore, realistic chemical oscillators in general can display a wider range of be-

haviour apart from simple oscillation, and exhibit complex or chaotic oscillations by changing

the system’s parameters [Scott, 1991]. This is in particular true for our candidate experimen-

tal system involving electrochemical oscillators. A natural question to ask for these systems

is whether chimera states can persist when the oscillatory behaviour changes, in the complex

or chaotic oscillatory regimes when the notion of periodicity is defined differently (or non-

existent) in comparison to simple oscillation. Hypothetically, since synchronization (or a lack

thereof) is defined through relationships in the phase, and the mechanism responsible for the

appearance of a chimera state can be described through dynamics of the phase, one would

expect that even under complex or chaotic oscillations, these notions will continue to hold

as long as the notion of phase can be well-defined in these regimes. Therefore, we consider a

model of nonlocally coupled complex or chaotic oscillators in which the phase of oscillation

can be meaningfully defined, and analyze the model in 2D. We study the phenomenology of

patterns corresponding to the chimera state in the complex or chaotic oscillatory regimes of

this model.

1.6 Outline

In this thesis, we will address the questions posed in Section 1.5 following the outline we

give below. In Chapter 2 we will first establish the mathematical background necessary

to understand the origin of nonlocally coupled models, as well as its description in terms of

evolution of the phase variable. We then go on to review particular results from the literature

15
relevant to the models studied in this thesis.

In Chapter 3, we present our results for a variant of the nonlocally coupled complex

Ginzburg-Landau equation in a 1D ring, described in Section 3.2. We introduce time-delay

in the coupling, which is a constraint present in the experimental setup. We also introduce

heterogeneity in the intrinsic frequencies of the oscillators, to account for the systematic

error encountered in the experiment. We find that the time delay can be used to control the

synchronization behaviour of the system, and by tuning the time delay, the system transitions

back and forth from completely synchronized states to completely unsynchronized states in

a continuous manner, over a finite interval of the time delay. Between these transitions,

we encounter chimera states in which the synchronized subpopulation is slower or faster

than the unsynchronized subpopulation. These chimera states exist at different values of the

delay. By relating our model to the phase equation for nonlocally coupled systems, discussed

in Chapter 2, we give a qualitative prediction of the synchronization behaviour when the

parameters of the nonlocal complex Ginzburg-Landau equation are varied. In the presence

of small frequency heterogeneity, “noisy” chimera states are present in the system, with

coexisting subpopulations of synchronized and unsynchronized oscillators. In these chimera

states, we can still observe the necessary feature of spatial localization of the subpopulations.

However, at higher frequency heterogeneity, the localization is no longer observed, such that

we can no longer identify the observed states as a chimera state.

In Chapter 4, we present our findings for a nonlocally coupled Rössler model in 2D

with no-flux boundary condition, described in Section 4.1. To generate a chimera state, we

propose a method to identify suitable parameter values, based on our knowledge of the phase

equation for nonlocally coupled systems. We then vary the parameters of the Rössler model

and study the behaviours in the complex and chaotic oscillatory regimes. We prove the

existence of chimera states in these regimes, in particular, we find spiral wave chimerae with

synchronization defect lines, which are associated with locally complex or chaotic oscillations

16
in general. Our results is the first indication that the existence of chimera states can indeed

be generalized to include nonlocally coupled complex and chaotic oscillatory systems.

Lastly, in Chapter 5, we summarize the results presented in this thesis and discuss possible

avenues for future investigations arising from this thesis. We suggest and describe a candidate

experimental system, which is promising for the observation of spiral wave chimerae under

identical, nonlocal coupling.

17
Chapter 2

Mathematical Background and Literature Review

In this chapter we give a mathematical treatment of nonlocally coupled oscillatory systems, in

particular the phenomenology of chimera states in these systems. We shall see that chimera

states appear in the regime of weak interaction. For these nonlocally coupled systems,

a reduced equation that describes only the dynamics in the phase can be obtained when

the oscillators in the population interact weakly. As such, chimera states can be studied

under the framework of this phase description. We review the Kuramoto model, which is

an example of such reduction to phase dynamics for globally coupled oscillators. The same

reduction procedure performed on the Kuramoto model can be applied to obtain the phase

equation for nonlocally coupled oscillators. After deriving the phase equation for nonlocally

coupled systems and discussing its synchronization properties, we review specific results on

the phase equation, its variants, as well as a model of complex oscillatory system. These

results will give us relevant insights for the models investigated Chapters 3 and 4.

2.1 Definitions and Basic Terminologies

In the following sections, we take the time to mathematically define and extend some im-

portant concepts regarding oscillations first discussed in Chapter 1.

2.1.1 Notions of Synchronization

In Chapter 1, we introduced the simple firefly experiment to illustrate the principles of syn-

chronization. A bounded difference between the unwrapped phases of a pair of oscillators

indicates synchronization, and thus the oscillators take on a common (time-averaged) fre-

quency. This is the notion of synchronization known as phase locking, which we have

18
described previously using the simple example of the firefly experiment.

Otherwise, if this difference between a pair of oscillators is unbounded, then we cannot

say that their are synchronized in the sense of phase locking. However, a weaker notion of

synchronization called frequency locking can be defined, wherein a pair of oscillators take

on a common frequency. Note that phase locking implies frequency locking, but the opposite

is not true - a pair of oscillators can take on the same frequency through interaction but

their unwrapped phase difference can be unbounded.

2.1.2 Limit Cycle Oscillations

We have shown by examples some of the qualities that our oscillators should have: that the

oscillation is sustained without the presence of an external force or mutual interaction, the

oscillations should be robust with respect to external perturbation, and lastly the oscillations

should be able to easily synchronize under suitable conditions. Specifically, our use of the

word “oscillator” refers to the mathematical notion of limit cycle oscillator, which comes

from studies of dynamical systems. We discuss the concept of a limit cycle oscillator below.

Let X(t) = (X1 (t), ..., Xn (t)) be n dynamical variables describing the state of a system.

In general, the asymptotic behaviour of a set of differential equations dt X = F(ǫ, X) will

depend on some control parameter ǫ. At a given value of ǫ, suppose X0 is a fixed point

solution to the system, such that ∇F(X0 ) = 0, and we want to determine the behaviour

of nearby solutions. We do so by adding a small perturbation δX(t) to X0 . A common

technique is to insert the particular solution X′ (t) = X0 +δX(t) into our system, then expand

F(X0 + δX) as a Taylor series about X0 . This procedure of linearization yields a system

of equation dt (δX) = DF(X0 ) · δX describing the evolution of the perturbation, whether

it grows or diminishes. The object DF(X0 ) is the Jacobian matrix of F evaluated at the

point X0 . Diagonalization of DF(X0 ) yields eigenvalues λn = αn + iβn ∈ C which describe

the evolution in time, with the real parts αn indicating stability, since the solutions grow in

proportion to eαn t . If αn < 0 (αn > 0), small perturbations decay (grow) exponentially in

19
time, in which case we say that the fixed point X0 is linearly stable (unstable).

Rather than evolving to a steady-state solution given by a stable fixed point X0 , as the

control parameter is varied, an initially stable fixed point solution X0 can become unstable,

such that there is now an attracting, periodic solution Xp (t) (e.g. Xp (t) = Xp (t + T )) to

the system that forms a closed trajectory. The limit cycle is attracting in the sense that all

nearby solutions in an ǫ-neighbourhood of Xp approach it asymptotically [Strogatz, 2000b].

In the theory of nonlinear dynamics, this change in behaviour is called a supercritical Hopf

bifurcation, and it is one possible mechanism by which stable oscillatory behaviour emerges

from a steady-state.

In the simplest case of n = 2, the supercritical Hopf bifurcation corresponds to a pair of

complex conjugate eigenvalues crossing the half-plane from Re αn < 0 into Re αn > 0. At

the onset of this oscillatory behaviour, every system can be reduced to the normal form

by rescaling and change of variables [Strogatz, 2000b], into

dA(t)
= ǫA − η|A|2 A, (2.1)
dt
where A = reiθ and η are complex, and ǫ is the real parameter that controls the bifurcating

behaviour. The magnitude r = |A| is identified with the amplitude and θ = Arg z the phase

of the limit cycle oscillation. The presence of the nonlinear term −|A|2 A saturates the linear

growth proportional to the linear term A (for ǫ > 0). We further remark that the limit cycle

of Eq. (2.1) evolves with constant amplitude and frequency [Strogatz, 2000b]. This implies

that despite nonlinear terms being present in the original system dt X = F(X), at the onset

of oscillation, the limit cycle of the system resembles a perfect circle, and the phase on it

evolves uniformly. In summary, a limit cycle oscillator fits the necessary requirements of

robust, self-sustained oscillations that can become easily synchronized: i) the amplitude of a

limit cycle oscillator is stable with respect to small perturbations, due to the attractive nature

of the limit cycle, and ii) the phase is neutrally stable in the sense that perturbations in this

20
coordinate will neither grow nor decay - hence a population of mutually interacting limit

cycle oscillators can easily become adjusted to a common rhythm and mutually synchronize

[Pikovsky et al. , 2001].

2.1.3 Phase

We have introduced the unwrapped phase in Chapter 1 to illustrate basic concepts of syn-

chronization. In the firefly example, we illustrated how φf iref ly (t) evolves under external,

periodic forcing in Figure 1.1. Note that we did not give a mathematical description of the

phase φf iref ly (t), which is in general necessary to study synchronization of oscillators. Thus,

for general models of oscillators, we would like to be able to define a phase variable that

suitably reflects changes to the oscillations under perturbation or coupling.

The variable θ =Arg A in Eq. (2.1) for example, is a suitable definition of the phase of

oscillation in Eq. (2.1). It is given by θ = tan−1 (Y /X) ∈ [−π, π) and maps the position

(X, Y ) on the limit cycle uniquely to a phase θ(X, Y ). In particular, if we couple Eq. (2.1)

to an external periodic force, the values (X, Y ) evolve in a way to reflect the changes to the

natural motion of the system due to this forcing, and so does the phase θ(X, Y ).

A phase variable defined in a similar manner to θ is wrapped, since it is defined on a

bounded domain [a, b). We note that θ has a geometrical interpretation as the angle between

the real and imaginary components of A, which performs rotation in the complex plane

around the origin. We note that in general this geometrical way of defining a phase works

only when the limit cycle can be embedded in R2 , and when the interior region surrounded

by the limit cycle is simply connected. In general, the phase is more of a concept used to

describe oscillatory behaviour and not a unique quantity. A suitably defined phase variable

parametrizes the oscillation within each cycle in an unambiguous manner.

Note that the evolution of a phase variable is not necessarily linear, but one can always

make a transformation φ → Φ such that ∂t Φ = ωt evolves with constant frequency ω. We call

this unique definition the true phase [Pikovsky et al. , 2001]. In the remainder of this thesis

21
whenever we refer to the notion of phase, we will always make clear in the corresponding

text the definition we are using. However, when we speak of the phase difference between a

pair of oscillators, we will always mean the difference between the unwrapped phases.

2.1.4 Complex Ginzburg-Landau Equation

In the toy model example of Section 1.2, we argued that waves appear in the system when

the reaction inside each tiny beaker is oscillatory, and coupled to its nearest-neighbours.

This implies that the toy model system without coupling, dt u = F(u) describing the local

changes in the concentrations u must possess an attractive periodic solution u(t) = u(t + T ).

We also argued that in the limit of continuous space, the toy model leads to a general class

of models called reaction-diffusion equations, given by dt u(x, t) = F(u) + D∇2 u.

We know that nonlinear wave patterns can occur in reaction-diffusion systems. In ac-

tuality, the onset of oscillations in these spatially extended models is more complicated.

Spatially ordered patterns, such as standing waves or traveling waves, can emerge from an

initial spatial uniformity, due to instability of this spatially uniform state. The emergence

of spatially periodic patterns means that we can employ Fourier analysis to study their be-

haviours. A pattern is then associated with a particular wavelength selected just at the

onset of the instability, as the corresponding spatial Fourier modes become enhanced by the

instability and grow exponentially. At the same time, the instability can be either static or

oscillatory, leading to either Turing patterns or nonlinear waves [Cross & Greenside, 2009].

We wish to have a reduced description of the dynamics and the emergent patterns close

to the onset of an instability, that is, when a narrow band of spatial modes about the max-

imal spatial mode have just become unstable. A powerful mathematical method to treat

this problem is by writing down what is called an amplitude equation. Assuming a spa-

tially periodic structure has emerged in the system, one makes an ansatz by introducing

a complex amplitude A, which modulates the spatially periodic pattern slowly. The idea

behind this is based on the emergence of an “average” wavelength resulting from a contin-

22
uum of spatial modes having nearly the same wavelengths and amplitude, represented by A

[Cross & Greenside, 2009].

For an oscillatory instability at the zero wavelength, the corresponding amplitude equa-

tion is given by the following equation [Cross & Greenside, 2009, Pikovsky et al. , 2001]

∂t A(r, t) = A − (1 + ic3 )|A|2A + (1 + ic1 )∇2 A, (2.2)

known as the complex Ginzburg-Landau equation (CGLE). It is a universal description

of reaction-diffusion systems at the onset of an oscillatory instability, and can be obtained

from general reaction-diffusion equations via centre-manifold reduction [Kuramoto, 1984] or

multiple-scale analysis [Cross & Greenside, 2009]. Not surprisingly, Eq. (2.2) without the

Laplacian coupling is exactly the normal form given by Eq. (2.1), up to a rescaling of the pa-

rameters. The CGLE exhibits a wide variety of behaviours from simple, homogeneous oscilla-

tions to turbulence. For its richness in dynamics and universal quality, the CGLE is one of the

most well studied dynamical models of pattern forming systems [Aranson & Kramer, 2002].

2.2 Chimera States in 1D and 2D

We now discuss examples in the literature which provide evidence for the existence of chimera

states as a general phenomenon in nonlocally coupled systems. We start with the example

of an oscillatory system given by

Z
2
∂t A(r, t) = (1 + ic)A − (1 + ib)|A| A + K(1 + ia) G(r − r′ )[A(r′ , t) − A(r, t)]dr′ , (2.3)

for the complex field A = X + iY , and r ∈ D, a bounded domain in Rm , with m = 1 or

m = 2 typically. On the domain D, we impose either periodic boundary condition or no-flux

boundary condition (i.e. the gradient ∇n A = 0 evaluated on the boundary of D, where n

denotes the outward-pointing normal vector on the boundary of the domain). The method

23
of integration involving these boundary conditions used throughout this thesis, is discussed

in Appendix A.1. Much like the toy model example of reaction-diffusion systems, we can

discretize a model such as Eq. (2.3) by dividing the domain D into discrete points {ri }. In this

way, we can think of the local value of A(ri , t) at each point as an individual oscillator, and the

corresponding equation describes a population of interacting oscillators. This is necessarily

the point of view taken when we simulate systems of partial differential equations, that in

the limit of higher resolution (i.e. large number of oscillators), we expect the discretized

model to recover the behaviour of the continuous model [Cross & Greenside, 2009]. We will

use this concept to describe the systems we are studying throughout.

For its resemblance to the CGLE, we call Eq. (2.3) the nonlocal CGLE. Note that Eq. (2.3)

without the coupling term is also the normal form of Hopf bifurcation, Eq. (2.1), up to a

suitable rescaling. For this reason, we might also think of the nonlocal CGLE as a reduction

of more general models of the form

Z
∂t u(r, t) = F(u) + K G(r − r′ )[u(r′ , t) − u(r, t)]dr′ , (2.4)

where the system ∂t u(r, t) = F(u) is at the onset of oscillation. This fact has been demon-

strated in [Kuramoto, 1997].

We will specify the exact form of G in the following sections, but for now we take it to be

function that depends on r−r′ through the magnitude of the difference |r−r′ |. Additionally,

G(r) is a unimodal, symmetric function that decays to zero at r → ∞, and is normalized such
R
that G(r)dr = 1. Assuming these general properties will suffice for our present discussion.

We can see how the coupling term in Eq. (2.3) fits the description of nonlocal coupling which

we have given in Chapter 1 - the coupling is essentially a convolution of G with A centred at

r, minus the self-coupling coming from A(r). Since the function G decays as |r−r′ | increases,

the oscillator at r is coupled more strongly to an oscillator at r1 than an oscillator at r2 if

|r − r1 | < |r − r2 |. This coupling scheme is also identical everywhere in space, in the same

24
way as global or local coupling we discussed in Chapter 1. Furthermore, if interaction is

absent in the system, i.e. K = 0, the intrinsic frequency of each oscillator is identical, since

the same parameters (b, c) govern the behaviour of each oscillator.

When the system given by Eq. (2.3) is self-oscillatory for a suitable set of parameters

(a, b, c), and K > 0 is sufficiently large, the system evolves to a state of complete mu-

tual synchronization, and we expect to observe either a spatially uniform oscillatory state,

or the presence of waves, similar to the scenario we have described in Section 1.2. How-

ever, if K is lowered past some critical value Kc , the exotic chimera state can appear in

the nonlocal CGLE, depending on the initial condition of the system. We have repro-

duced simulations of the nonlocal CGLE in 1D [Kuramoto & Battogtokh, 2002] and in 2D

[Shima & Kuramoto, 2004], in Figure 2.1 and Figure 2.2, respectively.

Figure 2.1 shows a snapshot of the phase pattern generated from a simulation of Eq. (2.3)

on a 1D ring. Values of the phase at each point in the spatial domain [−1, 1] are defined as

φ(r) = Arg A(r) = tan−1 [Y (r)/X(r)], as discussed in Section 2.1.3. Figure 2.1 shows a sub-

population of oscillators that take on nearly the same phase during evolution, and another

subpopulation that take on phase values distributed in [−π, π). The first subpopulation can

only lag by 2nπ in the unwrapped phase representation (n ∈ Z), hence it is synchronized in

the sense of phase locking, described in Section 2.1.1. Similarly, oscillators in the unsynchro-

nized subpopulation have to take on each phase value in [−π, π) according to a probability

density that is inversely proportional to their velocities [Kuramoto & Battogtokh, 2002].

The subpopulation with phase distributed in [−π, π) therefore corresponds to unsynchro-

nized oscillators. A more revealing representation of the chimera state can be seen in the

bottom panel of Figure 2.1, showing a plot of the long-time averaged frequencies in the pop-

ulation, measured from the evolution of the phase φ(r). We clearly observe a subpopulation

of synchronized oscillators taking on a common frequency (on average), as implied by phase

locking, while a subpopulation of unsynchronized oscillators that take on a distribution of

25
frequencies.

On the other hand, in a 2D system with no-flux boundary condition, Figure 2.2 shows

rigidly rotating spiral wave about a “core” in the centre that shows a discontinuous variation

in real component X(r, t) = Re A(r, t). The behaviour of the oscillators in the core are in

direct contrast to those in the smooth, spatially periodic pattern in the spiral. By definition,

the appearance of a wavelength in the system implies bounded phase differences, which

applies to the oscillators belonging to the spiral wave pattern. The discontinuous variation

of the real and imaginary componenets of A(r, t) in the centre implies that the phase field

φ(r) is distributed in [−π, π). By the same argument as for the 1D case, the oscillators in

the core are considered to be unsynchronized. Again, Figure 2.3 shows a more revealing

presentation by the long-time averaged frequencies, taken in a cross-section intersecting the

core of desynchronized oscillators, and we observe identical behaviour in the frequencies as in

the 1D system. Following after [Martens et al. , 2010], we also refer to the 2D state depicted

in Figure 2.2 a spiral wave chimera.

When a chimera state appears, one important, common feature we observe in both the

1D and 2D system is the spatial localization of the synchronized and unsynchronized sub-

populations. This particular behaviour seems natural if we consider the nonlocal coupling

given by the function G - it is in fact counter-intuitive that the subpopulations should be-

come delocalized in space, since oscillators that are strongly coupled to each other should

come to behave in a similar manner, and these are the oscillators that are closer to each

other. We will see below in Section 2.4.2 that this is exactly the intuition presented by the

self-consistency solution to the phase equation of nonlocally coupled systems, which we will

discuss below in this chapter. In Chapter 3 we will use this localization criterion as an iden-

tifiable feature of chimera states, when we study a nonlocally coupled model with frequency

heterogeneity.

The states shown in Figure 2.1 and Figure 2.2 are generated from particular initial

26
conditions. As an example, initial phase configurations leading to chimera states in 1D

can be generated by modulating a uniform, noisy state by a Gaussian envelope function

[Abrams & Strogatz, 2006]. Within one standard deviation σ of the Gaussian, the noise is

mostly unaffected, but outside of σ the noise is exponentially damped - such configurations

resemble the phase snapshot shown in Figure 2.1. The observation that chimera states can

only be generated from a distinct set of initial conditions is associated with the behaviour

of bistability, which we will discuss below in Section 2.4.

3
2
1
0
φ

-1
-2
-3
1.06
1.04
<ω>

1.02
1.00
0.98
0.96
-1 0 1
x
Figure 2.1: Simulation of nonlocal CGLE, Eq. (2.3), in a population of N = 128 oscillators
arranged on the interval [−1/2, 1/2] with periodic boundary condition on a 2D domain.
Subpopulations of synchronized and unsynchronized oscillators can be clearly seen from the
snapshot of phase φ(x) (black) and the long-time averaged frequencies hωi (red). Parameters
used in the simulation are (a, b, c) = (-1.0, 0.88, 2.0) and K = 0.1.

