Synchronization in Oscillatory Networks: Grigory V. Osipov Jürgen Kurths Changsong Zhou

S P R I N G E R S E R I E S Springer ..
I N S YN E R G E T I C S COMPLEXITY
Grigory V. Osipov
Jürgen Kurths
Changsong Zhou
Synchronization
in Oscillatory
Networks

Springer Complexity
Springer Complexity is an interdisciplinary program publishing the best research
and academic-level teaching on both fundamental and applied aspects of complex
systems – cutting across all traditional disciplines of the natural and life sciences,
engineering, economics, medicine, neuroscience, social and computer science.
Complex Systems are systems that comprise many interacting parts with the abil-
ity to generate a new quality of macroscopic collective behavior the manifestations
of which are the spontaneous formation of distinctive temporal, spatial or functional
structures. Models of such systems can be successfully mapped onto quite diverse
“real-life” situations like the climate, the coherent emission of light from lasers,
chemical reaction-diffusion systems, biological cellular networks, the dynamics of
stock markets and of the internet, earthquake statistics and prediction, freeway traf-
fic, the human brain, or the formation of opinions in social systems, to name just
some of the popular applications.
Although their scope and methodologies overlap somewhat, one can distinguish
the following main concepts and tools: self-organization, nonlinear dynamics, syn-
ergetics, turbulence, dynamical systems, catastrophes, instabilities, stochastic pro-
cesses, chaos, graphs and networks, cellular automata, adaptive systems, genetic al-
gorithms and computational intelligence.
The two major book publication platforms of the Springer Complexity program
are the monograph series “Understanding Complex Systems” focusing on the vari-
ous applications of complexity, and the “Springer Series in Synergetics”, which is
devoted to the quantitative theoretical and methodological foundations. In addition
to the books in these two core series, the program also incorporates individual titles
ranging from textbooks to major reference works.
Springer Series in Synergetics
Founding Editor: H. Haken
The Springer Series in Synergetics was founded by Herman Haken in 1977. Since
then, the series has evolved into a substantial reference library for the quantitative,
theoretical and methodological foundations of the science of complex systems.
Through many enduring classic texts, such as Haken’s Synergetics and In-
formation and Self-Organization, Gardiner’s Handbook of Stochastic Methods,
Risken’s The Fokker Planck-Equation or Haake’s Quantum Signatures of Chaos,
the series has made, and continues to make, important contributions to shaping
the foundations of the field.
The series publishes monographs and graduate-level textbooks of broad
and general interest, with a pronounced emphasis on the physico-mathematical
approach.
Editorial and Programme Advisory Board

Péter Érdi
Center for Complex Systems Studies, Kalamazoo College, USA,
and Hungarian Academy of Sciences, Budapest, Hungary
Karl Friston
Institute of Cognitive Neuroscience, University College London,
London, UK
Hermann Haken
Center of Synergetics, University of Stuttgart, Stuttgart, Germany
Janusz Kacprzyk
System Research, Polish Academy of Sciences, Warsaw, Poland
Scott Kelso
Center for Complex Systems and Brain Sciences, Florida Atlantic University, Boca Raton, USA
Jürgen Kurths
Nonlinear Dynamics Group, University of Potsdam, Potsdam, Germany
Linda Reichl
Center for Complex Quantum Systems, University of Texas, Austin, USA
Peter Schuster
Theoretical Chemistry and Structural Biology, University of Vienna, Vienna, Austria
Frank Schweitzer
System Design, ETH Zurich, Zurich, Switzerland
Didier Sornette
Entrepreneurial Risk, ETH Zurich, Zurich, Switzerland
Grigory V. Osipov Jürgen Kurths
Changsong Zhou
Synchronization
in Oscillatory Networks
With 221 Figures

Professor Dr. Grigory V. Osipov
Department of Radiophysics
Nizhny Novgorod University
23, Gagarin Avenue
603600 Nizhny Novgord, Russia
Professor Dr. Jürgen Kurths

Institute of Physics
University of Potsdam
Am Neuen Palais 19
14469 Potsdam, Germany
Dr. Changsong Zhou

Institute of Physics
University of Potsdam
Am Neuen Palais 19
14469 Potsdam, Germany
Library of Congress Control Number: 2007927818

ISSN 0172-7389
ISBN 978-3-540-71268-8 Springer Berlin Heidelberg New York
This work is subject to copyright.All rights are reserved,whether the whole or part of the material is concerned,
specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction
on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is
permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version,
and permission for use must always be obtained from Springer. Violations are liable to prosecution under the
German Copyright Law.
Springer is a part of Springer Science+Business Media
springer.com
© Springer-Verlag Berlin Heidelberg 2007
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply,
even in the absence of a specific statement, that such names are exempt from the relevant protective laws and
regulations and therefore free for general use.
Typesetting by the authors and SPi using a Springer LATEX macro package
Cover design: WMXDesign, Heidelberg
Printed on acid-free paper SPIN: 10997659 54/3180/SPi 543210
To my father Vladimir G.O.
To my mother Gerda J.K.
To my wife Wei C.S.Z.
Preface
The formation of collective behavior in large ensembles or networks of coupled

oscillatory elements is one of the oldest problem in the study of dynami-
cal systems. Nevertheless, it is an actually challenging field for a theoretical
understanding as well as for applications in various disciplines, ranging from
physics, chemistry, earth sciences via biology and neuroscience to engineering,
business and social sciences. Due to the large number of effective degrees of
freedom in spatially extended systems, a rich variety of spatiotemporal regimes
is observed. Three main types of collective behavior are distinguished (1) a
fully incoherent state or highly developed spatiotemporal disorder; (2) par-
tially coherent states, where some of the participants in the network behave in
some common rhythm, forming clusters; (3) a fully coherent state or a regime
of globally synchronized elements. The basic phenomenon of these structure
formations is synchronization, i.e. regime of coherent activity, which is uni-
versal in many dynamical systems and can be understood from the analysis
of common models of oscillatory networks.
Cooperative phenomena in ensembles of globally (mean-field) coupled
phase equations were studied first by Winfree and Kuramoto. They showed
that if the coupling is strong enough mutual synchronization emerges. In
contrast to the mean-field Winfree and Kuramoto models, Ermentrout and
Kopell’s classic works deal with chains of phase equations. Besides cluster
and global synchronization effects, the main results there involve travelling
waves. A main part of this book presents different aspects of synchronization
in chains and lattices of locally interconnected nonidentical nonlinear oscilla-
tory elements. Main kinds of collective behavior are discussed for prototypical
systems: limit cycle oscillators (van der Pol generators), chaotic oscillators
(Rössler and Lorenz oscillators), phase oscillators (pendulum-like systems)
and discrete in time systems – circle maps. The typical synchronization phe-
nomena in such networks are quite similar because these models under some
assumptions, such as weak coupling, can be reduced to the general model
of coupled oscillators – a network of continuous or discrete phase oscilla-
tors. In general, the amplitude evolution should be taken into account. Then
VIII Preface
amplitude-dependent effects, for instance, oscillations death and amplitude

turbulence, can be observed. The response of ensembles of excitable elements
to external forces resembles synchronization phenomena in oscillatory ensem-
bles. These effects as well as the constructive role of noise in synchronization
problems and collective effects in complex networks, such as small-world and
scale-free are discussed as well.
These target models describe many real objects, such as electric power grid,
lasers, phase arrays, the heart muscle, etc. Therefore, the proposed research
issues in the complex behavior, synchronization and control in networks of
nonlinear oscillators allow to investigate novel applications of synchronization
and control of dynamical systems in signal processing, communications, power
systems and biological engineering. This book will be interesting for students,
postgraduate students, and specialists in physics and applied mathematics,
but also in various applied areas, such as engineering, biology, and social life
sciences.
Contents
Part I: Basics on Synchronization and Paradigmatic Models
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Synchronization Phenomena in Nature, Physics,
and Engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Goal of the Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Terminological Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Bibliographical Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Basic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 Harmonic Oscillator: Amplitude, Frequency
and Phase of Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Van der Pol Oscillator: Quasi-Harmonic
and Relaxation Limit Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Rössler Oscillator: From Phase-Coherent
to Funnel Chaotic Attractors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Lorenz Oscillator: “Classic” and Intermittent Chaotic
Attractors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Phase Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5.1 First-Order Phase Oscillator (Active Rotator) . . . . . . . 21
2.5.2 Second-Order Phase Oscillator
(Pendulum-Like System) . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.3 Third-Order Phase Oscillator (Chaotic Rotator) . . . . . 24
2.5.4 Discrete-Time Rotator (Circle Map) . . . . . . . . . . . . . . . . 24
2.6 Discrete Map for Spiking–Bursting Neural Activity . . . . . . . . . . 28
2.7 Excitable Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.7.1 Hodgkin–Huxley Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.7.2 FitzHugh–Nagumo Model . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.7.3 Luo–Rudy Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
X Contents
3 Synchronization Due to External Periodic Forcing . . . . . . . . . 35

3.1 Synchronization of Limit-Cycle Oscillator
by External Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.1.1 Weak Forcing: Phase Description . . . . . . . . . . . . . . . . . . 36
3.1.2 Synchronization of a van der Pol Oscillator
by External Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Phase Synchronization of a Chaotic Rössler Oscillator
by External Driving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3 Imperfect Phase Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 Transition to the Regime of Chaotic Phase Synchronization:
The Role of Unstable Periodic Orbits . . . . . . . . . . . . . . . . . . . . . . 45
3.5 External Phase Synchronization
of Chaotic Intermittent Oscillators . . . . . . . . . . . . . . . . . . . . . . . . 47
3.5.1 Forced Model Quadratic Map . . . . . . . . . . . . . . . . . . . . . 47
3.5.2 Forced Lorenz Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.6 Synchronous Response of Excitable Systems
to a Periodic External Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4 Synchronization of Two Coupled Systems . . . . . . . . . . . . . . . . . . 55

4.1 Synchronization of Regular Systems . . . . . . . . . . . . . . . . . . . . . . . 55
4.1.1 Phase Dynamics Approach . . . . . . . . . . . . . . . . . . . . . . . . 56
4.1.2 Synchronization of Two Coupled van der Pol
Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.1.3 Synchronization of Coupled Active Rotators . . . . . . . . . 66
4.2 Synchronization of Coupled Chaotic Oscillators . . . . . . . . . . . . . 68
4.2.1 Phase Synchronization of Rössler Oscillators . . . . . . . . 68
4.2.2 Synchronization of Coupled Intermittent Oscillators . . 77
4.2.3 Oscillatory and Rotatory Synchronization
of Chaotic Phase Systems . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.3 Synchronization of Coupled Circle Maps . . . . . . . . . . . . . . . . . . . 90
4.3.1 Regular Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.3.2 Chaotic Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Part II: Synchronization in Geometrically Regular Ensembles
5 Ensembles of Phase Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

5.1 General Model and Malkin’s Theorem . . . . . . . . . . . . . . . . . . . . . 104
5.2 Unidirectional Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.3 Synchronization Phenomena in a Chain
of Bidirectionally Coupled Phase Oscillators . . . . . . . . . . . . . . . . 112
5.3.1 Synchronization, Clustering and Multistability
in Chains with Linearly Distributed
Individual Frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Contents XI
5.3.2 Synchronization Transitions in Chains

with Randomly Distributed Individual Frequencies . . . 119
5.4 Influence of Non-Uniform Rotations on the Synchronization . . 121
5.5 Mutual Entrainment in Populations of Globally Coupled
Phase Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.6 Synchronization Phenomena in a Chain of Coupled
Pendulum-Like Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6 Chains of Coupled Limit-Cycle Oscillators . . . . . . . . . . . . . . . . . 129

6.1 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.2 Synchronization Clusters and Multistability at Linear
Variation of Individual Frequencies Along the Chain . . . . . . . . . 130
6.2.1 Model Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
6.2.2 Global Synchronization in an Assembly, Stationary
Phase Distributions, Synchronization area . . . . . . . . . . 133
6.2.3 Regimes of Cluster Synchronization . . . . . . . . . . . . . . . . 135
6.2.4 Multistability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.3 Oscillation Death . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.4 Effects of Nonuniformity of the Frequency Mismatch
Gradient in the Formation of Synchronized Clusters . . . . . . . . . 145
6.4.1 Sensitivity of the Structures to Regular
Nonuniformities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.4.2 The Effect of Random Dispersion of Individual
Frequencies on Cluster Synchronization . . . . . . . . . . . . . 146
6.5 Synchronization in a Chain of van der Pol Oscillators . . . . . . . . 147
6.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
7 Ensembles of Chaotic Oscillators with a Periodic-Doubling

Route to Chaos, Rössler Oscillators . . . . . . . . . . . . . . . . . . . . . . . . 151
7.1 Synchronization Effects in a Homogeneous Chain
of Rössler Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
7.2 Basic Model of a Nonhomogeneous Chain, Phase
and Frequency Definitions, and Criteria
of Phase Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
7.3 Phase Synchronization in a Chain with a Linear Distribution
of Natural Frequencies, Phase-Coherent Rössler Oscillators . . . 154
7.3.1 Theoretical Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
7.3.2 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
7.4 Synchronization in a Chain with Randomly Distributed
Natural Frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
7.5 Phase Synchronization of Rössler Oscillators
with the Funnel Attractor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
XII Contents
7.6 Anomalous Collective Behavior of Coupled

Chaotic Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
7.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
8 Intermittent-Like Oscillations in Chains

of Coupled Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
8.1 Model of Coupled Intermittent Maps, Phase
and Frequency, Synchronization Criteria . . . . . . . . . . . . . . . . . . . 170
8.2 Linearly Distributed Control Parameters, Soft Transition
to Global Synchronization Regime . . . . . . . . . . . . . . . . . . . . . . . . 171
8.3 Randomly Distributed Control Parameter, Transition
to Spatiotemporal Intermittency . . . . . . . . . . . . . . . . . . . . . . . . . . 173
8.4 Collective Oscillations in a Chain of Spiking Maps . . . . . . . . . . 177
8.5 Synchronization in Ensembles of Globally
Coupled Bursting Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
8.5.1 Mutual Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . 180
8.5.2 External Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . 182
8.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
9 Regular and Chaotic Phase Synchronization

of Coupled Circle Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
9.1 Common Model for a Chain of Coupled Circle Maps . . . . . . . . 188
9.2 Synchronization in a Chain of Identical Circle Maps . . . . . . . . . 189
9.2.1 Symmetrically Coupled Maps . . . . . . . . . . . . . . . . . . . . . 190
9.2.2 Effect of Asymmetry of Coupling . . . . . . . . . . . . . . . . . . 195
9.2.3 Synchronization in Lattices of Coupled Maps . . . . . . . . 197
9.3 Ensembles of Coupled Nonidentical Circle Maps
and Criteria of Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
9.4 Synchronization and Clustering in a Chain of Regular CMs . . 200
9.4.1 Linear Distribution of Individual Frequencies . . . . . . . . 200
9.4.2 Random Distribution of Individual Frequencies . . . . . . 206
9.5 Chaotic Phase Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
9.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
10 Controlling Phase Synchronization

in Oscillatory Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
10.1 General Principles of Automatic Synchronization . . . . . . . . . . . 214
10.2 Two Coupled Poincaré Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
10.3 Coupled van der Pol and Rössler Oscillators . . . . . . . . . . . . . . . . 217
10.4 Two Coupled Rössler Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . 220
10.5 Coupled Rössler and Lorenz Oscillators . . . . . . . . . . . . . . . . . . . . 223
10.6 Principles of Automatic Synchronization
in Networks of Coupled Oscillators . . . . . . . . . . . . . . . . . . . . . . . . 224
10.7 Synchronization of Locally Coupled Regular Oscillators . . . . . . 225
10.8 Synchronization of Locally Coupled Chaotic Oscillators . . . . . . 228
Contents XIII
10.9 Synchronization of Globally Coupled Chaotic Oscillators . . . . . 230

10.10 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
11 Chains of Limit-Cycle Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . 233

11.1 Introduction and Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
11.2 Mechanism of Localized Structure Formation . . . . . . . . . . . . . . . 235
11.3 Dissipative Coupling (Zero “Dispersion”) . . . . . . . . . . . . . . . . . . 235
11.3.1 Desynchronization of Front Propagation . . . . . . . . . . . . 235
11.3.2 Localized Synchronization Structures . . . . . . . . . . . . . . . 237
11.3.3 Nonlocal Synchronization
in Nonhomogeneous Chains . . . . . . . . . . . . . . . . . . . . . . . 238
11.3.4 Fully Incoherent (Turbulent-Like) Oscillations . . . . . . . 239
11.4 Nonscalar (Dissipative and Conservative) Coupling . . . . . . . . . . 241
11.4.1 Bursting Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
11.4.2 Nonpropagation to Propagation Transition
via Intermittency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
11.4.3 Noise Influence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
11.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
12 Chains and Lattices of Excitable Luo–Rudy Systems . . . . . . . 251

12.1 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
12.2 Cardiac Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
12.3 Methods: Theoretical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
12.4 Computational Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
12.4.1 One-Dimensional Simulations . . . . . . . . . . . . . . . . . . . . . 255
12.4.2 Two-Dimensional Simulations . . . . . . . . . . . . . . . . . . . . . 261
12.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Part III: Synchronization in Complex Networks and Influence

of Noise
13 Noise-Induced Synchronization in Ensembles of Oscillatory

and Excitable Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
13.1 Degrading Effects of Noise: Noise-Induced Phase Slips . . . . . . . 270
13.2 Noise-Induced CS and PS in Uncoupled Chaotic Oscillators . . 273
13.2.1 Noise-Induced CS of Identical Chaotic Oscillators . . . . 273
13.2.2 Noise-Induced PS of Nonidentical Uncoupled
Chaotic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
13.3 Noise-Enhanced PS in Weakly Coupled Chaotic Oscillators . . . 288
13.3.1 Noise-Enhanced PS of a Chaotic Laser
Due to Periodic Forcing . . . . . . . . . . . . . . . . . . . . . . . . . . 289
13.3.2 Noise-Enhanced PS of Two Coupled Rössler
Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
XIV Contents
13.3.3
Noise-Enhanced PS in Arrays of Globally Coupled
Rössler Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
13.3.4 Experimental Observation of Noise-Enhanced PS . . . . 297
13.4 Noise-Enhanced Synchronization-Like Phenomena
in Arrays of Coupled Excitable Cells . . . . . . . . . . . . . . . . . . . . . . 305
13.4.1 Phase Synchrony in Chains of Coupled Noisy
Excitable Neurons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
13.4.2 Noise-Enhanced PS of Coupled Excitable Neurons
by External Forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
13.4.3 Resonant Pattern Formation in 2D Arrays . . . . . . . . . . 313
13.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
14 Networks with Complex Topology . . . . . . . . . . . . . . . . . . . . . . . . . . 317

14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
14.2 Dynamical Equations and Stability Analysis . . . . . . . . . . . . . . . . 320
14.3 Phase Synchronization in Small-World Networks
of Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
14.4 Synchronization in Scale-Free Networks of Oscillators . . . . . . . . 324
14.5 Mean-Field Analysis of Hierarchical Synchronization . . . . . . . . 331
14.6 Synchronization Properties of Weighted Networks . . . . . . . . . . . 332
14.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
Part I
Basics on Synchronization and Paradigmatic

Models
1
Introduction
1.1 Synchronization Phenomena in Nature, Physics,

and Engineering
Fortunately an important feature in our living world is the tendency to achieve
common rhythms of mutual behavior, or is, in other words, the tendency
to synchronization. This phenomenon of synchronization is extremely wide
spread in nature as well as in the realm of technology and society. Synchro-
nization is usually understood as the capacity of objects of different nature to
form a common operation regime due to interaction or forcing. The fact that
various objects seek to achieve order and harmony in their behavior, which is
a characteristic of synchronization, is a manifestation of the natural tendency
to self-organization existing everywhere in nature [1–5, 7–10].
Synchronization processes can be met in physics, chemistry, engineering
and mechanics, life sciences and medicine, economics, or social life. Synchro-
nization is possible if at least two elements are coupled but it much more often
happens in ensembles including hundreds, thousands, and even more subsys-
tems. Examples of large biological ensembles include collections of flashing
fireflies [12, 13], pacemaker cells in the heart [14–16], insulin-secreting cells
of the pancreas [17], crickets that chirp in unison [18], cells in the mam-
malian small intestine [19], rhythmic activity in the brain due to synchronized
firing of a huge number of neurons [20–25], synchronization during epilep-
tic seizures [26], groups of women whose menstrual period become synchro-
nized [30], or population dynamics [31, 32]. From an engineering perspective
the collective behavior of laser and power generator arrays is of special prac-
tical importance [38–44]. Mutual coherent behavior allows to generate much
stronger power than can be obtained by a single laser. Arrays of microwave
oscillators [33] and arrays of super-conducting Josephson junctions [34,35] are
another object of intensive research. In mechanics synchronization has found
wide application in the construction of various vibro-technical devices [3]
and robot manipulators [36, 37]. In radio-physics, radio-engineering, radio-
location, radio-measurements and radio-communication, synchronization is
4 1 Introduction
employed for frequency stabilization of generators, for synthesizing frequen-

cies and demodulation of signals in Doppler systems, in exact time systems,
by designing phase antenna arrays [57]. The construction of modern radio
systems and complexes is, in fact, impossible without making use of synchro-
nization. Several secure and efficient communication schemes are based on
chaotic synchronization [158–161]. Applications also include chemical oscilla-
tions or waves in the Belousov–Zhabotinsky reaction [55, 56], and synchro-
nization phenomena in production networks [54]. Synchronization problems
are also crucial for the design of computers with parallel architecture [53].
Formation of “Mexican wave” during football match and simultaneous hand
clapping are also manifestations of synchronous behavior.
The study of synchronization has become a main field in non-linear science.
The theory of synchronization can be subdivided into three main parts:
(1) Classical theory of synchronization, which studies phenomena in coupled
periodic self-oscillatory systems
(2) Theory of chaotic synchronization which deals with the cooperative
behavior of chaotic subsystems
(3) Theory of synchronization-like phenomena in noise and excitable systems
Historically, a systematic analysis of synchronization phenomena was
started with the following problem. Let us take a self-oscillatory system (i.e.,
it exhibits a sustained oscillatory motion without any external driving) which
displays periodic oscillations. Then it is additionally driven by an external
periodic force. As a result of such forcing one can observe that the forced
systems becomes to follow the driving signal, namely, the frequency (period)
of oscillations in the forced systems becomes equal to the forcing frequency
(period). This phenomenon is called external synchronization.
Two or many weakly mutually coupled periodic self-oscillatory systems
with slightly different parameters can oscillate with the same frequency
(period). This mutual synchronization was first observed in the 17th century
by the Dutch physicist Christian Huygens for two coupled pendulum clocks
hanging on the same wooden beam [137]. It is interesting to note that he
called this effect sympathy. Huygens’ finding was the beginning of Nonlinear
Sciences.
After the discovery of deterministic chaos the search for synchronization
has moved to chaotic systems. This has caused strongly increasing research on
the theory of chaotic oscillations and its applications due to the high interest
in synchronization of chaotic oscillators.
Now we distinguish three main types of synchronization in chaotic systems:
(1) Complete (or full) synchronization (CS)
(2) Generalized synchronization (GS)
(3) Phase synchronization (PS)
Complete synchronization of identical systems occurs when the initially
different states of all coupled systems become identical [1]. Generalized
1.2 Goal of the Book 5
synchronization implies that the output of a driving system is associated with

a given function of the output of another one, the slave system [1]. Chaotic
phase synchronization is somewhat similar to the synchronization of periodic
oscillations and is manifested in the appearance of certain relations between the
phases of interacting nonidentical systems and results in the establishment of
a coincidence of characteristic time scales of the coupled systems. Remarkably
the amplitudes of oscillations often remain chaotic and practically uncorre-
lated [1]. Exactly this type of synchronization will be in the main focus of this
book.
The main necessary ingredient for chaotic systems to become phase syn-
chronized is the existence of characteristic time scales or rhythms which allows
to observe and to investigate synchronization and its characteristics for cou-
pled periodic and chaotic systems from a common point of view. For dynam-
ical systems, that will be considered here, synchronization problems will be
formulated in terms of coincidence of their characteristic time scales (charac-
teristic frequencies). For periodic units they are just the periods (frequencies)
of oscillations but for chaotic or stochastic units they are the averaged time
intervals (averaged frequencies) between some repeated events. Hence, we will
study conditions for frequency entrainment as one (weak) of the criteria to
identify synchronized behavior. Besides the characteristic frequency of oscil-
lations, a rhythmic behavior offers to introduce the phases of oscillations –
another crucial characteristic of motion. So the phase locking can be regarded
as another – sometimes even stronger – criterion for synchronization. In the
frames of such phase and frequency approach it is quite natural, that syn-
chronization processes in various systems of different nature will have close
similarities and can be studied by using common tools.
We will focus in this monography on both the classical theory of syn-
chronization as well as the phase synchronization theory of chaotic systems.
A crucial point is to explain how the phase dynamics can be approximated
in such complex and oscillatory systems. This is first given for periodically
forced and two coupled systems (Part I) and then for regular and complex
networks of oscillators (Part II and III).
1.2 Goal of the Book
There are some instructive monographs on synchronization but they are

mainly related to synchronization of a few coupled subsystems, or focus on
certain field of applications, or are more popular (see Sect. 1.4). As explained
in Sect. 1.1, most synchronization examples in our living world are related to
an interplay of many subsystems, i.e., ensembles or networks of coupled units.
The study of such networks has become nowadays a main topic in a broader
research on complex systems and is in the focus of this book. In the last
years, much progress has been achieved in this direction. But so far there is
no general theory of dynamic behavior of nonlinear synchronization networks.
6 1 Introduction
i=1,...,N
1,1 1,2 1,N
2,1 2,2 2,N
j=1,...,M
M,1 M,2 M,N
Fig. 1.1. Lattice structure of nearest-neighbor coupled systems
Even the possibility of the existence of a total synchronization regime in a

network is still unclear. Structural complexity, connections diversity, dynam-
ical complexity, etc. make the network study even with the possibilities of
modern computers very difficult and studies of synchronization in networks
with complicated topology are in its infancy. Therefore, we mainly confine
ourselves to some variants of suitable geometrically regular network configu-
ration. Hence, a lattice model will be chosen as a basic scheme for networks in
Part II. As one basic scheme of coupling between the elements we take nearest-
neighbor coupling (Fig. 1.1). Also long-range and global coupling schemes will
be presented. Additionally, we mostly suppose that all units are nonidentical
(but not too different).
The common mathematical model of such a network is
Ẋi,j = F (Xi,j ) + d1 H(Xi,j+1 − Xi,j ) + d2 H(Xi,j−1 − Xi,j )
+ d3 H(Xi+1,j − Xi,j ) + d4 H(Xi−1,j − Xi,j ), (1.1)
i = 1, ... , N, j = 1, ... , M,
where Xi,j is an n-dimensional vector of the (i, j)th oscillator variables.
F (Xi,j ) : Rn → Rn is a vector function, n the dimension of the individual
oscillator, d1 , d2 , d3 , d4 are the coupling parameters determining the coupling
between the oscillators in the four directions of the lattice. In dependence
on the coupling coefficients, we can consider uni- or bidirectional coupling.
H is the n × n matrix, which determines the function of the coupling and by
1.3 Terminological Remarks 7
which the variables of the oscillators are coupled, N and M define the size of
the lattice. The mainly studied boundary conditions are free ends.
In the case of one-dimensional in space structure the network is reduced
to a chain and the model (1.1) can be rewritten as:
Ẋj = F (Xj ) + d1 H(Xj+1 − Xj ) + d2 H(Xj−1 − Xj ),