The presence of chimera states is by no means limited to the nonlocal CGLE model we

have discussed. As demonstrated in [Shima & Kuramoto, 2004], spiral wave chimerae also

27
0 1

10
0.5

20

X(r, t)
0
y

30

−0.5
40

50 −1
0 10 20 30 40 50
x
Figure 2.2: Snapshot of X(r, t) = Re A(r, t) from a simulation of the nonlocal CGLE,
Eq. (2.3), with no-flux boundary condition. Parameters used were (a, b, c) = (0, 0.4, 2.0) and
K = 0.4. A spatial domain of L2 = 50 × 50 is discretized into a lattice of N 2 points, where
N = 512.

exist in a nonlocally coupled 2D system of relaxational oscillators, given by the FitzHugh-

Nagumo (FHN) model (in the oscillatory regime). Relaxational oscillators exhibit alternating

branches of oscillations within a single cycle, where in one branch the phase advances much

faster than the other. Unlike the type of oscillation present in the normal form of the Hopf

bifurcation, Eq. (2.1), the phase and amplitude evolution in the FHN model are strongly

nonlinear, and so the system is far from the onset of the Hopf bifurcation. This is a demon-

stration that chimera states can exist in more general models of oscillators with high degree

of nonlinearity.

28
1.7

1.68

1.66
<ω>

1.64

1.62

1.6
21 22 23 24 25 26 27 28
x
Figure 2.3: Long-time averaged frequencies of the nonlocal CGLE model, at a cross-section
y = 28.8, using same parameters as Figure 2.2. Units on the horizontal axis given in terms
of the real length of the system.

Despite these instances of observation of chimera states in systems of oscillators under-

going simple and relaxational oscillation, as well as knowledge of several sets of parameters

to generate these chimera states in these systems, there is currently no way to determine

whether any given set of parameters (in the oscillatory regime) will lead to the formation of

chimera states. As an example, the nonlocal CGLE simulation in 2D uses a set of parameters

given in [Kuramoto, 2003], which are different from those used in the 1D simulation given in

[Kuramoto & Battogtokh, 2002]. Both of these parameter sets lead to self-sustained oscilla-

tion in the system regardless of the dimensionality of space - but based on our own simulations

of the nonlocal CGLE, the parameters used in 1D do not lead to spiral wave chimerae in the

2D simulation, and vice versa. Similarly, the study in [Shima & Kuramoto, 2004] simulated

the nonlocal FHN model in 2D, and our own simulations found that the parameters given

29
in the study do not lead to a chimera state in a 1D system. Given that only a few sets of

parameters have been described in the literature, this problem of identification of parameters

is highly relevant to the experimental studies of chimera states, but has not been studied ex-

tensively. Hypothetically, an experimental setup of nonlocally coupled oscillators can employ

either a ring (1D) or lattice (2D) configuration, and different types of oscillators, which can

be close or far from the onset of oscillations as in the case of the CGLE or FHN oscillators.

Thus, it is important to have a definite criterion by which suitable sets of parameters can be

selected, leading to formation of chimera states in different dimensions and in different types

of oscillators. In Chapter 4, we will propose and demonstrate a method to select parameters

based on the phase equation.

2.2.1 Derivation of Nonlocal Coupling

To give some insights into how chimera states arise in the weak-coupling limit, and to derive

the mathematical form of nonlocal coupling present in Eq. (2.3), we now examine a simple

model that exemplifies the intuition behind nonlocally coupled systems, based on the example

found in [Shima & Kuramoto, 2004]. Nonlocal coupling can arise from the introduction of

a diffusive, auxiliary field to the original system of equation. Consider the following three-

component system defined on R2

∂t X(r, t) = f (X, Y ) + K(B − X) (2.5)

∂t Y (r, t) = g(X, Y ) (2.6)

τ ∂t B(r, t) = −B + D∇2 B + X. (2.7)

With K = 0, parameters of the system are chosen such that (∂t X, ∂t Y ) = (f, g) undergo

limit cycle oscillations at each position r in the medium. We introduce a third scalar field B

as the diffusive, auxiliary field. In a chemical context, the intuition behind a coupling term

proportional to (B − X) can be thought of as the following: the species B mediates changes

30
in the concentration of X by diffusing to other parts of the reaction medium, whereas the

concentrations of the non-diffusive species X and Y are only modified by reactions described

by f and g implying that the spatial influences of these species are extremely confined. The

variable τ is a characteristic time which measures how long it takes for changes in the local

concentration of B to occur. If τ → 0, the local concentration of B changes extremely fast

in comparison to X and Y . This leads to a separation of time scale, so the component B can

be adiabatically eliminated from Eq. (2.5) by solving the equation −B + D∇2 B + X = 0.

This gives

Z
B(r, t) = G(r − r′ )X(r′ , t)dr′ , (2.8)

where in 2D, the Green function G is given by a modified Bessel function of the second kind,

properly normalized:

1
G(r) = K0 (r/r0 ), r = |r − r′ |. (2.9)
2πr02
We note that in 1D, the Green function solution to −B + D∇2 B + X = 0 takes the much

simpler form of a normalized exponential, G(r) ∝ e−κr . In either case, a characteristic

length can be identified in the Green function - r0 in the 2D solution and κ−1 in the 1D

solution. The characteristic length corresponds to a spatial distance, outside of which the

function G vanishes much more rapidly. In the remainder of this chapter, whenever the

Green function G appears in an equation or is referred to in the discussion, we always

refer to the solution in the appropriate spatial dimension, found by solving Eq. (2.7) in the

limit τ → 0. The use of the term ”nonlocal” to describe the coupling introduced in this

section can be evidently interpreted as the presence of a large separation of timescale. As

suggested by Kuramoto, the nonlocality is “effectively” introduced in this manner into the

model [Shima & Kuramoto, 2004].

When a spiral wave is present in the system described by Eqs. (2.5), (2.6) and (2.8),

31
there are two parameters which dictate the wavelength of the resulting pattern. On one

hand, the strength of the coupling K scales the wavelength of the pattern, which may be

compared to the characteristic coupling range r0 = D in the Green function, Eq. (2.9),

which is another spatial scale present in the system. In the long wavelength limit, when K

is large, the coupling K(B − X) may be approximated by diffusion coupling D̃∇2 X with a

new diffusion constant D̃ = KD. On the other hand, if K < Kc is sufficiently weak, then the

diffusion approximation described above is no longer valid, and the appropriate behaviour is

described by the nonlocally coupled system. In this limit, variations in the wavelength below

the coupling range r0 begin to appear, which leads to a pattern such as the one depicted in

Figure 2.2 [Kuramoto, 2003] - the spatially discontinuous core is a consequence of this weak

coupling.

To obtain the nonlocal CGLE, Eq. (2.3), for example, we start with the following system,

given by Eq. (2.10). A diffusive field B governed by Eq. (2.11) is coupled to it in the same

way as in Eq. (2.7)

∂t A(r, t) = (1 + ic)A − (1 + ib)|A|2 A + K(1 + ia)(B − A) (2.10)

τ ∂t B(r, t) = −B + D∇2 B + A, (2.11)

where B is also complex. Again, in the absense of the coupling term, Eq. (2.10) is (up to a

rescaling) the normal form of the Hopf bifurcation given in Eq. (2.1). As in the adiabatic

elimination, we let τ → 0. By linearity, the real and imaginary components of B both satisfy

Eq. (2.11). Inserting the Green function into Eq. (2.10), we get the desired result, Eq. (2.3).

While we have obtained the nonlocally coupled CGLE model, Eq. (2.3) is difficult to

analyze mathematically. We would like to somehow extract intuition from it and study the

synchronization behaviour of the system, since its self-oscillation resembles the normal form

of the Hopf bifurcation, and so should be a universal model for other systems of coupled

oscillators close to the onset of oscillation. A procedure called the phase reduction can be

32
performed in the limit of K ≪ 1, when the amplitude degree of freedom in the system can be

neglected. It turns out that the mathematical description of phase dynamics is analytically

manageable and grants us insights into the behaviour of synchronization of weakly-coupled

oscillators. In the next section, we review the mathematical basis of the phase reduction.

2.3 Phase Model

The goal in this section is to motivate the perturbative method that reduces Eq. (2.3) to a

coupled equation involving only the dynamics of a phase variable. This perturbative method

was first used to derive the Kuramoto model, which is a model of globally coupled limit cycle

oscillators and describes only the evolution of the phase. This method of phase reduction is

valid under the assumption of weak coupling. The derivation and analytical treatment of the

Kuramoto model, which we study below in the next sections, will also illustrate the intuition

behind synchronization phenomena in populations of interacting limit cycle oscillators, as

was first pointed out in [Winfree, 1967]. The ideas introduced in the treatment of the

Kuramoto model can be used to study the phase model reduced from general, nonlocally

coupled systems, which we will discuss in Section 2.4.1.

2.3.1 Kuramoto Model

In his classic text, Yoshiki Kuramoto studied a system of globally coupled limit cycle oscilla-

tors, now widely known as the Kuramoto model [Kuramoto, 1984]. We denote by the set of

equations dXi /dt = F(Xi ), for i = 1, ..., N, corresponding to the uncoupled motions of the

N oscillators in the population. Thus, for each i, the system has a periodic solution Xip ∈ Rn .

Under sufficiently weak coupling, the dissipative effect of attraction to the limit cycle quickly

stabilizes any changes to the amplitude of the oscillation. Thus, the evolution of the system

remains close to the unperturbed limit cycle, and the problem now involves the consistent

definition of a phase in the neighbourhood of the “true” closed, periodic trajectory.

33
We discuss the problem first for a single oscillator perturbed by a small external source,

and then generalize the formulation to a mutually interacting population. Let C denote the

limit cycle of a system of differential equations dX/dt = F(X), i.e. C = {Xp (t)|t ∈ [0, T )},

where T is the period of the solution Xp such that Xp (t) = Xp (t + T ). To each X ∈ C

we assign a value φ(X), the true phase of the limit cycle, defined so that dφ/dt = 1. If we

introduce a small perturbation to the system as dX/dt = F(X)+ǫp(X), for some value ǫ ≪ 1

indicating the smallness of disturbance characterized by p, the solution no longer exactly

following C as in the unperturbed system. Instead, given the smallness of ǫ, the perturbed

solution will undergo periodic motion in the vicinity of C. A way to consistently define

the notion of phase on C, as well as in a tubular neighbourhood G ⊃ C is by the notion of

isochrons, a one-parameter family of curves on which the phase φ is constant [Winfree, 1980].

Let X ∈ C and X′ ∈
/ C, where X′ represents the perturbed solution. Let both X and X′ be

on the same isochron I(φ), then φ(X) = φ(X′ ). When X(t) goes around C, so does X′ (t),

in such a way that φ(X(t)) = φ(X′ (t)) for all t. This means that these periodic trajectories

have the same period. Additionally, the motion of X′ is also described by dX′ /dt = F(X′ ),

since X and X′ belong to the same isochron. We can then write down the following equation

on G, for a general X ∈ G.

dφ(X) dX
= ∇X φ ·
dt dt
= ∇X φ · [F(X) + ǫp(X)]

= 1 + ǫ∇X φ · p(X), (2.12)

which describes the change in phase at a general point in the neighbourhood of C. We used

the identity dφ(X)/dt = 1 = ∇X φ · F(X) for X ∈ G, having established the fact that the

free evolution of X ∈ G is also given by F.

The value of φ alone is insufficient for locating X along the isochron of φ, but the equation

above can be closed by substituting Xp ∈ C for X in the intersection of C and I(φ), in the

34
limit ǫ ≪ 1 when |X − Xp | is small. Thus


≃ 1 + ǫΩ(φ), (2.13)
dt

where

Ω(φ) = Z(φ) · p(Xp (φ)) (2.14)

Z(φ) = (∇X φ)|X=Xp . (2.15)

One can then apply a change of variable φ = t + ψ to Eq. (2.13), where ψ represents the

deviation to the original phase variable φ due to the perturbation. The change of variable

leads to


= ǫΩ(t + ψ). (2.16)
dt
Note that ǫ ≪ 1 means ψ is a “slow” variable, changing slowly over the course of one period

of oscillation. Hence, we can approximate the instantaneous value of dt ψ by hdt ψi over this

time scale, where the angular braces denote taking the time average. As such, we take the

time average of Eq. (2.16) over one period. Furthermore, we set ǫ = 1, which is equivalent

to redefining time as t → t′ = ǫt. We obtain

T
1
Z
ω = Ω(t′ )dt′ (2.17)
T 0

= ω. (2.18)
dt

Using the formalism above, we can now consider a population of N mutually interacting

oscillators. Let our system be expressed in the form

dXi
= F(Xi ) + ǫpi (2.19)
dt
X
pi = Vij (Xi , Xj ) + fi (Xi ), (2.20)
i6=j

35
for i = 1, ..., N. Note that the vectors pi contain a sum of all pairwise interactions given by

the potentials Vij , and functions fi which expresses the non-identical nature of the oscillators

in the population. Otherwise analogous to the one oscillator equation, there is now a sum

appearing on the right hand side

" #
dφi X
=1+ǫ Z(φi )Vij (φi , φj ) + gi (φi ) , (2.21)
dt j

where gi (φi ) = Z(φi )fi (Xp (φi )). We now average over one period the quantity in square

brackets and set ǫ = 1 as before. Subtituting φi = t + ψi we obtain

dψi X
= ωi + Γij (ψi − ψj ) (2.22)
dt j
Z T
1
Γij (ψi − ψj ) = Z(t + ψi )Vij (t + ψi , t + ψj )dt (2.23)
T 0
1 T
Z
ωi = gi (t + ψi )dt. (2.24)
T 0

Furthermore, Kuramoto made the following assumptions on Γij to obtain a mathematically

treatable model: i) the form of the function Γij ≡ Γ is identical for every pair of oscillators i

and j, and ii) Γ(ψi − ψj ) is simply a sinusoidal function of the phase difference ψi − ψj . The

function Γ is called the phase response curve and will play an important role in Chapter 3

and Chapter 4. While the first assumption can be thought of as a mean-field approximation,

the second approximation can be motivated by the fact that Γ is a periodic function in both

φi and φj . The function Γ(φi − φj ) is given by

σ
Γij (ψi − ψj ) = − sin(ψi − ψj ), (2.25)
N
where K > 0, this yields the Kuramoto model of globally coupled phase oscillators

N
dψi σ X
= ωi − sin(ψi − ψj ), (2.26)
dt N j=1

36
where the frequencies ωi are distributed according to a density g(ω), which is taken to be

symmetric and unimodal. Note that the smallness factor ǫ which appears in Eq. (2.19) plays

the role of K in in nonlocal CGLE model, Eq. (2.3). As we shall see in Section 2.4.1, it is

the smallness of K which allows these models to be reduced to a description only in terms

of phase dynamics. The factor σ appearing in the Kuramoto model, on the other hand, can

be thought of as the amplitude of the phase response function Γ (i.e. the magnitude of the

frequency response), but it is introduced into the model by hand rather than the consequence

of an exact analytic result.

Finally, we note that Eq. (2.26) is given in terms of the phase deviations ψi , which we

can easily transform back into the set of of original phases φi = t + ψi by substitution and

redefining ωi′ = 1 + ωi as the intrinsic frequencies and using ǫ to make the transformation

back to the initial time variable. Hence there is no loss of generality in the form that appears

in Eq. (2.26), and in refering to ψi as equivalent description of the phase of each oscillator.

2.3.2 Characterization of Synchronization in the Kuramoto Model

We now analyze the Kuramoto model to reveal insight into its synchronization behaviours.

We follow the derivation given in [Acebron et al. , 2005], which provides a more rigorous

argument by introducing the continuity equation, in comparison to the original text by

Kuramoto. We begin by defining the complex order parameter

N Z π N
!
1 X iψj 1 X
reiΘ = e = eiψ δ(ψ − ψj ) dψ. (2.27)
N j=1 −π N j=1

Interpreting eiψj as a unit vector in the plane, Eq. (2.27) becomes an averaged vectorial sum,

and its magnitude r measures how coherently the phases evolve on the unit circle. If the

phases ψj are close to each other during evolution, then r(t) ∼ 1 and otherwise r(t) ∼ 0.

Although the complex exponential wraps the phase into [−π, π), synchronization in the sense

of phase locking can still be observed: if a pair of oscillators have the same value of phase

37
wrapped into [−π, π) and during the evolution we consistently observe a zero (wrapped)

phase difference, then in the unwrapped representation these oscillators have on average a

phase difference of 2nπ where n is an integer. This fits the definition of phase locking since

the unwrapped phase difference of these oscillators do no increase monotonically in time.

Inserting Eq. (2.27) into Eq. (2.26), we get

dψi
= ωi + σr sin(θ − ψi ), (2.28)
dt
which is the one-oscillator equation where each oscillator is now coupled to the others via

the quantity r and θ. This can be integrated to yield the phases {ψi }, which themselves

determine r through Eq. (2.27) and must be consistent with the value r that appears in the

above Eq. (2.28). In the limit of infinitely many oscillators, we rewrite Eq. (2.27) as

Z π Z ∞

re = dψ dω eiψ ρ(ψ, ω, t)g(ω), (2.29)
−π −∞

as the sum of delta functions N −1 N


P
i=1 δ(ψ − ψi ) becomes replaced by a continuous distri-
R
bution ρ̃(ψ, t) = ρ(ω, ψ, t)g(ω)dω. Note that we consider ψ ∈ [−π, π) as a wrapped phase.

We can now formulate Eq. (2.28) as a continuity equation describing the evolution of the

density ρ

∂ρ ∂
+ {[ω + σr sin(Θ − ψ)]ρ} = 0, (2.30)
∂t ∂ψ

with the constraint that −π
ρ(ψ, ω, t)dψ = 1. We find some solutions which satisfy ∂t ρ = 0,

corresponding to stationarity of the density distribution of the phases. Assuming that σ > 0,

we have a non-trivial solution in setting ∂t ψ = ω + σr sin(Θ − ψ) = 0, corresponding to a

fixed point ψ ∗ which exists when |ω| < σr and −π/2 ≤ ψ − Θ ≤ π/2, i.e. the phases are

bounded with respect to each other. If |ω| > σr, then ∂t ψ 6= 0, so we have a trivial solution

for the density ρ = (2π)−1, satisfying the normalization constraint, together with r = 0.

This implies that the oscillators take on all phase values in [−π, π) equally likely, and such

38
a scenario means that the unwrapped phase differences between a pair of oscillators cannot

be bounded.

We do not proceed further into the derivation, which eventually yields the critical value

σc such that the completely desynchronized state becomes linearly unstable when σ > σc

[Acebron et al. , 2005]. We look at the implication of the criteria of entrainment given by ω,

σ and r. If the intrinsic frequencies of the population is identical, ωi = ω, then all oscillators

follow either |ω| < σr or |ω| > σr. These two cases correspond to mutual synchronization in

the system, or a lack thereof. In the case of identical oscillators, we can only then observe

asymptotic states with values of r ∼ 1 or r ∼ 0. If the frequencies in the population are

distributed, however, then both |ω| < σr or |ω| > σr can be satisfied in the population - this

means that the population is partially-synchronized. Unlike in the chimera states described

in Section 2.2, which are partially-synchronized states of a system of identically coupled,

identical oscillators, the partially-synchronized states in the Kuramoto model typically do

not have the feature that the synchronized and unsynchronized subpopulations of the system

are localized in space. In summary, the Kuramoto model puts Winfree’s intuition of syn-

chronization behaviour in a population of globally coupled oscillators on solid, mathematical

ground. For its analytical simplicity, one can easily introduce variations into the Kuramoto

model to include time delay [Yeung & Strogatz, 1999], noise as well as varying connectivity

in the coupling topology [Acebron et al. , 2005].

2.4 The Phase Equation for Nonlocally Coupled Systems

2.4.1 Derivation of the Phase Equation

The method developed to treat globally coupled oscillators in Section 2.3 can be modified to

study nonlocally coupled systems of identical oscillators. In this section, we use the nonlocal

CGLE, Eq. (2.3) as an example, and obtain from it a coupled set of equations involving the

dynamics of the phase only. As mentioned previously, such procedure is amenable when the

39
coupling is weak, which is exactly the regime in which chimera states appear in nonlocally

coupled systems.

To obtain the reduced phase equation, we need to first compute the phase response curve

following the procedure in Section 2.3. In the nonlocal CGLE model, the phase response

curve Γ is given by

Γ(φ − φ′ ) = Z(φ) · p(φ′ , φ), (2.31)

since the product Z(φ) · p(φ′ , φ) turns out to be a function of the phase difference φ − φ′

already, there is no need to perform the average over one period. The primed and unprimed

variables indciate evaluation of the scalar field at r′ and r, respectively, and p is given by

the real and imaginary components of the difference A(r′ , t) − A(r, t)

 
′ ′
X − X − a(Y − Y )
p= . (2.32)
′ ′
Y − Y + a(X − X)
Since p is to be evaluated on the limit cycle, we substitute X = cos φ and Y = sin φ

 
′ ′
cos φ − cos φ − a(sin φ − sin φ)
p(φ′ , φ) =  . (2.33)
′ ′
sin φ − sin φ + a(cos φ − cos φ)

The vector Z(φ) can be computed according to [Kuramoto, 1984], based on Floquet the-

ory, a method for solving systems or ordinary differential equations with periodic coefficients.

We simply give the result

Z(φ) = ω −1(−b cos φ − sin φ, −b sin φ + cos φ), ω = c − b. (2.34)

Multiplying Z with p, we obtain

Γ[φ(r, t) − φ(r′ , t)] = −ω −1 [(a − b)(1 − cos(φ(r, t) − φ(r′ , t))) + (1 + ab) sin(φ(r, t) − φ(r′ , t))].

(2.35)

40
Eq. (2.35) represents the frequency change as a function of a pair of oscillators with difference

φ − φ′ in their phases. We have to also consider the fact that each oscillator φ(r) is affected

by an integral of Γ[φ(r) − φ(r′ )], weighted by the Green function G(r′ − r) that appears in

the nonlocally coupled system. Hence, for a system of identical oscillators with a common

intrinsic frequency ω, we obtain the following equation:

Z
∂t φ(r, t) = ω + G(r − r′ )Γ[φ(r, t) − φ(r′ , t)]dr′ , (2.36)

Even if Γ is not known analytically, we know that it is a periodic function in both φ and

φ′ . As such, we can expand it in terms of a Fourier series. If Γ is “suitably sinusoidal”

[Kuramoto, 1997, Kuramoto & Shima, 2003, Shima & Kuramoto, 2004] and does not con-

tain higher order harmonics, we can take the same approach as in the Kuramoto model and

assume that it is given by a sinusoidal function

Z
∂t φ(r, t) = ω − G(r − r′ ) sin[φ(r, t) − φ(r′ , t) + α]dr′ . (2.37)

We refer to Eq. (2.37) as the phase equation, a simplified phase description of the nonlocally

coupled system. The only part of Eq. (2.37) which depends on the dimensionality of space

is the Green function G, which is given according to Section 2.2.1. If the Fourier expansion

of Γ contains only the lowest-order terms proportional to sin(φ − φ′ ) and cos(φ − φ′ ), then

we can expect Eq. (2.37) to be a good description of the system’s behaviours in the weak-

coupling limit. Note that the presence of α indicates exactly this simplification - expanding

− sin[φ(r, t) − φ(r′ , t) + α] using trigonometric identities, we get a linear combination of

sin[φ(r, t) − φ(r′ , t)] and cos[φ(r, t) − φ(r′ , t)], with coefficients given as sinusoidal functions

of α.

41
2.4.2 Analytical Treatment of the Phase Equation

There exist numerous analytical results on the phase equation, the most classical being

the self-consistency criteria for the phase equation [Kuramoto & Battogtokh, 2002], which

is quite similar to the procedure described in Section 2.3.2 for the Kuramoto model. The

procedure we describe below is valid for the phase equation in both 1D and 2D

We first define a complex order parameter

Z

W (r, t) ≡ R(r, t)eiΘ(r,t)
= G(r − r′ )eiφ(r ,t) dr′ , (2.38)

Eq. (2.38) is just a sum of the phases weighted by the function G. Assuming a chimera state

has formed as depicted in Figure 2.1 or Figure 2.2, let the origin of the coordinate system

r = 0 be located at |W (r)| = 0 (which should coincide with the centre of the unsynchronized

subpopulation shown in these figures). Inserting Eq. (2.38) into Eq. (2.37) yields an equation

∂φ(r, t)
= ω − R(r, t) sin[φ(r, t) + α − Θ(r, t)]. (2.39)
∂t
At this point, symmetry considerations can be made for the mean field variables R and Θ

in the chimera states. Let Ω be the constant rotation frequency of the mean field modulus

R(r, t). The value Ω has the interpretation of the common frequency taken by the synchro-

nized subpopulation in the 1D chimera state, or the rotation frequency of the spiral wave

in the 2D chimera state, which we assume to exist in seeking a time-independent (in the

statistical sense) solution. Note that by definition of the complex order parameter W , R is

radially symmetric (since G is) as well as constant along the boundary of disks with radius

r. In addition, Θ(r, t) = Ωt + Θ0 (r, t) where Θ0 expresses the difference between the steady

rotation at a frequency Ω and the actual rotation of the mean field phase Θ. Note that

under the assumption of steady rotation, the quantity Θ0 (r, t) is also time-independent, but

not space-independent, so the dependence of Θ(r, t) on space is through the spatial profile

Θ0 (r).

42
Introducing a new phase variable ψ(r, t) = φ(r, t) −Ωt and applying these considerations,

we get

∂ψ(r, t)
= ω − Ω − R(r, t) sin[ψ(r, t) + α − Θ0 (r)], (2.40)
∂t
in which case, the definition of the order parameter becomes
Z
iΘ0 (r,t) ′
R(r, t)e = G(r − r′ )eiψ(r ,t) dr′ . (2.41)

Eq. (2.40) expresses the evolution of the phase variable from the rotating frame of reference

of frequency Ω, driven by the mean field magnitude R. The vanishing of the mean field

towards r = 0 implies the appearance of two groups of oscillators, one synchronized by the

central force of the mean field, and another one which fails to do so. This is revealed by

further analyzing the fixed points of Eq. (2.40). For different values of ω, Ω and R(r, t), we

see that ∂t ψ(r, t) has a pair of stable and unstable fixed points in the cases |ω − Ω| < R(r, t).

In this case, the solution flows to the stable fixed point ψ0 (r) found by implicitly solving

Eq. (2.40), meaning the phase difference between the phase field and that of the co-rotating

frame is bounded. Groups of oscillators satisfying this condition are phase synchronized and

rotate with a frequency identical to Ω. Otherwise, when |ω − Ω| > R(r, t) we find that

∂t ψ(r, t) 6= 0, so no fixed point in the co-rotating frame is possible. The phase difference

ψ(r) between the steady co-rotating frame and the phase field φ(r) becomes unbounded.

Groups of oscillators satisfying this condition are unsynchronized.

Although we will not study in the full detail the self consistency criteria, we describe

it briefly. The remainder of the method involves the assumption of time independence

of R and Θ and replacing the phase ψ on right hand side of Eq. (2.41) by the average

in time, hψi. Assuming ergodicity, the average can be replaced by an ensemble average

using a probability density p(ψ), defined separately for oscillators that satisfy |ω − Ω| <

R(r, t) or |ω − Ω| > R(r, t). This yields the self consistency criteria and the functions R

and Θ can be numerically calculated. The numerically obtained values of R and Θ were

43
found to be in excellent agreement with those obtained directly via simulations of Eq. (2.37)

[Shima & Kuramoto, 2004].

The synchronization criteria described in this section characterize and predict the exis-

tence of a partially-synchronized state, and necessarily lead to the chimera state patterns seen

in Figure 2.1 and Figure 2.2 where the subpopulations of synchronized and unsynchronized

oscillators are spatially localized. The reasoning behind this can be seen from Eq. (2.40),

which has the interpretation of a single oscillator being driving by a periodic driving force.

At first, a single oscillator at r = r0 appears to become uncoupled from the rest of the

population, where r0 is the point such that R(r0 , t) = 0. Oscillators near r0 on the other

hand, feel a small amount of force, but they can’t be synchronized since |ω − Ω| > R(r, t),

failing the synchronization criteria. This leads to the formation of an unsynchronized sub-

population. In particular, the spatial localization of the synchronized and unsynchronized

subpopulations is maintained since the amplitude of the force R(r, t) is radially symmetric

and depends on the phase field ψ(r) through Eq. (2.38). This can be seen from the fact that

oscillators in the unsynchronized subpopulation are driven by small values of R(r, t), and to

compute R(r, t) one averages the phases of all oscillators, but nearby oscillators are weighted

more strongly by the Green function. Since oscillators in the unsynchronized subpopulation

observe almost all phase values in [−π, π), the value R(r, t) computed for these oscillators

will be consistently small - hence they will remain unsynchronized. The same argument

applies to oscillators in the synchronized subpopulation, except R(r, t) is consistently large

in this subpopulation, since the phases of these oscillators varies smoothly.

2.4.3 Simulation of the Phase Equation

Simulations of the phase equation in 1D and 2D have been performed using 0 < α < π/2

as the control parameter [Kuramoto & Battogtokh, 2002, Martens et al. , 2010], and results

similar to those of Figure 2.1 and Figure 2.2 have been observed. In the phase equation,

chimera states only appear for and α →− π/2 in 1D and α →+ 0 in 2D, though the exact

44
mechanism that lead to this behaviour remains elusive [Martens et al. , 2010]. It has recently

been shown in numerical simulations of the 2D phase equation, that chimera state patterns

with symmetries different from Figure 2.2 can be observed by controlling the effective radius

of the Green function G as well as α [Omel’chenko et al. , 2012]. Their observations suggest

that the system has very strong dependence on initial conditions.

A useful observation made in [Martens et al. , 2010] is the fact that formation of chimera

states does not depend on the Green function used in the simulation. To study the dy-

namical properties of chimera states, we do not necessarily need to use the Green functions

obtained from adiabatic elimination of the auxiliary diffusive components, as discussed in

Section. 2.2.1. In the 2D simulations of Eq. (2.37), spiral wave chimerae were observed for

G(r) taking the forms of an exponential, a Gaussian or a step function [Martens et al. , 2010].

As long as the chosen function G(r) is unimodal, symmetric and decays to zero at infinity,

it captures the essential behaviour of nonlocal coupling. The freedom to choose the ac-

tual form of G(r) has also be exploited in several analytical studies [Abrams et al. , 2008,

Martens et al. , 2010], in particular in the derivation of the phase diagram of the 1D phase

equation, discussed in [Abrams & Strogatz, 2004, Laing, 2009a], which we will discuss below

in Section 2.5.2. We will exploit the same convenience when we define the function G(r) for

our models in Chapters 3 and 4.

2.5 Results on the 1D Phase Equation

Having established the main theoretical results on nonlocally coupled systems and the phe-

nomenology of chimera states, we discuss some results on the phase equation in 1D that are

pertinent to the discussion of our experimentally-motivated model in Chapter 3.

45
2.5.1 Chimera State Solutions in |α| < π

We would like to remark on our own simulations of the phase equation in a 1D ring. Figure 2.4

shows that there exist solutions on the whole interval −π < α < π, corresponding to different

synchronization behaviours. The vertical axis of Figuire 2.4 shows long-time averaged values

of r given by Eq. (2.27), defined by


M N
1 1 X X iφj (tk )
hri = · e , (2.42)
T N k=1 j=1

as a function of different values of α, measured during the time interval of length T = tM −t1 .

Although the order parameter r was defined for the Kuramoto model, we can use it generally

as a measure of synchronization in 1D systems of oscillators. We measure its long-time

average instead as an approximation to the asymptotic value taken by a stable solution of the

phase equation. In Figure 2.4, we integrate the 1D phase equation with periodic boundary

condition for different values of α ∈ (−π, π) and measure the corresponding asymptotic

values of hri. We see regions of α corresponding to different synchronization behaviours. A

value of hri = 1 indicate that the phases φj are tightly bunched on the unit circle during

time evolution, and the opposite is true if hri = 0. These two values of hri correspond to

the population of oscillators being mutually synchronized, or mutually unsynchronized.

At approximately values 0.6 < hri < 0.8 we observe an aggregation of chimera state

solutions, near α ∼ ±π/2, which we have indicated in Figure 2.4 as blue symbols. States

that have lower values of hri < 0.5 correspond to unstable, “modulated drift” states at the

onset of appearance of spatial structure [Abrams & Strogatz, 2006]. These are indicated by

the red symbols in the same figure. On the branch α > 0, the chimera states are similar to

the one depicted in Figure 2.1. On the other hand, for α < 0, the chimera states behave

different - the synchronized subpopulation now take on a common frequency faster than those

of the unsynchronized subpopulation. These behaviours are revealed in the plot of the phase

snapshots and long-time averaged frequencies in Figure 2.5, for α = ±1.51. The relevance

46
of this result will arise in Chapter 3 when we discuss the model for our experimental system

of electrochemical oscillators.

0.8

0.6
<r>

0.4

0.2

0
-3 -2 -1 0 1 2 3
α
Figure 2.4: Simulation of the phase equation, Eq. (2.37) over the range of |α| < π/2.
Blue diamonds (X’s) indicate chimera states in which the synchronized subpopulation is
faster (slower) than the desynchronized subpopulation, while red diamonds (X’s) indicate
modulated drift states.

2.5.2 Bistability of Solutions

In [Abrams & Strogatz, 2004, Laing, 2009a], an exactly solvable case of the 1D phase equa-

tion, Eq. (2.37) involving a Green function of the form G(x) = (2π)−1 (1 + A cos x) was

introduced. The form of the Green function enabled the analysis of the phase diagram

of the system, from which the stability of chimera states, as well as other solutions corre-

sponding to complete mutual synchronization can be inferred. The beauty of the proce-

dure resulted in reducing the difficult task of solving for the two unknown functions R(x)

47
3 3
2 2
1 1
0 0
φ

-1 -1
-2 -2
-3 -3
1.6
1.5 2.5
1.4 2.4
<ω>

1.3 2.3
1.2
2.2
1.1
1 2.1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
x x
Figure 2.5: Plots of the phase snapshots and long-time averaged frequencies, computed
from simulations of the phase equation on a 1D ring, using N = 256 oscillators, for α = 1.51
(left column) and α = −1.51 (right column), showing different chimera state solutions.

and Θ(x) of the self-consistency criteria to solving for two unknown complex coefficients

[Abrams & Strogatz, 2006]. This is done by making an ansatz for the order parameter, of

the form R(x)eiΘ(x) = c + a cos x, where a and c are complex numbers. The assumption is

based on the observation that the spatial profile of R(x) corresponding to chimera states

resembles a sinusoidal function. The real part of a can be used as an order parameter of the

system. If Re {a} = 0, then R(x) is spatially uniform, and and the system is in a completely

synchronized state. Otherwise if Re {a} > 0, then spatial structure is present in R(x), as

required by the entrainment criteria discussed in Section 2.4.2, implying the emergence of a

chimera state.

From numerical simulations, one can infer that the stable chimera state is born from

48
a pitchfork bifurcation, and on the other hand destroyed in collision with the unstable

chimera state via saddle node bifurcation. From the bifurcation diagram of Re {a} versus

the phase lag β ≡ π/2 − α as the control parameter, it can be shown that there is a

region of β corresponding to bistability of solutions, where stable, completely synchronized

states coexists with stable chimera states [Laing, 2009a]. In some cases, the stability of the

chimera state is currently under debate. A recent study showed that in some cases chimera

states only exist as chaotic transients in a small, finite system, and these transients persist

for an amount of time proportional to the size of the system [Omel’chenko et al. , 2010,

Wolfrum & Omel’chenko, 2011]. This observation would then imply that in the limit of

infinite system size, chimera states exist as extremely long-lived transients, rather than

stable fixed points of the dynamics.

As stated in [Motter, 2010], the bistability regime observed in nonlocally coupled systems

implies that the final state depends on the initial conditions chosen. This bistability can be

an indicator that chimera states exist as solutions that are reached from a set of initial

conditions, different from those that lead to completely synchronized states. We will make

use of this observation in Chapter 3 to study chimera states in the time-delayed, nonlocally

coupled model of oscillators.

2.5.3 Robustness With Respect to Frequency Heterogeneity

Using the analytical method of Ott and Antonsen [Ott & Antonsen, 2008], Laing gave an

analytical confirmation of the stability results of [Abrams & Strogatz, 2006]. In addition,

Laing also extended their results to system with frequency heterogeneity, in which the fre-

quencies ω = ω(x) of the oscillator at a position x ∈ [0, 2π] are distributed according to a

Lorentzian function g(ω) = D/[π(ω 2 + D 2 )] [Laing, 2009b]. In the absence of heterogeneity,

49
the results of [Abrams & Strogatz, 2006] are recovered. Laing’s model is given by


dφ(x, t)
Z
= ω(x) − G(x − y) cos[φ(x, t) − φ(y, t) − β]dy, (2.43)
dt 0

where β = π/2−α, so that the definition of Eq. (2.43) is identical to Eq. (2.37). Additionally,

the function G was chosen to be the same as in the study in [Abrams & Strogatz, 2006].

Using the full width at half maximum D of the Lorentzian function (which is a measure of

heterogeneity in the model) as a bifurcation parameter, a cross-section of the phase diagram

was examined. At a value of β = 0.15 where the completely synchronized state is stable at

D = 0, Laing slowly increased D to reach different stability regions of the phase diagram. He

found that at an intermediate amount of heterogeneity, D = 0.02, the chimera state persists,

but only “noisy” synchronized states exist at higher heterogeneity D = 0.06 [Laing, 2009a,

Laing, 2009b] - a somehow surprising result, since one might expect a larger amount of

heterogeneity will disrupt synchronization further. Even in the presence of heterogeneity,

there is a range of values of β for which the chimera state solution and the (noisy) completely

synchronized solutions are both stable. However, for other values of β, other scenarios can

also be observed, in which the only stable solution is the chimera state, or the (noisy)

completely synchronized state.

2.5.4 Chimera States in Systems with Time Delay

Chimera states have also been observed in systems that are fundamentally different from the

ones we have discussed so far. [Omel’chenko et al. , 2008] studied a globally coupled model

of Stuart-Landau oscillators with a time-delayed stimulus on the domain [−1, 1] of the form

∂W
= (1 + iω − |W |2)W + C(W̄ (t) − W ) + S(x, t), (2.44)
∂t
R1
where W is a complex field and W̄ = 21 −1 W (x, t)dx is the “mean-field” linearly coupled

in the system. W̄ is then delivered with a time delay in the stimulus given by S(x, t) =

50
Kρ(x)W̄ (t − τ ). The function ρ(x) ∝ e−ax is properly normalized. The corresponding phase

equation is given by [Tass & Haken, 1996]

∂ψ C 1
Z
= ω− sin[ψ(x, t) − ψ(x′ , t)]dx′
∂t 2 −1
K 1
Z
− ρ(x) sin[ψ(x, t) − ψ(x′ , t − τ )]dx′ . (2.45)
2 −1
Numerical simulations of this equation for C, K ≪ 1 exhibited chimera states at differ-

ent mutually exclusive ranges of time delays [τ1 , τ1 + ǫ1 ), (τ2 − ǫ2 , τ2 ] ⊂ [0, T ], where T is

the period of the uncoupled system and τ1 < τ2 . At τ1 (τ2 ), chimera states in which the

synchronized population takes on a slower (faster) average frequency than the randomized

population. Outside these ranges, in the intervals [0, τ1 ) and (τ2 , T ], the system is completely

synchronized, whereas in the interval (τ1 + ǫ1 , τ2 − ǫ2 ) there exist completely desynchronized

states as well as unstable chimera states.

Note the coupling in the model in Eq. (2.45) is different from what had previously been

described for nonlocally coupled systems in Section 2.2.1. The stimulation profile ρ(x) ∝ e−ax

in Eq. (2.45) is a function of space rather than a function of relative spatial distance (as in the

Green function G(r) appearing in nonlocally coupled systems), meaning that the coupling

in the model of [Omel’chenko et al. , 2008] is non-identical, since the time-delayed coupling
R1
term proportional to −1 sin[ψ(x, t) − ψ(x′ , t − τ )]dx′ is multiplied by ρ(x).

2.5.5 Other Coupling Topologies

A discrete variant of the phase equation has been studied in [Abrams et al. , 2008] has a

different coupling topology than what we have been discussing so far. The model is given by

2 N
∂φni X Kmn X
=ω+ sin(φm n
j − φi + α), (2.46)
dt m=1
N j=1

which consists of two groups of globally coupled oscillators, each containing N mutually

coupled oscillators. The first N oscillators belong to the first group and the remainder

51
belong to the second. As we can see from Eq. (2.46), each oscillator is coupled to the N − 1

oscillators in its own group by a coupling strength Knm , where n = m, and coupled to all

N oscillators in the other group by a coupling strength Kmn , where n 6= m. The coupling

parameters are defined such that K11 = K22 = µ and K12 = K21 = ν, where µ > ν and

µ+ν = 1. This means that oscillators belonging to the same group are more strongly coupled

to each other in comparison to oscillators in different groups. By changing the discrepancy

between these coupling strengths, chimera states can be observed in the overall population

of 2N oscillators, where all oscillators in one group become mutually synchronized, while

those in the adjacent group become unsynchronized [Abrams et al. , 2008]. Although it is

a more restrictive realization of nonlocal coupling since it involves assigning the oscillators

into two populations a priori, the coupling topology of Eq. (2.46) is easy to implement and

has inspired a recently reported experimental demonstration of chimera states in a chemical

system using coupled relaxational oscillators [Tinsley et al. , 2012]. Furthermore, the results

derived for this model does not depend on the spatial dimensionality of the system. Unlike

Eq. (2.46), the nonlocal coupling we have described in this chapter leads to typical patterns

observed in Figure 2.1 and Figure 2.2, which certainly exhibit dependence on dimensionality.