(1.2)
j = 1, . . . , N
The model structure of (1.1) or (1.2) is analogous to network models of

continuous media and lattice models used to study problems of turbulence,
spin glasses, or cellular automata [95]. In this sense (1.1) is quite common and
can be regarded as, generally speaking, a discrete model of nonequilibrium
dissipative media. From this point of view, its exploration is of interest as an
approach of understanding complex phenomena in continuous nonequilibrium
media such as turbulence and structure formation.
Many real-world systems displays very complicated coupling topology and
some basic types of them will be discussed as well from the viewpoint of
synchronization in Part III.
The main problems we will focus in our book are:
(1) Existence, formation and stability of global and cluster synchronization
(2) Characteristics of synchronization regime such as the distribution over
the network of residual errors in the phase and the frequency of synchro-
nization
(3) Conditions and ways of entering different synchronization regimes and
conditions and ways of their violation (de-synchronization) in time and
space
(4) Characteristics of non-synchronous regimes
(5) Possibilities of controlling synchronization
(6) Influence of specific oscillatory properties of single elements on the syn-
chronization transitions
(7) Common principles of synchronization phenomena for different networks
(8) Influence of coupling schemes on synchronization
We will also focus on effects of noise, which is typical in many natural
and experimental systems, on synchronization and synchronization-like pheno-
mena in networks of oscillatory and excitable elements and will demonstrate
that noise can even play a constructive role, i.e., it can induce or enhance
synchronization (Part III).
1.3 Terminological Remarks

To avoid misunderstandings and make reading easier, we now give a few basic
terms applied in this book.
We refer synchronization as a process of an adjustment of rhythms of oscil-
lations of two or many systems due to (weak) coupling or forcing.
8 1 Introduction
We will characterize the rhythm of oscillations by the characteristic time

scale or frequencies which are:
(1) For periodic oscillations – the period (T ) or frequency (Ω) of the oscilla-
tions
(2) For chaotic oscillations – the averaged time (T ) or frequency (Ω) of
appearance of some repeating events
If it will not be specified as oscillators we consider self-oscillatory systems.
We call a regime synchronous when the following conditions are fulfilled:
– We will conclude that two arbitrary (not necessary neighboring) continu-
ous in time oscillators i and j are m : n synchronized, when
mΩi = nΩj , (1.3)

−1
where m and n are integers and Ωi,j ∼ Ti,j .
– Besides this criterion of frequency entrainment, another – sometimes
stronger – criterion of m : n synchronous regime between oscillators i
and j is phase locking:
(1) Strict phase locking (the phase shift is constant)
|mφi (t) − nφj (t)| = Const, (1.4)
(2) Non-strict phase locking (the phase shift is bounded)
|mφi (t) − nφj (t)| < Const, (1.5)
where φi (t) and φj (t) are somehow-introduced instantaneous phases of

ith and jth oscillators.
If these conditions are satisfied for all network elements, we have global
synchronization. If they are fulfilled only for some groups of elements, then
cluster synchronization appears.
If conditions (1.3)–(1.5) are not fulfilled, the dynamics of the ith and jth
oscillators is non-synchronous.
De-synchronization is the process of transition out of the synchronous
regime.
Phenomena very similar in their manifestation to synchronization of cou-
pled oscillatory systems can be observed in coupled excitable systems and
uncoupled oscillatory systems which are subject to a common external force.
The corresponding synchronization-like phenomena we will call synchrony.
1.4 Bibliographical Remarks

Now we briefly describe the place of our monograph in the large research
field dealing with synchronization problems by comparing with other books
on that.
1.4 Bibliographical Remarks 9
First, we would like to mention the recently published book by A. Pikovsky,

M. Rosenblum, and J. Kurths [1] devoted to all now-known synchronization
phenomena: synchronization of a periodic oscillator by external force, synchro-
nization of two oscillators, synchronization of chaotic systems, synchronization
in the presence of noise, synchronization in oscillatory media. But ensembles
and networks are not in the focus of [1].
Synchronization without formulae illustrated with many biological and
technological examples is given in the book by S. Strogatz [96].
A rather popular introduction to the synchronization theory illustrated by
many examples can be found in [7, 97].
Synchronization by external force, synchronization of two and many oscil-
lators, the influence of noise on synchronization, etc. are studied in some parts
the monographs [9, 11].
The phase approximation approach was developed by Y. Kuramoto [8] and
presented in his classical book. He studied synchronization in large popula-
tions of phase oscillators and in continuous media.
Synchronization from the view point of biological application is studied
in [5] and chaotic synchronization for engineers is discussed in [27–29].
I.I. Blekhman [2, 3] addresses his books to synchronization in mechanical
oscillators, electronic and quantum generators, etc.
Noisy influenced synchronization in periodic systems was developed by
R.L. Stratonovich [98] and A.N. Malakhov [99].
The only book solely devoted to synchronization problems in networks of
coupled oscillators is that by V.S. Afraimovich, V.I. Nekorkin, G.V. Osipov,
and V.D. Shalfeev [4], but it is restricted to networks of identical phase locked
loops (PLL) systems. Synchronization phenomena in complex systems was
also considered in [6].
There are also some reviews on synchronization, e.g., [100,101] and special
issues in journals [45, 46].
In the focus of this book are synchronization phenomena in networks of
coupled oscillators.
2
Basic Models
Synchronization phenomena studied in this book are mainly based on the

dynamics of phases and frequencies of oscillations. Because there is no gen-
eral way to introduce a phase and a frequency for arbitrary oscillatory system,
we will present in this chapter several often used phase and frequency defini-
tions. We start with classical definitions of phase and frequency of oscillations
introduced for the simplest case, the harmonic oscillator (Sect. 2.1). Then we
describe some paradigmatic models from nonlinear dynamics, which will be
used in the following to demonstrate synchronization effects. For all these
models we show their typical behavior. Especially we describe how the phases
and frequencies of oscillations can be introduced also in complex systems.
The first model is the van der Pol oscillator – the basic model of a limit-cycle
oscillator (Sect. 2.2). Next, we briefly describe the famous Rössler (Sect. 2.3)
and Lorenz (Sect. 2.4) systems as paradigmatic chaotic oscillators. In Sect. 2.5
we present several models of phase oscillators which allow to demonstrate
both simple limit-cycle type dynamics as well as complex chaotic dynamics.
We also introduce two maps (difference equations) used to model intermittent
behavior and spiking–bursting activity in neurons (Sect. 2.6). Finally, three
typical excitable systems are given to show some similarities of synchroniza-
tion phenomena in self-oscillatory and excitable systems (Sect. 2.7).
2.1 Harmonic Oscillator: Amplitude, Frequency

and Phase of Oscillations
The harmonic oscillator described by the linear equation:
ẍ + ω02 x = 0 (2.1)
has the solution

x(t) = A cos(ω0 t + φ0 ), (2.2)
12 2 Basic Models
where A is the constant amplitude of oscillations,
φ(t) ≡ ω0 t + φ0 (2.3)
is the linear increasing instantaneous phase, φ0 is the initial phase, and ω0 is

the frequency. The phase can be also found from x(t) and its time derivative
y(t) ≡ ẋ(t):
y(t)
φ(t) = − arctan + πk, (2.4)
ω0 x(t)
where k is an integer. (In the following we will omit the term πk in the phase
definitions.) This classical phase definition is based on the transition to polar
coordinates, the radius vector A and the angle φ. It shows phase evolution
in time during one rotation period T = 2π/ω0 . After each rotation the phase
becomes larger by 2π. We will show in the next sections that similar phase
definition can be used for regular and chaotic self-sustained oscillators.
2.2 Van der Pol Oscillator: Quasi-Harmonic

and Relaxation Limit Cycles
A basic model in nonlinear dynamics is the van der Pol oscillator:
ẋ = y,
(2.5)
ẏ = −ω 2 x + μ(1 − x2 )y,
where ω is the natural frequency of the oscillations and the parameter of

nonlinearity μ ≥ 0 governs the form of the oscillation. Equation (2.5) was
introduced already in 1927 by the Dutch engineer van der Pol [48] to describe
voltage and current evolution in a triode generator. Equation (2.5) has become
a paradigmatic model in the theory of nonlinear oscillations and nonlinear
dynamics [1, 47, 49, 50] because it is a basic self-sustained oscillator, i.e., (2.5)
exhibits a sustained periodic motion without any external driving. This oscil-
lator generates two main regimes for weak (μ 1) and strong (μ 1)
nonlinearity, respectively (Figs. 2.1 and 2.2).
In the power spectrum of a weekly nonlinear – quasi-harmonic – van der
Pol oscillator (μ 1) only one frequency (Fig. 2.2a), the natural frequency
ω, is strongly dominating; while the power spectrum of the strongly nonlinear
oscillator (μ 1) is much richer (Fig. 2.2b)and contains a large number of
linear combinations of harmonics and subharmonics of ω.
The convergence rate to the limit cycle for week and strong nonlinearities
is quite different (Fig. 2.1a, b). In the latter case we have a relaxation type of
oscillations, i.e., one can distinguish slow and fast motions (Fig. 2.2d). This
distinction implies a profound difference for the formation of synchronization
of coupled or driven van der Pol oscillators.
2.2 Van der Pol Oscillator: Quasi-Harmonic and Relaxation Limit Cycles 13
4 15
(a) (b)
10
2
5
0 0
y
−5
−2
−10
−4 −15
−4 −2 0 2 4 −4 −2 0 2 4
x x
Fig. 2.1. Phase portraits for the van der Pol oscillator (2.5) for μ = 0.12 (a) and
μ = 7 (b)
1
10
0
10
−1
10
−2
10
−3
10
−4 (a)
10 ( b)
−5
10
0.0 0.5 1.0
0.0 0.5 1.0
Ω /2π
Ω / 2π
4.0
0.0
x
(c) (d)
−4.0
100
(e) (f )
50
φ
0 0 50 100
0 50 100
t t
Fig. 2.2. Power spectra (a, b), time series x(t) (c, d), and phases (e, f ) for the
system (2.5) for μ = 0.12 (a, c, e) (quasi-harmonic type) and μ = 7 (b, d, f )
(relaxation type)
First phase definition. In a straightforward generalization of the harmonic

oscillations we get in both cases:
y(t)
φ(t) = − arctan (2.6)
x(t)
In such a way defined phase increases monotonously in time, the growth
is 2π per cycle as in the harmonic oscillator (2.4). However, the phase
evolutions in time for weak and strong nonlinearity strongly differ
(Fig. 2.2e, f). In the quasi-harmonic case the phase grows practically
uniformly (linear), while for strongly nonlinear oscillations the evolution
14 2 Basic Models
of the phase φ(t) looks like intermittency consisting of relatively large

epochs of practically constant phase and short intervals of phase jumps of
2π, called phase slips. The instantaneous frequency ω(t) of the oscillations
can be defined as:
dφ
ω(t) = . (2.7)
dt
First frequency definition. For the periodic oscillations of quasi-harmonic
and relaxation types the frequency of oscillations Ω can be defined through
the period length T of oscillations, i.e.,
Ω = 2π/T. (2.8)
2.3 Rössler Oscillator: From Phase-Coherent to Funnel

Chaotic Attractors
Inspired by a taffy-pulling machine, the German physico-chemist Otto Rössler
proposed in 1976 the self-oscillatory chaotic system [58]:
ẋ = −ωy − z,
ẏ = ωx + ay, (2.9)
ż = b + z(x − c),
where the parameters a, b, c, ω are positive. Here we set b = 0.1 and c = 8.5.
For the parameter range a ∈ [0.15 : 0.3] (2.9) governs two strongly differ-
ent regimes of chaotic oscillatory dynamics, which are illustrated by their
phase portraits (Fig. 2.3), time series (Fig. 2.4a, b, c) and power spectra
(Fig. 2.4d, e, f).
(1) When a is below a critical value ac (ac ≈ 0.186 for ω = 0.98), the chaotic
trajectories always cycle around the unstable fixed point (x0 , y0 ) ≈ (0, 0)
in the (x, y) subspace, i.e., max(y) > y0 (Fig. 2.3a). In this case straight-
forwardly, the rotation angle (2.6) can be defined as the phase, which
increases almost uniformly, i.e., the oscillator has an almost coherent phase
dynamics. Hence, this attractor of system (2.9) is called phase-coherent
attractor [1]. Therefore, the phase definition (2.6) is appropriate for some
limit-cycle attractors as well as for some chaotic attractors. Loosely speak-
ing, (2.6) can be applied to all complex systems, for which a projection of
the phase trajectory on some properly chosen plane looks like a smeared
limit cycle with positive curvature.
(2) Beyond this critical value ac , the trajectories no longer always completely
cycle around (x0 , y0 ), but some max(y) < y0 occur, which are associated
with faster returns of the orbits (Fig. 2.3b, c). The attractor is now no
longer phase-coherent but it becomes a funnel one. Such earlier returns
in the funnel attractor happen more frequently with increasing a (Fig.
2.3c). For funnel attractors, the straightforward phase definition (2.6) is
not appropriate.
2.3 Rössler Oscillator: From Phase-Coherent to Funnel Chaotic Attractors 15
(a) (b) (c)
(d) (e) (f)
0.05
(g) (h) (i)
P(T)
0.00
1 7 1 7 1 7
T T T
Fig. 2.3. Upper panel (a–c): projections of the attractors of the Rössler system
(2.9) onto the plane (x, y); middle panel: (d–f ): projections onto (ẋ, ẏ); lower panel
(g–i): distribution of the return times T . The parameters are ω = 0.98 and a = 0.16
(a, d, g), a = 0.22 (b, e, h) and a = 0.28 (c, f, i)
Hence, we have to use other concepts for an appropriate phase definition [60].
One is based on the general idea of curvature of an arbitrary curve [61]. For
any two-dimensional curve r1 = (u, v) the angle velocity at each point is
ds
ν= /R, (2.10)
dt
where
ds/dt = u̇2 + v̇ 2 (2.11)
is the speed along the curve and
R = (u̇2 + v̇ 2 )3/2 /[v̇ ü − v̈ u̇] (2.12)
is the radius of the curvature. If R > 0 at each point, then

dφ v̇ü − v̈ u̇
ν= = 2 , (2.13)
dt u̇ + v̇ 2
is always positive and therefore the variable φ defined as

v̇
φ = νdt = arctan , (2.14)
u̇
is a monotonically growing angle function of time and can be considered as
a phase of oscillations. In this the projection r2 = (u̇, v̇) is a curve cycling
16 2 Basic Models
monotonically around a certain point, i.e., we have found an effective projec-

tion of the phase trajectory.
The above definitions (2.13) and (2.14) of frequency and phase are general
for any dynamical system if the projection of the phase trajectory on some
plane is a curve with a positive curvature. It is applicable to a large variety of
chaotic oscillators, such as the Lorenz system [62], Chua circuit [63], Colpits
oscillator [64], extended Bonhöffer–van der Pol oscillator [65], Anishchenko–
Astakhov generator [66], the model of the ideal four-level laser with periodic
pump modulation [67], or electrochemical oscillators [68], and by analysis of
phase dynamics between El Niño/Southern Oscillation and the Indian Mon-
soon [69].
This leads us to:
Second phase definition. For phase-coherent as well as funnel attractors
of (2.9) projections on the plane (ẋ, ẏ) always rotate around the origin
(Fig. 2.3d, e, f) and the phase can be calculated according to (2.14)
ẏ
φ = arctan . (2.15)
ẋ
Second frequency definition. From (2.13) we can get the characteristic
mean frequency
ẏẍ − ÿ ẋ
Ω= ν = 2 , (2.16)
ẋ + ẏ 2
where . denotes time averaging.
Third frequency definition If the phase φ(t) of oscillations is defined, then
the mean frequency of oscillations can be also defined as an average speed
of the phase increase:
φ(T ) − φ(0)
Ω = lim , (2.17)
T →∞ T
where T is observation time.
To describe the influence of the noncoherence of the orbits on the
timescales of the oscillations, we calculate the distribution of the return
time T of the orbits between two successive crossings of the Poincaré section
ẏ = 0, ẋ > 0. For the phase-coherent attractor (e.g., a = 0.16), T is located
in a relatively narrow interval (Fig. 2.3g). When the attractor becomes non-
coherent, those earlier returns are associated with much smaller T , and the
distribution becomes rather broad (Fig. 2.3h, i). For intermediate a (e.g.,
a = 0.22) one characteristic time scale is still dominant, while for large a
(e.g., a = 0.28) two well-pronounced timescales are formed. All mentioned
properties are also clearly seen from power spectra (Fig. 2.4d, e, f).
The degree of noncoherence on timescales can be also quantified by the
phase diffusion coefficient Dφ as
2Dφ t = (φ(t) − φ(t) )2 , (2.18)
2.3 Rössler Oscillator: From Phase-Coherent to Funnel Chaotic Attractors 17
25
15
5
x
−5
−15 (c)
(a) (b)
−25
0 50 100 0 50 100 0 50 100
t t t
(d) (e) (f)

100
10−2
10− 4
0.0 0.5 1.0 0.0 0.5 1.0 0.0 0.5 1.0
Ω/2π Ω/2π Ω/2π
Fig. 2.4. Upper panel (a–c): time series x(t); lower panel: (d–f ) power spectra of
the Rössler system (2.9). The parameters are ω = 0.98 and a = 0.16 (a, d), a = 0.22
(b, e) and a = 0.28 (c, f )
100
10−1
10−2
Dφ
10−3
10−4
10−5
0.15 0.20 0.25 0.30
a
Fig. 2.5. Phase diffusion coefficient Dφ (2.18) for the Rössler system vs. a, ω = 0.98
where · denotes the ensemble average. Dφ increases in general with a inter-

rupted by Dφ = 0 in periodic windows. For a phase-coherent chaotic attractor,
the phase increases approximately uniformly and Dφ is very small. But for a
funnel chaotic attractor the increase of the phase is strongly nonuniform and
Dφ can be a few orders larger in magnitude (Fig. 2.5), which makes synchro-
nization more difficult as will be discussed.
Third phase definition. The phase of oscillations can also be defined
based on an appropriate Poincaré section with which the chaotic orbit
crosses once for each rotation (Fig. 2.3a, d–f). Successive crossing with
the Poincaré section can be associated with a phase increase of 2π and
the phases in between can be computed with a linear interpolation, i.e.,
t − τk−1
φ(t) = 2πk + , (2.19)
τk − τk−1
where τk is the time of the kth crossing of the flow with the Poincaré
section. As seen in Fig. 2.3d–f, the successive maxima or minima of the
18 2 Basic Models
chaotic time series ẏ(t) (or y(t) in Fig. 2.3a) correspond to a particular
Poincaré section. This means that in this case the phase can be defined
equivalently by examining the maxima or minima of the scalar chaotic
time series without reconstruction of the dynamics in a higher dimensional
phase space and finding a Poincaré section. Note that this approach does
not work always.
Fourth frequency definition. We emphasize that the mean frequency of
chaotic oscillations Ω can be also calculated as
M
Ω = lim 2π (2.20)
t→∞ t
where M is the number of rotations of the phase point around the origin
or the number of crossing of the flow with some Poincaré section during
time t. This method can be directly applied to observed time series, when
one e.g., takes for M the number of maxima of some variable (e.g., y(t))
in (2.9).
Fourth phase definition. The phase in any system can be rather generally
defined by using the Hilbert transform. This approach is based on the
analytic signal concept [71] and was introduced by Gabor [70]. Given a
signal s(t), the analytic signal ζ(t) is a complex function of time defined as
ζ(t) = s(t) + js̃(t) = A(t)ejφ(t) (2.21)
where the imaginary part s̃(t) is the Hilbert transform of s(t)

∞
s(τ )
s̃(t) = π −1 P.V. dτ (2.22)
−∞ t − τ
(here the integral is taken in the sense of the Cauchy principal value).
The instantaneous amplitude A(t) and the instantaneous phase φ(t) of
the signal s(t) are thus uniquely defined from (2.21).
To conclude this section we want to note that for some typical classes of oscil-
lators, e.g. phase-coherent oscillators, the phases and the frequencies defined
in different ways are practically equivalent, and they give the same results in
the study of phase synchronization. However, for some other classes of sys-
tems, e.g., oscillators with funnel attractors, only one or two of the phase and
frequency definitions are applicable.
2.4 Lorenz Oscillator: “Classic” and Intermittent

Chaotic Attractors
To describe some qualitative behavior in thermo-convection, Ed Lorenz
derived in 1963 another paradigmatic chaotic oscillator consisting of three
ordinary differential equations:
2.4 Lorenz Oscillator: “Classic” and Intermittent Chaotic Attractors 19
50 50
(a)
25 25
z
0 0
−20 0 20 −5 15 35
x u
Projections of the “classical” Lorenz attractor on the planes (x, z) (a) and
Fig. 2.6.
(u = x2 + y 2 , z) (b). σ = 10, b = 8/3, and r = 28
ẋ = −σ(x − y),
ẏ = (r − z)x − y, (2.23)
ż = −bz + xy,
with the parameters σ (a Prandtl number), r (a Rayleigh number), and b > 0.

(2.23) also arises in models of one-mode lasers and dynamos [340], and it
describes the convection in a vertical loop [341].
System (2.23) demonstrates several types of chaotic behavior; two of them
are of special interest for synchronization:
(1) The famous Lorenz attractor is obtained for σ = 10, b = 8/3, and r = 28
(Fig. 2.6a). However, here the phase definition is not so straightforward
because the attractor projections on the planes (x, y), (x, z), or (y, z)
do not show rotations around a unique center. But, this complication
can be overcome by using the projection on the plane (u = x2 + y 2 , z)
(Fig. 2.6b). Analogous to (2.6), it allows now to introduce a phase as:
z − z0
φ = arctan , (2.24)
u − u0
where u0 = 12 and z0 = 27 and the mean frequency as:
Ω =< φ̇ > (2.25)
(2) Another typical behavior of the Lorenz system is type-I intermittent

chaotic dynamics (Fig. 2.7) which takes place for σ = 10, b = 8/3, and
r = 166.1. Interestingly, the resulting oscillations have a well pronounced
characteristic time scales as well. For type-I intermittency long-lasting
laminar stages with a duration τ0 are typically interrupted by a very short
turbulent stage (sometimes, just one jump) with a duration T and the next
20 2 Basic Models
60
0
x
−60
100
0
y
−100
250
z
50
0 10 20 30 40 50 60 70 80 90 100
time
Fig. 2.7. Intermittent chaotic oscillations in the Lorenz oscillator for σ = 10, b =
8/3, and r = 166.1
laminar stage begins. The average length of the laminar stage (ALLS) for
any intermittent type-I chaotic oscillator is calculated as [72, 73]:
1
τ0 ∝ √ , (2.26)
r − rcr
where r is a bifurcation parameter and rcr is the critical value when chaos
sets in1 . We use now the onset of each laminar epoch tn as a marker event
in order to introduce a phase of these intermittent oscillations. Then the
interval [tn , tn+1 ], which covers the whole cycle of one laminar and one
turbulent stage, is attributed to a 2π phase increase and in between we
use a linear interpolation (see also 2.19):
t − tn
φ(t) = 2π + 2πn, tn ≤ t < tn+1 . (2.27)
tn+1 − tn
A similar intermittent chaotic behavior is demonstrated by the simple

quadratic map [74]:
xn+1 = f (xn ), (2.28)
where f (x) consists of the standard quadratic part and a special return part
that acts as a turbulent stage and ensures chaos:

ε + x + x2 , if x ≤ 0.2,
f (x) = (2.29)
g(x − 0.2) − ε − 0.24, if x > 0.2
1
Note, that often because of τ /T 1 the time of the full cycle Tc = τ + T , i.e., the
time between the beginnings of two sequential laminar stages, practically equals
to τ . Therefore, the coincidence of the averaged τ leads to the coincidence of the
averaged Tc .
2.5 Phase Oscillators 21
0.3
(a)
0.2
0.1
x(n)
0
−0.1
−0.2
0 1000 2000 3000 4000 5000
0.3
0.2
(b)
0.1
x(n)
0
−0.1
−0.2
0 1000 2000 3000 4000 5000

n
Fig. 2.8. Intermittent chaotic oscillations in the quadratic map (2.28-2.29). ε =
0.00005, g = 2 (a) and g = 7 (b)
Here g regulates the coherence properties of the chaotic attractor: for g < 5
the laminar stage duration is distributed in a rather narrow band, i.e. the
chaotic behavior is highly coherent, but for g > 5 this distribution is rather
broad (Fig. 2.8).
2.5 Phase Oscillators

A wide and important class of dynamical systems are phase systems, also
called phase oscillators or rotators. A phase system has only the phase variable
or angle variable and its time derivatives in continuous-time systems and
phase variable or angle variable and the differences in discrete-time systems
as variables. Note that in these systems the phase variable can exhibit not only
monotonous growing. It can increase and decrease as well. Therefore, it does
not have the same meaning as the phases introduced in previous paragraphs.
But for simplicity sometimes we will call this variable as a phase. We mainly
distinguish the following types:
2.5.1 First-Order Phase Oscillator (Active Rotator)

The first-order equation also called Adler equation
φ̇ + sin φ = ω, (2.30)
22 2 Basic Models
180
160
140
ω =2
120
100 ω =1.5
φ
80
60
ω =1.1
40
20 ω =1.01
0
0 10 20 30 40 50 60 70 80 90 100
time
Fig. 2.9. Phase variable evolution in (2.30) for different ω
with the parameter ω ≥ 0 describes the behavior of a first-order phase oscil-

lator, called rotator. System (2.30) has two steady states for ω < 1: a stable
state φ̄s = arcsin ω and an unstable one φ̄u = π − arcsin ω. When ω = 1, the
steady states merge to the point φ̄ = π/2. For ω > 1 there are no steady states
but rotations. Their main characteristic is a growing rate of φ which strongly
depends on γ = ω − 1. If γ is close to zero, the evolution of the phase variable
φ(t) looks like intermittency of relatively large epochs of practically constant
phase and short intervals of phase jumps of 2π, called phase slips (Fig. 2.9 for
ω = 1.01). With increasing γ, the length of these plateaus becomes shorter
(Fig. 2.9). If γ is relatively large, the phase evolution is close to a linear one,
i.e., the phase grows almost uniformly (Fig. 2.9 for ω = 2).
Equation (2.30) arises in various areas of science and engineering, such as:
(1) Biology. Oscillating neurons, firefly flashing rhythm, human sleep–wake
cycle (e.g., [25, 97])
(2) Condensed-Matter Physics. Josephson junction, charge-density waves
(e.g., [34, 35])
(3) Mechanics. Over-damped constant-forced pendulum (e.g., [3])
(4) Electronics. Phase-locked loops (e.g., [57])
2.5.2 Second-Order Phase Oscillator (Pendulum-Like System)

A more complex dynamics is generated by a second-order phase system:
φ̇ = y,
(2.31)
ẏ = ω − λy − sin φ,
where the parameters λ ≥ 0 and ω ≥ 0. The phase space of this system

is cylindrical (φ, y). Therefore there are two types of motions in (2.31) (a)
rotations, i.e. φ increases unbounded and (b) oscillations, i.e., the range of
φ is limited to some band which is narrower than 2π. System (2.31) is often
called pendulum or pendulum-like system because it describes the dynamics of
a damped pendulum with the constant rotatory moment ω. Throughout the
book these terms are used as synonyms.
The parameter plane (ω, λ) is subdivided into the domains D1 , D2 , and
D3 corresponding to different types of dynamical behavior (Fig. 2.10). They
are characterized in the phase plane (φ, y) (Fig. 2.11). For ω < 1, the phase
plane contains two equilibrium states φ̄s = arcsin ω, ȳ s = 0 and φ̄u = π -
arcsin ω, ȳ u = 0. For the parameters belonging to domain D1 , the image point
acquires a stable steady state of O1 (φ̄s , 0) for arbitrary initial conditions on the
phase cylinder (excluding the separatrices S2 , S4 , entering a saddle O2 (φu , 0))
(Fig. 2.11, left panel). For the parameters of domain D2 , depending on the
initial conditions, the image point on the (φ, y)-surface, goes either to a stable
steady-state or to a stable limit cycle L of rotatory type – a cycle which
envelopes the cylinder (Fig. 2.11, middle panel). D1 and D2 are separated by
the curve ωL (Fig. 2.10). The parameter values satisfying the curve ωL are
corresponding to the saddle separatrix loop. The straight line ω = 1 on the
(ω, ε)-plane is also bifurcational and corresponds to the formation of a saddle-
node as a result of merging of the stable O1 and unstable O2 equilibrium states.
1.2
D3
1.0
0.8 D2
ω
0.6
ωL
D1
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
λ
Fig. 2.10. Parameter plane (ω, λ) showing different types of behavior of (2.31)
24 2 Basic Models
Fig. 2.11. Characteristics of domains corresponding to Fig. 2.10. D1 : only one

attractor exists: a stable steady-state (arcsin ω, 0). D2 : in dependence on initial
conditions the image point tends either to a stable steady-state or to a stable limit
cycle. D3 : there is only one attractor – a stable limit cycle
A part of the straight line ω = 1 lying close to the ε = 0-axis corresponds

to a bifurcation of the saddle-node separatrix loop. In D3 only rotations are
realized (Fig. 2.11, right panel), i.e. a stable rotatory limit cycle L exists.
System (2.31) describes a lot of mechanical, electrical, and engineering
systems. Among them are (a) pendulum (e.g., [132]), (b) Josephson junction
(e.g., [34, 35]), (c) phase locked loop (e.g., [57]), etc.
2.5.3 Third-Order Phase Oscillator (Chaotic Rotator)

A third-order phase system that is able to demonstrate complex regular as
well as chaotic behavior is described by:
φ̇ = y,
ẏ = z, (2.32)
μż = ω − sin φ − y − εz,
where μ ≥ 0, ε ≥ 0 and ω ≥ 0.
System (2.32) can have three types of regular and chaotic solutions [51] (a)
rotations, (b) oscillations and (c) oscillations–rotations, which are presented
in dependence on the parameters in Sect. 4.2.3.
(2.32) models a Josephson junction with a constant biased current and
subject to a load with inductance, resistance, and capacitance [52] or a phase-
locked loop (PLL) system with the simplest second-order filter [82].
2.5.4 Discrete-Time Rotator (Circle Map)

A basic discrete-time model of a phase system is the discrete-time rotator or
circle map (CM)
φk+1 = ω + φk − F (φk ), (2.33)
where φk is the phase variable at discrete times k = 1, 2, ...; ω ∈ [0; 2π] can
be interpreted as frequency; F (φ) is a piecewise linear 2π-periodic function of
the form:
F (φ) = cφ/π (2.34)
defined in the interval [−π, π], and c ∈ [−π, π] is the control parameter. One
can use any 2π-periodic function F (φ), e.g., F (φ) = sin φ; but our choice (2.34)
of a piecewise linear function F (φ) is motivated not only by the simplicity of
consideration (see [4, 75]) but also by the requirement that there exists chaos
for c < 0 and there are no stable periodic orbits for any c < 0.
First, we shortly describe basic properties of this circle map. It has for
ω < |c| a unique fixed point φ̄ = ωπ/c which is stable if φ̄ ∈ [0; π] and
unstable if φ̄ ∈ [−π; 0].
The dynamics of a CM can be mainly described by the rotation number
ρ, which is defined as the average growth rate of the phase (compare with
(2.16)):
1 φM − φ1
ρ= lim , (2.35)
2π M →∞ M
where M is the number of iterations. (2.35) is valid for regular (c ≥ 0) and
chaotic c < 0 dynamics and defines characteristic time scale of rotations. The
regions of some rational ρ on the plain (c, ω) are marked in Fig. 2.12 with
different levels of gray color.
The parameter c controls the coherence properties of the motions. As a
measure of the degree of coherence analogous to (2.18), we use the variance
D of the phase dynamics which is defined for large k as:
D = (φk+1 − φk − φk+1 − φk )2 (2.36)
. denotes time averaging. So for c = 0, the rotations are uniform, i.e.

completely coherent, and D = 0. If |c| grows, the non-coherence of the
rotation increases. The dependence of D on c for different ω is shown in
Fig. 2.13. The phase evolution in the cases of regular (c = 0, 0.5) and chaotic
(c = −0.5, −1.0, −1.5, −2.0, −2.5, −3.0) motions are presented in Fig. 2.14.
The behavior can be also characterized by the ratio γ of the iteration number
Ng where the phase is increasing and that Nd where the phase is decreasing:
γ = Nd /Ng . (2.37)
Then, we easily see that for c > −ω, γ = 0, but otherwise γ = 0 (Fig. 2.15).
System (2.33) is one of the basic models in nonlinear dynamics, and it
has been studied in many mathematical (cf., [76]), physical (cf., [77–79]) and
engineering issues (in particular, in the theory the digital phase-locked loops
[57, 74, 81]).
26 2 Basic Models
6.283
1
6/7
5/6
4/5
3/4
2/3
3/5
1/2
2/5
1/3
1/4
1/5
1/6
1/7
0 0
−3.141 0.0 3.141
c
Fig. 2.12. Distribution of rotation numbers of the circle map (2.33) on the plain
(c, ω). Several regions where the rotation numbers are rational (ρ = p/q) are pre-
sented. From bottom to top different gray level regions are ordered as shown on
the right side. Region in which ρ = 0 is white. Region in which ρ = 1 is black.
Between these regions there exist (but not presented) relatively small regions with
other rational rotation numbers
5.0
0.6
1.0
1.2 1.4
4.0 1.8
2.2
2.6
0.8
3.0
D
0.4
−1.9 −1.7 −1.5 −1.3
2.0
1.0
0.0
−3.0 −2.0 −1.0 0.0 1.0 2.0 3.0
c
Fig. 2.13. The variance D of φk (2.36) vs. c for the map (2.33) at different values
of the frequency parameter ω
180.0
c = 0.0
−0.5
−3.0
0.5
120.0 −2.5
phase
−1.0
−1.5
60.0 −2.0
0.0
0 50 100 150 200 250 300
k
Fig. 2.14. Phase variable evolution at ω = 0.6 for different c for the map (2.33)
1.5
0.6
1.0
1.4
1.8
2.2
2.6
1.0
γ
0.5
0.0
−3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0
c
Fig. 2.15. Ratio γ (2.37) of the duration of phase decreasing intervals to the dura-
tion of phase increasing intervals in dependence on c for different values ω in the
map (2.33)
28 2 Basic Models
2.6 Discrete Map for Spiking–Bursting Neural Activity
Biological neurons are able to exhibit a plethora of dynamical behavior, from

regular to chaotic one, and especially a mixing of spiking and bursting (e.g.,
[86, 87]). The main interesting collective phenomena in ensembles of neurons
is synchronization. One of the most challenging and inspiring problems here
is an explanation of synchronization and revealing its role in neurobiological
systems, where multiple timescale (spiking and bursting) chaotic behavior is
often observed. Typical examples are (a) the Central Pattern Generator [302],
which is known to produce common rhythmic bursting, while its individual
neurons, if isolated, would show irregular bursts (b) thalamic neurons dur-
ing periods of drowsiness, inattentiveness, and sleep [83,84], and (c) midbrain
dopaminergic neurons, which exhibit spiking and bursting [85]. Coherent oscil-
lations in such systems are thought to play a substantial role in information
processing [86]. During the last decade mutual synchronization in small and
large neural ensembles has been studied in some detail. It has been observed in
numerical [88, 89], electronical [89], and real neurobiological experiments [90].
The study of spiking–bursting activity of neurons are usually based on either
detailed channel based models or on phenomenological (conceptual) models.
The first class consists of many differential equations [86], whereas the second
class can be even written as maps (difference equations). Such a discrete model
for spiking–bursting neural activity was recently proposed by Rulkov [91]:
α
x(n + 1) = 1+x(n) 2 + y(n),
(2.38)
y(n + 1) = y(n) − σx(n) − β,
where x(n) and y(n) are, respectively, the fast and slow dynamical variables.
The slow evolution of y(n) is due to small values of the positive parameters β
and σ (each one of the order 10−3 ). The parameter α controls the dynamics
of the fast variable x(n). Typical regimes of temporal behavior of the map
(2.38) are shown in Fig. 2.16.
The bursting activity is characterized by a distinct timescale which allows
to introduce the phase and the frequency of bursting in each oscillator. The
phase of bursting oscillations φ(n) increases linearly between the moments nk
at which the kth burst starts and gains a 2π growth over each time interval
nk+1 − nk
n − nk
φ(n) = 2πk + 2π (2.39)
nk+1 − nk
(see 2.19 and (2.27)). The mean frequency of the burst dynamics is an average
speed of the phase increase (see 2.17)
φ(n) − φ(0)
Ω = lim . (2.40)
n→∞ n
2.7 Excitable Systems 29
(a)
1.5
0.5
x(n)
−0.5
−1.5
−2.5
(b)
1.5
0.5
x(n)
−0.5
−1.5
−2.5
3000 4000 5000
n
Fig. 2.16. Typical waveforms of spiking–bursting behavior generated by the map
(2.38) for: (a) α = 4.1 and (b) α = 4.4, and β = σ = 0.001
2.7 Excitable Systems

Besides self-oscillatory systems another important class of dynamical systems
are excitable systems. They are paradigmatic models of various dynamical
processes in biological systems. Here we present three basic models of excitable
systems which will be used later to demonstrate synchronization-like phenom-
ena in excitable media.
2.7.1 Hodgkin–Huxley Model
The Hodgkin-Huxley (HH) model describes how action potentials in neurons

are initiated and propagated. It is a set of nonlinear ordinary differential
equations that approximates the electrical characteristics of excitable cells.
Alans Lloyd Hodgkin and Andrew Huxley described the model in 1952
to explain the ionic mechanisms underlying the initiation and propagation of
action potentials in the squid giant axon [492]. They received the 1963 Nobel
Prize in Physiology or Medicine for this work.
The standard Hodgkin–Huxley (HH) model [462] of a neuron considered
the membrane current arises mainly through the conduction of sodium (Na)
and potassium (K) ions through voltage dependent channels in the membrane,
Id and Ir , respectively. However, spike-independent oscillations [463] have
been observed in some types of neurons. The HH model of thermally sensitive
neurons proposed in [464] modified the classical HH neuron model to include
two additional slow currents Isd and Isr according to the experimental findings
of spike-independent oscillations [463]. This extended model can mimic spike
train patterns observed in electroreceptors from dogfish and catfish, and from
30 2 Basic Models
20
−20
V (mv)
−40
−60
−80
0.0 2.0 4.0 6.0 8.0 10.0
t
Fig. 2.17. Typical time series in HH model (2.41)
facial cold receptors and hypothalamic neurons of the rat, etc. The equations
of this specific HH model read:
dV
CM = −Il − Id − Ir − Isd − Isr + Dξ(t)
dt
ar φ(T )(ar∞ − ar )
= ,
dt τr
asd φ(T )(asd∞ − asd )
= ,
dt τsd
asr φ(T )(−ηIsd − θasr )
= , (2.41)
dt τsr
with Id = ρ(T )gd ad∞ (V −Vd ), and Ik = ρ(T )gk ak (V −Vk ), (k = r, sd), where
(T −T0 )/10 (T −T0 )/10
ak∞ = [1 + exp(−sk (V −V0k ))]−1 and ρ(T ) = A1 , φ(T ) = A2 .
Here V is the membrane potential, and Il is the leakage current. ak is the
activation variable, and ρ(T ) and φ(T ) are temperature-dependent scaling
factors. A more detailed description of the model, its parameters, such as
CM , gk , τk . . ., and comparison with the experimentally observed temperature
dependence of spike train patterns can be found in [464]. It has been shown
that this system exhibits a homoclinic bifurcation, where the interspike inter-
val becomes very long, when the control parameter T , the temperature, is
varied [465]. While the classical HH model can only display periodic spiking
behavior, this HH model of thermally sensitive neurons can display chaotic
spiking–bursting behavior as illustrated in Fig. 2.17.
2.7.2 FitzHugh–Nagumo Model
The FitzHugh-Nagumo (FHN) model [92, 93]
x3
ẋ = F (x, y) = x +− y,
3 (2.42)
ẏ = εG(x, y) = ε(ax + by − c),
Fig. 2.18. The original Nagumo circuit is held as a trust in the laboratory of Dr.
K. Aihara in University of Tokyo. Photo provided by H. Suetani and K. Aihara with
permissions.
where a, b, ε are parameters, is a widely used model in the theory of excitable

media and it has been studied and applied in many mathematical and biolog-
ical works. The FitzHugh-Nagumo model is a two-dimensional simplification
of the HH model of spike generation. Here, x is the membrane potential, and
y is a recovery variable. This system was suggested by FitzHugh in 1961, who
called it “Bonhoeffer-van der Pol model”. The model can be described by
an equivalent circuit, which was built by Nagumo et al. in 1962 (Fig. 2.18).
Because ε << 1 is very small, all oscillations in (2.42) can be divided in slow
and fast motions.
The shapes of the functions F (x, y) and G(x, y) are typically as shown in
Fig. 2.19. The null-cline F (x, y) = 0 has an N -like form. We denote the left-
decreasing branch of this null-cline as h− (y), the right decreasing branch as
h+ (y) and the middle-increasing branch as h0 (y). The left and right branches
correspond to stable regions of the slow motion and they have their own basins
of attraction. The null-cline G(x, y) = 0 is monotonic and for chosen values
of parameters it intersects the left h− (y) branch of the null-cline F (x, y) = 0
at a single point (x0 , y 0 ).
The point (x0 , y 0 ) of intersection of the monotonic null-cline G(x, y) = 0
and (h− (y)) of the null-cline F (x, y) = 0 (Fig. 2.19) is the steady state of
(2.42). It is quasi-stable under small perturbations. But relatively large noise
perturbations can excite the system by kicking it away from the stable point
into the basin of attraction of the right slow branch. The system then moves
along the latter, jumps back to the left branch and returns into the vicinity
of the steady state. This process of excitation repeats, and the resultant time
series x(t) looks like a sequence of pulses of excitation (Fig. 2.20).
32 2 Basic Models
1.5
1
E D
0.5
h_(y)
ho(y) h+(y)
0
Y
−0.5
C
A
B
−1
G(x,y)=0 F(x,y)=0
−1.5
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
X
Fig. 2.19. Typical phase diagram for the excitable FitzHugh–Nagumo system,
showing the null-clines F (x, y) = 0 (dashed line) and G(x, y) = 0 (solid line), and
stochastic trajectories produced by the system (2.42) in the presence of an additive
noise term in the first equation
3.0
2.0
1.0
0.0
x
−1.0
−2.0
−3.0
0.0 10.0 20.0 30.0 40.0 50.0
t
Fig. 2.20. Typical time series in FHN model (2.42) in the presence of an additive
noise term in the first equation
In the phase plane (x, y) we can hence distinguish two main types of
motion:
(1) Activation, i.e., motion between the null-cline G(x, y) = 0 in the vicinity
of the steady state and the line x = 0 (approximately between the points
A and B)
(2) Excursion, i.e., motion after the first passage of the line x = 0 along the
branches h+ (y) and h− (y) till the first passage of the null-cline G(x, y) = 0
(approximately from point B through points C, D, E, to point A).
2.7.3 Luo–Rudy Model
Now we present a special model of an excitable system in cardiology; it

describes a single cardiac cell and reads:
dV
Cm = −Iion − Istimulus , (2.43)
dt
where V is the membrane voltage, Istimulus is the input stimulus, and Iion is
the sum of six ionic currents:
Iion = Ina + Isi + Ik + Ik1 + Ikp + Ib (2.44)
as specified in the Luo–Rudy phase I model [94]. For simplicity, we will call
system (2.43) Luo–Rudy model. The form of the sodium current Ina , the slow
inward calcium current Isi , and the potassium current Ik , is given by:
Ii = Gi gi (V, t)(V − Ei ), (2.45)
where Gi is the maximum constant conductance of the ion, gi (V, t) is the

product of one or more gating variables, and Ei is the reversal potential of
the ion. The dynamics of each gating variable is modelled as
dgi g∞ − g i
= , (2.46)
dt τgi
where
g∞ = αgi /[αgi + βgi ] (2.47)
is the steady-state value, τgi = 1/[αgi + βgi ] is the time constant, and the
α’s and β’s are functions of the membrane voltage. The form of the time-
independent potassium current Ik1 , the plateau potassium current Ikp , and
the background current Ib , is given by
Ii = Ai (V − Ei ), (2.48)
where Ei is the reversal potential of the ion and Ai is a scaling factor. Note that
Ai is a function of voltage for Ik1 and Ikp . The conductances (Gi (mS/cm2 ))
for each current are the following: Gna = 23, Gsi = 0.07, Gk = 0.705, Gk1 =
0.6047, Gkp = 0.0183, Gb = 0.03921. The reversal potentials (Ei (mV)) are:
Ena = 54.44, Ek = −77, Ek1 = −87.23, Ekp = −87.23, Eb = −59.87. The
reversal potential of Isi depends upon the internal calcium concentration,
which varies with time. More details can be found in [94].
34 2 Basic Models
40
20
−20
V
−40
−60
−80
−100
0 20 40 60 80 100
time
Fig. 2.21. Typical response of the Luo–Rudy model (2.43)–(2.48) on a super-
threshold force
It is clear that a single cell without any perturbation, has a stable steady
state. If some subthreshold force is applied (in (2.43) Istimulus = 0), the system
becomes excited (depolarization stage). After the excitation the system comes
back to the steady state (repolarization stage). The response of the single cell
has a pulse-like form as illustrated in Fig. 2.21.
The Luo–Rudy phase 1 model has been constructed based on experimental
data. Due to its high accuracy in reconstructing the membrane potential and
the ion currents, it is widely used in numerical experiments. Two- and three-
dimensional simulations with the Luo–Rudy model allows to reproduce many
known features, such as single spiral wave or spiral wave chaos, observed in
heart tissue (see special issue [393]).
3
Synchronization Due to External
Periodic Forcing
In this chapter we describe synchronization phenomena of periodic and chaotic

oscillators subject to an external periodic force (Fig. 3.1). In the classical
theory of synchronization the periodically driven oscillator is the main and
historically first studied model. The electrical engineers Appleton [102] and
van der Pol [48] were the first to show the possibility of synchronization of
triode generator by a weak external periodic signal. Then, external synchro-
nization of self-oscillatory systems was theoretically studied by the Russian
physicists Andronov and Vitt [103, 104], Mandelshtam and Papaleksi [105].
For self-oscillators in a relaxation regime external synchronization was inves-
tigated by Cartwright and Littlewood [106, 107]. The classical theory of
synchronization was finally worked out till the 1960s. Now there are compre-
hensive reviews and books where this theory is presented (cf. [1]). Therefore,
in this and the next chapters, we give only a short description of the classical
case and start with periodically forced limit-cycle oscillator (Sect. 3.1).
Our main intention, however, is to discuss synchronization in chaotic sys-
tems. Hence, we present the studies which have extended the external phase
synchronization to chaotic systems. It is important to emphasize that differ-
ent topologies of chaotic attractors (e.g., phase-coherent and funnel attrac-
tors in Fig. 2.3 or intermittent systems in Figs. 2.7 and 2.8) require different
approaches to the synchronization problem (see phase definitions in Chap. 2).
But in all cases a necessary condition which allows to formulate the problem
of chaotic synchronization is the existence of a characteristic time scale in
the considered systems. Then, this problem becomes quite similar to the case
of classical synchronization and can be commonly formulated for both cases:
Under which conditions the observed (for some chaotic systems – mean) fre-
quency of oscillations in an externally periodically driven system will come
into coincidence with the frequency of the driving? Usually these conditions
are quantified in terms of the amplitude and the mismatch between the fre-
quency of the external force and the natural (internal) frequency of the oscil-
lator. Considering the more complex chaotic phase synchronization, we have to
differentiate into perfect and imperfect synchronization as demonstrated for
36 3 Synchronization Due to External Periodic Forcing
the paradigmatic examples of the Rössler (Sect. 3.2) and Lorenz (Sect. 3.3)
oscillators, respectively. A detailed description of the phase synchronization
mechanism is presented in Sect. 3.4. Section 3.5 is devoted to the transi-
tion to phase synchronization in chaotic intermittent oscillators. Synchronous
response of excitable systems is discussed in Sect. 3.6. In Sect. 3.7 we conclude
this chapter.
3.1 Synchronization of Limit-Cycle Oscillator

by External Force
We will consider here separately two main cases of the self-sustained periodic
oscillators in Fig. 3.1 (1) a weakly forced limit-cycle oscillator and (2) an
arbitrarily forced van der Pol oscillator.
3.1.1 Weak Forcing: Phase Description
We start with a brief description of the synchronization phenomena for the

simplest example (for a detailed description cf. [1]), a weakly periodically
forced limit-cycle system:
ẋ = F (x) + εp(t), (3.1)
where x and F are n-dimensional vectors, p(t) is a periodic (with period T

and frequency ω = 2π/T ) external force with the small amplitude ε. Suppose
that the unforced system (ε = 0) has a stable period T0 , limit-cycle x0 (t) =
x0 (t + T0 ) with a uniformly growing phase φ – a coordinate along this cycle.
Hence,
φ̇ = ω0 , (3.2)
where ω0 = 2π/T0 is the natural frequency of the oscillations.
If the external force is small (ε 1), system (3.1) can be converted to the
corresponding phase model:
φ̇ = ω0 + εq(φ − ωt), (3.3)
where q is a 2π-periodic function of phases φ and ωt.
SELF−
EXTERNAL PERIODIC FORCE SUSTAINED
OSCILLATOR
Amplitude A Mean
Frequency ω frequencyω0
Fig. 3.1. Oscillator under external periodic force. In synchronous regime the fre-
quency Ω of driven oscillator is equal to the frequency of external force ω
3.1 Synchronization of Limit-Cycle Oscillator by External Force 37
Introducing the difference between the phase of the oscillations φ and the
phase ωt of the external force:
θ = φ − ωt (3.4)
and averaging (3.3) over the period of oscillations, we obtain:
θ̇ = −Δ + εq(θ), (3.5)
where Δ = ω − ω0 is the frequency mismatch. In the simplest case q(θ) = sin θ