While the experiment described in [Tinsley et al. , 2012] is an important first step, our work

in Chapter 3 aims to provide a rigorous correspondence between experiment and theoretical

model under the general form of nonlocal coupling discussed in this chapter.

2.6 Complex Oscillatory Systems

In many oscillatory systems including chemical systems, oscillations can vary from sim-

ple, complex to chaotic [Scott, 1991]. In complex oscillations, as we illustrate below, the

branching of amplitude maxima is observed. This means that a value of the wrapped phase

φ ∈ [0, 2π) within one cycle of oscillation no longer corresponds to a unique value of the

amplitude - the relationship is instead one-to-many. In the case of chaotic oscillation, the

52
correspondence between phase and amplitude is even more complicated, as the oscillation

becomes aperiodic. Figure 2.6 shows time series that demonstrate these different cases of

oscillatory behaviours.

10

-5

75 80 85 90 95 100
Time

Figure 2.6: Time series of oscillatory behaviour showing simple (black), complex (red and
green) and chaotic (blue) behaviours.

As such, a canonical model like the CGLE described in Section 2.1.4 can only approximate

the oscillatory behaviours and patterns of reaction-diffusion systems in the simple oscillatory

regime. As described in Section 1.2, the observation of wave patterns in these spatially-

extended systems indicate synchronization in the medium. Since the concepts of phase

and synchronization are not foreign to complex and chaotic oscillators, a spatially-extended

system in these regimes can also become synchronized and lead to the emergence of oscillatory

patterns. It is then necessary to study these systems under locally complex or chaotic

oscillations, in order to build an encompassing understanding of the possible behaviours.

We first describe a set of ordinary differential equations called the Rössler model, which

53
exhibits oscillations in simple, complex and chaotic regimes. We then discuss a reaction-

diffusion system based on the Rössler model and the types of behaviours and oscillatory

patterns observed in a spatially-extended system undergoing complex and chaotic oscilla-

tions.

2.6.1 Rössler Model

One-dimensional maps and ordinary differential equations can become chaotic via an infinite

series of period-doubling bifurcations [Strogatz, 2000b]. We restrict ourselves to a discussion

of the latter case, a scenario possible in ordinary differential equations of dimension greater

than or equal to three. The Rössler model is given by

∂t X = −Y − Z (2.47)

∂t Y = X + aY (2.48)

∂t Z = b + Z(X − c), (2.49)

Plots of the solutions of the Rössler model for a = 0.2, b = 0.2 and increasing values of c

inducing successive period-doubling are shown Figure 2.7. The corresponding projections

onto the X-Y plane can be seen in Figure 2.8, where the effect of period-doubling can be

observed. At c = 2.0, the projection is that of a simple limit cycle in the X-Y plane, but

at higher values of c the projections becoming increasingly more self-intersecting, and this

is a consequence of the period-doubling. As the system becomes chaotic at higher values

of c, a countably infinite number of amplitude branches appear due to and infinite cascade

of period-doubling, as is shown for c = 5.0. Chaos is manifested in the behaviour that the

trajectory visit the different amplitude branches in an aperiodic manner. By the definition

of determinstic chaos, the solution depends sensitively on initial condition, as indicated by

a positive Lyapunov exponent [Strogatz, 2000b].

After m − 1 ≥ 1 period-doubling has occured, the oscillation has period Tm−1 = 2m−1 T

54
is lengthened, where T is the period in the simple oscillatory regime. We call the resulting

complex oscillation the period-m regime. In this regime, a wrapped phase φ ∈ [0, 2π) can

be defined to parametrize oscillation within a time interval [0, T ), but such definition fails

to be a unique correspondence between the phase and amplitude on the whole interval of

one period [0, 2m−1 T ) of period-m oscillation. A way to remedy this is by defining instead

Φ(t) = φ(t) + 2πj, where j is an integer that increments by one every time φ(t) intersects

zero. The value j keeps track of the amplitude branch on which the oscillation takes place.

For instance, period-2 oscillation can be described unambiguously by Φ(t) ∈ [0, 4π), and

period-m oscillation is described by Φ(t) ∈ [0, 2πm). In the Rössler model, the phase φ(t)

can be defined as φ(t) = tan−1 (Y (t)/X(t)), similar to θ described in Section 2.1.3, since the

interior domain surrounded by the trajectories shown in Figure 2.8 are simply connected

regions in the X-Y plane [Pikovsky et al. , 2001].

2.6.2 Diffusively Coupled Rössler Model and Synchronization Defect Lines in 2D

The diffusively coupled Rössler model in 2D was studied in [Goryachev & Kapral, 2000,

Davidsen et al. , 2004]. It is given by the set of equations

∂t X(r, t) = −Y − Z + D∇2 X (2.50)

∂t Y (r, t) = X + aY + D∇2 Y (2.51)

∂t Z(r, t) = b + Z(X − c) + D∇2 Z, (2.52)

with r ∈ R2 . When this spatially extended system is synchronized and the local oscillation

is complex, or period-n, a possible manifestation is the appearance of a target or spiral

pattern together with synchronization defect lines (SDLs), which are the loci of points

that undergo period-m dynamics, where m < n. The presence of SDLs in complex and

chaotic oscillatory regimes of the BZ reaction have been observed in experiments such as

[Park & Lee, 1999]. Moreover, SDLs also arise in the important context of physiological

55
3 8

2.5
6
2
4
1.5

1 2

0.5
0
0 5
5

0
−5
6 8
4 4 6
2 2
0 0
−5 −2 −10 −4 −2
−4 −6

14
20
12

10 15

8
10
6

4
5
2

0 0
10 10
5 5
0 0 15
10 10
−5 5 −5 5
0 0
−10 −5 −10 −5
−10 −10

Figure 2.7: Plots show the components of (X, Y, Z) with a = 0.2, b = 0.2 and c = 2.0 (top
left), 3.0 (top right), 4.0 (bottom left) and 5.0 (bottom right). Increase of the parameter c
shows successive period-doubling of the trajectory.

and medical sciences, where alternans (action potential waves exhibiting temporal alterna-

tion in the durations of consecutive cycles) in cardiac tissue can lead to the appearance of

SDLs, which induce meandering motions in the spiral wave. Such phenomenon has been at-

tributed as one of the precursors to ventricular fibrillation, a life-threatening heart condition

[Karma & Gilmour, 2007]. The experimental observation of SDLs in cardiac cells has been

reported in [Kim et al. , 2007].

Imagine a spiral wave present in the system, which is a particular solution in the simple

oscillatory regime of the Rössler reaction-diffusion model given above. In this regime, we

56
2 4
2
0 0
-2
-2
-4
-4 -6
Y

-4 -2 0 2 4 -4 -2 0 2 4 6
6
4 5
2
0 0
-2
-4 -5
-6
-8 -10
-5 0 5 -5 0 5 10
X
Figure 2.8: Projections onto the X-Y plane of the plots in Figure 2.7, shown in the same
order for the same parameters. The twisting in 3D space can be seen as self-intersection in the
plane, which increases the winding number of the curve. The regimes shown correspond to
simple oscillatory (top left), period-2 (top right), period-4 (bottom left) and chaotic (bottom
right) behaviours.

observe waves emitting from the centre of the spiral. Unlike the case of the ordinary dif-

ferential equations, period doubling in the partial differential equations Eqs. (2.50)-(2.52) of

an oscillatory system takes place via a different mechanism known as a 2:1 resonant Hopf

bifurcation [Sandstede & Scheel, 2007]. When the system has undergone the first period-

doubling, the local oscillations are now period-2. For instance, a spiral wave pattern in the

period-2 regime is shown in Figure 4 of [Goryachev & Kapral, 2000]. We describe the simple

scenario of a period-1 SDL in a period-2 system in the next paragraph.

For this discussion, let the centre of the spiral wave be the origin of our coordinate system.

At a point x0 located along a radial path from the centre of the spiral, we observe incoming

wavefronts in alternation of a large and small maxima. The appearance of a period-1 SDL

57
can be characterized by the following example. Suppose at one instance in time t0 , a large

maximum arrives at x0 , at this instance, at a point x1 located one wavelength away from x0

along the same radial path, we should then observe the arrival of the small maximum. Now,

if we follow the level set of the large maximum, in the angular direction starting from x0 , we

arrive at a point x2 such that |x2 | = |x1 |, but the wavefronts at these two points are shifted by

±T in time, or ±2π in phase with respect to each other, where T is the period of the simple

oscillatory regime. There is now a locus of points on which the local oscillations is period-1,

and separates the points x1 and x2 - we call this the period-1 SDL in a period-2 system. In

general, the appearance of period-m SDLs can be attributed to the alternation of amplitude

branches, which has been studied in [Goryachev & Kapral, 2000] as a combinatorial problem.

In general, the time series across a period-m SDL is shifted by ±2πm in phase with respect to

each other, and this is associated with the alternation between the large and small maxima

across the SDL. In the chaotic regime, the local amplitude oscillation becomes aperiodic,

and SDLs in the this regime are dynamical objects generated and vanish spontaneously

[Davidsen et al. , 2004].

After an infinite series of period-doubling, the initial spiral wave pattern in the system

is still preserved (for instance, Figure 22 of [Goryachev & Kapral, 2000] shows an intact

spiral pattern in the phase in the chaotic regime of the diffusively coupled Rössler model)

even though the local dynamics becomes chaotic. This means that it is still possible for the

system to become phase-locked and synchronized in the sense described in Section 2.1.1. For

the spatially-extended system, we can define a phase at each point in the system by φ(r, t) =

tan−1 [Y (r, t)/X(r, t)] ∈ [−π, π). Even though φ(r, t) does not describe unambiguously the

full cycle of oscillation [0, 2m−1 T ), it can still be used to characterize synchronization in

complex or chaotic oscillatory system. Phase locking in terms of φ(r, t) in complex or chaotic

oscillatory systems indicates that the local solutions (X(r, t), Y (r, t)) at each r come to evolve

with only finite lags between each other. Chimera states shown in Figure 2.1 and Figure 2.2

58
demonstrate that they can be characterized only by the phase-locking relations (or a lack

thereof) between oscillators in the population, and as we have seen in Section 2.4.2, chimera

states in nonlocally coupled systems can be characterized using a phase-only description.

If the local dynamics of a system resembles the kinds depicted in the Rössler model, in

Figure 2.8, where the notion of phase and synchronization are well-defined, one would expect

to observe chimera states even in the regime where the oscillators become chaotic. This is

indeed what we show in this thesis, for the first time, as discussed in Chapter 4.

59
Chapter 3

1D Ring of Nonlocally Coupled Oscillators

3.1 Electrochemical Oscillators

We motivate the work presented in this chapter by briefly discussing the experimental setup

involving an array of coupled electrochemical oscillators, which has been used extensively in

the past to study synchronization phenomena [Kiss et al. , 2002]. Oscillations in the system

are based on the electrochemical dissolution of nickel in sulfuric acid. The current produced

by the dissolution can be measured to infer the rate of reaction, which is a function of the

circuit potential of the electrochemical cell containing all reactants. The potential of each

electrochemical oscillator is maintained externally via a potentiostat.

At low circuit potential, nickel spontaneously dissolves into nickel ions. As this potential

is increased, the dissolution becomes impeded by the formation of nickel oxide or sulfate,

but there also exist regimes of higher potential in which the nickel can tunnel through the

oxide layer, enhancing dissolution. At even higher voltage, a region of negative differen-

tial resistance occurs, in which the relationship between current and voltage through the

electrochemical system becomes nonlinear. In this regime, when resistance is added to the

system, self-oscillation in the dissolution current occurs in the system. Further variation of

the potential will lead to complex or chaotic oscillations [Haim et al. , 1992].

An array of N electrochemical cells can be coupled as in the schematic representation

shown in Figure 3.1. The oscillators are connected individually in parallel to resistors Rp

that induce oscillations, which are then connected to a set of zero resistance ammeters

(ZRA in Figure 3.1). At their common ground point, a resistor Rs is added to the system,

inducing global coupling. A differential potential δV (t) is calculated by measuring the current

produced by each oscillator, and is then fed back into the potentiostat with some delay in

60
Counter Electrode
11
00
00
11
00
11 1
0
0
1
Reference 0
1
0
1
0
1
1
0
0
1
Electrodes 0
1
0
1
Potentiostat
1111111111
0000000000
0011
1100 1
011
00 0
1
0000000000
1111111111
00
11 1
0
0011
1100
00 1
11 0
011
100
0000000000
1111111111
00
11
0000000000
1111111111 0
1
0
1
00
11
00
1100
11
00
11 0
1
0
100
11
0000000000
1111111111
00
11 0
1
0000000000
1111111111
0011
1100 1
011
00 H2 SO 4 0
1
solution V(t) = V 0 + δV
1
0
Rp 0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
ZRA
Rs

Computer
I(t) Feedback ( δV )

Figure 3.1: Schematic representation of the electrochemical experiment setup.

time. This creates global coupling in the form of a voltage difference

N
KX
δV = h(xi (t)), (3.1)
N i=1
¯
where h is a polynomial function. The quantity xi (t) = Vi (t) − I(t)Rp is calculated for

¯ is the
oscillator i, where Vi (t) is the voltage applied to the particular oscillator and I(t)

mean current of the population, calculated with a time delay τ . Note that the coupling

parameter K depends on Rs and Rp .

In order to have nonlocal coupling, the computer program controlling the experimental

apparatus can be modified so that in a ring arrangement, only the M/2 left and right neigh-
¯ [Blaha, 2012].
bours of an oscillator are used in the computation of the mean current I(t)

This results in a step-function coupling similar to the one used in various studies of chimera

states in the 1D phase equation [Wolfrum & Omel’chenko, 2011]. In the next section, we

61
discuss a model system which can approximate the population of nonlocally coupled elec-

trochemical oscillators, in the regime where the motion of each oscillator is simple periodic.

3.2 Time-delayed, Nonlocal Complex Ginzburg-Landau Equation

For N oscillators evenly arranged on the interval [−1, 1] with periodic boundary condition,

the evolution of the complex amplitude Ai = Xi + iYi is given by

N
X
2
∂t Ai (t) = (1 + ici )Ai − (1 + ib)|Ai | A + K(1 + ia) Gij Aj (t − τ ), (3.2)
j=1

which we refer to as the time-delayed, nonlocal CGLE. In Eq. (3.2), Ai (t) = A(xi , t) is the

complex amplitude of an oscillator at position xi = −1 + 2i/N at time t. The matrix Gij is

normalized such that the row sum N


P
i=1 Gij = 1. The model Eq. (3.2) is aimed at modeling

the simple oscillatory regime of the electrochemical oscillator experiment [Blaha, 2012], but

in consideration of the fact that Eq. (3.2) without the coupling term is the normal form of

Hopf bifurcation, Eq. (2.1), this model should be able to describe any nonlocally coupled

system at the onset of oscillation as long as the coupling term remains identical to Eq. (3.2.

An array of oscillators (as many as up to 60 in the experiment) of a chosen intrinsic

frequency is coupled by the function


 1/(2R),

if |xi − xj | ≤ R
Gij = G(|xi − xj |) = (3.3)
 0,

otherwise,

with R ∈ (0, 1]. In the experimental setup, the sign and strength of the coupling constant

K can be tuned, and higher polynomial powers of Aj (t − τ ) can be added to the coupling

function, but the coupling is delivered with a non-zero time delay τ , which can be adjusted.

Due to experimental limitations, the intrinsic frequency in the population can only be es-

tablished with roughly 5-10% variation from their average frequency. Let ri be a uniform

random variable drawn from the interval [−1/2, 1/2] and ǫ a small constant. In the numerical

62
simulation of Eq. (3.2), we mimic the frequency heterogeneity by defining ci = c + ǫri , since

the intrinsic frequency of each oscillator is given by ωi = ci − b [Kuramoto, 1984].

3.3 Effects of Time Delay

We first focus on the system Eq. (3.2) by studying its behaviours under the assumption of

identical oscillators, setting ǫ = 0. Fixing (a, b, c) = (0.0, 0.2, 2.0), K = 0.1 and R = 0.7, we

numerically integrate the system using a first order Euler scheme, with dt = 0.01 or 0.003.

Note that in particular we have chosen a = 0, so the prefactor K(1 + ia) in front of the

coupling term is purely real. This is done in consideration of the fact that the coupling in

the experimental setup described in the previous section behaves the same way. The value

N = 60 is chosen throughout so as to match the number of oscillators in the experimental

setup. We prepare the system in an initial phase configuration given by

"  2 #
2i
φi (t = 0) = 6 · si · exp −10 −1 + , (3.4)
N
as described in Section 2.2. Each si is a random variable draw from the uniform distribution,

on the interval [−1/2, 1/2]. We then set Xi (t = 0) = cos(φi (t = 0)) and Yi (t = 0) =

sin(φi (t = 0)) as the real and imaginary components of the system.

For each time delay τ in the interval [0, T ], where T = 2π/(c − b) = 3.49 is the period for

each uncoupled oscillator, we integrate the system and measure the long-time averaged value

of hri, defined in Eq. (2.42), in a similar manner to Figure 2.4. In the absence of heterogeneity,

we find that as τ varies in the interval [0, T ], there are regions of τ corresponding to states

of complete synchrony, as well as states of completely asynchrony. This is shown in the

measurement of hri as a function of τ , in Figure 3.2. The transition from complete synchrony

to asynchrony (and vice versa) takes place over finite intervals of τ , in which the observed

values of hri ∈ (0, 1) - it is in these intervals where we look for the presence of chimera states.

We sample different values of τ in the intervals over which transitions take place, and ob-

63
1

0.8
<r>

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
τ/T
Figure 3.2: The time-averaged order parameter hr(t)i as a function of τ is measured over
a period of roughly 100 oscillations, in a simulation of the time-delayed, nonlocal CGLE,
Eq. (3.2) using (a, b, c) = (0, 0.2, 2.0), K = 0.1 and ǫ = 0.

serve chimera states depicted in Figure 3.3, showing the phase snapshots φ(x) and their long-

time averaged frequencies, hω(x)i for the population off oscillators. At a value of time delay

τ1 /T ∼ 0.224, we observe a chimera state similar to those previously reported in the 1D phase

equation with no time delay [Kuramoto & Battogtokh, 2002, Abrams & Strogatz, 2004]. From

the plot of the averaged frequencies in the lower left panel of Figure 3.3, we can clearly see

that the synchronized subpopulation takes on an average frequency at ωsync ∼ 1.718, slower

than the frequencies in drifting subpopulation, which has a “bump” distribution localized

in space. With 60 oscillators we can observe a sufficiently smooth profile of this frequency

distribution.

On the other hand, at time delay τ2 /T ∼ 0.695, we observe different chimera states that

64
3 3
2 2
1 1
φ

0 0
-1 -1
-2 -2
-3 -3

1.78 1.88

1.76 1.86
<ω>

1.74 1.84

1.72
1.82
-1 0 1 -1 0 1
x x
Figure 3.3: Snapshots of phase φ and time-averaged frequencies hωi taken over roughly 170
oscillations in a simulation of the time-delayed, nonlocal CGLE, Eq. (3.2). Parameters are
the same as in Figure 3.2. At time delays τ1 /T ∼ 0.224 (left panels) and τ2 /T ∼ 0.695 (right
panels). The order parameter of the states lie in the range 0.65 < hri < 0.75.

behave in the opposite manner: the lower right panel of Figure 3.3 indicates that the syn-

chronized subpopulation takes on an average frequency at ωsync ∼ 1.88, which is faster than

the frequencies in the unsynchronized subpopulation. The observation of chimera states in

the region of time delay where the system smoothly transitions synchrony to asynchrony, and

vice versa, has also been reported [Omel’chenko et al. , 2008], for a globally coupled phase

model with spatially-dependent time-delayed stimulation, which we have given in Eq. (2.45).

As mentioned in Section 2.5.4, the coupling in this model is non-identical, and chimera states

are generated because of this spatial inhomogeneity [Omel’chenko et al. , 2008].

65
3.3.1 Variation of Parameters: Relation to the Phase Equation

We now keep ǫ = 0, and vary the parameters a, b and c of the system. Figure 3.4 shows

a comparison between the behaviour of hri as a function of τ , for b = 0.2 and 0.8, fixing

all other parameters. We observe a shift in the intervals of τ where the system takes on

values 0 < hri < 1, and hence a shift in the values of τ where chimera states are observed.

In Section 2.4, we measured the behaviour of hri versus α for the 1D phase equation on

the domain |α| < π, shown in Figure 2.4. The dependence of hri on the control parameter

is surprisingly similar to that of Figure 3.2, apart from the fact that the two systems have

different control parameters.