and then (3.5) yields the Adler equation
θ̇ = −Δ + ε sin θ. (3.6)
This first-order phase equation has been described in Sect. 2.5.1. Its stable
steady-state solution
Δ
θ̄ = arcsin (3.7)
ε
exists if
|Δ| < |ε| (3.8)
and results in a constant phase shift (strict phase locking) between the phases
of the driving and the driven (the whole) systems. Therefore, external syn-
chronization takes place if condition (3.8), which defines the synchronization
area (the band of synchronization), is fulfilled and is defined through the ratio
of the frequency mismatch Δ and the amplitude A of the external force. The
transition from synchronization regime to beating regime, which takes place
for Δ = A, occurs through the collision and disappearance of the stable and
unstable steady states.
3.1.2 Synchronization of a van der Pol Oscillator by External Force
Now we treat an arbitrary (not small like in Sect. 3.1.1) external force applied
to a limit-cycle oscillator, namely to a van der Pol oscillator. From the various
studied cases we present results of an external synchronization on a weakly
nonlinear van der Pol oscillator (2.4). The model system is then:
ẋ = y,
(3.9)
ẏ = −ω02 x + μ[(1 − x2 )y + 2A cos ωt],
where ω0 is the natural frequency of the oscillator, μ is a small parameter,

i.e., the nonlinearity is weak, A and ω are the amplitude resp. the frequency
of the external force. For an unforced (A = 0) oscillator, a stable limit-cycle
with the period T0 = 2πω0 exists in the phase space. The case of an arbi-
trary amplitude A does not allow to restrict our treatment based only on
the phase dynamics (Sect. 3.1.1). But the investigation of the stated problem
can be again performed analytically by using asymptotic methods (cf. [108]).
There, the analysis of the synchronization problem (3.9) can be reduced to

the analysis of the equation:
ȧ = −iΔa + a − |a|2 a − iA, (3.10)
where a is the complex amplitude of the oscillations and Δ = (ω 2 − ω02 )/2ω

is the normalized frequency mismatch. A steady state of (3.10) corresponds
to synchronous behavior in the original system (3.9), and a limit cycle in
(3.10) corresponds to a beating regime in (3.9). We omit here rather simple
arithmetical calculations and present only the main features. The approximate
bifurcation diagram of (3.10) shows two qualitatively different transitions from
(to) synchronization regime to (from) beating regime (Fig. 3.2). They depend
on the parameter values of the external force:
1. If the driving amplitude A is small, the synchronous behavior disappears
through a saddle-node bifurcation (Sect. 3.1.1). By the transition from
region O (where three steady states exist: stable node, saddle, and unsta-
ble focus) to region C (where only one unstable focus exists) the node
and the saddle collide and by the distraction of the separatrix loop of the
saddle-node a stable limit cycle appears.
2. If the driving amplitude is rather large with increasing frequency mismatch
Δ, firstly (transition B → D) an Andronov–Hopf bifurcation happens. In
the beginning the amplitude of the born limit cycle is small, and the ori-
gin lies outside of this cycle. That means the external synchronization
1.2
D D
1.0
B
0.8
0.6
A
0.4 C C
O
0.2
0.0
−2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0
Δ
Fig. 3.2. Bifurcation diagram for system (3.10) corresponding to a harmonically

forced weakly nonlinear van der Pol oscillator. In region O there exist three steady
states: stable focus (or node) corresponding to a synchronous state in the original
system (3.9), saddle, and unstable focus (or node). In region B only one stable
focus (or node) exists. In regions C and D a stable limit cycle exists. In C(D) the
origin lies inside (outside) of this cycle. The cycle in C corresponds to the beating
regime in (3.9). The transition from O to C, i.e., from synchronization regime to
beating regime, occurs via a saddle-node bifurcation (bold curve). The Andronov–
Hopf bifurcation line (B → D) is shown by the dashed curve. The transition from
the phase entrainment to the nonsynchronized regime (D → C). In regions O and
B synchronization is strict (1.4), but in region D synchronization is nonstrict (1.5)
3.2 Phase Synchronization of a Chaotic Rössler Oscillator 39
still exists but only in the average, i.e., the difference between the phase
of the oscillations and the phase of the external force is not constant but
bounded. By the transition D → C, a point on the limit cycle becomes
rotating around the origin, i.e., a beating regime occurs. This transition
can be better understood by a bifurcation analysis of system (3.10) rewrit-
ten in the real coordinates R and θ (a = R(t)exp[iθ(t)]). In the cylindrical
phase space (R, θ) by the transition D → C the oscillatory limit cycle
disappears but a rotatory limit cycle appears – the cycle which envelops
the cylinder.
3.2 Phase Synchronization of a Chaotic Rössler

Oscillator by External Driving
Similarly to the external synchronization of periodic oscillators, we will
describe in Sects. 3.2–3.5 phase synchronization of chaotic oscillators by exter-
nal force.
First we take the periodically driven Rössler oscillator [58] as an illustrative
example:
ẋ = −ω0 y − z + A sin ωt,
ẏ = ω0 x + ay,
ż = f + z(x − c), (3.11)
with the parameters ω0 = 0.97, f = 0.2, and c = 10. When setting a = 0.04,
the free Rössler oscillator exhibits a periodic motion with the frequency ω ∗ =
0.981. Figure 3.3a shows that this periodic motion is locked with the ratio
n : m = 1 : 1 to the rather weak external periodic signal with A = 0.4 and
ω = 1.0. The whole Arnold tongue for the 1 : 1 synchronization regime is
shown in Fig. 3.3b.
Now we demonstrate phase synchronization of a chaotic oscillator by a
periodic driving. Here we illustrate the synchronization behavior for the sys-
tem (3.11) in a chaotic regime with a = 0.165. For this a the attractor in the
20 0.5
10 0.4
0.3
0
A
x
0.2
−10 0.1
(a) (b)
−20 0.0
0 50 100 0.95 0.97 0.99 1.01
time ω
Fig. 3.3. (a) Synchronization of periodic oscillation (solid line) to a weak periodic
driving signal (dotted line). (b) Arnold tongue of the 1 : 1 synchronization
(a) (b)
10
y 0
−10
−20
−20 −10 0 10 20 −20 −10 0 10 20
x x
Fig. 3.4. Stroboscopic plot of the phase-coherent Rössler system projected on

the (x, y) plane (filled cycles) at each period of the driving signal (3.11). The dot-
ted background is the unforced chaotic attractor. (a) A = 0.15, ω = 1.0, phase is
synchronized. (b) A = 0.15, ω = 1.02, phase is not synchronized
0.20
A
0.10
0.00
0.98 1.00 1.02
ω
Fig. 3.5. Synchronization region of the chaotic Rössler oscillator by an external

periodic force (3.11)
Rössler oscillator is phase coherent (see Sect. 2.3, Figs. 2.3 and 2.4). Here the
stroboscopic technique is very useful. Its idea is the following: We observe the
variable(s) of the driven system only at the moments tk = kT , where T is
the period of the external force and k = 1, 2, . . .. As shown in Fig. 3.4, when
the system is phase locked to the driving signal, the stroboscope of the sys-
tem state (x, y) at each period of the driving signal is restricted to an arc
area of the chaotic attractor, while it is distributed relatively uniformly over
the whole attractor when the system is out of the phase locking region. The
whole synchronization region, shown in Fig. 3.5, is very similar to the Arnold
tongue of the periodic oscillators in Fig. 3.3b. These properties were firstly
reported in [109] and studied more intensively in [110]. In the marked region
the mean frequency of the chaotic oscillations is equal to the frequency of the
external force, and the difference between the phases is bounded for all time.
This example shows that by using of a periodic external force, we are able to
control the spectral properties of complex signals.
Intuitively, we can expect this similarity because we have noted that
the phases of phase-coherent chaotic oscillations increase almost linearly
3.2 Phase Synchronization of a Chaotic Rössler Oscillator 41
as in quasiharmonic limit-cycle oscillators. To obtain a deeper insight into

this similarity and to reveal new features, it is very helpful to study phase
synchronization of chaotic oscillators in terms of unstable periodic orbits
(UPOs). As it is well known, these orbits build a skeleton of a chaotic attrac-
tor [79]: chaotic trajectory spend sometime in the close vicinity of these orbits.
Each of these unstable trajectories has its own period. And if we apply an
external force each UPO can be synchronized in full analogy of external peri-
odic synchronization. Therefore we will have a number of Arnold tongues
corresponding to phase locking regions with each UPO. The region where all
these tongues are overlapped is the region of chaotic phase synchronization.
Indeed, for a chaotic trajectory of the attractor wandering in a vicinity of
UPOs, the difference between the external phase and the current phase is
always bounded.
Calculating the phase locking regions of the UPOs embedded in the
continuous-time Rössler attractor, it has been found that the results quan-
titatively agree with the above consideration [119]. Additionally under the
influence of the force, some UPOs of the autonomous oscillator leave the bulk
of the attractor and may be visited extremely rarely.
The investigation of chaotic phase synchronization in terms of UPOs is
very useful to understand special features which are not present in the syn-
chronization of periodic oscillations. More details will be discussed in the
next sections in the context of imperfect phase synchronization (Sect. 3.3)
and phase synchronization transition (Sect. 3.4).
In conclusion of this section we want to emphasize the following. In PS of
chaotic oscillators, the timescales of one oscillator are entrained by another
chaotic or periodic oscillator. One would expect that the distribution of these
timescales plays an important role in the synchronization behavior. For exam-
ple, in the chaotic phase-coherent Rössler attractor discussed earlier, the
return times, i.e., the times between successive intersections of an appro-
priately chosen Poincaré surface vary, but the relative variation is rather
small (Fig. 2.3). As a result of this narrow-band distribution of timescales, this
chaotic Rössler oscillators can be rather easily synchronized by a weak forcing
with a period close to the average return time [109, 110, 118, 119]. In view of
UPOs, these orbits have narrowly distributed periods, and consequently, the
phase locking regimes are similar and all corresponding Arnold tongues have
an overlapping to guarantee complete PS of the chaotic attractor. In general,
a similar behavior can be observed in other systems with chaotic attractors
originating from the periodic-doubling scenario. However, as pointed in [111],
it may not be possible to synchronize some phase-coherent systems, even if
they exhibit very small phase diffusion Dφ . This may occur when the phase
has a sensitive dependence on the amplitudes in the presence of an external
force.
3.3 Imperfect Phase Synchronization

However, the scenario changes if the system has a rather broad distribution
of timescales of its periodic orbits. An external signal with a given frequency
then may not be able to entrain all the timescales in the system. As a result
phase slips occur when the chaotic flow comes to oscillate with timescales
outside the synchronization region of the driving signal and imperfect phase
synchronization is observed.
References [112–114] have carried out a comparative study of perfect and
imperfect PS for a periodically driven Lorenz system,
ẋ = 10(y − x),
ẏ = rx − y − xz, (3.12)
ż = xy − 2.667z + A cos ωt.
The unforced Lorenz system exhibits rich bifurcations when the parameter
r is varied [61]. For r = 210, chaotic oscillations result from a periodic-doubling
scenario, and, as mentioned above, the system can then be synchronized per-
fectly by a periodic driving signal with a frequency close to the average fre-
quency Ω = 24.92 of the chaotic oscillations, as it can be seen by the plateau
of the vanishing frequency difference Ω − ω = 0 in Fig. 3.6. The situation
becomes more complicated for r = 28 where a certain plateau of Ω − ω = 0
appears; however, this plateau is neither perfectly horizontal nor lies at zero.
As a result, PS in this case is not perfect, as shown by 2π or 4π phase slips
in Fig. 3.7.
The reason for this imperfect phase synchronization lies in the broad dis-
tribution of the timescales of the UPOs at r = 28. For this parameter, there is
a saddle point (0, 0, 0) embedded (situated very close) in the chaotic attractor.
The trajectory slows down considerably when passing near the saddle point,
24.85 24.9 24.95
ω (r = 210)
0.1
0.05
Ω−ω
−0.05 ω (r = 28)
8.2 8.3 8.4
Fig. 3.6. Perfect and imperfect phase synchronization for periodically forced Lorenz
systems (3.12); solid line: r = 28, A = 6; dotted line: r = 210, A = 3
3.3 Imperfect Phase Synchronization 43
16π
5895 5900
12π
2πN(t) - ωt
8π
4π
0
0 2000 4000 t 6000 8000
Fig. 3.7. Phase slips of imperfect phase synchronization in the periodically driven
Lorenz system at r = 28 (3.12)
Fig. 3.8. Individual frequencies of unstable periodic orbits embedded into the
Lorenz attractor at r = 28 (3.12); dashed line: mean frequency of the autonomous
chaotic motion
while the oscillation is much quicker when the trajectory rotates around one of
the two unstable foci [61]. Thus, the timescales have a relatively large variation
around the average value, as can be seen in the distribution of the frequency
of different UPO (Fig. 3.8). Due to this large variation of the frequencies,
the phase locking regions of the UPOs do not all overlap to produce a full
PS region of the chaotic attractor. For a given frequency and amplitude of
the external driving signal, there exist certain unstable orbits which are not
locked and imperfect phase synchronization characterizes this complicated
locking behavior.
Using the approach of UPOs, it is easy to demonstrate that a phase

slip develops when the system comes close to a certain UPO which is
not in a 1:1 phase locking with the driving signal. More interestingly,
a careful study reveals that during such a phase slip, the system is in
fact phase locked with the driving signal, however, with another locking
ratio, such as (l − 1) : l or (l − 2) : l. Possible values of l for these cases are
l ≥ l0 and l ≥ 2l0 , respectively, where l0 is determined by the largest
and smallest frequencies of the UPOs, i.e., l0 = ωmax /(ωmax − ωmin )
(smaller l become possible due to the broadening of the Arnold tongues).
In such cases, the 1:1 phase locking regions of some periodic orbits over-
lap with the (l − 1) : l or (l − 2) : l phase-locking regions of other periodic
orbits, as explicitly illustrated in Fig. 3.9. As a result, the system actu-
ally establishes phase locking to the driving signal with another locking
ratio rather than the 1:1 phase locking during the phase slip. During the
time course, the system alternates among different locking ratios when it
visits periodic orbits with overlapping Arnold tongues of different locking
ratios.
This makes clear that PS is a result of adjusting timescales by interac-
tion. Synchronization regimes with different locking ratios can be established
when the system is shifting among different timescales. In the Lorenz system,
a drifting of timescales occurs when the chaotic trajectory accesses closely to
different UPOs. In other cases, the drifting can be induced by a nonstation-
arity of the process, such as in the human cardiorespiratory system, where
alternating locking ratio have been also observed [115, 116].
6
A
0
8.25 ω 8.5
Fig. 3.9. Overlapping of phase locking regions of different locking ratios in the
periodically driven Lorenz system at r = 28 (3.12). Solid line: l = 7 (1 : 1), dashed
line: l = 15 (14 : 15); dotted line: l = 20 (18 : 20)
3.4 Transition to the Regime of Chaotic Phase Synchronization 45
3.4 Transition to the Regime of Chaotic Phase

Synchronization: The Role of Unstable
Periodic Orbits
Now, we discuss in more detail the transition to PS. To obtain a deeper
insight into this problem, we again study PS of chaotic oscillators in terms
of UPOs [117, 119]. A stroboscopic recording of the amplitude (denoted by x
now) and the phase φ of the chaotic signal at each period of the external force
yields the reduced discrete in time system [117]
x(n + 1) = f (x(n), φ(n)), (3.13)

φ(n + 1) = φ(n) + Ω + cos[2πφ(n)] + g(x(n)) (3.14)
which is essentially the circle map coupled to a chaotic map f (see Sect. 2.5.4).
In this presentation, Ω denotes the difference between the natural frequency
of the chaotic oscillator and the frequency of the driving force, is the cou-
pling strength which is proportional to the amplitude of the driving force,
and g(x(n)) corresponds to the nonuniformity of the phase rotations in the
chaotic oscillators as a result of chaotic fluctuations of the amplitude x. The
average growth rate of the phase, i.e., ΔΩ = limn→∞ [φ(n) − φ(0)]/n, corre-
sponds to the phase rotation number, and ΔΩ = 0 indicates synchronization
of the chaotic oscillator to the external force. Without loss of generality, the
periodically forced chaotic tent map f (x, φ) = 1 − 1.9|x| + 0.05 sin[2πφ(n)]
and g(x) = 0.05x were studied in [117].
The analysis of (3.14) is based on the presentation of a chaotic attractor
through its UPOs embedded in it [79]. A periodic orbit of period N has its
real period T ≈ T0 N , where T0 is the average return time of the period
“1” periodic orbit. For different periodic orbits, T0 shows fluctuations around
the average return time of the chaotic oscillations, as is modeled in the map
system by the term g(x). Due to these fluctuations, each periodic orbit has its
individual phase locking region under the periodic external forcing (Fig. 3.10).
In this illustrative mapping model, the region of complete PS is given by the
overlapping region of the synchronization regions (in the forms of Arnold
tongues) of all the UPOs.
Now we consider, how the transition from phase synchronization regime
looks from the viewpoint of the bifurcation theory. In the PS region, and for
a particular UPO, the uncoupled circle map (in (3.14 g(x(n)) = 0) has a
stable fixed point φs and an unstable fixed point φu , and accordingly, each of
the unstable periodic orbit is associated with an attractor and a repeller in
the direction of φ. In the generalized phase space (x, φ), the attractor (x, φs )
and the repeller (x, φu ) are well separated, as shown in Fig. 3.11a for the
mapping system (3.14). In this generalized phase space with the unwrapped
phase variable (the phase variable is viewed on the real line rather than on
the circle, and phase points separated by 2π are not regarded as the same).
The repellers are periodic orbits on the basin boundary of the attractors [118].
Fig. 3.10. Phase locking regions for periodic orbits with period 1–5 in system
(3.14). The region of full phase synchronization regime, where all the phase locking
regions overlap, is delineated with black
Fig. 3.11. Stable (in the φ direction) (pluses) and unstable (filled circles) peri-
odic orbits with periods 1–8 forming the skeletons of the attractor and repeller,
respectively (3.14). (a) Inside the full phase locking region, where the attractor and
the repeller are distinct. (b) Just beyond the border of the phase locking region at
which the attractor and repeller collide, the chaotic attractor as a whole is no longer
attractive in the φ direction
When the parameter moves close and crosses the phase locking boundary, the
attractor and the repeller of a few UPOs approach to each other, coalesce and
annihilate as a result of a saddle-node bifurcation, as illustrated in Fig. 3.11b.
Hence, these UPOs are not locked by the external force and phase slips may
3.5 External Phase Synchronization of Chaotic Intermittent Oscillators 47
occur. The dynamics in the weakly unstable direction φ is the same as in the
usual saddle-node bifurcation.
We emphasize that this desynchronization bears much similarity to the
riddling bifurcation in coupled identical systems where a certain UPO becomes
unstable in the transverse direction [120].
The scaling properties obtained in the transitions to (from) the PS regions
will be discussed in Sect. 4.2.1.
3.5 External Phase Synchronization of Chaotic

Intermittent Oscillators
Now we study external phase synchronization of chaotic systems with a

type-I intermittency. This type of intermittency has been observed in various
experimental fields, such as lasers [121–123], fluid dynamics [124], oxidation
processes [125], semiconductors [126], RV Tauri stars [127], plasma [128], or
electronic circuits [129].
From the viewpoint of synchronization it is important to emphasize, that
a chaotic intermittent motion does have its characteristic time scale – the
average length of the laminar stage τ (ALLS) or the average length between
two laminar stages (see Sect. 2.4). Hence, one can introduce a phase of the
intermittent oscillations (see Sect. 2.4), attributing to each interval between
the beginnings of the laminar stage a 2π phase increase.
The presence of the characteristic time scale allows to consider synchro-
nization problems. For externally forced systems we expect the following
synchronization behavior: Locking of a frequency, which corresponds to the
characteristic time scale (ALLS) of the system with intermittency, takes place
under sufficiently strong external driving.
We firstly treat a quadratic 1D map that exhibits an intermittency type-I
route to chaos which is subjected to external periodic driving. Then we go to
the forced Lorenz system.
3.5.1 Forced Model Quadratic Map
Let us start with a well-studied 1D map that exhibits the type-I intermittency
route to chaos (see Sect. 2.4) under external periodic forcing:
xn+1 = f (xn ) + A cos ωn, (3.15)
where A and ω are the amplitude and the frequency of the external force,
respectively, and f (x) consists of the standard quadratic part and a somewhat
arbitrary chosen return part that acts as a turbulent stage and ensures chaos

ε + x + x2 , if x ≤ 0.2,
f (x) = (3.16)
g(x − 0.2) − ε − 0.24, if x > 0.2.
First we will focus on the case of a coherent chaotic attractor and let g = 2.
We remind, that without external force the map (3.15) demonstrates a type-I
intermittent behavior for ε > 0, i.e., εcr = 0.
As mentioned before, ALLS τ can be viewed as the characteristic time
scale in systems with intermittent behavior. In order to determine ALLS,
we rewrite the map in the form of a first-order time-continuous differential
equation, which is obtained from (3.15) and (3.16) [73, 130]:
ẋ = ε + x2 + A cos ωt (3.17)
Using the following change of the variable x = −u̇/u, we obtain the Mathieu
equation:
ü + (ε + A cos ωt)u = 0. (3.18)
We will study in (3.18) the well-known cases of parametric resonance.
Parametric resonance of the kth order can be achieved when the following
relation is maintained [132]:
√ k
ε ≈ ω, k ∈ N (3.19)
2
Then the solution is characterized by harmonic oscillations with the frequency
k
2 ω and the exponentially growing amplitude
k
u = a cos ( ωt + φ)epk t , (3.20)
2
where a and φ are some constants, and pk depends upon the number of the
zone of the parametric resonance and the parameters of the systems. After
transformation to the original variable, the parametric instability vanishes
and one gets x = k2 ω tan ( k2 ωt + φ) − pk , which yields for the ALLS in (3.15)
and (3.16) in the synchronous regime:
2
τs ∝ (3.21)
kω
and finally: √
2 ε τ0
τs = , (3.22)
kω
where τ0 corresponds to the case of the autonomous map. So, inside a zone of
parametric resonance that takes place in (3.18), the exponential growth does
not affect the solution of (3.17), which is our concern, and all that matters is
the frequency of the solution. On the opposite, being outside of a resonance
zone, one gets a two-frequency solution in (3.18) and no synchronization exists
k
in the original system (3.17): u = aei 2 ωt eiΩk t + c.c. For the first Mathieu zone
(k = 1) the boundaries are given by
ω
A = 4ε| √ − 1|. (3.23)
2 ε
Outside the first zone of synchronization the beating frequency is easy to

write as: 2
ω ω A2
Ω1 = 2− √ − 2 (3.24)
4 ε 4ε
which gives a quadratic scaling law that is typical for phase synchronization
on the border of the synchronization region and refers to Arnold tongues.
Now we present numerical simulations of the system (3.15) and (3.16), and
compare them with our theoretical √
results. We use the irrational frequency
of external force ω = 0.001 · 2π 5−1 2 , unless another value is specified. In
Fig. 3.12a, b we show locking of the ALLS for different values of ε as the ampli-
tude of the driving A is increased. It is easy to see that the theoretical result
for the duration of the laminar stage of the synchronized motion (3.22) is quite
well satisfied. Note that for the considered case it holds τs ≈ Tcs = 2π/ω.1
Moreover, this relation also remains valid when the assumptions d ε and
(3.19) are not fulfilled. By approaching the synchronization plateau the curves
in Fig. 3.12a show a quadratic convergence, which confirms the quadratic
scaling law predicted by (3.24). In addition, the phase locking, as shown in
Fig. 3.12b, confirms the phase nature of the observed chaotic synchronization
(the phase of the intermittent oscillations is defined according to (2.27)).
(a) (b)
2000 90
ε=2.64*10−6
3.77*10−6
4.9*10−6 70 A=1.9*10−6
1800
50
φ2−φ1
<τ>
30
A=2.1*10−6
1600
10
A=2.3*10−6
1400 −10
0 1*10−6 2*10−6 3*10−6 0 150000 300000
A time
Fig. 3.12. (a) Locking of the ALLS by external periodic driving in (3.15) and (3.16).
(b) Phase difference evolution in nonsynchronous (A = 1.9 × 10−6 , 2.1 × 10−6 ) and
synchronous (A = 2.3 × 10−6 ) regimes in (3.15) and (3.16) for ε = 2.64 × 10−6
1
Note
√ that from the last equation we have the ALLS for autonomous map < τ0 >≈
π/ ε.
(a) (b)
4*10−5 6*10−6
3*10−5 S3
4*10−6 S1+ S1−
2*10−5
S2
ε
ε
2*10−6
1*10−5
S1
0
Ioff 0
Ioff
−1*10−5
0 1*10−5 2*10−5 2*10−6 0 2*10−6 4*10−6
A A
Fig. 3.13. (a) Three first zones of synchronization (Sk , k = 1, 2, 3) and the region
where intermittency is absent (Iof f ) in (3.15) and (3.16). (b) Regions of synchroniza-
tion (the first zone) with a positive (S1+ ) and a negative (S1− ) Lyapunov exponent
(LE). The theoretical border of the first synchronization zone (3.23) is the curve
marked by “o”
In Fig. 3.13 the first three zones of synchronization Sk , k = 1, 2, 3 and

the region of absence of intermittency Iof f are presented in the (d, ε)-plane.
Calculations show that the points of the synchronization regions that join the
ε-axis are positioned with accordance to (3.19). In Fig. 3.13 the regions S1
and Iof f are presented in more detail. The boundaries of the first zone of
synchronization defined by relation (3.23) (shown by the “o”-marked curve)
give a remarkable coincidence with our numerical results.
We find that the region of synchronization consists of two qualitatively
different parts: In S1+ the LE is positive, while in S1− it becomes negative, i.e.,
no chaos exists.
When the chaotic attractor in the autonomous map (3.15) and (3.16)
(A = 0) is strongly noncoherent, synchronization is more difficult to achieve.
Indeed, in this case the ALLS are broadly distributed in a long range of
values, so a priori the possibility of adjusting the motion of this type may be
problematic. Still, we carried out numerical simulations of (3.15) and (3.16) in
case g = 11. In accordance with [73] two typical timescales in the autonomous
map ((3.15) and (3.16) (A = 0)) are observed. The first one is inherited from
the case of g < 5 and presents itself a long-time laminar motion. The second
one is a short-time laminar motion (about ten iterations of the map) which
becomes more pronounced as g being increased. It turns out that synchroniza-
tion of the long timescale persists, while, quite naturally, the short time one
is not synchronized. If one measures ALLS by taking into account only the
long timescale and neglecting the fast passages, synchronization by external
driving is clearly observed. It allows to claim the existence of imperfect phase
synchronization, if the driven system behaves nonphase-coherent [112].
3.5.2 Forced Lorenz Oscillator
Now we present the results of numerical simulations of the Lorenz system (that
also exhibits type-I intermittency for r ≈ 166.06 [71]) under multiplicative
external driving which may also be regarded as a modulation of the bifurcation
parameter r [133]:
⎧
⎨ ẋ = σ(y − x),
ẏ = −y − xz + (r + A) cos ωt, (3.25)
⎩
ż = −bz + xy,
where b = − 83 , σ = 10, ω = 0.04177. To test whether a laminar or a tur-

bulent stage is observed, we calculate the sequences {yn } corresponding to
intersections of the trajectory with the plane {x = 0, ẋ > 0} and compare
each value with the correspondent fixed point in the autonomous system on
the edge of the tangent bifurcation. In the computations of the ALLS τ ,
a discrete time is used (one unit corresponds to the continuous-time interval
between intersections with the selected plane). In Fig. 3.14 we present results
for different values of r (including subcritical ones). There exist plateaus of
synchronization of the ALLS and they are similar to those derived for the
map (3.16). However, differences are also observed and resemble the imper-
fections of synchronization that we have discussed in case of the strongly
nonphase-coherent map (3.15) and (3.16). As the amplitude of the driving is
0.025
r−rc = −0.0005
0.020 0.0005
0.0015
0.015
1/<τ>
0.010
0.005
0.000
0.000 0.005 0.010 0.015 0.020
A
Fig. 3.14. Synchronization plateaus for the Lorenz system under external driving.
Here rc = 166.06149, τ1 = 0 corresponds to nonintermittent motion
increased, synchronization gradually disappears (first comes a shallow slope,

then a quadratic-like one).
3.6 Synchronous Response of Excitable Systems

to a Periodic External Force
Excitable systems are very typical in nature, e.g., neuronal networks and
cardiac tissue. Without any perturbations these systems are in the resting
state. But their response to external periodic force looks very similar to
the external synchronization of oscillatory systems. In this section we study
the effects of external force of a single excitable element governed by the
Luo–Rudy model (2.43)–(2.48). The equation for voltage evolution:
∂V
Cm = −Iion − Istimulus , (3.26)
∂t
where Istimulus is the input stimulus and Iion is the sum of six ionic currents
Iion = Ina + Isi + Ik + Ik1 + Ikp + Ib (3.27)
(a)
50
Istimulus
30
10
−10
(b)
50
voltage
−50
−100
(c)
50
voltage
−50
−100
3000 3500 4000 4500 5000
time
Fig. 3.15. Synchronous response of the Luo–Rudy system on the external peri-
odic stimulation. (a) The train of external pulses. (b) Voltage evolution for 2:1
synchronous excitation. (c) Voltage evolution for 1:1 synchronous excitation. The
parameters are amplitude of the external stimuli =50, pulse duration = 10 ms. The
initial conditions are different
3.7 Conclusions 53
0.07
0.06
1:1
0.05
Ω
0.04
2:1
0.03
0.02
0.045 0.050 0.055 0.060 0.065 0.070 0.075 0.080
ω
Fig. 3.16. Hysteresis by the synchronous response of the Luo–Rudy system on the
external periodic stimulation. The response frequency Ω vs. the input frequency ω
for parameters as in Fig. 3.15 is plotted
As the input stimulus Istimulus , we apply a periodic sequence of rectan-

gle pulses of fixed amplitude and duration. As the control parameter we take
the frequency ω of the forcing. The series of numerical experiments show the
following results. When ω is small, 1:1 synchrony resulted. This means that
each pulse produces an action potential, i.e., cell becomes excited. With an
increase of this pacing frequency, the transition to 2:1 synchrony is observed
at the critical value ω 2 . If we move toward the opposite direction, i.e., we
decrease the pacing frequency, the transition from 1:1 synchrony to 2:1 syn-
chrony happens at another critical value ω 1 < ω 2 , i.e., a hysteresis is formed
(Figs. 3.15 and 3.16). This result is obtained by the following adaptation pro-
cedure: The final values of variables obtained in the previous simulations are
taken as the initial conditions for the next experiment. Therefore, the response
of the system depends on the current state in the moment when the external
stimulus is applied (see also [134]).
As we will show in Chap. 11, the effect of bistability of periodic rhythms is
a reason of the limited success for suppressing spiral wave chaos in 2D media
using overdrive pacing.
3.7 Conclusions
In this chapter external synchronization of periodic and chaotic oscillators is

presented. The main finding are the following:
– Chaotic oscillators can be synchronized in terms of frequency and phase.

This is shown for chaotic attractors which occur via period-doubling
scenario (Rössler oscillator), via intermittency (Lorenz oscillator and
corresponding modeling map), and in a hard way (“classical” Lorenz
attractor).
– There are many similarities in the synchronization of regular and chaotic
oscillators. But in the latter case we have to distinguish perfect and imper-
fect phase synchronization in dependence on the spectrum of the system’s
UPOs.
– Using external periodic forcing, one can govern the characteristic time
scales of chaotic oscillators. This can be used in the theory of chaos control,
and can be applied, e.g., for the design of electronic and radio devices with
controlled spectra.
– Synchronization-like phenomena can be observed for externally forced
excitable systems.
Finally we would like to mention other interesting phenomena arising in

the problems of external force on the regular and chaotic oscillators:
1. Large force applied to a strongly nonlinear periodic oscillator can already

cause a chaotic behavior [10, 135].
2. Periodic external force acting on a chaotic system can destroy chaos and
as a result a periodic regime appears. This effect occurs for a relatively
strong forcing as well [10, 136].
4
Synchronization of Two Coupled Systems
In this chapter we go the next step and analyze synchronization phenomena in

systems of two coupled elements. We start (Sect. 4.1) with the classical case,
i.e., two coupled regular systems: We analyze two weakly coupled arbitrary
limit-cycle oscillators (Sect. 4.1.1), the nonscalar coupled van der Pol oscil-
lators (Sect. 4.1.2) and finally two coupled active rotators (Sect. 4.1.3). Then
(Sect. 4.2), synchronization of coupled chaotic systems is presented. Start-
ing with the analysis of chaotic Rössler resp. intermittent chaotic oscillators
(Sects. 4.2.1 and 4.2.2), Sect. 4.2.3 is devoted to the investigation of coupled
chaotic phase oscillators. Finally (Sect. 4.3) we treat two coupled circle maps
as a prototypical discrete in time system.
4.1 Synchronization of Regular Systems

Synchronization of two self-excited oscillatory system is a classical problem in
the theory of synchronization. Mutual synchronization of two quasiharmonic
self-oscillators was firstly studied by Mayer [138] and Gaponov [139]. For self-
oscillators in a relaxation regime, it was investigated by Bremsen and Feinberg
[140] and Teodorchik [141].
Although the problem of mutual synchronization of limit-cycle oscilla-
tors was intensively investigated in the first half of the 20th century, several
recent studies show the active interest on many open problems in this topic.
Among several recent findings, we would like to emphasize: limits of strong
and small (compared to damping in an isolated oscillator) coupling [142,143],
oscillator death effects in ensembles of globally coupled oscillators with ran-
domly distributed frequencies [144], or synchronization of (isochronous and
nonisochronous as well) limit-cycle oscillators (in case of coupling being of
the same order as damping in the isolated oscillator) quite scrupulously stud-
ied for some types of coupling [145,146]. Synchronization and oscillator death
have been studied in oscillators coupled via time delay [147–151].
56 4 Synchronization of Two Coupled Systems
This section is devoted to the investigation of the synchronization

phenomena in coupled regular oscillators. First (Sect. 4.1.1) we present a
common approach to the investigation of locking phenomena in a pair of
weakly coupled oscillators. In this case the problem can be reduced to the
analysis of the phase dynamics. Then (Sect. 4.1.2), we study the more general
case: coupled limit-cycle oscillators without the restriction of small coupling.
And finally (Sect. 4.1.3) we discuss specific synchronization properties of
coupled rotators.
4.1.1 Phase Dynamics Approach
To begin with we treat a system of two weakly coupled arbitrary limit-cycle

oscillators:
ẋ1 = F1 (x1 ) + εP1 (x1 , x2 ),
(4.1)
ẋ2 = F2 (x2 ) + εP2 (x1 , x2 ),
where x1 ∈ Rn1 and x2 ∈ Rn2 , and ε 1,1 and having nearly equal natural
frequencies ω1,2 = ω + εω1,2 . The analysis of a 1:1 synchronization behavior
of such oscillators can be reduced to the study of the coupled phase equations
[1, 8, 20, 152]:
φ̇1 = ω1 + εq1 (φ2 − φ1 ),
(4.2)
φ̇2 = ω2 + εq2 (φ1 − φ2 ),
where φ1,2 are the deviations from the phase ψ = ωt. ω1,2 denote small
(rescaled) deviations from the common frequency ω and q1,2 are 2π-periodic
functions in both phases. In the case of n : m synchronization the functions
q1,2 have the forms: q1 = q1 (nφ2 − mφ1 ) and q2 = q2 (mφ1 − nφ2 ), respectively.
For the difference between the phases of both oscillators θ = φ2 − φ1 ,
system (4.2) can be rewritten in the form:
θ̇ = Δ − 2εq(θ), (4.3)
where Δ = ω2 − ω1 and
q(θ) = q2 (−θ) − q1 (θ). (4.4)
Steady states of (4.3) defined from:
Δ
q(θ) = (4.5)
2ε
correspond to a synchronous behavior in (4.2) and hence also in the original
system (4.1). Note that here we have not made the restriction to weakly non-
linear oscillators. Therefore, model (4.3) works for relaxation oscillators as
1
Here and in other chapter we use letter ε if the coupling between elements is
small.
4.1 Synchronization of Regular Systems 57
well [153]. In the latter case the function q(θ) may have a discontinuity at the
origin that leads to the rapid appearance of an in-phase (θ ≈ 0) synchroniza-
tion regime.
The simplest coupling function is the sine, i.e., q1 (θ) = −q2 (θ) = sin θ.
Then the phase model (4.3) is:
θ̇ = Δ − 2ε sin θ, (4.6)
i.e., the first-order Adler equation (for a detailed description of this equation
see Sect. 2.5.1)
Its steady state
Δ
θ̄ = arcsin (4.7)
2ε
exists if
|Δ| < |2ε| (4.8)
and is stable. It defines a constant phase shift between the phases of oscillators.
Therefore, if the frequency mismatch |Δ| is less than some critical value Δcr =
|2ε| synchronization regime exists.
If the synchronization regime condition (4.8) becomes destroyed, e.g., due
to the increase of the frequency mismatch Δ, the phase difference is no more
bounded and in the original system (4.1) a beating regime appears. This regime
is characterized by the beating frequency Ωb which can be defined from (4.3)
for an arbitrary function q(θ) as:
2π −1
dθ
Ωb = 2π (4.9)
0 2εq(θ) − Δ
or especially for q(θ) = sin θ the beating frequency is:

Ωb = Δ2 − 4ε2 (4.10)
which is presented in Fig. 4.1.

In the vicinity of the critical frequency mismatch Δcr , the beating fre-
quency Ωb has a square-root dependence on Δ, i.e.,2

Ωb ≈ Δ − Δcr (4.11)
Corresponding to this frequency, the period Tb = 2π/Ωb , i.e., the period

of rotations in the model (4.6), is, therefore, proportional to (Δ − Δcr )−1/2 .
Close to the bifurcation point Δcr , rotations are strongly nonuniform (see
Sect. 2.5.1), i.e., the phase trajectory spends a lot of time in the vicinity of
the point θ̄ = π/2 (for sine as coupling function) and then makes a jump of
2π: a phase slip takes place. Hence, Tb is the period between these phase slips.
2
Note that this law does not depend on the special form of the coupling function
q(θ).
2.0
1.0
Ωb
0.0
−1.0
−2.0
−2.0 −1.0 0.0 1.0 2.0
Δ
Fig. 4.1. Beating frequency Ωb vs. frequency mismatch Δ for the coupling function
q(θ) = sin θ and ε = 0.5 in (4.3)
Conclusion
In this section we have shown that for weakly coupled oscillators the synchro-
nization problem can be solved only via the analysis of the phase dynamics
of the coupled systems. At that it is possible to say that the synchronization
regime appears for any large frequency mismatch (see conditions 4.8).
4.1.2 Synchronization of Two Coupled van der Pol Oscillators
Now we analyze a system of two nonscalar-coupled nonidentical van der Pol

oscillators (for the single case see Sect. 2.2), which, when isolated, differ in
frequency and amplitude of their stationary regimes [154].
Mathematical Model
In our study the coupling is supposed to contain dissipative (reactive) as well

as conservative (inductive) terms. So we get the following equations:
ẍ1 + x1 = μ(1 − x21 )ẋ1 + μα(x2 − x1 ) + μβ(ẋ2 − ẋ1 ),

(4.12)
ẍ2 + (1 + μΔ)x2 = μ(1 + γ − x22 )ẋ2 + μα(x1 − x2 ) + μβ(ẋ1 − ẋ2 ),
where we assume μ 1, α and β regulate the rate of the conservative and dis-
sipative coupling, respectively, γ and Δ define the amplitude and the frequency
mismatches. Averaging (4.12) over the period of possible synchronous oscilla-
tions we obtain:
ż1 = z1 (1 − z1 z̄1 ) + (β − iα)(z2 − z1 ),

(4.13)
ż2 = z2 (1 + γ − z2 z̄2 ) + iΔz2 + (β − iα)(z1 − z2 ),
z1 and z2 being complex amplitudes of oscillations. The conversion to real

amplitudes R1,2 and phases φ1,2 by z1,2 = 21,2 e−iφ1,2 leads to
R
⎧
R12
⎪
⎪ Ṙ 1 = R 1 1 − β − + R2 (α sin(φ1 − φ2 ) + β cos(φ1 − φ2 )),
⎪
⎪ 4
⎪
⎪
⎨ Ṙ = R 1 + γ − β − R22 + R (−α sin(φ − φ ) + β cos(φ − φ )),
2 2 4 1 1 2 1 2
⎪
⎪ R1 φ˙1 = −αR1 + R2 (α cos(φ1 − φ2 ) − β sin(φ1 − φ2 )),
⎪
⎪
⎪
⎪
⎩
R2 φ˙2 = −(α + Δ)R2 + R1 (α cos(φ1 − φ2 ) + β sin(φ1 − φ2 )).
(4.14)
Writing this system in terms of the phase difference θ = φ1 − φ2 yields a
system that we are going to deal with in order to study synchronization in
the original system (4.12)
⎧
⎪ R2
⎪
⎪ Ṙ1 = R1 1 − β − 41 + R2 (α sin θ + β cos θ),
⎪
⎨
R2
Ṙ2 = R2 1 + γ − β − 42 + R1 (−α sin θ + β cos θ), (4.15)
⎪
⎪
⎪
⎪
⎩ θ̇ = Δ + α R2 − R1 cos θ − β R2 + R1 sin θ.
R1 R2 R1 R2
As long as nonzero steady states (R1,2 = 0) of (4.15), which are solutions of

⎧ 2

⎪
⎪ R 1 − β −
R1
+ R2 (α sin θ + β cos θ) = 0,
⎪
⎪ 1 4
⎪
⎨
2
R2
R2 1 + γ − β − 4 + R1 (−α sin θ + β cos θ) = 0, (4.16)
⎪
⎪
⎪
⎪
⎪
⎩ Δ + α R2 − R1 cos θ − β R2 + R1 sin θ = 0,
R1 R 2 R 1 R 2
correspond to synchronization regimes of (4.12), we focus our attention on

analyzing these rest states of (4.15). To determine regimes of synchronization,
we first calculate a bifurcation diagram of (4.15). Finding bifurcation curves,
where the rest states change their stability, emerge or disappear, allows to
identify regions of qualitatively different behavior in the (α, β, γ, Δ) parameter
space. Additionally, we get the boundaries, where limit cycles of (4.15), that
correspond to quasiperiodic solutions of (4.12) (i.e., beating regimes), appear,
and then we investigate the stability of the zero rest state (R1 = R2 = 0),
which is evidently not captured by (4.15) and (4.16), by performing a linear
analysis in (4.13). Moreover, we focus our attention on the dependence of the
phase, frequency, and amplitude concerning synchronization regime on these
parameter values.
In the case γ = 0, α, β 1 an analytical approach has proved to be

successful [143]. As the rest states near the unperturbed (α = β = 0) values
0 0
of the amplitudes R1 = R2 = 2 are sought, let
R1 = 2 + r1 ,
(4.17)
R2 = 2 + r2 ,
where r1 , r2 2. Then

r1 = α sin θ + β(cos θ − 1),
(4.18)
r2 = −α sin θ + β(cos θ − 1),
gives a reduction of (4.15) to O(α2 , αβ, β 2 ):
Δ − α2 sin 2θ − 2β sin θ = 0 (4.19)
According to [143], (4.19) can have four, two, or no solutions at all. They
correspond to steady states of (4.15), and among them one or two may be
stable and these stable states refer to the synchronized regimes in the original
system (4.12). These synchronization solutions can be roughly classified into
two types:
1. In-phase if |θ| ≤ π/2
2. Antiphase (otherwise) synchronization regime
We illustrate the dynamics of the studied system (4.12) by plotting the
bifurcation diagrams in the (β, Δ)-plane for a sequence of α values (Fig. 4.2).
(a) (b)
7 1.4
6 α=1 OD 1.2
5
0S1
1
α = 0.3
4 B 0.8
Δ
α=1
Δ
3 0.6 α=0.5
α =0
2 0.4
α = 0.1
S
1 0.2 S2 α=0.33
0 0
−1 0 1 2 3 4 0 0.05 0.1 0.15 0.2 0.25 0.3
β α =0.27 β
Fig. 4.2. For the system (4.13): (a) Boundaries of the regions of the synchronization
regime (marked by S, for each value of α the region lies below the bifurcation curves).
The beating regime region is marked by B and the oscillator death region by OD.
(b) Boundaries of the bistability regime (for each value of α the bistability region
lies below the bifurcation curves and is marked by S 2 , the region of monostability
is marked by 0 S 1 (the left upper index refers to the in-phase regime: |θ| ≈ 0))
The evolution of the in-phase synchronization regime and the bistability

regimes of in-phase and antiphase synchronization regime is clearly seen there.
The region, where at least one synchronization regime exists, is marked by
S (Fig. 4.2a) and lies beneath the bifurcation curves that separate it from
the beating regime region B (to the left) and the oscillator death region
(OD) – the region of absence of oscillations (in (4.15) steady state with
R1 = R2 = 0 is stable) (to the right). This comparison clearly exhibits
that the increase of the strength of the conservative coupling α makes the
synchronization region growing at the expense of the oscillator death and
the beating ones. According to the presented results, an optimal strength
of dissipative coupling β can be claimed to exist. At this strength, the
synchronization regime holds for an arbitrarily large frequency mismatch
Δ and the “channel” around the asymptote β = 1 broadens as conserv-
ative coupling is increased. At the same time, the region, where only one
stable (in-phase) synchronization regime exists (0 S 1 , Fig. 4.2b), is partially
substituted by the region of bistability S 2 . This S 2 is situated beneath the
bifurcation curves (for different α, respectively) that divide it from the former
one. Note, that while the height of S 2 gradually increases, its width cannot
exceed β = 14 .
Our numerical simulations also exhibit the existence of very small regions,
where in dependence on the initial conditions four different synchronization
regimes may be observed.
Characteristics of Synchronization Regimes
The analysis of (4.15) and (4.16) allows to determine the frequency ωs and
the amplitude characteristics of the synchronized behaviors.
If α, β 1, then one can easily get the frequency of the synchronized
solution
Δ
ωs = −φ̇1 = − − α(1 − cos θ). (4.20)
2
By neglecting the terms of O(αβ) the frequency, observed in the system (4.12),
is ω̃s = 1 + μωs
2 .
According to (4.20), the in-phase solution implies that the synchronization
frequency ωs is equal to the mean frequency of both uncoupled oscillators:
Δ
ωs = −φ̇1 = −φ̇2 ≈ (4.21)
2
and that of the antiphase solution is
Δ
ωs = −φ̇1 = −φ̇2 ≈ 2α + . (4.22)
2
Numerical simulations yield that there is no qualitative difference in the
case when α and β are not small (Fig. 4.3a–d). This leads to the important
(a) (b)
1.8 −2.55
α =0.1 α= 0.27, Δ= 0.04
α =0.27 −2.6 α= 0.33, Δ= 0.04
1.6 α =0.33 α= 0.5, Δ=0.2
α =0.5 −2.65 α = 1, Δ= 0.8
1.4
−2.7
1.2 −2.75
φ
−2.8
φ
1
−2.85
0.8 −2.9
0.6 −2.95
−3
0.4
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 −3.05
0 0.05 0.1 0.15 0.2 0.25
β β
(c) (d)
0.515 2.5
α =0.1
0.51 α=0.27, Δ= 0.04
α =0.27
α =0.33 α=0.33, Δ= 0.04
0.505 α =0.5 α=0.5, Δ =0.2
2 α= 1, Δ =0.8
0.5
0.495
ωs
0.49
ωs
1.5
0.485
0.48
0.475 1
0.47
0.465
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.5
0 0.05 0.1 0.15 0.2 0.25
β β
(e) (f)
2.2
2.1
2
2 1.9
1.8
R1,2
R1,2
1.7
1.6
1.5
1.5
1.4
1.3
1.2
1 0 0.05 0.1 0.15 0.2
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
β β
Fig. 4.3. Phase difference θ̄, synchronization frequency ωs , and amplitude char-
acteristics R̄1,2 of the in- and antiphase solutions of (4.13) in dependence on the
dissipative coupling β. (a, c) stable in-phase solution for fixed Δ = 1, (b, d) stable
antiphase solution, typical amplitude values for (e) in-phase (α = 0.5, Δ = 1) and
(f ) antiphase (α = 0.5, Δ = 0.2) synchronization regimes (R1 solid lines, R2 dashed
lines)
conclusion that the synchronization phenomena as well as the phase, fre-

quency, and amplitude of the synchronization regime remain qualitatively the
same in a broad range of parameter values, and show deviations only near the
saddle-node bifurcation. In particular, the frequency is also in a quite good

quantitative agreement with (4.21) and (4.22).
Frequency asymmetry. Now let us turn to the area between the beating
regime and the oscillator death regions with a characteristic value of β near 1
(Fig. 4.2a) that was described before as an optimum coupling value. The corre-
spondent boundary curves have an asymptote β = 1; the width of this region
decreases to zero for Δ → ∞. Increasing Δ gradually, we find numerically
that the synchronization properties change dramatically. The synchronization
frequency deviates from the mean frequency and approaches that of the first
oscillator, i.e.,
ωs = α. (4.23)
Hard and soft transitions to the synchronous regime. As mentioned in
Sect. 4.1.1, another important characteristic of interacting systems, where syn-
chronization can take place, is the beating frequency Ωb .
Three types of qualitatively different oscillatory behavior may be observed
in the averaged system (4.13):
• θ = φ1 − φ2 = const corresponds to the synchronization regime, Ωb = 0.
Regime of strong synchronization takes place
• |θ| = |φ1 − φ2 | ≤ const means that a stable limit cycle exists, and changes
in the phase difference of the solution are limited. This is called “phase
entrainment” [155] or “phase trapping” [145]; Ωb is still zero, i.e., this is a
synchronization regime as well, but this regime is nonstrong
• Finally, if |φ| = |φ1 − φ2 | grows unbounded, Ωb = 0 now, then a stable
limit cycle (also called a “libration orbit” [155]) corresponds to a phase
drift regime. Thus, in the original system (4.12) a beating regime takes
place.
The transition from the second to the third type of behavior was the
subject of study in [155] for the special case α = 0, Δ = 0, β = γ = 0.
Next, we describe the routes of losing the synchronization regime appearing
in this system. By numerically performed simulations of the original system
(4.12) for μ = 0.1, the averaged beating frequency Ωb is calculated for γ = 0,
varying Δ, a fixed α, and a sequence of β values (Fig. 4.4a) and vice versa
(Fig. 4.4b). Indeed, when a stable rest state disappears through a saddle-node
bifurcation, a stable limit cycle with a phase difference growing unbounded is
born instead. There are two kinds of transitions (1) As long as the flow slows
down near the location of the former rest state, the period of this cycle is
very large near the bifurcation point, making the averaged beating frequency
increasing continuously (Fig. 4.4a), i.e., a soft transition to (from) synchro-
nization regime takes place. Nevertheless, the larger β, the sharper the slope
of the frequency curve is. (2) A quite opposite scenery is observed, when an
Andronov–Hopf bifurcation occurs. A limit cycle with a bounded variation
(a) 2.5
(b)
2.5
β=0.1 α=0.1
β=0.3 α=0.3
2 β=0.5 2 α=0.5
β=0.7 α=0.7
β=0.9 α=0.9
1.5 1.5
Ωb
1
Ωb
0.5 0.5
0 0
0.5 −0.5
0 1 2 3 4 5 0 1 2 3 4 5
Δ Δ
Fig. 4.4. Beating frequency Ωb , calculated for the original system (4.12) for μ = 0.1
and γ = 0, illustrates soft and hard transitions from the synchronous to the beating
regime. (a) fixed α = 0, (b) fixed β = 0.5
of the phase difference appears and it requires larger frequency mismatches