As suggested in [Kuramoto & Battogtokh, 2002], in the phase reduction of the nonlocal

CGLE, an “effective” α ≡ α(a, b) can be computed as a function of the parameters of the

original system. We hypothesize that there also exists a relationship between α and τ , as

well as the parameters (a, b, c) for our time-delayed nonlocal CGLE model. The parameter α

appearing in the phase equation can be interpreted as a “phase lag” that introduces a shift in

the function Γ(φ −φ′ ), which governs the frequency change due to a pair of oscillators having

phase difference φ − φ′. Similarly, time delay in the coupling is proportional to Aj (t − τ ), and

this delay can be related to a phase shift, since the quantity Aj is periodic. If the frequency

of oscillation is independent of time, as is the case when the system is in a stable state, the

coupling term Aj (t − τ ) can be related to Aj (t) by a rotation that depends on the effective

frequency of this particular oscillator and the magnitude of the time delay τ . In addition

to a phase lag caused by time delay, the parameters (a, b, c) themselves contributed to the

overall lag, and this contribution can be calculated according to [Kuramoto, 1984].

The procedure to compute α(τ, a, b, c) is as follow. First, we perform phase reduction

on our system, Eq. (3.2), according to Section 2.4.1 by analytically computing Γ(φ − φ′ ),

for given values of (a, b, c) and τ . We then approximate it by the sinusoidal coupling of the

form − sin(φ − φ′ + α), appearing in the phase equation, for |α| < π. In turn, this will allow

66
us to invert for the function α ≡ α(τ ) for fixed a, b, and c, and map the synchronization

behaviour hri as a function of τ into a function of α. For notational convenience, below we

use unprimed and primed to denote different oscillators, as opposed to using indices.

b = 0.2
b = 0.8
1

0.8
<r>

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
τ/T
Figure 3.4: Comparison between time-averaged order parameter hr(t)i as a function of τ ,
taken over a period of roughly 100 oscillations, using b = 0.2 or b = 0.8. The parame-
ters used throughout in the simulations of the time-delayed, nonlocal CGLE, Eq. (3.2) are
a = 0, c = 2.0, K = 0.1 and ǫ = 0.

For the nonlocal CGLE, Eq. (2.3), Γ(φ−φ′) can be computed analytically as a function of

φ − φ′ , so there is no need to average over a period as in Eq. (2.31). We find that this is also

true for the time delayed model, so that Γ(φ − φ′ ) becomes the direct dot product of Z(φ)

and p(φ, φ′), where the primed and unprimed variables denote the phases of two different

oscillators. The quantity Z(φ) is calculated in [Kuramoto, 1984] and given in Eq. (2.34).

Associating the real and imaginary parts of a complex quantity with components of a vector

in R2 , the vector denoting the coupling p(φ, φ′) in our case is the term (1 + ia)A′ (t − τ ),

67
which does not depend on φ. In the limit of weak coupling, we rewrite A′ (t − τ ) as


A′ (t − τ ) = |A′ |eiφ (t) · e−i∆(τ ) = A′ (t)e−i∆(τ ) , (3.5)

where ∆(τ ) denotes a phase shift induced by a particular time delay τ . Assuming that

|A′ | → 1, which is the case for weak coupling and the amplitude relaxes quickly to the limit

cycle, we can then rewrite the complex quantity (1 + ia)A′ (t)e−i∆(τ ) as

p(φ′ ) ≃ e−i∆ (cos φ′ − a sin φ′ , sin φ′ + a cos φ′ )T

= (cos(φ′ − ∆) − a sin(φ′ − ∆), sin(φ′ − ∆) + a cos(φ′ − ∆))T , (3.6)

in which the term e−i∆ can be thought of as a rotation applied to a point on the periodic

trajectory A(φ) = cos(φ) + i sin(φ). Taking the dot product Z(φ) · p(φ′ ) we get

Γ(φ − φ′ ) = −ω −1 [(b − a) cos(φ − φ′ + ∆) + (1 + ab) sin(φ − φ′ + ∆)], (3.7)

where ω = c−b. Using trigonometric identities, we expand cos(φ−φ′ +∆) and sin(φ−φ′ +∆)

as

cos(φ − φ′ + ∆) = cos ∆ cos(φ − φ′ ) + sin ∆ sin(φ − φ′ )

sin(φ − φ′ + ∆) = cos ∆ sin(φ − φ′ ) + sin ∆ cos(φ − φ′ ).

At the same time, − sin(φ − φ′ + α) = − cos α sin(φ − φ′ ) − sin α cos(φ − φ′ ). Matching the

coefficients (functions of α) with Eq. (3.7), we find that

cos α = ω −1[(1 + ab) cos ∆ − (b − a) sin ∆] (3.8)

sin α = ω −1[(1 + ab) sin ∆ + (b − a) cos ∆]. (3.9)

68
We can invert for α by thinking of Eq. (3.8) and Eq. (3.9) as complex numbers

eiα = ei∆ ω −1 [(1 + ab) + i(b − a)]. (3.10)

We the match the arguments of the exponential on both sides. For complex numbers z, w,

Arg zw = Arg z + Arg w, so

 
−1 b−a
α = ∆(τ ) + tan = ∆(τ ) + β, α ∈ (−π, π). (3.11)
1 + ab
Since ∆(τ ) remains unknown, we make a first approximation by ∆(τ ) ∼ ωτ , since for

sufficiently weak coupling, the effective frequencies in the population do not vary significantly.

This yields the curves plotted in Figure 3.5 for parameters corresponding to Figure 3.4.

Using this figure, we can identify values of τ corresponding to particular values of α, which

themselves can be identified with values of hri according to Figure 2.4. In this manner,

the curve α(τ ) allows us to predict for which values of τ we can observe a completely

synchronized, completely unsynchronized, or chimera state in the system. We find very

good agreement between the locations of the synchrony-asynchrony transitions in Figure 3.4

and the prediction given by Figure 3.5. Moreover, Eq. (3.11) predicts that the curve α(τ )

characterizing the transitions does not depend on c, and this is confirmed by our simulations

(not shown). On the other hand, simulations also confirm that varying a results in a shift

in hri vs τ , as predicted by Eq. (3.11) (not shown).

3.4 Effects of Frequency Heterogeneity

Having characterized the behaviours of the system of identical oscillators, we now introduce

heterogeneity in the system by setting the small parameter ǫ > 0. For comparison, we set

ǫ = 0.01 and ǫ = 0.05. Figure 3.6 depicts the effect of such heterogeneity on the behaviour

of hri. The general dependence of hri on τ remains unaffected, but due to the presence

of heterogeneity, the synchronized states in the system of identical oscillators now become

69
3 b = 0.2
b = 0.8
2

0
α

-1

-2

-3
0 0.2 0.4 0.6 0.8 1
τ/T
Figure 3.5: The map α(τ ) from Eq. (3.11) is plotted for parameter values (a, b) identical to
those in Figure 3.4. The red dashed curves indicate boundaries at −π/2 and π/2. In the
vicinity of the intersection of α(τ ) with these boundaries we expect chimera states to occur
(compare with Fig. 3.4).

“noisy”. States with hri = 1 in the ǫ = 0 system now correspond to slightly lower values

of hri in the presence of heterogeneity. On the other hand, unsynchronized states with

hri = 0 in the ǫ = 0 system now take on slightly higher values of hri. The presence of

heterogeneity affects both the synchronized and unsynchronized states of the system, and

this can be seen by these fluctuations in the order parameter. We still identify thse states

with hri ∼ 1 and hri ∼ 0 as the (noisy) synchronized and unsynchronized states in the

system with heterogeneity.

We investigate the presence of chimera states by sampling points in the regimes of τ where

hri ∈ (0, 1). At ǫ = 0.01, we sample points τ1 /T ∼ 0.201 and τ2 /T ∼ 0.702 and produce plots

70
1

0.8

0.6
<r>

0.4

0.2
ε = 0.0
ε = 0.01
ε = 0.05
0
0 0.2 0.4 0.6 0.8 1
τ/T
Figure 3.6: Using the same values of (a, b, c) and K as in Figure 3.2 for the simulation of
the time-delayed, nonlocal CGLE, Eq. (3.2), the frequency heterogeneity is varied. The plot
of hr(t)i as a function of τ , showing the effect of frequency heterogeneity on the synchro-
nization order parameter. Frequency heterogeneity appears to diminish both the completely
synchronized states and the chimera states. At the same control parameter τ , we see that
hriǫ>0 < hriǫ=0.

similar to Figure 3.3, shown in Figure 3.7. We observe distinguishable spatial structures in

the plots of the time-averaged frequencies, hω(x)i, but with fluctuation to the smooth profiles

shown in Figure 3.3 for the system of identical oscillators. In the bottom left (bottom right)

of Figure 3.7, we can observe a subpopulation of synchronized oscillators with ωsync ∼ 1.718

(ωsync ∼ 1.773) that are slower (faster) than the subpopulation of unsynchronized oscillators,

which take on a distribution of frequencies.

Because the experimental setup can be subjected to higher level of heterogeneity, we

increase the small parameter to ǫ = 0.05 to see if we can observe similar behaviour shown

71
in Figure 3.7. At ǫ = 0.05, we sample points τ1 /T ∼ 0.189 and τ2 /T ∼ 0.731. Figure 3.8

shows the phase snapshots and time-averaged frequencies at τ1 and τ2 in the case of non-

identical oscillators, and we no longer observe structures similar to the 1D chimera states.

The plots of average frequencies still give indication of similar behaviour as in the case of

identical oscillators, namely the presence of synchronized subpopulations - which now become

delocalized in space. The lower left (right) panel of Figure 3.8 shows a slower (faster) ωsync ∼

1.72 (ωsync ∼ 1.875) in the synchronized subpopulation, and as a result of delocalization, the

unsynchronized subpopulations now appear distributed in an unstructured manner. The lack

of localization of the subpopulations reminds us of the behaviour in the Kuramoto model

of non-identical oscillators, in which partially-synchronized states can appear in a similar

manifestation as those of Figure 3.8. Thus, there is no evidence of a chimera state here.

As we have mentioned in Section 2.4.1, spatial localization of the synchronized and un-

synchronized subpopulations is a feature in the self-consistent chimera state solution of the

phase equation for identical oscillators. This feature can be observed even in the presence of

frequency heterogeneity [Laing, 2009a]. The results in [Laing, 2009a, Laing, 2009b] revealed

that chimera states can exist at an intermediate level of heterogeneity, but too much het-

erogeneity can destroy these chimera states. However, as pointed out in Section 2.5.2, for

some values of the control parameter, bistability of solutions can be observed, when both

the chimera state solutions and the completely synchronized solutions are stable. This fea-

ture suggests dependence of the system’s final state on its initial condition. This may be

a sufficient condition on observing the chimera state, and we investigate this in the next

section.

3.4.1 Test For Bistability

We first simulate our system with ǫ = 0 using two different initial conditions. We use an

initial state prepared by setting φi (t = 0) = const. + ni , where ni is a small noise chosen from

the uniform distribution on the interval [−0.01, 0.01]. We set the other initial state to the

72
3 3
2 2
1 1
φ

0 0
-1 -1
-2 -2
-3 -3

1.78 1.88

1.76 1.86
<ω>

1.74 1.84

1.72 1.82
-1 0 1 -1 0 1
x x
Figure 3.7: Snapshots of phase φ and time-averaged frequencies taken over roughly 170
oscillations, hωi with heterogeneity in the frequencies. Parameters used for the simulation
of the time-delayed, nonlocal CGLE, Eq. (3.2)are same as in Figure 3.2, except ǫ = 0.01, at
time delays τ1 /T ∼ 0.201 (left panels) and τ2 /T ∼ 0.702 (right panels).

usual initial condition, given by Eq. (3.4). We distinguish these two cases as “uniform IC”

and “Gaussian IC”, respectively. In the top left panel of Figure 3.9, we observe the expected

bistability behaviour, when ǫ = 0. For the two different initial conditions, there is a clear

separation in the behaviours of hri in the regimes τ1 ∈ (0.23, 0.25) and τ2 ∈ (0.65, 0.7), where

we can see that there exist both chimera state solutions and completely synchronized states.

However, as soon as heterogeneity is turned on, the hri versus τ curves collapse onto each

other, even for as little heterogeneity as ǫ = 0.01. In the previous section, we’ve shown

that the solutions for ǫ = 0.01 and ǫ = 0.05 are qualitatively different, and in the former

we can still identify localization of the synchronized and unsynchronized subpopulations.

When heterogeneity is present, we can observe localization feature but not bistability of

73
3 3
2 2
1 1
φ

0 0
-1 -1
-2 -2
-3 -3

1.8 1.88

1.78 1.86
<ω>

1.76 1.84

1.74 1.82

1.72 1.8
-1 0 1 -1 0 1
x x
Figure 3.8: Snapshots of phase φ and time-averaged frequencies taken over roughly 170
oscillations, hωi with heterogeneity in the frequencies. Parameters used for the simulation
of the time-delayed, nonlocal CGLE, Eq. (3.2)are same as in Figure 3.2, except ǫ = 0.05, at
time delays τ1 /T ∼ 0.189 (left panels) and τ2 /T ∼ 0.731 (right panels).

solutions, unlike the 1D phase equation studied in [Laing, 2009a], which can exhibit both

features under frequency heterogeneity. We also confirm that under heterogeneity, states

with similar values of hri do not behave qualitatively different.

3.5 Discussion

Numerical simulations of the nonlocal, time-delayed CGLE reveal to us a new insight, which

is that the time delay can be used as a control parameter of the system to change the

synchronization behaviour. As we remarked in Chapter 2, in the systems in which chimera

states were observed under weak coupling, not all parameter values of the oscillatory regime

74
1 1
0.8 0.8
<r>

<r>
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
τ/T τ/T
1 ε = 0, Gaussian IC
ε = 0, uniform IC
0.8
ε = 0.01, Gaussian IC
<r>

0.6
ε = 0.01, uniform IC
0.4
ε = 0.05, Gaussian IC
0.2 ε = 0.05, uniform IC
0
0 0.2 0.4 0.6 0.8 1
τ/T
Figure 3.9: Plot of hr(t)i as a function of τ using (a, b, c) = (0.0, 0.2, 2.0) and K = 0.1 for the
simulations of the time-delayed, nonlocal CGLE, Eq. (3.2). Comparisons are made between
the cases of identical and non-identical oscillators, with varying amount of heterogeneity
ǫ = 0, 0.01 and 0.05, using two different initial conditions (explained in text).

will lead to the formation of chimera states. In particular, parameter values that lead to

chimera states in 1D do not also lead to chimera states in the 2D system, and vice versa.

In an experimental setup of a 1D ring or 2D lattice configuration of a system of nonlocally

coupled oscillators, if we are left without the knowledge of the correct parameter values, but

there is the option of tuning the time delay at which the nonlocally coupling is delivered,

we can instead observe chimera states by varying the time delay. We have put this idea

to use in a simulation of a 1D ring of nonlocally coupled FHN oscillators, for which only a

set of parameters for the 2D simulation is known. With time delay built into the nonlocal

coupling, we can achieve in the population of relaxational oscillators the same behaviours

75
shown in Figure 3.2 and Figure 3.3 (not shown).

In the case of identical oscillators, we observed chimera states in a regime of τ between

the completely synchronzied and unsynchronized states of the system. The two branches of

chimera states in which the synchronized subpopulation is either slower or faster than the

unsynchronized subpopulation can also be seen in the phase equation with |α| < π, shown

in Figure 2.4. This observation motivated us to find a correspondence between the two in

the weak coupling limit. The result is a qualitative prediction of transition regions from

synchrony to asynchrony (and vice versa), where the chimera states are located, as well as

their dependence on the parameters a, b and c. Our result is in qualitative agreement with

the observed behaviour of hri versus τ in our own simulations, as well as in the experiments

[Blaha, 2012]. We should note that this picture is incomplete, since it involves reducing the

system to a phase equation, and so this result does not depend on the coupling strength

K. If K is increased in a continuous manner, the system should change in a way until it is

able to completely synchronized at all values of τ . Otherwise, if K is decreased, the system

eventually comes to behave as if the oscillators were uncoupled.

The full behaviour of our system in terms of K and τ resembles the phase diagram derived

in [Omel’chenko et al. , 2008], despite our model being drastically different from theirs. We

recall that the study in [Omel’chenko et al. , 2008] was based on a time-delayed globally cou-

pled model, in which two coupling terms control the behaviour of the system. Their model

has a global coupling, controlled by the coupling strength parameter C, and a time-delayed

coupling controlled by the strength parameter K. Spatial variation is introduced by multi-

plying the time-delayed term with a function ρ(x) ∝ e−ax , meaning that unlike our nonlocal

coupling scheme, the coupling is nonidentical and depends on the position of an oscillator.

This spatial dependence implies that oscillators near the centre of the population x ≃ 0 feel

a stronger effect of time-delay coupling, so the model of [Omel’chenko et al. , 2008] assumes

anisotropy. On the other hand, the formulation of nonlocal coupling we have introduced in

76
this thesis is a spatially homogeneous model, and assumes an identical coupling scheme be-

tween identical oscillators. The finding in [Omel’chenko et al. , 2008] is the first observation

that anisotropy, as opposed to nonlocal coupling, can also induce chimera states, but this

may be a generic effect considering that nonidentical nature of the anisotropic coupling.

We emphasize that the different theoretical studies on the phase description, including

[Omel’chenko et al. , 2008], rely on finding self-consistent solutions of R and Θ numerically,

then confirming these solutions by numerical integration of the phase equation and computing

these order parameters in a manner similar to the analysis in [Kuramoto & Battogtokh, 2002].

In these studies, no attempts were made to make correspondence between their results to

the original system from which the phase equation was reduced. We studied this corre-

spondence in a model with both amplitude and phase degrees of freedom, and related this

model to the phase-reduced model. In our calculation, we compared the original system to

the phase equation, Eq. (2.37), noting that the behaviour of the original system can be well

approximated in the weak-coupling limit, since the phase response curve Γ(φ −φ′ ) is suitably

sinusoidal. Our calculation demonstrated the point of view that different parameters of a

nonlocally coupled system contribute to an equivalent phase lag parameter α present in the

phase equation. In particular, the dependence on parameters for our time-delayed, nonlocal

CGLE model can be predicted correctly. Even though this phase reduction reveals some

insights on the original model, it points out one weakness of the procedure, which results in

the loss of some descriptive power since the coupling strength of the original system is not

taken into account anymore, when in fact the coupling strength does control the system’s

behaviour, however weak it might be.

In the case of nonidentical oscillators, at a higher amount of heterogeneity ǫ = 0.05, we

did not observe the localization of synchronized and unsynchronized subpopulations, which

is the usual manifestation of chimera states, and such localization is required by the theo-

retical prediction on the phase equation. This prompted us to look for a different signature,

77
the bistability of solutions, which is sometimes observed when chimera states appear, as re-

ported in [Laing, 2009a]. The study on the 1D phase equation in [Laing, 2009b] showed that

a chimera state can exist under a moderate amount of heterogeneity, but noisy, complete

synchronization takes over a high amount of heterogeneity. Furthermore, at moderate het-

erogeneity, the chimera state is also bistable with a (noisy) completely synchronized state.

Obviously, in our model as ǫ → 0, the localization of the synchronized and unsynchronized

subpopulations becomes more and more pronounced, as Figure 3.7 shows. However, when

we study the dependence on initial conditions in Figure 3.9, as soon as we introduce fre-

quency heterogeneity in our model, the observable regions of bistability vanished, even for

the small heterogeneity ǫ = 0.01 we have used. This is further indicating that bistability of

solutions is only a sufficient condition for chimera states to occur in a heterogeneous system.

Recent experimental work on the chimera state also has revealed behaviours different from

those predicted by the phase equation model [Tinsley et al. , 2012]. This, together with our

findings, may suggest that the current theory of chimera states based on the phase equation

is incomplete.

In our simulations of the time-delayed, nonlocal CGLE, one can quantify the true fre-

quency heterogeneity for fixed ǫ in the system as a standard error of the mean frequency

calculated over many realizations. However, the heterogeneity parameters we have chosen

ǫ = 0.01 and ǫ = 0.05 should yield deviation from the mean consistently below that of the

experimental setup. In the experiment, states similar to the ones depicted in Figure 3.8 were

observed, which as we explained, cannot be qualified as chimera states due to a lack of spa-

tial localization of the subpopulations. While at a lower heterogeneity ǫ = 0.01 we observe

a behaviour predicted in [Laing, 2009a] for the phase equation, as discussed in the previous

paragraph, we cannot make a direct comparison between his theoretical prediction and the

results observed in our model. As stated in the final words of [Laing, 2009a], “the challenge

remains to discover similar results for oscillators not described by a single variable”.

78
Recent experiments have followed the guidance provided by various theoretical works,

in producing chimera-like and chimera states under different setups. The first experimen-

tal example was provided by [Hagerstrom et al. , 2012], which is a realization of the system

of coupled-map lattices described in [Omelchenko et al. , 2011]. These discrete, chaotic os-

cillators produced a chimera-like states which showed spatial disorder, but the oscillations

are synchronous, and hence are drastically different in nature in comparison to the nu-

merical results shown in Figures 2.1, which shows that a subpopulation of oscillators have

unbounded phase differences with respect to all others. The chemical experiment reported

in [Tinsley et al. , 2012] reproduced the coupling configuration given by Eq. (2.46), which

couples two groups of phase oscillators. The groups are globally coupled, but a distinc-

tion is made between the intra- and inter-group coupling strengths K11 = K22 = µ and

K12 = K21 = ν, where µ > ν. This results in a chimera state for the overall popula-

tion, in which one group becomes completely synchronized, while the other is completely

desynchronized. Our investigation of the chimera state focused on its existence under the

more general form of nonlocal coupling that depends on the dimensionality of space, which

leads to chimera states with different manifestations. So far, chimera state has not been

demonstrated experimentally using the more general form of nonlocal coupling.