Δ to show a nonzero frequency (Fig. 4.4b), i.e., a hard transition is observed.
When the phase of the solution becomes eventually drifting, the period is far
from being infinitely large. Consequently, the discontinuity of the frequency
characteristic corresponds to a manifold in the parameter space. By crossing
of that the beating regime distinctively changes its properties which gives rise
to a limit cycle with unbounded phase growth (drift).
Oscillators with Frequency and Amplitude Mismatches

and Their Response Characteristics
Now effects of the amplitude mismatch (γ = 0) on the synchronization regimes

are included, too. Main properties are given in the summarized synchroniza-
tion diagram (Fig. 4.5) for different values of γ.
As before, the region, where at least one synchronization regime exists,
is marked by S (Fig. 4.5a–c) and either has two separate parts (Fig. 4.5a, b)
(if γ < 2α) or forms one continuous area (Fig. 4.5a), which has a branch
between the beating regime (B) and the oscillator death (OD) regions.
A comparison clearly shows that an increase of the amplitude mismatch γ
makes the synchronization region expanding strongly. Its width for a fixed
Δ 1 is close to γ. At the same time, the region of bistability S 2 is gradually
substituted by the region, where the only stable rest state is the antiphase one
π 1
S (Fig. 4.5d, e), which eventually merges with the region of the in-phase syn-
chronization regime 0 S 1 and the region S 1 , where the only stable rest state
varies its phase difference characteristic from in- to antiphase continuously
(Fig. 4.5f).
(b)
25
(a) 1.5
20
1 B
15
OD
Δ
Δ
B 0S1
10
0.5
5 S2
S
0 0
0 0.5 1 1.5 2 2.5 3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
β β
25
(c) 1.5 (d)
20 B
1
15
S
Δ
πS1
Δ
B OD 0S 1
10
0.5
5
S S2
0 0
0 0.5 1 1.5 2 2.5 3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
β β
20 (f) 1.5
(e)
18
16
B
14
B OD 1
12 S
10
Δ
Δ
8
0.5 S1
6
4
2
0 0
0 0.5 1 1.5 2 2.5 3 3.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
β β
Fig. 4.5. Influence of the amplitude mismatch γ in the system (4.13). (i) Regions
of synchronization (S 1 ) for α = 0.5 and (a) γ = 0, (c) γ = 1, (f ) γ = 2; (ii) regions
of bistability (S 2 ), surrounded by regions of in- (0 S 1 ) and antiphase (π S 1 ) synchro-
nization regime. α = 0.5 and (b) γ = 0, (d) γ = 0.1, (e) γ = 1.2
Conclusions
We summarize the main results of this section:

1. Nonzero frequency mismatch (Δ = 0) and zero amplitude mismatch
(γ = 0)
– There exist regimes of mono-, bi- and multistability of synchronization,
oscillator death, and beating regimes.
– Almost in the whole region, where each synchronization regime exists,
its frequency remains approximately constant. Thus, we can mark out
three frequency values associated with the in- (ωs = Δ 2 ) and antiphase
(ωs = Δ 2 + 2α) regimes and synchronization near the frequency of the
first oscillator (ωs = α).
– Two qualitatively different types of synchronization regimes break up:
soft and hard transitions take place. For increasing Δ from zero (syn-
chronization regime) to its limit value, the beating frequency demon-
strates either a discontinuity or a smooth transition.
– In general, the region where synchronization takes place enlarges
monotonously with strengthening of a conservative coupling.
2. Nonzero frequency and amplitude mismatches
– Under certain conditions the oscillators become qualitatively unequal
in course of interaction and the second one dominates the first one, all
the while being only slightly “stronger” in the amplitude value, when
isolated.
– Generally, the region of synchronization enlarges as the amplitude
mismatch increases. The shape of this region acquires a new feature,
specifically, there appears a branch between the beating regime and
the oscillator death regions, which broadens following the increase of
the amplitude mismatch (the width of it was shown to be γ).
4.1.3 Synchronization of Coupled Active Rotators
Finally, we briefly describe synchronization phenomena in two coupled active

rotators (cf. Sect. 2.5.1):
φ̇1 = ω1 − a sin φ1 + d sin(φ2 − φ1 ),

(4.24)
φ̇2 = ω2 − a sin φ2 + d sin(φ1 − φ2 ),
where ω1,2 and d is the coupling strength. The parameter a characterizes

the nonuniformity of rotations. When a = 0, the synchronization problem in
(4.24) is special form of (4.2) which was solved above: both oscillators are
synchronized with the constant phase difference shift φ2 − φ1 = arcsin(Δ/2d)
if Δ < 2d (see (4.8)).
Due to the nonuniform rotations (a = 0 in (4.24)), the systems described

by (4.24) cannot be synchronized with a constant phase difference shift.
Phase diagrams of different regimes in (4.24) are presented on the plane
(Δ = ω2 − ω1 , d) in Fig. 4.6. With an increase of the rotation nonuniformity
parameter a, more and more n : m synchronization regions appear. There
the main region of 1 : 1 synchronization becomes slightly smaller, while the
other n : m synchronization regions become to cover a larger area in the
plane (Δ, d).
Δ
(a) (b) (c)
1.0
0.8 -
0.6 -
0.4 -
0.2 -
0.0
0. 1. 0. 1. 0. 1.
d d d
Fig. 4.6. Phase diagrams of different synchronization regimes in (4.24) are presented
on the plane (Δ = ω2 − ω1 , d). Parameters are: ω1 = 1, a = 0.5 (a), a = 0.8 (b), and
a = 0.95 (c). Synchronization regions are indicated in gray colors, beating regions
are white. In all diagrams the main gray region corresponds to 1 : 1 synchronization
regime. The gray regions to the left and to the top correspond to 2 : 1, 3 : 1, 4 : 1,
etc. synchronization regimes
Conclusions
The main finding of this section is that nonuniformity of rotations can lead
to various n : m synchronization while for the uniform rotation only 1 : 1
synchronization is possible.
4.2 Synchronization of Coupled Chaotic Oscillators
For coupled chaotic systems, different synchronization regimes have been

found. The most important of them are (1) complete or identical synchroniza-
tion (CS) [156–158], (2) phase synchronization (PS) [109, 110, 118, 162–165],
and (3) generalized synchronization (GS) [167, 168].
In this section we pay main attention to chaotic phase synchronization
(CPS) which is rather close to the synchronization of periodic oscillators. We
will present three basic routes to PS.
4.2.1 Phase Synchronization of Rössler Oscillators
We demonstrate that two nonidentical chaotic Rössler oscillators are able to

synchronize their phases due to coupling [59]. The phenomenon of CPS is
a direct extension of the classical definition of synchronization of periodic
oscillators, where only phase locking is important, while no restriction on the
amplitudes is imposed. CPS can be defined as the appearance of entrain-
ment between the phases of interacting systems, while the amplitudes remain
chaotic and, often, noncorrelated. Similar to the synchronization of periodic
oscillators, CPS already sets in for very weak or even vanishing coupling if
the detuning between the interacting oscillators is small.
CPS has been firstly reported for chaotic attractors with rather coher-
ent phase dynamics. These attractors have a relatively simple topology of
oscillations and a well-pronounced (narrow-band) peak in the power spec-
trum, which allows to introduce the phase and the characteristic frequency of
motions (cf. Figs. 2.3a and 2.4a). Thus, CPS in this case is a direct extension
of PS of periodic oscillations and it is very similar to PS of periodic oscillations
in the presence of small noise. However, such a phase-coherent dynamics is a
strong restriction for applications. We often meet noncoherent attractors and
rather broadband power spectra (Fig. 2.4f). Then a phase of the oscillations
may not be defined straightforward, and there in general does not exist only
one single characteristic time scale. In contrast to phase-coherent attractors,
the analogy with noisy periodic oscillators is then no longer evident. To treat
such noncoherent systems, we introduce in the following a rather general
approach to define a phase as that described in Sect. 2.3. This allows to study
CPS in systems of coupled chaotic oscillators with even strongly noncoherent
phase properties.
4.2 Synchronization of Coupled Chaotic Oscillators 69
We demonstrate the different types of transition to CPS for the paradig-

matic system of two bidirectionally coupled nonidentical Rössler oscillators:
ẋ1,2 = −ω1,2 y1,2 − z1,2 ,
ẏ1,2 = ω1,2 x1,2 + ay1,2 + d(y2,1 − y1,2 ), (4.25)
ż1,2 = f + z1,2 (x1,2 − 8.5),
where d is the coupling strength. ω1,2 determine the mean frequency of the
oscillators in the case of phase-coherent attractors. In our study we take f =
0.1, ω1 = 0.98, and ω2 = 1.02. The parameter a is changed in the interval
[0.15; 0.3].
As already mentioned in Sect. 2.3, for phase-coherent as well as funnel
attractors occurring in the Rössler oscillator, projections of chaotic trajecto-
ries onto the plane (ẋ, ẏ) always rotate around the origin and therefore the
phase can be defined as3
ẏ
φ = arctan , (4.26)
ẋ
while the instantaneous frequency is:
ẏẍ − ÿ ẋ
ν= . (4.27)
ẋ2 + ẏ 2
We then use again two criteria to detect the existence of CPS4 :
1. Locking of the mean frequencies
Ω1 = ν1 = Ω2 = ν2 , (4.28)
2. Bounded phase difference
|φ2 (t) − φ1 (t)| ≤ const. (4.29)
To illustrate the transition to PS for different values of the parameter a,

we inspect the change of the mean frequency ratio Ω2 /Ω1 , the phase difference
and the spectrum of the LEs vs. the coupling strength d.
Routes to Chaotic Phase Synchronization
In dependence on the coherence of chaotic attractors three regimes of quali-

tatively different transitions to CPS can be clearly distinguished in the phase
diagram (Fig. 4.7).
3
We have to note that for the funnel chaotic attractors the coupling may change
their topology. As a result the strong cyclic structure of orbits projection in the
(ẋ, ẏ)-plane may be destroyed and phase measurement (4.26) fails occasionally
for intermediate values of coupling. But for small coupling and for coupling near
the transition to CPS the phase is well defined by (4.26). See also [169].
4
Note that phase and frequency locking are not always identical in chaotic and
stochastic systems.
0.20
l1
0.15 l2
l3
0.10
d
0.05
0.00
0.15 0.20 0.25 0.30
a
Fig. 4.7. Critical coupling curves of the Rössler systems (4.25). l1 corresponds to
the onset of CPS, i.e., below this line the oscillations are not synchronized, and above
both the phase- (4.29) and frequency- (4.28) locking conditions are fulfilled; l2 to the
transition of one of zero LEs to negative values and l3 to zero crossing of one of the
positive LEs. Note: In this figure we do not separate cases, where synchronization
occurs between regular and chaotic oscillations
Strong Coherence
Let us start with the simplest case in the interval [0.15; 0.186], where both
oscillators have phase-coherent chaotic attractors (see Figs. 2.3a and 2.4a, d).
For these attractors the zero LEs are associated with the phase dynamics.
Due to the high degree of coherence of the motions, i.e., a very small diffusion
constant Dφ (see Fig. 2.3), phase and frequency locking occur shortly after the
transition of one of the zero LEs to a negative value. Note that the two largest
LEs remain positive, i.e., hyperchaos remains. Hence, the amplitudes of the
oscillators are only weakly correlated. A strong correlation of the amplitudes
sets in only at a much larger coupling (d > dl3 ) where one of the positive
LEs becomes negative, and the two systems achieve a special type of almost
complete synchronization (CS).5 Between phase and complete synchronization
usually it is possible to observe lag synchronization [166], where the motions
of two chaotic systems are nearly identical, but one oscillator lags in time to
the other.
PS of phase-coherent chaotic Rössler oscillators can be better understood
by separating the original system into the dynamics of amplitude and phase.
By introducing6
φ = arctan(y/x), ρ = (x2 + y 2 )1/2 , (4.30)
5
Complete synchronization (x1 (t) = x2 (t), y1 (t) = y2 (t), z1 (t) = z2 (t)) is impossi-
ble because the oscillators are nonidentical. But in the following for the shortness
we will call this regime “complete synchronization.” In this regime with increase
of coupling the states of two (or more) systems becomes closer and closer.
6
This approach is valid only for the phase-coherent case.
we get
ρ̇1,2 = aρ1,2 sin2 φ1,2 − z1,2 cos φ1,2 +

d(ρ2,1 cos φ2,1 cos φ1,2 − ρ1,2 cos2 φ1,2 ),
φ̇1,2 = ω1,2 + a sin φ1,2 cos φ1,2 + z1,2 /ρ1,2 sin φ1,2 −
d(ρ2,1 /ρ1,2 cos φ2,1 sin φ1,2 − cos φ1,2 sin φ1,2 ),
ż1,2 = f − cz1,2 + ρ1,2 z1,2 cos φ1,2 (4.31)
In this system phases are the slowest variables in comparison to the other
variables ρ1,2 and z1,2 . Therefore we can use averaging over rotations of the
phases. Introducing the phase difference θ = ψ1 − ψ2 of the “slow” phases
ψ1,2 according to φ1,2 = ω0 t + ψ1,2 , the frequency difference Δ = ω1 − ω2 and
averaging the equations for them, one gets
dθ d ρ2 ρ1
= Δ − ( + ) sin θ. (4.32)
dt 2 ρ1 ρ2
When we neglect the fluctuations of the amplitudes, (4.32) is reduced to the
Adler equation which has the stable steady-state solution
2Δρ1 ρ2
θ = arcsin (4.33)
d(ρ21 + ρ22 )
when the coupling strength d is larger than the critical value
dps = 2Δρ1 ρ2 /(ρ21 + ρ22 ). (4.34)
For the case of slightly different oscillators, i.e., ρ1 ≈ ρ2 ,
dps = Δ (4.35)
dps is then the onset of PS for phase-coherent attractors. This makes clear that
CPS is very similar to the classical case of phase synchronization of coupled
periodic oscillators (see Sect. 4.1).
Intermediate Coherence
A quite different route to CPS takes place for a ∈ [0.195, 0.25] where both
chaotic attractors are of the funnel type (see Figs. 2.3b and 2.4b, e). Here
the curves l1 and l2 in Fig. 4.7 are clearly separated, but both lie below the
curve l3 . Hence, the two largest LEs remain positive during the transition
to CPS, which means that no bifurcation of the hyperchaotic attractor can
be associated to the locking of the phases. CPS occurs here via a crisis-like
transition inside the hyperchaos, i.e., via an interior crises – strong widening
(or narrowing) of a preexisting chaotic attractor [70] of the chaotic set. Such
a change of the internal structure of the attractor is seen by the projection
of the phase trajectory on the plane (φ1 , φ2 ) for d outside (Fig. 4.8a) and
φ1 φ1
Fig. 4.8. Projections of trajectories of the Rössler systems (4.25) for a = 0.22 on
the plane (φ1 , φ2 ) for the coupling strength d outside (a) (d = 0.055) and within
(b) (d = 0.075) the synchronization region
within (Fig. 4.8b) the synchronized regime. In the case of a nonsynchronized

state (but near the transition point) the trajectory covers practically the whole
plane (φ1 , φ2 ) with different density. The dense band corresponds to relatively
long epochs of synchronized motion. The presence of a trajectory in the rest
of the plane indicates the existence of phase slips. In the synchronized state
the plot is restricted to narrow bands.
Strong Noncoherence
In the interval a ∈ [0.25, 0.3], the curve l1 lies above the curve l3 , showing
that CPS occurs only after one of the positive LEs passes to a negative value,
i.e., the transition to CS. It is important to note that, by CS, both oscillators
have established a rather strong cross-correlation. However, such a strong
relationship is an average property over the whole attractors, while locally,
phase slips associated to different number of oscillations in the two oscillators
in a period of time may occur for coupling strengths d shortly after l3 , as
seen in Fig. 4.9 as the typical behavior for d ∈ [dl3 , dl1 ]. CPS appears now
as a manifestation of CS. This property is in contrast to the above regimes,
where CPS is a much weaker degree of synchronization compared to CS, and
the phases become locked before a strong correlation of the amplitudes is
established. Thus, for highly noncoherent oscillations due to the existence of
two distinct characteristic time scales, a rather strong coupling is necessary
to keep both oscillators in small or larger cycles simultaneously in order to
maintain the phase locking. Otherwise, a phase slip develops quickly due to the
very different time scales when the two oscillators are on a small and a large
10
(a)
6
φ2−φ1
−2
20
(b)
10
dy1/dt, dy2/dt
−10
−20
27100 27150 27200 27250 27300 27350
time
Fig. 4.9. CPS in the strong noncoherence regime. (a) Time evolution of phase
difference. (b) Variables ẏ1,2 in system (4.25) for a = 0.2925 and d = 0.179 ∈
[dl3 , dl1 ]. Solid and dotted lines correspond to the first and the second oscillator,
respectively. In the time interval between dashed lines the first oscillator produces
four rotations in the (ẋ1 , ẏ1 )-plane around the origin, but the second one generates
only three rotations, which leads to a phase slip in (a)
cycle, separately. Consequently, phase locking becomes impossible without a

strong correlation in the amplitudes.
Scaling Properties of Synchronization Transitions
To get a deeper insight into the routes from nonsynchronized to synchronized

state, we study the scaling properties of intermittent phase slips near the
CPS transition point. Very close to the transition point, phase slips occur
very rarely and the average time interval τ between two successive slips is
superlong and can be described by τ ≈ exp(k|d − dl1 |−1/2 ) [117, 118, 170,
171]. But away from the critical point, quite different scaling properties of
the transitions to CPS are observed for phase-coherent and funnel attractors
(Fig. 4.10). For phase-coherent attractors, due to the existence of only one
well-pronounced time scale, a synchronous state is achieved through a smooth
frequency locking, and the behavior can be described by a type-I intermittency
scaling law. While for the funnel attractors, there exist small and large loops of
1.08
a=0.16
1.05
a=0.22
Ω2/Ω1
a=0.28
1.02
0.99
0.00 0.05 0.10 0.15 0.20
d
Fig. 4.10. Mean frequency ratio Ω2 /Ω1 vs. coupling d for the Rössler systems (4.25).
The fitting curve (∼ |d − dl1 |0.5 ) for a = 0.16 (dashed line) and fitting straight lines
for a = 0.22 (solid line) and a = 0.28 (dotted line) are presented
the trajectory rotations. When the coupling is weak, it appears very often that
the first oscillator is on the small loops, while the second one is on the large
ones, or vice versa. This leads to an unpredictable, oscillatory-like evolution of
the Ω2 /Ω1 -ratio. Relatively close to the critical point, rather strong coupling
makes both oscillators stay simultaneously on small or large loops for a long
period of time. The Ω2 /Ω1 -ratio decreases monotonous and can be fitted by
a straight line in a large range below the critical point.
The difference in the transitions to CPS may also be understood by means
of UPOs [119]. It was shown that CPS takes place in a parameter region where
all pairs of UPOs, embedded in the chaotic attractors, are synchronized. For
phase-coherent attractors the period of all UPOs are close to each other (see
Fig. 2.3g) and the boundaries of the synchronization regions in the form of
Arnold tongues corresponding to the locking of different pairs of UPOs lie in
a relatively narrow region. The approach to CPS, thus, is associated to an
effective saddle-node bifurcation with small noise and a type-I intermittency
occurs, resulting in a relatively hard transition to CPS. For the funnel chaotic
attractors the distribution of the mean periods of UPOs is rather broad (see
Fig. 2.3h, i). Therefore, the coupling strengths corresponding to the onset of
synchronization of different pairs of UPOs are distributed in a rather large
interval. With an increase of d, the system crosses continuously a series of well-
separated Arnold tongues, resulting in a much slower convergence to CPS. The
scaling lows presented in Fig. 4.10 support these treatment.
Phase Diagram of Possible Regimes
The phase diagram of different regimes observed in the system of coupled

phase-coherent (a = 0.16) Rössler attractors for varying the coupling d and
frequency mismatch Δ exhibits three regions of qualitatively different behavior
(Fig. 4.11) (compare with the phase diagram for coupled regular oscillators in
Fig. 4.2):
0.4
OD
0.3
NS
Δ
0.2
S
0.1
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
d
Fig. 4.11. The regions of nonsynchronous (NS) and synchronous (S) motion and
absence of oscillations (oscillation death) (OD) in system (4.25). The diagram is
approximate, e.g., windows of periodic behavior in regions I and II are not shown.
Compare with the phase diagram in Fig. 4.2 for the van der Pol oscillators
– The synchronization region S, where the conditions of frequency and phase

locking are fulfilled. It is important to note that there is no threshold of
synchronization; this is a particular feature of the highly phase-coherent
Rössler attractor.
– The region of nonsynchronized (NS) oscillations, where |Ω1 −Ω2 | = Ωb > 0.
In analogy to the case of periodic oscillators, this frequency Ωb can be
interpreted as beating frequency.
– The region where the interaction results in a suppression of self-oscillations
(OD) in both systems. This effect of oscillation death is similar to the
observed one for coupled periodic oscillators (see Sect. 4.1.2).
Note that the boundaries between these different regimes are slightly dis-
persed, and windows of periodical behavior are present, although for a large
domain of parameters in the regimes I and II the amplitudes of the oscillations
are chaotic.
Next, we discuss the behavior of the beating frequency Ωb , when the fre-
quency mismatch Δ changes (Fig. 4.12). We see that for weak coupling (small
d), the beating frequency Ωb smoothly depends on Δ. This means that a
regime, where the frequencies of interacting oscillators differ by a rather small
value, is possible. For sufficiently strong coupling the transition from synchro-
nous to a nonsynchronous state is rather sharp: a virtual jump in dependence
0.20
d=0.002
0.05
0.1
0.15 0.14
Ωb
0.10
0.05
0.00
0.00 0.05 0.10 0.15 0.20
Δ
Fig. 4.12. Dependence of the beating frequency Ωb on the frequency mismatch Δ
for different values of coupling d in (4.25). In full analogy to the classical case of
two coupled periodic oscillators, the transition to (or from) synchronization regime
occurs smoothly or practically by a jump for weak and strong coupling, respectively
of Ωb on Δ is observed. Thus the frequencies of the interacting oscillators

cannot be close: they either coincide or differ by a finite value. These two
types of transitions to (or from) the synchronization regime are very similar
to those for periodic oscillators.
Hence, we can conclude that the phase dynamics of chaotic phase-coherent
Rössler oscillators is similar to that of the well-studied classical systems, e.g.,
the van der Pol oscillator (see Sect. 4.1.2).
Conclusions
There are three types of transition to CPS in dependence on the coherence

properties of motions quantified by diffusion of the phase which should be
general for chaotic oscillators where a phase can be properly defined.
For small diffusion, the onset of CPS is accompanied by the transition of
one of the zero LEs to negative values. If the diffusion of the phase is rather
strong, phase locking only occurs after the onset of almost CS (one positive
LE becomes negative). For intermediate diffusion CPS sets in via an inverse
interior-like crises of the hyperchaotic attractor.
4.2.2 Synchronization of Coupled Intermittent Oscillators
In Sect. 3.5 the effect of external CPS of intermittent chaotic oscillators due
to a periodic driving was demonstrated. As the examples of such oscillators
the Lorenz system and quadratic map have been used. Now we will show that
the mutual synchronization of such systems is possible as well. Because both
mentioned systems have their well-pronounced characteristic time scale, with
increase of coupling, it is possible to expect the coincidence of averaged length
of laminar stages, which means the existence of synchronous regimes.
Synchronization of Maps
Let us consider the system of two coupled nonidentical maps generating

chaotic intermittent behavior (for individual dynamics see Sect. 2.4):

xn+1 = f1 (xn ) + d(yn − xn ),
(4.36)
yn+1 = f2 (yn ) + d(xn − yn )
where

ε1,2 + x + x2 , if x ≤ 0.2,
f1,2 (x) = (4.37)
g(x − 0.2) − ε1,2 − 0.24, if x > 0.2
For this system of coupled maps for different ε1,2 the transition to synchro-
nous regime is presented in Fig. 4.13. This regime appears then the average
lengths of the laminar stages in both maps becomes equal. Time series of the
synchronized maps are shown in Fig. 4.14. One can see that xn and yn are
shifted practically on the constant time value.
Synchronization of the Intermittent Lorenz systems
Coupled Lorenz systems are described by the model:
ẋ1 = σ(y1 − x1 ),
ẏ1 = r1 x1 − y1 − x1 z1 + d(y2 − y1 ),
ż1 = −bz1 + x1 y1 ,
(4.38)
ẋ2 = σ(y2 − x2 ),
ẏ2 = r2 x2 − y1 − x2 z2 + d(y1 − y2 ),
ż2 = −bz2 + x2 y2 ,
where b = − 83 , σ = 10, r1 = 166.1, r2 = 166.12. Let us remind that for the

calculation of the laminar stage durations, we calculate the sequences {yn }
corresponding to intersections of the trajectory with the plane {x = 0, ẋ > 0}
and compare each value with the correspondent fixed point in the autonomous
system on the edge of the tangent bifurcation. In the computations of τ ,
1580
ε1=4*10−6
ε2=4.5*10−6
1560
1540
<τ 1,2>
1520
1500
1480
0 0.0001 0.0002
d
Fig. 4.13. Synchronization of intermittent maps (4.36) and (4.37) at g = 2 and

different ε1,2 . Average lengths of the laminar stage < τ1,2 > in dependence on the
coupling
0.25
0.20
0.15
0.10
0.05
xn, yn
0.00
−0.05
−0.10
−0.15
−0.20
−0.25
50000 55000 60000 65000 70000
time
Fig. 4.14. Time series for synchronized maps (4.36) and (4.37). xn (solid line) and
yn (dashed line). Parameters are: g = 2, ε1 = 4 ∗ 10−6 , ε2 = 4.5 ∗ 10−6 , d = 0.00015
a discrete time is used (one unit corresponds to the continuous-time interval

between intersections with the selected plane). In Fig. 4.15 results of numerical
simulations are presented. One can see, that with increasing of coupling the
averaged durations of laminar stages in both subsystems becomes closer. There
is some interval of coupling where the chaotic intermittency disappears. Then
(dcr ≈ 0.01) CPS appears. At that two LEs in (4.38) are positive.
0.16
1/t1
1/t2
0.14
0.12
0.1
1/ti
0.08
0.06
0.04
0.02
0
10−4 10−3 10−2 10−1 100 101
d
Fig. 4.15. Synchronization of intermittent chaotic Lorenz systems (4.38) for

b = − 83 , σ = 10, r1 = 166.1, r2 = 166.12. The dependence of the inverse averaged
laminar stage durations on the coupling is plotted. First two systems are nonsyn-
chronized. Then (d ∈ [6 ∗ 10−3 : 10−2 ]) intermittency disappears. For d ≥ 10−2 both
systems are synchronized in phases
4.2.3 Oscillatory and Rotatory Synchronization of Chaotic

Phase Systems
Here we explore synchronization phenomena for chaotic phase systems. The
dynamics of two coupled chaotic phase systems is governed by the following
equations:
φ̇1,2 = y1,2 ,
ẏ1,2 = z1,2 ,
μ1,2 ż1,2 = γ1,2 − sin φ1,2 − y1,2 − εz1,2 + d1 (y2,1 − y1,2 ) + d2 (z2,1 − z1,2 ),
(4.39)
where d1,2 are the couplings coefficients.
As mentioned in Sect. 2.5.3, the uncoupled system can demonstrate three
types of chaotic behavior in dependence on the parameter values ω, ε, and μ
[51] (1) rotations (Fig. 4.16a), (2) oscillations (Fig. 4.17a), and (3) oscillations–
rotations (Fig. 4.19a). We analyze synchronization phenomena for all these
types of chaotic dynamics [172].
The synchronization of coupled chaotic oscillators and the synchronization
of coupled chaotic phase systems have much in common. But there are also
2
(a)
0.02 (b)
1.5
λ
y
0
0.5
0
−3 −1.5 0 1.5 3 0 0.005 0.01 0.015
φ d1
300 (d)
1
d1=0.0065 (c)
200
d1=0.007
φ2−φ1
Ω1/Ω2
100
d1=0.0072
0.995
d1=0.008
0
0 50000 100000 150000 200000 0 0.005 0.01 0.015
time d1
Fig. 4.16. Synchronization of rotatory phase variables. (a) Projections of the typical
rotatory trajectory of the uncoupled system in (4.39) on the plane (φ, y). Parameters
are: γ = 0.645, μ = 3.0. In Fig. 4.16b–d parameters are: γ1 = 0.645, γ2 = 0.667,
μ = 3.0, and d2 = 0. (b) The four largest LEs, one of which is always zero.
(c) Difference of phase variables φ2 − φ1 for nonsynchronous (d1 = 0.0065; 0.007;
0.0072) and synchronous (d1 = 0.008) regimes. (d) The mean frequency ratio Ω1 /Ω2
vs. coupling
interesting differences due to the special nature of phase systems. Especially

the following two properties of an uncoupled chaotic phase system are im-
portant for the study of peculiar synchronization processes in coupled phase
systems (1) one of the variables is already the phase variable, i.e., we have
not to retrieve phase dynamics here and (2) chaos possesses a zero LE, i.e.,
in the chaotic parameter regime, the dynamics has a zero LE. Due to the
first property, we can distinguish two types of CPS (1) “real” chaotic phase
synchronization (RCPS) and (2) “complete” (almost) chaotic phase synchro-
nization (CCPS). In the case of RCPS the well-known conditions of phase and
frequency locking ((4.29) and (4.28)) of two coupled systems should be fulfilled
and we have hyperchaos there, i.e., two positive LEs exist. In the case of CCPS
only one LE remains positive, although the phase- and the frequency-locking
conditions are fulfilled. There is another type of synchronization, complete
(almost) chaotic synchronization (CS), at which only one LE is positive but
phase and/or frequency locking does not exist. It is important to remain that
the negativity of the LEs is only a necessary condition for the stability of the
1 0.04
(a)
(b)
0.03
0.5
0.02
λ
y
0.01
0
0
−0.5 −0.01
0 1 2 3 0 0.005 0.01 0.015
φ d
600 1.01
(c) (d)
d=0.006
400
d=0.007
Ω1/Ω2
ψ1−ψ2
200
d=0.008
0
d=0.009 1
0 50000 100000 150000 200000 0 0.005 0.01 0.015
time d
Fig. 4.17. Synchronization of oscillatory phase variables. (a) Projections of a typical

oscillatory trajectory of the uncoupled system in (4.39) on the plane (φ, y) for the
parameters: γ = 0.83, μ = 3.3. In (b–d) the parameters are: γ1 = 0.815, γ2 = 0.83,
and μ1,2 = 3.3. (b) The four largest LEs. (c) Difference of the phase variables ψ1 −ψ2
of the y- and z-coupled subsystems in (4.25) (d1 = d2 = d) for nonsynchronous
(d = 0.006; 0.007), nearly synchronous (d = 0.008), and synchronous (d = 0.009)
regimes. (d) The mean frequency ratio Ω1 /Ω2 vs. coupling
synchronous state. But usually [173, 174] the transition to CS is very close to
the point when one of the positive LEs becomes negative. Therefore, we will
infer the onset of CS from this property.
Due to the second property, i.e., chaos in continuous in time system pos-
sesses a zero LE, there are many properties in common between phase synchro-
nization of autonomous chaotic oscillators (see previous sections) and phase
synchronization of autonomous chaotic phase systems.
As for periodic synchronization, the appearance of CPS is affected by the
frequency mismatch of the coupled subsystems and by the coherence property
of the motions. We will characterize this coherence, i.e., the diffusion of the
phase variables, by their variances Dφ1,2 that are defined for large times as
(φ̇1,2 − φ̇1,2 )2 = Dφ1,2 . (4.40)
We will show below that these Dφ1,2 of both coupled subsystems, as well as
their frequency mismatch, play the crucial role in the transitions to phase
synchronization regime.
Phase Synchronization of Rotatory Phase Variables
In this case the phase variables φ1,2 unboundedly increase and φ̇1,2 are always
(or almost always) positive. A projection of the chaotic phase rotating trajec-
tory on the plane (φ, y) (Fig. 4.16a) looks like a “smeared” periodic trajectory
with a monotonously (or almost always) increasing phase. Therefore, PS of
chaotic rotations is quite similar to periodic synchronization, i.e., in both
cases only the phase growth rate is important. The averaged growth rate of
the phases or the mean frequency of rotations can be defined as
Ω = φ̇ = y . (4.41)
In order to test for the existence of phase synchronization regime, we again

use two criteria. A chaotic 1:1 synchronization of the rotations occurs if the
mean frequencies characterizing the long-time scale behavior of the coupled
systems become equal:
Ω1 = Ω2 . (4.42)
On the short-time scale, i.e., inside the interval [−π; π], due to the high dif-
fusion of the phases, the transient phase differences can be rather large. The
second criterion we apply is the phase locking criterion:
|φ2 (t) − φ1 (t)| ≤ const (4.43)
that ignores the short-time scale behavior as well. Phase synchronization

regime according to criteria (4.42) and (4.43) can be observed for systems,
where the evolution of the phase variables behaves as an alternation of large
intervals, where the phase variable increases, with relatively small intervals,
where the phase variable decreases. We will demonstrate the existence of both
types of phase synchronization; RCPS and CCPS, namely for such type of
behavior.
These criteria are tested numerically. We set γ1 = 0.645, γ2 = 0.667,
μ1,2 = 3.0, and d2 = 0. For these parameters the diffusion of the phases is
relatively large in both systems (Dφ1 ≈ 0.219, Dφ2 ≈ 0.216), which affects
the occurrence of phase synchronization regime. To illustrate the correspond-
ing transition to PS, we plot the four largest LEs (Fig. 4.16b) and the mean
frequency ratio (Fig. 4.16d) vs. coupling, as well as the difference between
the phase variables φ1 − φ2 for different coupling strengths (Fig. 4.16c). One
can see that the RCPS occurs at d11 ≈ 0.0076 (Fig. 4.16d). For d1 > d11 the
frequency- and the phase locking conditions (4.42) and (4.43) are satisfied,
but the attractor is still hyperchaotic for a rather large interval.
As it was shown for phase-coherent attractors, phase synchronization
regime sets in approximately at that value of coupling, when one of the zero
LEs becomes negative. In our simulations we find (Fig. 4.16b) that one of the
zero LEs becomes negative already at d1 ≈ 0.003. But the transition to RCPS
in system (4.25) occurs for essentially larger coupling. The main reason for
this large difference is that PS takes place via a crisis transition of the struc-
ture of the hyperchaotic attractor, i.e., via an interior crises of the chaotic set.
The corresponding phase portraits look very similar to the phase portraits in
Fig. 4.8 obtained for coupled chaotic Rössler oscillators.
At larger coupling (d21 ≈ 0.0118), where one of the positive LEs becomes
negative, CCPS occurs. Due to the relatively high noncoherence properties,
the interval of values of coupling between the transitions to RCPS and to
CCPS L = [d11 ; d21 ] is small. As our numerical simulations show, the increase
of the parameters γ1,2 leads to a complication of the topological structure
of the chaotic attractors. The intervals, where the phase variables decrease,
become larger and the behavior transfers from a rotational to an oscillation-
rotational one. This leads to an increase of the noncoherence properties of
motion (diffusion of the phase variable increases) and as a result the width of
the interval L between RCPS and CCPS tends to zero. The reason for that
is the following. CPS is quite similar to the synchronization of periodic oscil-
lations in the presence of small noise [1, 163]. When noise increases, a larger
coupling is needed to achieve phase locking. By analogy, in order to suppress
large phase fluctuations by CPS, a stronger coupling has to be applied.
Phase Synchronization of Oscillatory Phase Variables
In this case in both subsystems in (4.39) the phase variable oscillates around
some constant value, i.e., φ1,2 are bounded (Fig. 4.17a). Synchronization of
such oscillatory phase variables is quite similar to the case of usual phase
synchronization of chaotic oscillators (see Sect. 4.2.1). Because of the simple
topology of the chaotic attractor, we can introduce a new “artificial” phase:
y
ψ = arctan , (4.44)
φ − arcsin γ
a new amplitude
A = ((φ − arcsin γ)2 + y 2 )1/2 (4.45)
and this yields the mean frequency
ψ(T ) − ψ(0)
Ω = ψ̇ = lim . (4.46)
T →∞ T
Here the conditions (4.42) and (4.43) applied to the new phase variables
ψ1,2 and the mean frequencies Ω1,2 can be used as criteria for synchroniza-
tion regime. Therefore, although the oscillatory and rotatory cases cannot be
generally reduced one to the other, two similar criteria for the existence of
phase synchronization regime can be used and as we will show, many sim-
ilar effects happen. For the chosen parameters: γ1 = 0.815, γ2 = 0.83, and
μ1,2 = 3.3, the coherence of the motions is significantly stronger compare to
the rotatory case (Dψ1 ≈ 0.075, Dψ2 ≈ 0.079). We consider y- and z-coupling
(in (4.39) d1 = d2 = d). As in the case of phase synchronization of the
rotatory phase variables, we compute the Lyapunov spectrum (Fig. 4.17b),

the frequency ratio (Fig. 4.17d), and the evolution of the phase variable dif-
ference (Fig. 4.17c). For the oscillatory phase variables both types of phase
synchronization, RCPS and CCPS, are found. With an increase of the cou-
pling the frequencies ratio ρ = Ω1 /Ω2 decreases to 1 smoothly (without any
jump), i.e., the transition to RCPS is a soft one. This is manifested in the
evolution of the phase variable difference, namely for a coupling close to the
critical value d1 = 0.0082 phase locking at large time intervals is observed
(Fig. 4.17c). Due to the high coherence of motions, i.e., small phase diffusion,
phase locking, and frequency entrainment occur approximately (shortly after)
at the same value of coupling for which one of the zero LEs becomes negative.
It should be mentioned that during the transition to synchronization regime of
the “artificial” phases ψ1,2 the new amplitudes A1,2 as well as the real phases
φ1,2 remain highly uncorrelated (Fig. 4.18). But some frequency entrainment
sets in, because the averaged number of oscillations per unit time, computed
easily as the number of maxima, coincides for both phases for d > d1 .
At essentially larger coupling (d2 = 0.043), CCPS and as a consequence a
strong correlation among all variables arises. The interval L = [d11 ; d21 ] between
the transitions to RCPS and to CCPS is now relatively large. As in the case
2.5 2.5
(a) (b)
2 2
1.5 1.5
φ2
φ2
1 1
0.5 0.5
0 0
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5
φ1 φ1
Fig. 4.18. Projections of the trajectories of the system (4.39) on the plane (φ1 , φ2 )
outside of the synchronization region – (a) (d = 0.008) and within the synchro-
nization region – (b) (d = 0.009). The parameters are: γ1 = 0.815, γ2 = 0.83, and
μ1,2 = 3.3
of rotatory synchronization, we observe that when the noncoherence increases

with an increase of the parameters γ1,2 , the interval L becomes smaller and
tends to zero.
Phase Synchronization of Oscillatory–Rotatory Phase Variables
A quite different situation happens in the case of oscillatory–rotatory behavior

of the phase variables (Fig. 4.19a). The existence of phase synchronization
regime is in general a nontrivial effect, because the phase variables φ1,2
increase nonmonotonously. Their evolution is an alternation between time
intervals, where the phase variable increases, and time intervals, where the
phase variable decreases. Due to the similar lengths of both types of intervals,
in this case it is impossible to separate the evolution of the phase variables
into two different time scales. In order to achieve synchronization regime, it
is obviously necessary to have synchronization of both subtypes of behavior:
2
(a)
0.02 (b)
1
0
0
λ
y
−0.02
−1
−0.04
−2 −0.06
−3 −1.5 0 1.5 3 0 0.005 0.01 0.015
φ d
300
(c) (d)
d=0.007 1.005
200
φ1−φ2
Ω1/Ω2
d=0.0078
100
d=0.008
1
0
0 50000 100000 150000 200000 0 0.005 0.01 0.015
time d
Fig. 4.19. Synchronization of oscillatory–rotatory phase variables. CCPS and CS

occur practically simultaneously at d ≈ 0.0082. (a) Projections of the trajectory
of an uncoupled system in (4.39) on the plane (φ, y). Parameters are: γ = 0.34,
μ = 5.0. In Fig. 4.19(b–d) parameters are: γ1 = 0.34, γ2 = 0.37, and μ1,2 = 5.0.
y- and z-coupled subsystems (in (4.39) d1 = d2 = d). (b) The four largest LEs.
(c) Difference of phase variables φ1 − φ2 for nonsynchronous (d = 0.007; 0.0078;
0.008) and synchronous (d = 0.0085) regimes. (d) The mean frequency ratio Ω1 /Ω2
vs. coupling
rotations as well as oscillations. As our numerical simulations show, the occur-

rence of RCPS is possible only for a very small parameter mismatch between
both subsystems in (4.39). If the parameters mismatch is large enough CCPS
and CS set in almost simultaneously (Fig. 4.19) or CS occurs even before
CCPS (Fig. 4.20).
Let us first consider the case when CCPS and CS are achieved almost at
the same critical coupling. We chose the parameters (γ1 = 0.34, γ2 = 0.37, and
μ1,2 = 5.0) in such a way that the noncoherence of motions in both subsystems
in (4.39) is very high. So we have Dφ1 ≈ 0.94 and Dφ2 ≈ 1.084. Figure 4.19
indicate that the transitions to CCPS and CS occur almost simultaneously at
d ≈ 0.0082, because both criteria are fulfilled.
On the contrary to the presented examples, where with an increase of
the coupling phase synchronization regime sets in before or simultaneously
with complete synchronization regime, we will show the possibility that phase
synchronization regime emerges clearly after the complete one [181]. We
take a relatively large parameters mismatch (γ1 = 0.34, γ2 = 0.39, and
μ1,2 = 5.0). One of the positive LEs λ2 becomes negative at d ≈ 0.0046,
i.e., complete synchronization regime sets in (Fig. 4.20). But the conditions
(4.42) and (4.43) for frequency and phase locking are fulfilled only beyond
d ≈ 0.012. Therefore complete synchronization is much weaker than phase
0.02
Ω1−Ω2, λ
0.01
−0.01
0 0.005 0.01 0.015
d
Fig. 4.20. Synchronization of oscillatory–rotatory phase variables. CS occurs before
CCPS. The parameters are: γ1 = 0.34, γ2 = 0.39, μ1,2 = 5.0, and d1 = d2 = d in
(4.39). The three largest LEs and the mean frequency difference Ω1 − Ω2 (circles)
vs. coupling are given
synchronization in this case. The LE λ2 demonstrates an interesting feature.

It increases rapidly and almost jumps to zero (but does not reached it), if
the coupling is close to the critical value d corresponding to the transition
to CCPS.
We have to note that if the noncoherence properties are very large, phase
synchronization regime cannot be achieved for any coupling.
Hard and Soft Transitions to Phase Synchronization
We have reported that phase synchronization regime of two coupled phase

systems (4.39) can appear or vanish in two ways: soft and hard. The soft
transition described in many examples in the previous sections is characterized
through a smooth locking of the observed frequencies. Also the topological
changes in the phase space appear smoothly. But for the hard transition to
phase and frequency locking quite another situation takes place, as illustrated
in Fig. 4.21.
0.03 1.01
(a) (b)
0.02
Ω2/Ω1
0.01
λ
−0.01
1
−0.02
0 0.005 0.01 0.015 0 0.005 0.01 0.015
d1 d1
(c)
d1=0.0
1000
d1=0.008
φ2−φ1
500
d1=0.0084
d1=0.0088
0
0 50000 100000 150000 20000
time
Fig. 4.21. Hard transition to RCPS in (4.39). The parameters are: γ1 = 0.645,
γ2 = 0.636, μ1 = 3.0, μ2 = 3.05, and d2 = 0. (a) The four largest LEs. (b) The mean
frequency ratio Ω1 /Ω2 vs. coupling. (c) Difference of the phase variables φ2 − φ1 for
nonsynchronous (d1 = 0.0; 0.008; 0.0084) and synchronous (d1 = 0.0088) regimes.
The diffusion constants are Dφ1 ≈ 0.219 and Dφ2 ≈ 0.218
Fig. 4.22. Projections of the trajectories of system (4.39) on the planes (φ1 , φ2 ) ((a)
and (b)) and (y1 , y2 ) ((c) and (d)) for γ1 = 0.645, γ2 = 0.636, μ1 = 3.0, μ2 = 3.05,
and d2 = 0 outside of the synchronization region – ((a) and (c)) (d1 = 0.0084) and
within the synchronization region – ((b) and (d)) (d1 = 0.0088)
The relatively large jump in the mean frequency ratio ρ = Ω2 /Ω1 from a
nonsynchronous (ρ = 1) to a synchronous (ρ = 1) hyperchaotic behavior at
d1 = 0.0088 can be regarded as a manifestation of a hard transition to phase
synchronization regime. Indeed, for very small changes in the coupling, strong
changes in the phase difference evolution (Fig. 4.21c) and in the phase portrait
(Fig. 4.22) are observed. For d1 = 0.0084, i.e., when d1 is very close to the crit-
ical value d11 , only very short intervals of synchronized epochs are observed in
the phase difference (compare with Figs. 4.16c, 4.17c, and 4.19c) that demon-
strate phase differences for the oscillatory case where the transition to PS is
hard. The projections of the hyperchaotic attractor on the planes (φ1 , φ2 ) and
(y1 , y2 ) slightly before and after the transition to phase synchronization regime
are presented in Fig. 4.22. For the synchronous regime the chaotic trajectory
lies within relatively narrow bands in the phase space (Fig. 4.22b, d), while
when synchronization regime is lost these bands smear and merge together
(Fig. 4.22a, c). Such a hard transition to a band-structured attractor can be
explained in terms of unstable periodic orbits. In [117] it was shown and in
previous sections described in detail that CPS regime occurs in that parameter
region, where all unstable periodic orbits embedded in the chaotic attractors
are synchronized. For the presented case the hard transition to phase synchro-
nization regime is caused by the fact that boundaries of the Arnold tongues,
which correspond to the synchronization of unstable orbits are very close to

each other. Another interesting result similar to that presented in Fig. 4.20
can be seen in Fig. 4.21a. When the coupling increases, one of the zero LEs
initially remains equal to zero, then it becomes negative and jumps back to
zero, without reaching it. This happens when the coupling is close to the crit-
ical value d11 corresponding to the transition to RCPS, and then beyond d11
this LE decreases again.
Conclusions
Rotatory, oscillatory, and oscillatory–rotatory synchronization regimes can

occur in two coupled autonomous chaotic phase systems. Three types of syn-
chronization have been studied:
1. Real chaotic phase synchronization (RCPS), which occurs when two LEs
are positive and when both the frequency- and the phase locking condi-
tions are fulfilled
2. Complete chaotic phase synchronization (CCPS), which occurs when only
one LE is positive and when both the frequency- and the phase locking
conditions are fulfilled
3. Complete chaotic synchronization (CS), which occurs when only one LE
is positive and when the frequency- and the phase locking conditions are
both not fulfilled.
In dependence on the coherence properties of the motions, which can be

measured by the diffusion of the phase variable, we observe four transitions
to phase synchronization regime:
1. For small diffusion the onset of phase synchronization regime is accom-

panied by the change of the Lyapunov spectrum (one of the zero LEs
becomes negative shortly before the onset)
2. If the diffusion of the phase variable is strong, then phase and complete
synchronization regimes (one of the positive LEs becomes negative shortly
before) occur simultaneously
3. Complete (almost) synchronization regime even sets in before phase syn-
chronization one. In this case the behavior of individual elements is so
complex that in order to achieve phase synchronous state, the complete
synchronous state should be achieved first
4. For intermediate diffusion phase synchronization regime appears via an
interior crises of the hyperchaotic set
Thus topological (e.g., coherence) properties of the motion play a crucial

role for the type and appearance of chaotic phase and complete synchroniza-
tion regime.
The effect of synchronization of chaos realized in a system of two coupled
phase-locked loops (PLL) generating chaotic signals (described by (4.39)) can
be used in secure communication applications. Important properties of PLL,

such as high accuracy of synchronization, or the possibility of rather simple
control, make PLL very promising for data communication using not only
regular but chaotic signals as well [175–179]. Unidirectionally coupled chaotic
PLLs analogous to (4.39) have been considered in [180,182]. In [180] CPS and
in [182] almost complete chaotic synchronization are presented.
4.3 Synchronization of Coupled Circle Maps

So far we have presented continuous in time systems to demonstrate phase
synchronization. In this section conditions for an onset of PS in a system of two
coupled discrete in time models, namely, nonidentical circle maps (CMs) [183],
are studied. Here we should note that like in continuous in time phase system
for CM the variable φ should be considered not like a phase but like a phase
variable which can decrease as well increase in time.
A pair of symmetrically coupled CMs (for a single map see Sect. 2.5.4) is
governed by the two-dimensional system:
φk+1
1 = ω1 + φk1 − F (φk1 ) + d sin (φk2 − φk1 ),
(4.47)
φk+1
2 = ω2 + φk2 − F (φk2 ) + d sin (φk1 − φk2 ),
where ω1,2 ∈ [0; 2π] are positive parameters which can be interpreted as
frequency, F (φ) is a piece-wise linear 2π-periodic function of the form
F (φ) = cφ/π defined in the interval [−π, π], and c is the control para-
meter.
System (4.47) can be regarded as a model of a coupled partial digital phase-
locked loop (DPLL) connected in parallel by phase-mismatching signals.
As in the case of time continuous systems, effects of mutual regular and
chaotic phase synchronization of two coupled systems can be characterized by
the ratio of their individual rotation numbers ρ1,2 (see Sect. 2.5.4) or by the
winding number: w = ρ2 /ρ1 and by the evolution of phase differences of CMs
in time. In this connection we can again use two criteria to test for m : n
synchronization regime, where m and n are integers.
1. m : n PS of chaotic rotations between two CMs is defined as phase
entrainment or locking
| m φk1 − n φk2 | < Const, (4.48)
for all k = 1, 2, ....
2. Synchronization regime of rotations is analogous defined as the coincidence
of their rotation numbers:
m ρ1 = n ρ2 . (4.49)
It is important to emphasize that there is a remarkable difference in the
synchronization for regular and chaotic CMs.
4.3 Synchronization of Coupled Circle Maps 91
4.3.1 Regular Synchronization
First synchronization of two coupled CMs, in regular regimes is tested.
Coherent Case (c=0)
We begin our analysis with the simplest case of coupled shift maps (c = 0),
i.e., system (4.47) reads now:
φk+1
1 = ω1 + φk1 + d sin (φk2 − φk1 ),
(4.50)
k+1
φ2 = ω2 + φk2 + d sin (φk1 − φk2 ).
For this system the critical value of coupling d+ corresponding to the

transition from nonsynchronous to 1:1 synchronous rotations can be easily
found from the conditions of the existence and stability of a fixed point for
the sine CM – the discrete analog of the Adler equation:
θk+1 = Δ + θk − 2d sin θk , (4.51)
where θk = φk2 − φk1 and Δ = ω2 − ω1 . This critical

value is d∗∗ = Δ/2. The
fixed point θ̄ = arcsin(Δ/2d) is stable for d < 1 + Δ2 /4 and corresponds to
a 1:1 regular strict synchronization regime in the original model (4.50) with
a constant in time phase difference θ̄. With increasing of the coupling d, a
period doubling cascade takes place which ends up in a chaotic behavior of
the oscillatory type (i.e., θk is bounded). Note that the rotation numbers of the
coupled CMs coincide here, i.e., a nonstrict synchronous motion takes place.
But at some critical value d∗ , the oscillatory chaotic attractor is changed into
a rotationally chaotic one (i.e., θk is unbounded) which leads to the loss of
the synchronization regime in the two-element model (4.50). This is a first
example in the book showed that increase of coupling can lead to desynchro-
nization. In Chaps. 8 and 9 we will demonstrate such transition for the chains
of coupled maps. The transition from oscillatory to rotational behavior of the
phase difference is accompanied with a sudden change in the geometry of the
chaotic set, i.e., an interior crises [70]. At this transition the 1:1 synchroniza-
tion regime in the model (4.51) is violated. In Fig. 4.23 the dependencies of
the critical values d∗∗ and d∗ on the frequency mismatch Δ are presented. As
will be shown below, these both values can be usually regarded as the lower
and the upper estimates of the boundaries of the 1:1 synchronization region
for both regular and chaotic rotations.
Noncoherent Case (c = 0)
Now we analyze the synchronization properties of regular CMs at different

values of the coherence parameter c. As one can see from Fig. 2.12 for fixed
values ω1 and ω2 with increasing parameter c, the individual rotation numbers
2.5 dcr
*
d
2.0
1.5
Δ
1.0
1:1 synchronization
0.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5
d
Fig. 4.23. Critical values of coupling d∗∗ and d∗ corresponding (i) to the transition
from nonsynchronous to synchronous motion (left curve) and (ii) to the transition
from synchronous to nonsynchronous motion (right curve) in the model (4.50) vs.
the frequency mismatch Δ. Between both curves there is the region of 1:1 synchro-
nization
ρ1 and ρ2 , and therefore, the winding number w can be varied. Hence, we

will study the following two cases of frequency distribution. We choose the
individual frequencies ω1,2 in such a way that for all considered values of
c, the rotation numbers of the uncoupled CMs are (1) identical or (2) can
be different. The structure of the synchronization regions, their number and
transitions from one synchronous region to another one can be very rich now.
The dependence of w on the coupling parameter d for both cases (Fig. 4.24a,
b) exhibits clearly the existence of a lot of phase locking regions; the number
and width of them are increasing with increasing of c. So for c = 0 only the 1:1
synchronization region (Fig. 4.24a, b) exists. For large enough values of c, the
synchronization between coupled CMs occurs not only as 1 : 1 synchronization
but also as m : n synchronization (i.e., m, n = 1). The m : n synchronization
is typical for coupled relaxation periodic oscillators for which the motions
are noncoherent. For instance, in two coupled, strongly nonlinear van der Pol
oscillators, a synchronization regime occurs firstly as m : n synchronization
and only for larger coupling as 1 : 1 synchronization. As we can see from
Fig. 2.13 with increasing c, the variance D of the phase increases too, and as a
result of that the region of 1:1 self-synchronization becomes smaller for the first
case of frequency parameters (Fig. 4.24a). If c is larger than some critical value
c∗ , 1:1 synchronization is impossible because of the very strong noncoherence
of the rotations. In the second case, the size of the 1:1 synchronization region
1.5 0.7
(a) (b)
0.6
1.4
0.5
1.3
0.4
w
0.3
1.2
c=0,0 c=0.0
0.1 0.2
0.2 0.4
0.4 0.2
0.6
0.6 0.8
1.1 0.9 1.2
1.0 1.6
0.1
1.0 0.0
0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.4 0.5
d d
Fig. 4.24. The winding number w vs. the coupling coefficient d at different c and
ω1 = π/2, ω2 = 2π/3 (a), respectively, ω1 = 4π/7, ω2 = 17π/23 (b) in system (4.47)
can surprisingly increase or decrease. In dependence on the values of ω1 and