79
Chapter 4

Nonlocally Coupled Complex Oscillatory System in 2D

4.1 Nonlocal Rössler Model

As discussed in Section 2.6, the notions of phase and synchronization are often well-defined in

regimes of complex or chaotic oscillators. In a population of locally coupled chaotic oscillators

in 2D, a spiral wave pattern in the phase can still be observed [Goryachev & Kapral, 2000].

Given these observations, and the fact that chimera states are defined through synchroniza-

tion (or a lack thereof) in the sense of phase locking between oscillators in a population, we

investigate the existence of chimera states in nonlocally coupled complex or chaotic oscilla-

tory systems in this chapter.

The model we choose to study is based on the Rössler model, discussed in Section 2.6.1.

The Rössler model can undergo complex or chaotic oscillations, in addition to satisfying

the requirement that it has a well-defined phase in these regimes. The nonlocally coupled

Rössler model in 2D is given by

Z
∂t X(r, t) = −Y − Z + K G(r − r′ )[X(r′ , t) − X(r, t)]dr′ (4.1)
Z
∂t Y (r, t) = X + aY + K G(r − r′ )[Y (r′ , t) − Y (r, t)]dr′ (4.2)

∂t Z(r, t) = b + Z(X − c), (4.3)

with only nonlocal coupling in X and Y . We note that the number of nonlocally coupled

scalar fields is arbitrary, but the particular choice in Eqs. (4.1)-(4.3) will be justified later.

The function G(r − r′ ) ∝ exp(κ|r′ − r|) is normalized, and κ = 1 is held fixed. The system

is discretized into N × N oscillators, with N = 512 throughout. With no-flux boundary

condition, the area of the domain is given by L × L in spatial units, typical value of L = 50

80
is used in our simulations. A time step of dt = 0.01 is used throughout. The algorithm

used to perform integration and implementation of the boundary condition are described in

Appendix A.1. For the parameters we will use below, one spiral rotation corresponds to on

average roughly six units of time (or six hundred integration time steps).

4.1.1 Evaluation of Parameters

Due to the lack of knowledge of parameter values, the first problem in simulating the non-

local Rössler system lies in generating a chimera state in the simple oscillatory regime. We

propose a method for the purpose of parameter evaluation based on our knowledge of the

phase equation, Eq. (2.37). As discussed, the phase response curve appearing in the phase

equation is of the general form Γ(φ − φ′ ), which may be computed numerically in cases

when the equations of the original system are known. We compute the function Γ for the

Rössler model in the simple oscillatory regime. The algorithm to perform this numerical

calculation can be found in Appendix A.2, and involves transforming a phase defined by

φ(t) = tan−1 [Y (t)/X(t)] ∈ [−π, π) (with no ambiguity for the simple oscillatory regime)

into a linearly evolving phase. If the resulting Γ is suitably sinusoidal, then Eq. (2.37) with

coupling of the form − sin(φ − φ′ + α) should approximate it faithfully. We find that it is

actually the case here for our system of equations if we do not have coupling in Z, since Z(t)

oscillates in sharp periodic pulses rather than being sinusoidal.

Under the observation that Eq. (2.37) is a good approximation of our nonlocal Rössler

model in the weak-coupling limit, our approach is to compute an equivalent parameter value

of α = α(a, b, c) as a function of the parameters. This is done by fitting − sin(x + α) to

the numerically computed Γ(x). In 2D, we know that spiral wave chimerae appear in the

phase equation for α sufficiently close to zero, so this should be a criterion of the evaluation

scheme. Additionally, since the parameter space of the Rössler model is complicated, once

we find a set of parameters (a, b, c) for which α(a, b, c) ∼ 0, we also need to have knowledge

of the direction in parameter space in which the system undergoes period-doubling. This

81
second criterion is necessary if we are to proceed to study chimera states in the complex or

chaotic regimes of the system.

In Figure 4.1 we depict some representative functions Γ(φ − φ′ ) at various parameter

values, and the corresponding α-values computed for two sets of parameters in different

simple oscillatory regimes of the Rössler model. The first set of parameters are a = 0.2,

c = 5.7 and b ∈ [1.5, 2.0] are shown in the top row, and the second are a = 0.2, b = 0.2 and

c ∈ [2.0, 2.8], shown in the bottom row. We can see a clear difference in the way Γ passes

through zero for each of set parameters. Interpreting Γ(φ − φ′ ) as a frequency response

based on the phase difference of a pair of oscillators, we can see that in the case when α

is small there are nearly-symmetric ranges of φ − φ′ over both sides of zero in which the

oscillator must speed up or slow down in order to facilitate synchronization. On the other

hand, this is not true for large values of α, and we see large intervals on which Γ < 0 in the

bottom right panel of Figure. 4.1. Based on the values of α obtained through our numerical

routine, our method suggests that the parameters a = 0.2, b = 0.2 and c ∈ [2.0, 2.8] (which

correspond to the simple oscillatory regime of the parameters in the studies of diffusively

coupled Rössler model, [Davidsen et al. , 2004, Goryachev & Kapral, 2000]) do not lead to

spiral wave chimerae in 2D under weak coupling. This is confirmed by our simulations

performed with these parameters.

4.2 Chimera State in the Nonlocal Rössler Model

Based on the outcome of the method described in the previous section, we simulate our

system with (a, b, c) = (0.2, 1.7, 5.7) and K = 1.0. Upon successively lowering the coupling

strength K, a spiral wave chimera emerges at K ∼ 0.05, as shown in Figure 4.2. This

demonstrates a case of success of the procedure described in the previous section.

We then lower b further, going through the first period-doubling at b ∼ 1.5. At b = 1.25

the system is further into the period-2 regime, and we recall from Section 2.6.2 that in the

82
-0.32 1 b = 1.9
b = 1.7
b = 1.6
-0.33

Γ(φ)/ω
0
-0.34
α

-0.35 -1

-0.36
1.5 1.6 1.7 1.8 1.9 2 -3 -2 -1 0 1 2 3
b φ
-0.8 0 c = 2.2
c = 2.4
c = 2.7

-0.85 -1
Γ(φ)/ω
α

-2
-0.9
-3
-0.95
-4
2 2.2 2.4 2.6 2.8 -3 -2 -1 0 1 2 3
c φ

Figure 4.1: Top row: Equivalent values of α computed by matching the numerically obtained
phase response curves for the Rössler model to the sinusoidal phase response of Eq. (2.37), for
parameters a = 0.2, c = 5.7 and varying b. Several representative phase response curves are
shown. Bottom row: Same as the top, except in a different parameter regime a = 0.2, b = 0.2
and varying c.

presence of a spiral wave, synchronization defect lines (SDLs) appear in system in order to

accommodate local complex oscillations. In general, SDLs in a period-n system are loci of

points that undergo period-m oscillations, where m < n, and across these period-m SDLs

the local time series are shifted in phase by ±2πm. Indeed, in the presence of a spiral wave

chimera, we find the emergence of a period-1 SDL in the period-2 regime of the nonlocal

Rössler model, as shown in Figure 4.3 and Figure 4.4, the snapshot of the field X(r) and the

corresponding image processed by our SDL detection algorithm described in Appendix A.3.

As opposed to the diffusively coupled system in [Davidsen et al. , 2004], where SDLs are

83
unambiguously attached to the centre of a spiral wave, here the period-1 SDL is attached to

the core of unsynchronized oscillators of the spiral wave chimera.

0
6

10 4

20 2

X(r, t)
y

0
30

−2
40
−4
50
0 10 20 30 40 50
x
Figure 4.2: Snapshot of the scalar field X(r, t) for the nonlocal Rössler model, exhibiting a
spiral wave chimera. Parameters used are (a, b, c) = (0.2, 1.7, 5.7), in the period-1 (simple
oscillatory) regime of the system.

To further characterize the patterns shown in Figure 4.2 and Figure 4.3 and confirm that

they are indeed chimera states, we measure the long-time averaged frequencies for over a

thousand spiral rotations, at a cross-section that intersects the spatially-discontinuous core

region in the centre of the pattern. The corresponding results are shown in Figure 4.5

and Figure 4.6. In the complex oscillatory regime, we can estimate the average frequency

ω(r) by counting the number of observed maxima n(r, t) at each point in the system up to

84
0

6
10
4

20 2

X(r, t)
y

30 0

−2
40
−4

50
0 10 20 30 40 50
x
Figure 4.3: Snapshot of the scalar field X(r, t) for the nonlocal Rössler model, exhibiting a
spiral wave chimera. Parameters used are (a, b, c) = (0.2, 1.25, 5.7), in the period-2 regime
of the system.

time t, i.e. ω(r) = limt→∞ 2πn(r, t)/t. Note that measuring the frequencies from φ(r, t) =

tan−1 [Y (r, t)/X(r, t)] yields equivalent results. Both figures reveal a bump structure, in

which the synchronized subpopulation in the spiral wave take on a common average frequency

ωsync = 1.023 and ωsync ∼ 1.03, respectively. In both cases, oscillators in the core follow a

smooth distribution of frequencies that are slower than ωsync . This behaviour is similar to

the chimera state in the time-delayed, nonlocal CGLE of identical oscillators in shown in

Figure 3.3, at higher time delay τ2 > τ1 . In fact, Figure 4.1 shows that the approximate

α-values for our choice of parameters is negative, and in Figure 2.5 we demonstrated that

85
0 1

10 0.8

20 0.6
y

30 0.4

40 0.2

50 0
0 10 20 30 40 50
x
Figure 4.4: Processed image showing the presence of a period-1 SDL together with a spiral
wave chimera, corresponding to Figure 4.3. The image processing algorithm is discussed in
Appendix A.3.

α < 0 chimera states in the phase equation correspond to the synchronized subpopulation

taking on a faster frequency than the unsynchronized subpopulation. This demonstrates

consistency of our parameter evaluation method.

From the parameters (a, b, c) = (0.2, 1.25, 5.7), we enter the period-4 and higher regimes

by changing one of the three parameters, while holding the others fixed. We remark that

for the Rössler ordinary differential equations, Eqs. (2.47)-(2.49), successive, infinitely many

period-doubling cascades can be observed all the way up to the chaotic regime. However, for

our model, which is a set of partial differential equations, and for the spatially distributed

model of [Davidsen et al. , 2004] given in Eqs. (2.50)-(2.52), only regimes up to period-4 can

86
1.025

1.02
<ω>

1.015

1.01

1.005
22 24 26 28
x
Figure 4.5: Time-averaged frequencies over ∼ 1700 oscillations recorded at the cross section
y = 27.05, in the corresponding Figure 4.2. Units on the horizontal axis given in terms of
the real length of the system.

be resolved, after which the system becomes chaotic.

We increase c from the initial value c = 5.7 and observe that the diameter of the core

diminishes as c is increased. In the theory of the phase equation in 2D, the area of the core

in the chimera states scales proportional to α [Martens et al. , 2010]. We see that changing

the value of c creates a similar effect on the area of the core, though in this case the diameter

is inversely proportional to the change in c. The same observation can also be made from

the method of parameter evaluation discussed in Section 4.1.1, in which we extract the

dependence α(c). However, in the case of complex oscillation, this correspondence is not

rigorous since the formalism of the phase response curve is only applicable to systems in the

87
1.03

1.025
<ω>

1.02

1.015

22 24 26 28
x
Figure 4.6: Time-averaged frequencies over ∼ 1700 oscillations recorded at the cross section
y = 28.125, in the corresponding Figure 4.3. Units on the horizontal axis given in terms of
the real length of the system.

simple oscillatory regime.

We observe that the spiral wave chimera is stable in the period-4 and chaotic regimes,

corresponding to c = 6.7 and c = 7.0, shown in Figure 4.7 and Figure 4.8. In the chaotic

regime, we confirm that the state depicted in Figure 4.8 with a spatially discontinuous

core in the centre is indeed a chimera state. In this regime, since we observe considerable

translational motion of the core subpopulation over the length of more than one thousand

spiral rotations, we do not measure the time-averaged frequencies at a cross section of space.

Instead, we observe the quantity n(r, t) over the entire system for over a thousand spiral

rotations. Figure 4.9 shows the snapshot of n(r, t) at time t = 1000 and t = 7500. The range

in the colour bars of corresponding to snapshots of n(r, t) at these two times indicate that

88
there is a distribution of periods in the core subpopulation. In the right panel corresponding

to t = 7500, we see a “trail” of oscillators where the core had passed through, becoming

re-synchronized with the subpopulation outside of the core.

In both panels of the figure, the subpopulation belonging to the spiral wave pattern

is synchronized, as there is consistently only a one period difference between oscillators in

this subpopulation. This difference can be attributed to the fact that the spiral wave has

passed through only part of the system during one rotation, so these locations observe new

maxima and n(r, t) increases by one, whereas the remainder of the system n(r, t) has not been

incremented. As time increases, the difference between the slowest and the fastest oscillators

is increasing with time, as indicated by the range in the colour bars, which implies that

oscillators in the core subpopulation are not phase locked with respect to the subpopulation

in the spiral wave, i.e. these oscillators correspond to an unsynchronized subpopulation that

move according to a distribution of frequencies - this is indicative of the presence of a chimera

state.

4.2.1 Disappearance of the Chimera State

We also investigate if there are other parameter values in the period-4 and chaotic regimes

of the system in which chimera states can be observed. Alternatively, we can reach different

period-4 and chaotic regimes by lowering b, upon which we observe an increase in the diameter

of the core subpopulation of the chimera state. This is opposite to the behaviour we observed

when c was increased.

At b ∼ 0.9, the system enters the period-4 regime. After a long transient has past, we

can no longer observe the presence of a core subpopulation exhibiting discontinuous spatial

variation. Instead, the spiral pattern is preserved, but we instead observe an elongated

“tip” structure in the centre of the spiral, shown in the snapshot of Figure 4.10. This

structure has also been documented in [Kuramoto & Shima, 2003] for the nonlocal CGLE

model. From our own simulations of the 2D phase equation, the same tip pattern can also

89
0
8

10 6

4
20
2

X(r, t)
y

0
30
−2

40 −4

−6
50
0 10 20 30 40 50
x
Figure 4.7: Snapshot of the scalar field X(r, t) for the nonlocal Rössler model, exhibiting a
spiral wave chimera. Parameters used are (a, b, c) = (0.2, 1.25, 6.7), in the period-4 regime
of the system.

be observed when the control parameter α is sufficiently far from zero. Our simulations of

the phase equation in 2D shows that as we increase α above some critical value αc , the core

subpopulation vanishes spontaneously and the tip pattern emerges. In this regime, we also

observe pronounced deformation of the wavefronts, although the overall spiral wave pattern

remains intact. Lowering b further to b = 0.75, the corresponding state in the chaotic regime

is shown in a snapshot, in Figure 4.11. Although the snapshots indicate that the spiral

wave is present, we observe significant meandering motion of the tip that takes place on a

bounded, curved trajectory. Unlike the motion of the core in the chimera state, the velocity

90
0
8
10 6

4
20

X(r, t)
2
y

30 0

−2
40 −4

−6
50
0 10 20 30 40 50
x
Figure 4.8: Snapshot of the scalar field X(r, t) for the nonlocal Rössler model, exhibiting a
spiral wave chimera. Parameters used are (a, b, c) = (0.2, 1.25, 7.0), in the chaotic regime of
the system.

of the meandering tip is much faster. Due to the strange topology of the tip structure, we

have not attempted to track and quantify exactly the nature of its meandering.

We further characterize the tip structure in the chaotic regime when b = 0.75, shown in

Figure 4.11. We would like to determine if there is still the presence of an unsynchronized

subpopulation of oscillators in the system. We measure n(r, t) over more than a thousand

spiral rotations in the same manner as in the previous section. In Figure 4.12, we show

the field n(r, t) at two different times t = 1000 and t = 7500. While the spiral pattern is

present in the system at those two times, it is not a rigidly rotating spiral wave and the

91
0 164 0 1226
1225
10 10 1224
163
1223
20 20
1222
162
y

y
1221
30 30
1220
161 1219
40 40
1218
50 160 50 1217
0 10 20 30 40 50 0 10 20 30 40 50
x x

Figure 4.9: Snapshots of observed number of maxima n(r, t) corresponding to Figure 4.8,
shown at two different times t = 1000 (left) and t = 7500 (right).

wavefront deforms. In addition, spiral wave tip performs meandering motion of its own. The

combination of wavefront deformation and tip meandering causes some parts of the system

to accumulate additional lags in values of n(r, t), unlike the case of rigidly-rotating spiral

waves mentioned in the previous paragraph. The appearance of three or four values of n(r, t)

shown in Figure 4.12 is due to these effects, rather than the appearance of a subpopulation

of unsynchronized oscillators, as in the spiral wave chimera.

We note the coincidence that decreasing b sufficiently has the same effect as increasing

α in the 2D phase equation. The qualitative features of wavefront deformation and the

appearance of the tip pattern that we observe in the regimes b = 0.9 and b = 0.75 are both

similar to α > αc .

4.3 Synchronization Defect Lines in the Period-4 and Chaotic Regimes

Having studied and identified period-4 and chaotic regimes where chimera states appear or

vanish, we turn out attention now to examine the behaviours of SDLs in these regimes, when

92
0 8

6
10
4
20 2

X(r, t)
y

0
30
−2
40 −4

−6
50
0 10 20 30 40 50
x
Figure 4.10: Snapshot of the scalar field X(r, t) for the nonlocal Rössler model. Parameters
used are (a, b, c) = (0.2, 0.9, 5.7), in the period-4 regime of the system.

many SDLs of different periodicities can appear in the system. We show processed images

of the SDLs in Figure 4.13 and Figure 4.14, corresponding to the two period-4 regimes

(a, b, c) = (0.2, 1.25, 6.7) and (a, b, c) = (0.2, 0.9, 5.7). A peculiarity that is immediately

noticeable in both Figure 4.13 and Figure 4.14 is the appearance of a period-2 SDL attached

on both ends to a period-1 SDL. This cannot be, since a SDL must end on the boundary or

at the centre of a spiral wave, so the only possibility is that these SDLs overlap each other.

If a period-1 and period-2 SDL are overlapping, then across this overlap, the local time series

are shifted by −2π or +6π (equivalently, +2π and −6π, depending on the sign of the lines

and the direction of spiral rotation) with respect to each other, which is indistinguishable

93
0
8

10 6

4
20
2

X(r, t)
y

0
30
−2

40 −4

−6
50
0 10 20 30 40 50
x
Figure 4.11: Snapshot of the scalar field X(r, t) for the nonlocal Rössler model. Parameters
used are (a, b, c) = (0.2, 0.75, 5.7), in the chaotic regime of the system.

from a ±2π phase shift according to our SDL detection scheme (as well as the scheme used in

[Davidsen et al. , 2004], so our scheme does not yield inferior performance in comparison to

others). We find that SDLs in these period-4 regimes shown in Figure 4.13 and Figure 4.14

are stationary, except for re-attachment events occurring at the core/spiral tip, and we do

not observe spontaneous generation of SDLs in the system after passing a long transient of

over a thousand spiral rotations.

In a complex oscillatory system with period-m oscillations in the presence of a spiral

wave, the number and periodicities of SDLs attached to the centre of the spiral wave must

be consistent with the following condition: the phase increase associated purely with cross-

94
0 167 0 1244

10 10

1243
20 20
166
y

y
30 30
1242

40 40

50 165 50 1241
0 10 20 30 40 50 0 10 20 30 40 50
x x

Figure 4.12: Snapshots of observed number of maxima n(r, t) corresponding to Figure 4.11,
shown at two different times t = 1000 (left) and t = 7500 (right).

ing these attached SDLs along a closed curve γ surrounding the centre must be equal to

2πmk − 2πnt , where k ∈ Z and the topological charge nt = ±1 is positive for a spiral ro-

tating in the counterclockwise direction, and negative otherwise (by the usual convention)

[Goryachev & Kapral, 2000]. We mentioned in Section 2.6.2 that the crossing of a period-

m SDL affects a ±2πm shift between the local time series on both sides of the SDL. The

condition not only tells us the sign of the phase shift associated with each SDL shown in

these configurations, but also corroborates with the picture of overlapping SDLs mentioned

in the previous paragraph. The condition given in [Goryachev & Kapral, 2000] confirms this

possibility, since there must be an overall number of SDLs attached to the core that im-

parts the correct total amount of phase shift. Since our spiral rotates clockwise, the total

phase shift associated with the SDLs attached to the core can be {2π, ±10π, ±18π, ...}. Of

these possibilities, only the choices of 2π and ±10π are consistent with the configurations

observed in the period-4 regimes, in Figure 4.13, but the choice of ±18π is also possible in

the configuration of Figure 4.14.

In the chaotic regimes, unlike in the cases shown in Figure 4.13 and Figure 4.14, SDLs

95
0

10
1.5

20
1
y

30

0.5
40

50 0
0 10 20 30 40 50
x
Figure 4.13: Processed image showing the presence of a period-1 (grey) and period-2 SDLs
(black) in the period-4 system, corresponding to Figure 4.7 in the presence of a spiral wave
chimera. The image processing algorithm is discussed in Appendix A.3.

can be spontaneously generated and disappear over time. In Figure 4.15 and Figure 4.16

corresponding to the chaotic regimes (a, b, c) = (0.2, 1.25, 7.0) and (a, b, c) = (0.2, 0.75, 5.7),

we also find several dominant SDLs attached to the core or spiral tip that are persistent

for over a thousand spiral rotations. In these regimes, we typically observe spontaneous

creations of closed period-2 SDL loops as well as period-2 SDLs entering the system from

the boundary. These dynamical SDLs will disappear from the system either by shrinking or

leaving at the boundary. When the core or the spiral tip moves in the system, the attached

period-1 SDL is elongated in the direction of motion. Unlike in [Davidsen et al. , 2004]

for the diffusively coupled Rössler model, Eqs (2.50)-(2.52), the motion of the core or the

96
0 2

10
1.5

20
1
y

30

0.5
40

50 0
0 10 20 30 40 50
x
Figure 4.14: Processed image showing the presence of a period-1 (grey) and period-2 SDLs
(black) in the period-4 system corresponding to Figure 4.10. The image processing algorithm
is discussed in Appendix A.3.

spiral tip do not appear to be ballistic. Occasionally, detachment and reattachment of the

attached period-2 SDLs from the tip can occur, and these behaviours are facilitated by SDL

loops emerging from the core or the spiral tip. Some of the possible configurations are

depicted in Figure 4.17. In general, the types of qualitative behaviours of the SDLs in our

system are similar to what has been described in the study of the diffusively coupled Rössler

model, but with a major differences in the SDL generated in the chaotic regime. In the

diffusively coupled Rössler model discussed in [Davidsen et al. , 2004], generation of period-

1 SDLs in the chaotic oscillatory regime dominates the overall SDL population, whereas in

for the particular parameters we have investigated, all of the SDLs generated are period-2.

97
0

10
1.5

20
1
y

30

0.5
40

50 0
0 10 20 30 40 50
x
Figure 4.15: Processed image showing the presence of a period-1 (grey) and period-2 SDLs
(black) in the chaotic system corresponding to Figure 4.8 in the presence of a spiral wave
chimera. The image processing algorithm is discussed in Appendix A.3.

Additionally, for the parameters used here, we have only observed configurations with a

single, persistent period-1 SDL.

4.4 Discussion

The pattern shown in Figure 4.8 is the first demonstration of chimera state in a system

of nonlocally coupled oscillators, where the oscillation is chaotic. Similar to the diffusively

coupled Rössler reaction-diffusion system, when a spiral wave is present in the system, SDLs

have to appear in the system in order to accommodate the effect of period-doubling. We

find the same scenario in the nonlocally coupled system, but in the limit of weak coupling,

98
0 2

10
1.5

20
1
y

30

0.5
40

50 0
0 10 20 30 40 50
x
Figure 4.16: Processed image showing the presence of a period-1 (grey) and period-2 SDLs
(black) in the chaotic system corresponding to Figure 4.11. The image processing algorithm
is discussed in Appendix A.3.

instead of the usual spiral wave pattern we find a spiral wave chimera together with the

presence of SDLs. In the diffusively coupled system, SDLs must form closed loops, or have

their ends attached to the boundary and to the centre of the spiral wave, which is a single

point in the system. In our nonlocal Rössler system, we find similar behaviours, except that

the SDLs are connected to the unsynchronized core. As we have suggested, the description

of chimera states in the phase equation, Eq. (2.37), requires the population of oscillators to

take on certain phase relations amongst themselves. These phase relations are manifested

as spatially localized subpopulations of synchronized and unsynchronized oscillators. Hence,

if these chimera states are to occur in systems of nonlocally coupled chaotic oscillators, the

99
Reattachment Events

A.

B.

Figure 4.17: Schematic illustrations of reattachment events in the chaotic regimes. The circle
represents the centre of the spiral wave to which the SDLs are attached. An SDL can be
detached from its connection and reattached at a closed loop emanating from the centre (A).
In another scenario, an SDL that is not connected to the centre initially can be attached to
it, but first detaching the other connected SDLs (B).

notion of phase and synchronization must be well-defined for these chaotic oscillators. We

find that this is the case for the Rössler model.

In the complex and chaotic oscillatory regimes of our nonlocal Rössler model, we find

exactly a synchronized subpopulation, corresponding to a spiral wave pattern, which takes

on a common average frequency, and a core of unsynchronized oscillators in the centre of

the spiral wave. We showed this by counting the number of cycles n(r, t) at every point in

the system. As we find that the difference |nspiral − ncore | between pairs of oscillators in the

core and in the spiral wave is increasing with time, we can argue that these subpopulations

100
correspond to unsynchronized and synchronized oscillators in the system. Additionally, in

the complex or chaotic oscillatory regimes, we observe in the system the presence of several

persistent period-2 SDLs, as well as a single persistent period-1 SDL together with a spiral

wave chimera.

We find a chimera state unlike the result reported in [Omelchenko et al. , 2011], which

studied a nonlocally coupled Rössler model in 1D, using parameters different from ours. The

“chimera-like” state reported in [Omelchenko et al. , 2011], however, appears to be only

spatially disordered, but the phases are mutually synchronized in the entire system, as

characterized by bounded phase differences between every pair of oscillators, so it is not a

chimera state in the sense we have discussed in Chapter 2. Our finding relaxes the assumption

and description of the chimera state in existing theories based on the phase equation. Even

though the Rössler model in the complex and chaotic oscillatory regimes is vastly different

from the phase equation, we observe some qualitatively similar behaviours in our spiral wave

chimera, such as the scaling of the size of the core with the system’s control parameters.

Additionally, there also exists regimes where instability of the chimera state is followed by

the appearance of the tip structure and the deformation of the spiral wavefronts.

In both chaotic regimes we have studied, we observe SDLs that are generated sponta-

neously as loops, entering the system, and disappear by leaving from the boundry or by

shrinking. In the regimes, we find SDL dynamics similar to those reported in the diffusively

coupled Rössler model [Davidsen et al. , 2004], such as detachment and re-attachment of

SDLs to the centre of the spiral wave, though in our case, the centre of the spiral wave

refers to the circular core of unsynchronized oscillators. As mentioned, we observe that the

SDL generation in our nonlocally coupled Rössler model is dominated by period-2 SDLs,

whereas the opposite is true in the diffusively coupled Rössler model, where the majority of

spontaneously generated SDLs are period-1. Further investigations will be necessary to see

if this behaviour is typical to the nonlocal system, and to understand its phenomenology. A

101
detailed quantitative comparison between the translational motion of the chimera core and

spiral tip in the nonlocally coupled Rössler model and its comparison to the diffusive system

would also be worthwhile.

We have also proposed and demonstrated with success a method based on matching the

sinusoidal coupling of the phase equation with an empirically obtained phase response curve.

As a first application, we used this method to evaluate parameters of the nonlocally coupled

Rössler model in the simple oscillatory regimes, to calculate an equivalent-α value of its

phase equation approximation. We expect such method to hold in the case where the phase

response function of the system is suitably sinusoidal, so that the empirically obtained phase

response curve is well-approximated by the phase equation in the weak-coupling limit. Our

method provides guidance to experimental studies, when there is a necessity for knowing the

parameter regimes that lead to chimera states in general models of oscillators.

102
Chapter 5

Conclusion

In this work, we have studied the existence of chimera states in systems of nonlocally cou-

pled oscillators undergoing simple, complex and chaotic oscillations. Previous studies have

mostly focused on chimera states in variants of models based the phase equation for non-

locally coupled systems in 1D or 2D, a reduced model that describes only the dynamics

in the phase of the oscillation amenable in the limit of weakly-coupling. Since one of our

goals is to motivate future experimental studies on the chimera state, we have considered in

this thesis models with both amplitude and phase degrees of freedom. This is a departure

from previous studies on the chimera states in the context of the phase equation, which is

not readily-implemented in an experimental setup of coupled oscillators. A major result

in this thesis is the extension of chimera states to complex and oscillatory systems, which

we have provided evidence for in a 2D model of nonlocally coupled oscillators given by the

Rössler model. In a simple oscillatory 2D system, chimera states are manifested in a pat-

tern consisting of a rigidly rotating spiral wave surrounding a spatially-discontinuous “core”,

corresponding to the synchronized and unsynchronized subpopulation of oscillators in the

system. As the oscillation becomes increasingly complex via period-doubling, synchroniza-

tion defect lines emerge in the system. These are loci of points with periodicities different

from the bulk of the system, and they appear in order to accommodate the existing spiral

wave chimera. Hence, the manifestation of chimera states in complex and chaotic oscillatory

system is similar to that of the spiral wave pattern in the diffusively coupled Rössler model

previously studied in [Davidsen et al. , 2004]: under successive period-doubling in the sys-

tem, the only modification to the patterns is the appearance of synchronization defect lines.

We gave the first demonstration that chimera states can exist under more relaxed assump-

103
tions, that the dynamics of the oscillators need not to take place on simple limit cycles. Since

the synchronized and unsynchronized subpopulations of the chimera state is characterized

through the relationship of phase locking, or a lack thereof, between the oscillators, the fact

that the notion of a phase can be generalized to chaotic systems means that these phase

relations can also emerge when the system undergoes aperiodic oscillations.

To mimic the experimental setting of a coupled array of electrochemical oscillators, we

studied chimera states in a model of simple, limit cycle oscillators coupled in a 1D ring con-

figuration, using a time delay in the coupling to control the system. We found that varying

the time delay in general induced change in the system’s synchronization behaviour, switch-

ing between completely synchronized states to completely unsynchronized states, and vice

versa. When the coupling is weak, these transitions are not sharp but occur within finite

intervals of time delays, and inside these intervals we can find chimera states. We related

the system’s time delay to the control parameter of the phase equation, which allowed us

to qualitatively understand the synchronization transitions in terms of the behaviour of the

phase equation. We also studied the robustness of chimera states under frequency hetero-

geneity, which is a physically relevant problem to experimental studies. We found that at

small level of heterogeneity, chimera states can persist in the system, and the spatial local-

ization of the synchronized and unsynchronized subpopulations can still be observed, in the

same way as previously studied cases of chimera states [Kuramoto & Battogtokh, 2002]. At

higher heterogeneity, this spatial localization feature is lost, and hence we no longer observe

chimera states. In the presence of any amount heterogeneity, we did not observe the bistabil-

ity of chimera state solutions and the completely synchronized solutions, although the lack

of bistability in the heterogeneous system is one of the scenarios predicted in [Laing, 2009a]

for the 1D phase equation. While the phase equation has been demonstrated to be able to

capture the typical dynamics of nonlocally coupled systems in the weak coupling regime,

in particular the synchronization behaviours of such systems, it does not capture all of the

104
stability features that we have observed for systems involving both amplitude and phase

degrees of freedom, such as our time-delayed, nonlocally coupled model, and in the experi-

mental system presented in [Tinsley et al. , 2012].

In our investigation of these models, we have also suggested more than one way to

control as well as determine the synchronization behaviours of our systems. The phase

equation contains a single parameter, but general oscillator models, including those that

are experimentally viable, usually have dependence on multiple parameters. It is not clear

how these parameters can be related to each other in an intuitive manner, which makes

experimental studies of the chimera states in general oscillator models difficult. In our study

of the nonlocally coupled Rössler model, we determined the relationship between the control

parameter of the phase equation and the parameters of the model, and from this we were

able to identify suitable parameters that lead to formation of a chimera state in the simple

oscillatory regime. The method we have used is expected to be applicable to general models

as well, so long as their phase reductions can be well-approximated by the phase equation.

As mentioned in the previous paragraph, in our study of the nonlocally coupled model with

time delay, we can observe chimera states in a suitable interval of delay. This implies that if

time delay is present in the experimental setup, we can use the delay as an effective control

parameter, as opposed to searching in the parameter space for a suitable set of chimera

state parameters. These two approaches should be of great relevance for future experimental

studies of chimera states.

5.1 Future Works

We conclude this thesis by outlining some meaningful research directions from the perspec-

tives of both theory and experiment.

105
5.1.1 Analytical Development

In the study of chimera states in nonlocally coupled systems, a rigorous correspondence

between theoretical predictions and experimental results has yet to be achieved. The inability

to do this is due to the lack of a description of synchronization for models possessing both

amplitude and phase degrees of freedom, so that reduction to phase equation in the weak-

coupling limit will not be necessary. As much as we have gathered from existing literature,

the only attempts to develop a theory with both amplitude and phase have been made

in [Laing, 2010] and [Bordyugov et al. , 2010], where the authors studied the stability of

chimera states in populations of identical Stuart-Landau oscillators.

5.1.2 Improved Parameter Space Search

The method we have proposed in Chapter 4 can be easily generalized to examine the full pa-

rameter space of any oscillator model (whose phase response curve Γ, needs to be sufficiently

sinusoidal), to search out parameters that lead to chimera states in a 1D or 2D system.

This will allow us to investigate chimera states experimentally or theoretically, using dif-

ferent models and different parameters, which will help us confirm and assess the generic

behaviours of chimera states we have observed (such as spatial localization or bistability), or

even potentially uncover new behaviours that are not observed in the phase equation model.

The automation of the parameter space search algorithm involves two separate routines.

The first only involves finding regions in the parameter space that corresponds to simple

oscillatory behaviour, since we are only interested in evaluating Γ in those regions. This

can be implemented by using a subroutine to examine the time series of the system for each

point in the parameter space. Then, one can compute the equivalent-α values by calculating

the phase response curve using the method of Appendix A.2, and find values of α that are

sufficiently close to ±π/2 or zero, depending on the dimensionality of space.

106
5.1.3 Candidate Experimental Systems for Chimera Spiral Waves

We briefly describe a chemical experiment that emulates closely the nonlocal coupling scheme

we have studied in this thesis. In particular, this setup potentially provides an opportunity

for observing spiral wave chimerae in 2D, which yet awaits experimental confirmation. The

chemical system is a BZ reaction system in a water-in-oil mixture covered with a surfactant

(aerosol OT) [Alonso et al. , 2011]. Chemical species are mixed and react inside the dispersed

water droplets covered by the surfactant, while they are freely diffusing in the oil phase of

the solution. This BZ-aerosol OT system has been experimentally demonstrated to show a

larger variety of patterns than the usual BZ reaction schemes, since there is the additional

ability to change the diffusion properties of the chemicals in the mixture by changing the

ratio of water, oil and the surfactant. Essentially, the water droplets covered by surfactants

have a much smaller diffusion constant compared to the chemical species in the oil phase of

the solution. This implies that once the species become trapped in a droplet, they diffuse at

the rate of the droplet instead. The experimental description we have just given is similar

to the picture in the derivation of nonlocal coupling, in Section 2.2.1, where we mentioned

the interpretation that nonlocal coupling can arise from a large separation of time scales

between the chemical species. This effect is mimicked in the BZ-aerosol OT system by

the water droplets, which traps and localizes some chemical species, forcing them to react

locally while allowing others in the oil mixture to be diffusive. If the diffusive species have a

much larger D ≫ Ddroplets , then it is similar to the adiabatic elimination introduced in the

derivation, when the local concentration of the diffusive species changes extremely fast in

comparison to the localizes species.

Given that chemical reactions, in particular BZ reactions can exhibit complex and chaotic

oscillations, the BZ-aerosol OT system is an excellent candidate experimental system for ob-

serving chimera states and synchronization defect lines in the complex and chaotic oscillatory

regimes.

107
Appendix A

Algorithms

A.1 Integration of Nonlocally Coupled Models

Integration of nonlocally coupled models, such as the nonlocal CGLE described in Chapter 2,

involves the computation of a convolution of the function G with the dynamical variables.

Instead of evaluating the convolution in a straightforward manner, we use instead the con-

volution theorem

f ⋆ g = F −1 [F [f ] · F [g]]. (A.1)

The Fourier transform is performed with the Fast Fourier Transform library, using the C++

wrapper class fftw3, which automatically assumes periodic boundary condition. Time inte-

gration is then carried out using an explicit first-order Euler method.

To illustrate the method, we consider the phase equation, Eq. (2.37). For convenience we

assume ω = 0. Using the identity sin(a + b) = sin a cos b + cos a sin b we split up the sinusoid

in the integrand on the right hand side of Eq. (2.37)

sin(φ(r, t) − φ(r′ , t) + α) (A.2)

= sin(φ(r, t) + α) cos φ(r′ , t) + cos(φ(r, t) + α) sin φ(r′ , t).

Using the convolution theorem, the integral can be written as

sin(φ(r, t) + α) · F −1 [F [G] · F [cos]] − cos(φ(r, t) + α) · F −1 [F [G] · F [sin]] (A.3)

The application of the same concept translates to the other models easily since there are

no further algebraic manipulations required. Note that then at each integration step when

108
∂t φ is computed, we only need to compute two forward and two backward Fourier transforms,

as the Fourier transform of G can be done once and stored at the start of integration.

A.1.1 No-flux Boundary Condition

If we wish to instead apply no-flux boundary condition on our systems, alteration of the

method described in the previous section is needed. No-flux boundary condition on a domain

D ⊂ R2 implies that the gradient in the direction of the outward normal n to the boundary

∂D is zero. We illustrate its implementation using the phase equation. In this case, the

domain D is a square of area L2 , which we discretized into N 2 points. The phase variable is

the set of points {φij |i, j = 0, ..., N − 1} evaluated at rij ∈ D.

Because the Fourier transform routine in fftw3 assumes periodic boundary condition,

the coupling will wrap around the square lattice, and we need to avoid this for implementing

no-flux boundary condition. We embed the field {φij } in a 2N × 2N lattice Θij , where for

i, j < N we have Θij = φij , while for i, j ≥ N we have Θij = 0. When the convolution integral

(evaluated via Fourier transform) in the phase equation is computed near the boundary at

i, j = N − 1, it receives no contributions from points i, j ≥ N [Press et al. , 2007].

If the convolution integral is evaluated in the region i, j ≥ N, points that are close to the

boundaries i, j = N − 1 and i, j = 2N − 1 are affected by the values of the field φij , so the

padding of zeros become “contaminated” after performing the Fourier transforms. According

to Eq. (A.3), at each integration step, we perform the convolutions f (φ) = F −1[F [G]·F [cos]]

and g(φ) = F −1 [F [G] · F [sin]] via the convolution theorem. To ensure that the outward

gradient φij is equal to zero at the true boundaries of the system i, j = N − 1, we perform

the following operations on the components fij = f (φij ) and gij = g(φij ) after each Fourier

transform step

109
fij += f2N −1−i,j + fi,2N −1−j + f2N −1−i,2N −1−j (A.4)

gij += g2N −1−i,j + gi,2N −1−j + g2N −1−i,2N −1−j (A.5)

for i, j < N. This is equivalent to the usual method of constructing a solution consistent

with no-flux boundary condition in the theory of partial differential equations. On a bounded

domain D ⊂ R2 , the particular solution consistent with no-flux boundary condition is created

by first solving the equations over R2 , then reflecting over the boundary ∂D, and added to the

solution restricted to D. This corresponds to exactly the operations Eq. (A.4) and Eq. (A.5).

After these operations are performed, we reset the zero padding Φij = 0, for i, j ≥ N.

A.2 Computation of the Phase Response Curve

We restate the formula for the phase response curve [Shima & Kuramoto, 2004]


1
Z

Γ(θ − θ ) = Z(λ + θ) · p(λ + θ′ , λ + θ)dλ, (A.6)
2π 0

where θ ≡ θ(r) and θ′ ≡ θ(r′ ). Z = (dX θ, dY θ, dZ θ) is the gradient of the phase. The vector

p = (pX , pY , pZ ) is given by the coupling function, which for the case of the nonlocally

coupled Rössler model, we have pX = X(r′ , t) − X(r, t), pY = Y (r′ , t) − Y (r, t) and pZ = 0.

Both quantities Z and p are to be evaluated on the limit cycle, so they may be parametrized

by the phase θ as Z ≡ Z(θ) and p ≡ p(θ′ , θ), as they appear in the integral Eq. (A.6).

Furthermore, this means that the oscillations at r and r′ can be described by the same orbit,

so we can instead write pX (θ′ , θ) ≡ X(θ(r′ )) − X(θ(r)) and pY (θ′ , θ) ≡ Y (θ(r′ )) − Y (θ(r)).

We compute the phase gradient Z according to [Brown et al. , 2004]. On the limit cycle

γ ∈ Rn , let θ(X) denote the phase of a point X ∈ γ, such that dt θ = ω evolves uniformly.

110
The definition of the gradient is given by

dθ δθ
= lim , (A.7)
dX |δX|→0 |δX|
where δX = X′ − X and X′ denotes a point not on γ. As in the formulation of a phase in the

Kuramoto model, discussed in Section 2.3, the phase is defined in a tubular neighbourhood

of γ via the notion of isochrons, so dt θ = ω evolves with a constant frequency even when the

trajectory X has been perturbed slightly off the limit cycle by a small disturbance in the

direction δX. Then, the phase difference δθ = θ(X + δX) − θ(X) may be computed in the

limit t → ∞ - it is the asymptotic difference as the perturbed trajectory returns to the limit

cycle γ.

Note that to perform the computation outlined in the previous paragraph, and to com-

pute the phase response curve, we need to have a uniformly evolving phase variable θ. As

mentioned in Section 2.1.3, this is not true for most definition of the phase, but we can

make a transformation θ → θ into a true phase variable. One way to do so is described in

[Kralemann et al. , 2008], which uses a transformation dΘ/dθ = σ(θ) where (2π)−1 σ(θ) is

the probability density distribution of the non-uniform phase θ, given by the inverse of the

averaged instantaneous frequency of the oscillation.

We use an approach to calculate the true phase based on linear interpolation. The method

described here is applicable to any general model. We discretize the time interval [0, N · dt]

into N points {t0 = 0, t1 = dt, ..., tN −1 = (N − 1)dt}, and let φ[i] = tan−1 (Y [i]/X[i]) denote

the value of the phase of the Rössler model at time ti (or a phase suitably defined for another

model that we are interested in), obtained by integrating Eq. (2.47)-(2.49) from the initial

condition (X0 , Y0, Z0 ). Similarly, We let Φ[j] = 2πj/M denote a linearly evolving phase at

discrete values of time tj ∈ [0, M · dt]. We let M > N and dt is given by the integration

step. For our algorithm we used dt = 0.005.

For each j, we can find a largest i such that φ[i] < Φ[j] < φ[i+1]. By linear interpolation,

111
we have Φ[j] = (1 − rj )φ[i] + rj φ[i + 1]. We then define an array m[j] as


m[j] = (rj + i), j = 0, 1, ..., M − 1 (A.8)
N
where rj and i are found by inverting linear interpolation of Φ[j], which is given above. We

now want to compute the transformed phase θ[i] from the non-uniformly evolving phase

φ[i]. We do this by performing a “reverse interpolation” using the map given by m[j].

For each i, we can write φ[i] = 2π(R + I)/M, where I is an integer and R is a fraction.

The largest integer j such that Φ[j] < φ[i] < Φ[j + 1] is equal to I = ⌊φ[i]M/(2π)⌋, and

R = φ[i]M/(2π) − I. This is essentially the reverse procedure of computing m[j]. Then, θ[i]

is given by the interpolation

θ[i] = (1 − R)m[I] + R · m[I + 1]. (A.9)

We make one remark here regarding the algorithm. We note that φ[i] can have discontinuities

in [0, N · dt] due to its definition. These discontinuities will carry over into the mapping

m[j] and produce a wrong result for θ[i]. Therefore, the interpolations performed in the

algorithm has to consider periodic boundaries for the phase variables. The asymptotic phase

difference δθ is measured following the steps described below. Let N be the number of

time steps corresponding to one period of oscillation in the Rössler model. We integrate

the system for K steps where K ≫ N, obtaining three arrays X[k], Y [k] and Z[k] (k =

0, 1, ..., K − 1), corresponding to the unperturbed trajectory. If this trajectory is subjected

to small perturbations dX, dY and dZ, applied independently to each component, at different

points on the limit cycle, we need to compute δθ at each point on cycle by following the

evolution of the unperturbed phase, as well as the evolution of the phase computed from

the perturbed trajectories. As such, we choose some large τ < K such that τ ≫ N. For

i = 0, 1, ..., N − 1 we set X ′ [i] = X[i] + dX, Y ′ [i] = Y [i] + dY and Z ′ [i] = Z[i] + dZ, and

define three sets of initial conditions: (X ′ [i], Y [i], Z[i]), (X[i], Y ′ [i], Z[i]) and (X[i], Y [i], Z ′ [i])

112
corresponding to independent perturbations. For each set of initial condition and each

i, we integrate the system for τ steps, and compute the corresponding perturbed phases

φdX [i], φdY [i] and φdZ [i] at time τ − 1. This can be visualized as subdividing the limit cycle

into N points, and integrating each point over τ steps, resulting in an overall rotation of the

initial values on the cycle.

Next, we need to uniformize the phases φ, φdX , φdY , φdZ → θ, θdX , θdY , θdZ by the method

described in the previous paragraphs, using φ to compute the map m, and using m in turn

to uniformize the phases. Since we want to compare the asymptotic differences between θ

and θdX , θdY , θdZ , we uniformize the phase φ[k] at the values k = τ, τ + 1, ..., τ + N − 1.

Since there is no coupling in the Z component of our nonlocal Rössler model, Eq. (4.1)-(4.3),

pZ = 0 and we need to only consider the first two components

dθ[i] θdX [i] − θ[i]


= (A.10)
dX dX
dθ[i] θdY [i] − θ[i]
= , i = 0, 1, ..., N − 1 (A.11)
dY dY

By defining ψ = λ + φ′ and ξ = φ − φ′ , we can rewrite Eq. (A.6) as


1
Z
Γ(ξ) = Z(ψ + ξ) · p(ψ, ψ + ξ)dψ, (A.12)
2π 0

which in terms of indices, is written as

N −1
1 X n dθ[(i + j) mod N]
Γ[|j − N/2|] = · (X[i] − X[(i + j) mod N])
N i=0 dX
dθ[(i + j) mod N] o
+ · (Y [i] − Y [(i + j) mod N]) , (A.13)
dY

where j = 0, 1, ..., N − 1. The value of Γ[|j − N/2|] corresponds to phase difference equal to

−π + 2πj/M. In our computation of the phase response curve, we have used values dX =

dY = 0.05, and the results are robust with respect to smaller values of these perturbations.

113
A.3 Detection of Synchronization Defect Lines

In Chapter 4, we employed a detection scheme that is different from [Davidsen et al. , 2004].

Our method is based on measuring the difference between the largest maxima between

neighbouring points in space. For each point r, we create a list T (r)[k] that saves the time

of occurrence of up to K most recent maxima of Z(r, t). This method is motivated by the

observation that Tmax (r) = maxk T (r)[k] should vary continuously for most points in the

system, while this quantity varies discontinuously across SDLs, and the difference in time

∆Tmax (r) = |Tmax (r) − Tmax (r′ )| for two adjacent points r and r′ on opposite sides of the

SDLs should reflect their periodicities. Recall from Chapter 2 that after m − 1 ≥ 1 period

doubling has occurred, the system is in the period-m regime and the oscillation has period

Tm−1 = 2m−1 T , where T is the period in the simple oscillatory regime. Across a period-1

and period-2 SDL, the value of ∆Tmax should correspond to roughly T and 2T . To use our

algorithm, we only need to have a rough estimate of T . We find in the nonlocal Rössler

model that T ≃ 6.00 on average, but there can be fluctuation by as much as ±0.2. The

detection method of [Davidsen et al. , 2004] depends on having a very stable value of T ,

which is not the case here, and we find that this method cannot resolve the defect lines as

clearly as our method.


To create the auxiliary field showing only the SDLs, in Figure 4.15 for example, we
calculate at N 2 points in our square domain the following quantities

∆Trow [i, j] = min(|Tmax [i + 1, j] − Tmax [i, j]|, KT − |Tmax [i + 1, j] − Tmax [i, j]|) (A.14)

∆Tcol [i, j] = min(|Tmax [i, j + 1] − Tmax [i, j]|, KT − |Tmax [i, j + 1] − Tmax [i, j]|), (A.15)

where Tmax (rij ) ≡ Tmax [i, j], and K is a parameter that specifies the number of most recent

maxima saved in the routine. The value KT is equal to the length of time in which we can

observe K maxima. The reason we defined ∆Trow [i, j] and ∆Tcol [i, j] in the above manner is

due to the special case when the spiral wavefront. During any time interval [t0 , t0 + KT ], the

wavefront separates points in the system at which the largest maxima occur close to t0 + KT

114
and points at which the largest maxima occur close to t0 . This happens because the spiral

wavefront has only reached part of the system when the detection routine is called. Figures

such as Figure 4.15 depicts an auxiliary field of N 2 points which we initially set to zero, and

assign values to each point (i, j) equal to ∆Trow [i, j]/T or ∆Tcol [i, j]/T , whenever ∆Trow [i, j]

or ∆Tcol [i, j] > T /2.

115
Bibliography

[Abrams & Strogatz, 2006] Abrams, D. M ., & Strogatz, S. H. 2006. Chimera states in a

ring of nonlocally coupled oscillators. International Journal of Bifurcation and Chaos,

16(1), 21–37.

[Abrams & Strogatz, 2004] Abrams, D. M., & Strogatz, S. H. 2004. Chimera states for

coupled oscillators. Physical Review Letters, 93, 174102.

[Abrams et al. , 2008] Abrams, D. M., Mirollo, R. E., Strogatz, S. H., & Wiley, D. A. 2008.

Solvable model for chimera states of coupled oscillators. Physical Review Letters, 101,

084103.

[Acebron et al. , 2005] Acebron, J. A., Bonilla, L. L., Vincente, C. J. Perez, Ritort, F., &

Spigler, R. 2005. The Kuramoto model: A simple paradigm for synchronization phenom-

ena. Review of Modern Physics, 77, 137–185.

[Alonso et al. , 2011] Alonso, S., John, K., & Bär, M. 2011. Complex wave patterns in an

effective reaction-diffusion model for chemical reactions in microemulsions. Journal of

Chemical Physics, 134, 094117.

[Aranson & Kramer, 2002] Aranson, I. S., & Kramer, L. 2002. The world of the complex

Ginzburg-Landau equation. Review of Modern Physics, 71(1), 99.

[Blaha, 2012] Blaha, K. 2012. personal communication.

[Bordyugov et al. , 2010] Bordyugov, G., Pikovsky, A., & Rosenblum, M. 2010. Self-

emerging and turbulent chimeras in oscillator chains. Physical Review E, 82, 035205.

[Brown et al. , 2004] Brown, E., Moehlis, J., & Holmes, P. 2004. On the phase reduction

and response dynamics of neural oscillator populations. Neural Computation, 16, 673 –

715.

116
[Buck, 1976] Buck, J. 1976. Synchronous fireflies. Scientific American, 234, May, 74.

[Buck, 1988] Buck, J. 1988. Synchronous rhythmic flashing of fireflies, II. The Quarterly

Review of Biology, 63(3), 265 – 289.

[Cross & Greenside, 2009] Cross, M., & Greenside, H. 2009. Pattern Formation and Dy-

namics in Nonequilibrium Systems. Cambridge: Cambridge University Press.

[Davidenko et al. , 1992] Davidenko, J. M., Perstov, A. M., Salomonsz, R., Baxter, W. T.,

& Jalife, J. 1992. Stationary and drifting spiral waves of excitation in isolated cardiac

muscle. Nature, 355, 349 – 351.

[Davidsen et al. , 2004] Davidsen, J., Erichsen, R., Kapral, R., & Chate, H. 2004. From

ballistic to Brownian vortex motion in complex oscillatory media. Physical Review Letters,

93, 018305.

[Engel et al. , 1991] Engel, A. K., König, P., Kreiter, A. K., & Singer, W. 1991. Interhemi-

spheric synchronization of oscillatory neuronal responses in cat visual cortex. Science,

252, 1177 – 1179.

[Goryachev & Kapral, 2000] Goryachev, A., & Kapral, R. 2000. Synchronization defect

lines. International Journal of Bifurcation and Chaos, 10(7), 1537.

[Hagerstrom et al. , 2012] Hagerstrom, A. M., Murphy, T. E., Roy, R., Hövel, P.,

Omelchenko, I., & Schöll, E. 2012. Experimental Observation of chimeras in couple-map

lattices. Nature Physics, 8, 658 – 661.

[Haim et al. , 1992] Haim, D., Lev, O., Pismen, L. M., & Sheintuch, M. 1992. Modeling

periodic and chaotic dynamics in anodic nickel dissolution. Journal of Physical Chemistry,

96(6), 2676 – 2681.

[Hanson, 1978] Hanson, F. E. 1978. Comparative studies of firefly pacemakers. Federation

Proceedings, 37, 2158.

117
[Huang et al. , 2004] Huang, X., Troy, W. C., Yang, Q., Ma, H., Laing, C., Schiff, S. J.,

& Wu, J. Y. 2004. Spiral waves in disinhibited mammalian neocortex. The Journal of

Neuroscience, 24(44), 9897 – 9902.

[Karma & Gilmour, 2007] Karma, A., & Gilmour, R. F. 2007. Nonlinear dynamics of heart

rhythm disorders. Physics Today, 60, 51 – 57.

[Kim et al. , 2007] Kim, T. Y., Woo, S.-J., Hwang, S.-M., Hong, J. H., & Lee, K. J. 2007.

Cardiac beat-to-beat alternations driven by unusual spiral waves. Proceedings to the Na-

tional Academy of Science, 104(28), 11639 – 11642.

[Kiss et al. , 2002] Kiss, I., Zhai, Z., & Hudson, J. L. 2002. Emerging coherence in a popu-

lation of chemical oscillators. Science, 296, 1676 – 1678.

[Kralemann et al. , 2008] Kralemann, B., Cimponeriu, L., Rosenblum, M., & Pikovsky, A.

2008. Uncovering interaction of coupled oscillators from data. Physical Review E, 77,

066205.

[Kuramoto, 1984] Kuramoto, Y. 1984. Chemical Oscillations, Waves, and Turbulence.

Berlin: Springer-Verlag.

[Kuramoto, 1997] Kuramoto, Y. 1997. Phase and center-manifold reductions for large popu-

lations of coupled oscillators with application to non-locally coupled systems. International

Journal of Bifurcation and Chaos, 7(4), 789 – 805.

[Kuramoto, 2003] Kuramoto, Y. 2003. Reduction methods applied to non-locally coupled

oscillator systems. Pages 209 – 227 of: Hogan, S. J., Champneys, A. R., Krauskopf, B.,

Bernardo, M Di, Wilson, R. E., Osinga, H. M., & Homer, M. E. (eds), Nonlinear Dynamics

and Chaos: Where Do We Go From Here? IOP Publishing Ltd.

[Kuramoto & Battogtokh, 2002] Kuramoto, Y., & Battogtokh, D. 2002. Coexistence of co-

herence and incoherence in nonlocally coupled phase oscillators: a soluble case. Nonlinear

118
Phenomena in Complex Systems, 5, 480.

[Kuramoto & Shima, 2003] Kuramoto, Y., & Shima, S. 2003. Rotating spiral wave without

phase singularity in reaction-diffusion systems. Progress of Theoretical Physics Supple-

ment, 150, 115 – 125.

[Laing, 2009a] Laing, C. R. 2009a. Chimera states in heterogenous networks. Chaos, 19,

013113.

[Laing, 2009b] Laing, C. R. 2009b. The dynamics of chimera states in heterogeneous Ku-

ramoto networks. Physica D, 238, 1569 – 1588.

[Laing, 2010] Laing, C. R. 2010. Chimeras in networks of planar oscillators. Physical Review

E, 81, 066221.

[Martens et al. , 2010] Martens, E. A., Laing, C. R., & Strogatz, S. H. 2010. Solvable model

of spiral wave chimeras. Physical Review Letters, 104, 044101.

[Motter, 2010] Motter, A. E. 2010. Spontaneous synchrony breaking. Nature Physics, 6,

164 – 165.

[Mukhametov et al. , 1977] Mukhametov, L. M., Supin, A. Y., & Polyakova, I. G. 1977.

Interhemispheric asymmetry of the electroencephalographic sleep patterns in dolphins.

Brain Research, 134, 581 – 584.

[Nicolis & Prigogine, 1977] Nicolis, G., & Prigogine, I. 1977. Self-Organization in Nonequi-

librium Systems. New York: Wiley.

[Omelchenko et al. , 2011] Omelchenko, I., Maistrenko, Y., Hövel, P., & Schöll, E. 2011.

Loss of coherence in dynamical networks: spatial chaos and chimera states. Physical

Review E, 21, 013112.

119
[Omel’chenko et al. , 2008] Omel’chenko, O. E., Maistrenko, Y. L., & Tass, P. A. 2008.

Chimera states: the natural link between coherence and incoherence. Physical Review

Letters, 100, 044105.

[Omel’chenko et al. , 2010] Omel’chenko, O. E., Wolfrum, M., & Maistrenko, Y. L. 2010.

Chimera states as chaotic spatiotemporal patterns. Physical Review E, 81, 065201(R).

[Omel’chenko et al. , 2012] Omel’chenko, O. E., Wolfrum, M., Yanchuk, S., Maistrenko,

Y. L., & Sudakov, O. 2012. Stationary patterns of coherence and incoherence in two-

dimensional arrays of non-locally-coupled phase oscillators. Physical Review E, 85, 036210.

[Ott & Antonsen, 2008] Ott, E., & Antonsen, T. M. 2008. Low dimensional behavior of

large systems of globally coupled oscillators. Chaos, 18, 056125.

[Park & Lee, 1999] Park, J.-S., & Lee, K. J. 1999. Complex periodic spirals and line-defect

turbulence in a chemical system. Physical Review Letters, 93, 5393.

[Pikovsky et al. , 2001] Pikovsky, A., Rosenblum, M., & Kurths, J. 2001. Synchronization:

A Universal Concept in Nonlinear Sciences. Cambridge, UK: Cambridge University Press.

[Press et al. , 2007] Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery, B. P.

2007. Numerical Recipes 3rd Edition: The Art of Scientific Computing. NY: Cambridge

University Press.

[Rattenborg et al. , 2000] Rattenborg, N. C., Amlaner, C. J., & Lima, S. L. 2000. Behav-

ioral, neurophysiological and evolutionary perspectives on unihemispheric sleep. Neuro-

science and Biobehavioral Reviews, 24, 817 – 842.

[Sandstede & Scheel, 2007] Sandstede, B., & Scheel, A. 2007. Peiod doubling of spiral waves

and defects. SIAM Journal on Applied Dynamical Systems, 6, 494 – 547.

[Scott, 1991] Scott, S. 1991. Chemical Chaos. NY: Oxford University Press.

120
[Shima & Kuramoto, 2004] Shima, S., & Kuramoto, Y. 2004. Rotating spiral waves with

phase-randomized core in nonlocally coupled oscillators. Physical Review E, 69, 036213.

[Singer & Gray, 1995] Singer, W., & Gray, C. M. 1995. Visual feature integration and the

temporal correlation hypothesis. Annual Review of Neuroscience, 18, 555 – 586.

[Strogatz, 2000a] Strogatz, S. H. 2000a. From Kuramoto to Crawford: exploring the onset

of synchronization in populations of coupled oscillators. Physica D, 143, 1–20.

[Strogatz, 2000b] Strogatz, S. H. 2000b. Nonlinear Dynamics and Chaos: With Applica-

tions to Physics, Biology, Chemistry, and Engineering. Cambridge, MA: Perseus Books

Publishing, LLC.

[Tass & Haken, 1996] Tass, P. A., & Haken, H. 1996. Synchronization in networks of limit

cycle oscillators. Zeitshcrift fur Physik B, 100, 303 – 320.

[Tinsley et al. , 2012] Tinsley, M. R., Nkomo, S., & Showalter, K. 2012. Chimera and phase-

cluster states in populations of coupled chemical oscillators. Nature Physics, 8, 662 – 665.

[Ts’o & Gilbert, 1988] Ts’o, D., & Gilbert, C. 1988. The organization of chromatic and

spatial interactions in the primate striate cortex. The Journal of Neuroscience, 8, 1712 –

1727.

[Ts’o et al. , 1986] Ts’o, D., Gilbert, C., & Wiesel, T. N. 1986. Relationship between hori-

zontal interactions and functional architecture in cat striate cortex as revealed by cross-

correlation analysis. The Journal of Neuroscience, 6, 1160 – 1170.

[Turing, 1952] Turing, A. 1952. The chemical basis of morphogenesis. Philosophical Trans-

actions of the Royal Society of London, 237, 37 – 72.

[Wiesenfeld et al. , 1996] Wiesenfeld, K., Colet, P., & Strogatz, S. H. 1996. Synchronization

transitions in a disordered Josephson series array. Physical Review Letters, 76, 404 – 407.

121
[Wiesenfeld et al. , 1998] Wiesenfeld, K., Colet, P., & Strogatz, S. H. 1998. Frequency lock-

ing in Josephson arrays: Connection with the Kuramoto model. Physical Review E, 57,

1563 1569.

[Winfree, 1980] Winfree, A. 1980. The Geometry of Biological Time. Berlin: Springer-

Verlag.

[Winfree, 1967] Winfree, A. T. 1967. Biological rhythms and the behavior of populations of

coupled oscillators. Journal of Theoretical Biology, 16, 15 – 42.

[Wolfrum & Omel’chenko, 2011] Wolfrum, M., & Omel’chenko, O. E. 2011. Chimera states

are chaotic transients. Physical Review E, 84, 015201(R).

[Yeung & Strogatz, 1999] Yeung, M. K. S., & Strogatz, S.H. 1999. Time delay in the Ku-

ramoto model of coupled oscillators. Physical Review Letters, 82, 648.

[Zhabotinsky, 1964] Zhabotinsky, A. M. 1964. Periodical oxidation of malonic acid in solu-

tion (a study of the Belousov reaction kinetics). Biofizika, 9, 306 – 311.

122

Вам также может понравиться