ω2 , the rotation number difference Δρ can increase or decrease. Therefore,
at fixed parameters ω1 and ω2 with increase of c, a synchronization regime
can occur sometimes at smaller coupling and sometimes at larger coupling
(Fig. 4.24b).
4.3.2 Chaotic Synchronization
We now carry out an analysis of synchronization for two coupled chaotic CMs
(c < 0). In this system some different synchronization properties are observed.
We perform numerical simulations for fixed ω1 = 0.6 and different values of ω2
(Fig. 4.25). Usually there exists only the region of 1:1 synchronization regime.
Only in rather small intervals of c, regions of m : n synchronization regime
can occur. It should be noted that in all our presented experiments only m : 1
synchronization regimes with different m = 2, 3, 47 are observed. Figure 4.25
indicates that the geometrical structure and the sizes of the synchronization
7
If we take larger value of ω2 , a synchronization regime with larger m takes place
too.
−3.0
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j)
−2.5 -
−2.0 -
−1.5 -
−1.0 -
−0.5 -
c=0 +
0. 1. 2.
d d d d d d d d d d
Fig. 4.25. Regions of chaotic phase synchronization for ω1 = 0.6 and different
values of ω2 : 0.8 (a), 1.0 (b), 1.2 (c), 1.4 (d), 1.6 (e), 1.8 (f ), 2.0 (g), 2.2 (h), 2.4 (i),
2.6 (j) in (4.47). The main gray regions correspond to 1:1 synchronization regime.
In columns (c–j) for relatively small −c small regions of 2:1 (c–g), 3:1 (f –h) and
4:1 (i, j) synchronization are presented. They are visible as small stripes in the left
bottom areas
regions strongly depend on c which, as discussed above, defines the complexity

of the behavior of the uncoupled elements. It is obvious that the processes of
rotation locking in the system of coupled elements cannot be exactly predicted
from the properties of motion of the uncoupled elements. As demonstrated in
the regular case, even a weak coupling can already lead to a strong change in
the mutual dynamics. But some common rules of appearance and destruction
of the synchronization regime can be obtained and explained by knowing
properties of the behavior of single elements. Considering the effect of c on
the synchronization properties, we can roughly distinguish three intervals of c:
1. Small −c where only a monotonous increase of the phases is possible in
the interacting elements; in our experiments this is the interval D1 : c ∈
[−ω1 = −0.6, 0]
2. Large −c for which the variance D is very large (see Fig. 2.13) and due
to a high noncoherence of the motions, synchronization regime cannot be
achieved; in our simulation this is the interval D3 : c < −2.25
3. Intermediate c that do not belong to the two previously defined intervals;
this is the interval D2 : c ∈ [−0.6; −2.25]
For each of these intervals we now analyze the influence of the three para-
meters on synchronization discussed in Sect. 2.5.4, i.e., the variance D, the
parameter γ characterizing the relative duration of intervals of phase increase
and phase decrease, and the rotation number difference Δρ = ρ2 − ρ1 .
Small and Large Noncoherence
In the interval D1 the difference of the rotation numbers Δρ plays the cru-
cial role in getting synchronization. As in the case of regular coherent CMs
(Sect. 4.3.1), the critical value of coupling d+ , at which the transition to 1:1
synchronization regime occurs, depends on the value of the rotation number
difference Δρ. At larger values Δρ, a larger value d is needed to achieve syn-
chronization. The sizes of the synchronization regions become smaller with
increase of c. This happens due to the increase of the noncoherence of the
rotation. At chosen values ω1 and ω2 in the interval D3 , synchronization is
in general impossible due to the highly noncoherent properties of rotations.
For instance, at c = −2.5 and ω1 = 0.6 resp. ω2 = 2.6, we find that imperfect
PS (i.e., seldom occurring phase slips are possible, cf. Sect. 3.3) exists at a
very small frequency mismatch Δ = 0.0001. Therefore, a very small Δ and
as a result of that a very small rotation number difference (remember that
it is the average characteristic) does not always guarantee the occurrence of
a phase synchronization regime, because the complexity, specifically the non-
coherence, of the behavior quantified by the variance D of rotations can be
crucial. The existence of time intervals with a strongly different phase growth
rate, which leads to the absence of well-pronounced characteristic time scale,
makes the processes of locking of rotations impossible.
Intermediate Noncoherence
A quite different situation is observed in the interval D2 . For relatively small

(Fig. 4.25a, b) as well as large (Fig. 4.25h–j) Δ the main influence on syn-
chronization is exerted by the rotation number difference Δρ = ρ2 − ρ1 and
the variance D. But for intermediate Δ (Fig. 4.25c–g) initially synchronization
regime is not achieved with an increase of c for any values of coupling. For
further increase of c, a synchronization regime can appear again. The existence
of such “islands” of synchronization can be qualitatively explained as follows:
As mentioned above for chaotic rotations, two types of phase increase are
possible: monotonous increase or alternation of intervals of phase increase
and phase decrease. The transition to the second type of rotation occurs for
−c > ω. So for the first element with ω1 = 0.6, this critical value is equal
to −0.6. Figure 2.14 indicates that γ (the ratio of the duration of the phase
decreasing intervals to the duration of phase increasing intervals for the first
element) becomes strongly increasing at c ≈ −1. If in the second element the
phase is still monotonically increasing, then the time interval, where the phases
rotate in opposite directions, are existing for coupled elements. This makes
the phase entrainment rather difficult and usually a phase synchronization
regime does not exist.8 If with an increase of c in the second element phase
decreasing intervals appear, rotations in both elements become more similar,
i.e., in both elements the phases can grow and vanish, and a phase entrainment
can happen. We expect that this mechanism leads to the existence of islands of
synchronous motions for several values of frequency mismatch (Fig. 4.25c–g),
because these islands appear at such values c that approximately correspond
to the transition to the second type of rotations in the second CM.
Synchronized Hyperchaos
Here we present another interesting synchronization phenomenon [184],

whereby we use the LEs to describe the occurrence of CPS. For system (4.47)
the LEs are given by:
λ1 = log | 1 − c
π |,
(4.52)
1
M
λ2 = limM →∞ M k=1 log | 1 − c
π − 2 d cos(φk2 − φk1 ) |.
Since the first LE λ1 is constant and positive for all values of d, we infer that
only the sign of the second LE λ2 is important for the occurrence of chaotic
PS. If both LEs are positive, there is a hyperchaotic regime that determines
usually a nonsynchronized regime. If, with increase of coupling, λ2 becomes
negative, there is a strong indication for the occurrence of CPS. This situation
takes place at the transitions to 1:1 synchronization regime in all simulations
presented in Figs. 4.26 and 4.27.
But this is not the only scenario for the transition from nonsynchronous
to synchronous behavior for which criteria (4.48) and (4.49) are satisfied.
This is illustrated by the plots showing the dependencies of the winding
number w = ρ2 /ρ1 and the second LE λ2 on the coupling coefficient d
(Fig. 4.26) and phase diagrams for nonsynchronous (Fig. 4.27a, b) and syn-
chronous (Fig. 4.27c) regimes. In the interval d ∈ [0.285, 0.32] the winding
number w = 3/1 that corresponds to a 3:1 synchronization regime, but the
second LE remains positive λ2 ≈ 0.05, i.e., synchronized hyperchaos exists.
Also there are intervals of d in which 2:1 and 1:1 hyperchaos synchronizations
8
It is interesting to note that as one can see from the conditions for the existence
and stability of a fixed point in the sine circle map, the 1:1 synchronization can
occur for CM with contrary rotating phases, e.g., for ω1 < 0 < ω2 .
4.0
w
3.0
λ2
2.0
1.0
0.0
−1.0
0.10 3.1
−2.0
−3.0
0.05 3.0
−4.0
−5.0
0.00 2.9
0.25 0.30 0.35 0.25 0.30 0.35
−6.0
0.0 0.5 1.0 1.5 2.0
d
Fig. 4.26. The winding number w = ρ2 /ρ1 and the second LE λ2 vs. the coupling
coefficient d for ω1 = 0.6, ω2 = 2.0, and c = −0.15 in (4.47). Regions of 3:1, 2:1, and
1:1 synchronization exist. Enlargements of the interval [0.25;0.35] are presented in
the insets (left: λ2 , right: w)
Fig. 4.27. Phase portraits of system (4.47) for ω1 = 0.6 , ω2 = 2.0, c = −0.15 and
different d within (c) (d = 0.3) and outside (a) (d = 0.25) and (b) (d = 0.275) of the
3:1 synchronization region. In all three cases the system (4.47) is in a hyperchaotic
regime (λ1 , λ2 > 0)
are observed. The transition to (or from) synchronized hyperchaos is accom-

plished with a drastic change in the structure of the chaotic set (Fig. 4.27). In
the case of nonsynchronous hyperchaos (Fig. 4.27a, b), the chaotic trajectory
covers practically the whole phase space; i.e., the square [−π : π; −π : π] with
different densities. When the value of coupling is close to the critical value
corresponding to the transition to a synchronized hyperchaos, we observe a
localization of areas visited by the chaotic trajectory. The appearance of more
30
d = 0.28
20
d = 0.282
10
φ2k−3φ1k
d = 0.3
0
−10 d = 0.322
d = 0.326
−20
−30
0 2000 4000 6000 8000 10000
k
Fig. 4.28. Evolution of the phase difference θk = φk2 −3φk1 for synchronous (d = 0.3)
and nonsynchronous (d = 0.28; 0.282; 0.322; 0.326) regimes of system (4.47). The
parameters are: ω1 = 0.6, ω2 = 2.0, and c = −0.15
dense bands of motions can be clearly seen (Fig. 4.27b, c). From the synchro-
nization point of view, the attendances of gaps between these bands are corre-
sponding to slips in the phase difference θk = φ2 − 3φ1 , i.e., jumps of 2π [117]
(see Fig. 4.28). A decrease of the number of slips exhibits the tendency of the
system to perfect PS where no slips exist. At synchronized hyperchaos, the
chaotic trajectory is concentrated only in relatively narrow bands in the phase
space (Fig. 4.27c). This transition to synchronous motions corresponds to the
transition of the phase difference θk = φk2 − 3φk1 from rotation to oscillation.
Thus the transition from nonsynchronous to synchronous behavior in a two-
element CMs system occurs through an interior crisis [79] of the hyperchaotic
set, i.e., in both regimes both LEs are positive.
Conclusions
The main finding of this section devoted to the synchronization phenomena

in two coupled circle maps are:
– Coupled circle maps demonstrate very rich collective dynamics which has
much in common with coupled continuous in time systems
(a) The increase of coupling usually leads to synchronization
(b) The parameter (frequency) mismatch and coherence properties of
motions play the crucial role in synchronization
(c) For strong coherence only 1:1 synchronization is possible, while n : m
synchronization is a typical feature when the coherence is weak
(d) In the chaotic regime the transition to a synchronized behavior can
appear via bifurcation or interior crises of chaotic set
– In coupled circle maps there are several effects which are not observed for
continuous in time oscillators
(a) Due to the discreteness the systems can have equal time scale charac-
teristics (rotation numbers) for different parameters;
(b) The increase of coupling leads to the destruction of synchronization
regime, i.e., there exists an interval of coupling for which a synchro-
nization regime takes place;
(c) Strong noncoherence of motions is a reason for nonpredictable
synchronization transitions.
Part II
Synchronization in Geometrically Regular

Ensembles
5
Ensembles of Phase Oscillators
In this chapter we are starting with the main part of this book, the treatment
of synchronization phenomena in ensembles or networks of coupled oscilla-
tors. First we treat networks of coupled first-order phase oscillators. This
choice is based on the fact that coupled phase oscillators are a basic model
for analyzing synchronization processes in large ensembles of oscillatory sys-
tems: limit-cycle oscillators (see Chap. 6) and chaotic oscillators (see Chap. 7).
Phase dynamics approximation can be successfully applied to any weakly cou-
pled oscillators [7, 8, 20, 186, 187]. Moreover, coupled phase systems are used
to model various systems in physics, engineering, and biology. Among them is
the discrete driven sine-Gordon equation in the under- or overdamped limits,
where the coupling between partial elements has a sinusoidal form. This has
been used as (1) models of coupled pendulum systems [2, 131, 188–193], (2)
Josephson-junction arrays [194–198], (3) magnetic Heisenberg models [199],
(4) one-dimensional (1D) chiral XY model [201], (5) granular superconduc-
tors [200], (6) phase-locked loops [4], (7) phase antenna arrays [202–204], etc.
Thus, their study has also basic importance for applications.
In the first part of this chapter we analyze the collective dynamics in
ensembles of first-order phase oscillators. In the beginning (Sect. 5.1), the
main model of phase dynamics is introduced in a rather general form. Then,
we study collective phenomena in chains of unidirectionally coupled systems
(Sect. 5.2). Section 5.3 is devoted to synchronous regimes in chains with lin-
early and randomly distributed individual frequencies of the rotation. Effects
of nonuniformity of rotations are considered in Sect. 5.4. The classical problem
of mutual synchronization in ensembles of globally coupled phase oscillators is
presented in Sect. 5.5. The second part of this chapter (Sect. 5.6) is devoted to
the study of synchronization phenomena in ensembles of second-order phase
oscillators, i.e., pendulum-like systems. In Sect. 5.7 we conclude this chapter.
104 5 Ensembles of Phase Oscillators
5.1 General Model and Malkin’s Theorem

The mathematical foundation of network models is based on Malkin’s
theorem. It gives general conditions under which weakly connected periodic
oscillators can be reduced to the analysis of the corresponding phase
model. This theorem has been applied to weakly connected systems, (e.g.,
Blekhman [2], Ermentrout and Kopell [186, 187], and Hoppensteadt and
Izhikevich [185]).
Following [185], this theorem is stated below.
Malkin’s theorem. Consider weakly connected oscillators of the form
Ẋj = Fj (Xj ) + εGj (X), Xj ∈ RN , ε 1, (5.1)
such that each uncoupled system
Ẋj = Fj (Xj ), (5.2)
has an exponentially orbitally stable T0 -periodic solution γj (t) ∈ Rm , j =

1, . . . , N . Let τ = εt be slow time and let φj (τ ) ∈ S 1 be the phase deviation
from the natural oscillation γj (t), t ≥ 0; i.e.,
Xj (t) = γj (t + φj (τ )) + O(ε). (5.3)
Then, there is an ε0 > 0 such that for all 0 ≤ ε ≤ ε0 the vector of the
phase deviation φ = (φ1 , . . . , φn )T , where T is an operator of transpose, is a
solution to
φ̇j = Hj (φ − φj ) + O(ε), (5.4)
with the vector φ − φj = (φ1 − φj , ..., φn − φj )T , and the function
T0
1
Hj (φ − φj ) = Qj (t)T Gj (γ(t + φ − φj ))dt, (5.5)
T0 0
where Qj (t) is the unique nontrivial T0 -periodic solution to the linear system
Q̇j = −{DFj (γj (t))}T Qj (5.6)
satisfying the normalization condition
Qj (t)T Fj (γj (t)) = 1 (5.7)
for some (and hence all) t.

According to this theorem, phase model of a chain of locally and weakly
coupled oscillators can be generally written as:
φ̇j = ωj + ε1 h(φj−1 − φj ) + ε2 h(φj+1 − φj ) (5.8)
where j = 1, ..., N , N is the number of oscillators, ωj are the individual

frequencies, ε1,2 characterize the “downstream” and “upstream” coupling cor-
respondingly (see Fig. 5.1), and h is a 2π-periodic in both phases coupling
5.1 General Model and Malkin’s Theorem 105
ε1 ε1 ε1 ε1
J −1 J J+1
ε2 ε2 ε2 ε2
Fig. 5.1. Schematic coupling of the j th element in the chain. ε1 corresponds to the
coupling strength with the left neighbor while ε2 corresponds to that with the right
neighbor
function. For simplicity we take identical couplings inside a direction, while in

some situation they can be vary. But, the analysis in the latter case is similar.
Let us take free-end boundary conditions:
φ 0 = φ1 , φN +1 = φN . (5.9)
The existence of phase equation allows very easily to solve the problem of
global synchronization in the considered system (5.8).
After introducing the phase difference between the neighbors
θj+1 = φj+1 − φj (5.10)
the original system (5.8) reads:
φ̇1 = ω1 + ε1 h(θ1 ),
(5.11)
θ̇j = Δj + ε1 h(θj−1 ) + ε2 h(θj+1 ) − (ε1 + ε2 )h(θj ),
where j = 1, ..., N − 1, Δj = ωj+1 − ωj is the frequency mismatch between

the neighbors.
After introducing the new parameters: d1,2 = ε1,2 /(ε1 +ε2 ), γj = Δj /(ε1 +
ε2 ) and a new time τ = t(ε1 + ε2 ), the last N − 1 equations in system (5.11)
can be rewritten as:
θ̇j = γj + d1 h(θj−1 ) + d2 h(θj+1 ) − h(θj ), (5.12)
where j = 1, ..., N −1 and θ0 = θN = 0. Analyzing the stationary states in this
system (5.12), one can solve the problem of global synchronization of weakly
coupled oscillators.
In this chapter and in all following ones we focus on synchronization effects
in chains of coupled oscillators for which the dependence on their individual
frequencies changes from monotonic, completely regular (for Δω ∗ = 0) to a
completely irregular one (for Δω ∗ = Δω) [205–209]
Δω ∗ (j − 1)(Δω − Δω ∗ )
ωj = ω0 + + + Δω ∗ ξj j = 1, . . . , N. (5.13)
2 N
Here, ξj are uniformly distributed random numbers in the interval [−0.5; +
0.5], N is the number of elements in the chain, and Δω is the interval of
frequency dispersion.
In all our models we take the simple coupling function h(θ) = sin θ.
5.2 Unidirectional Coupling

We start with the treatment of the collective dynamics in a chain where the
“upstream” coupling is absent (ε2 = 0 in (5.8)). The individual frequency is
linearly distributed along th e chain (Δω ∗ = 0 in (5.13)). Hence, the model
(5.12) for the phase differences can be written as:
θ̇j = γ − sin θj + sin θj−1 ;

(5.14)
j = 1, ..., N,
where γ > 0 is a constant frequency mismatch and the boundary condition

θ0 (t) = 0.
This model is a particular case of the more general model
θ̇j = γ − sin θj + d1 sin θj−1 ;

(5.15)
j = 1, ..., N,
where d1 is not necessary small “downstream” coupling strength. We will

analyze model (5.15) because its dynamics has a strong importance for
both the theoretical understanding of the development of different dynamical
regimes (especially chaotic ones) in the flow systems and for various applica-
tions in the theory of coupled phase-locked loops [4].
The system (5.15) evolves not only in time, but in space as well (with
changing j). That is why we are interested in both stationary regimes devel-
oping in different partial elements: first in synchronization regimes and then
in the propagation of these synchronization regime along the chain, as well as
in possible bifurcations resulting in quenching of the synchronization regime.
We referred to synchronization regime as a stationary regime (θ̇j = 0) for
which between every two partial elements there is a constant residual phase
difference θj . If synchronization develops in all elements, it is referred to as
global synchronization, but if it sets in only in some of the elements, the regime
is called cluster synchronization.
As far as the parameter d1 is concerned, both positive and negative
coupling coefficients are possible.
In dependence on the frequency mismatch γ we distinguish two cases:
(1) Let γ < 1. Then the following changes of downstream regimes takes
place: between the first and the second element a synchronization regime sets
in with θ1 = arcsin γ. If x2 = γ +d1 sin θ1 = γ(1+d1 ) < 1, the synchronization
regime appears between the second and the third element with the phase shift
θ2 = arcsin x2 . The picture is quite different if | x2 |> 1. Because the phase
difference θ2 (t) is increasing infinitely, nonsynchronous regimes between each
pair of elements in the rest of the chain occur. As j is increased, the change
of a synchronization regime into a nonsynchronous one means that there is a
spatial bifurcation of synchronization loss. In the general case reasoning this
way about two arbitrary elements, we conclude that a synchronization regime
is found in all rotators with j = 1, ..., j ∗ , if the condition
5.2 Unidirectional Coupling 107
⎡∗ ⎤
j −1

| xj |= γ ⎣ di1 ⎦ < 1 (5.16)
i=0
is fulfilled for all of them, while for all remaining j = j ∗ + 1, j ∗ + 2, . . . , N an

nonsynchronous regime occurs.
Let us consider what happens in the phase space of the system (5.15) with
increasing j. Following the changes in the behavior of the image points on the
phase lines θj , we note the following: The spatial bifurcation of the loss of
synchronization regime that occurs when passing from j = j ∗ to j = j ∗ + 1
is of the same kind as the bifurcation in the individual system which takes
place as the parameter γ is increased up to the value γ = 1 (see Sect. 2.5.1).
Thus the bifurcation corresponds to the merging and vanishing of stable and
unstable steady states and of developing a rotatory periodic motion. However,
owing to the fact that j is a discrete spatial variable, the phase of the merging
of equilibrium states may be omitted.
In the synchronization regime (θ˙j = 0 in (5.15)), the fixed values of the
phase shift are defined by θj = arcsin xj , where xj can be resulting from:
γ − xj + d1 xj−1 = 0, x0 = 0 (5.17)
leading to ⎧ j
⎨ γ(1 − d1 ) ,
⎪ for d1 = 1,
1 − d1
xj = (5.18)
⎪
⎩
γj, for d1 = 1
Figure 5.2 illustrates the distributions of θj as a function of j for fixed
γ = 0.1 and different d1 (the points for different j are connected by straight
lines).
Note that as follows from (5.16) and (5.18) if γ < 1 and |d1 | < 1, for all
equations in (5.15), there are stable steady states and, therefore, all elements
in the chain of arbitrary size are synchronized (Fig. 5.2b–e).
When γ < 1 and |d1 | ≥ 1 , only some chain elements with 1 ≤ j ≤ j ∗
are synchronized (Fig. 5.2a, f). For the elements with j > j ∗ , nonsynchronous
regimes are generated in the chain.
(2) Let γ > 1 and consider negative coupling strength d1 < 0. Then in the
first two elements of the chain a nonsynchronous regime with a growing phase
difference θ1 develops. What happens for the other phase differences θj (t)?
Depending on the value of the coupling parameter d1 , the rotations along the
chain evolve in different ways. For d1 close to zero and γ slightly larger than
unity, it is possible to suppress the nonsynchronous regime as j grows. In this
case, the nonsynchronous rotatory regime among the first two elements turns
into a nonstrictly synchronous one between the second and third elements,
i.e., θ2 is not constant but oscillates around a certain mean value. Analogous
oscillatory regimes appear between the next subsequent elements, and the
1.5
(a)
θj
0.0
1.5
(b)
θj
0.0
0.2
(c)
θj
0.1
0.10
(d)
θj
0.05
0.1
(e)
θj
0.0
1.5
(f)
θj
0.0
1.5
0 20 40 60 80 100
j
Fig. 5.2. Stationary phase differences θj for synchronous regimes for γ = 0.1,
N = 100 and different coupling strengths d1 in the model (5.15). (a) d1 = 1, (b)
d1 = 0.9, (c) d1 = 0.5, (d) d1 = −0.5, (e) d1 = −0.9, and (f) d1 = −1.1. Global
synchronization regime sets for (b–e), while cluster synchronization regime (only
for the elements in the beginning of the chain) sets in the cases (a) and (f). In the
latter case, steady states in (5.15) exist only for a few elements in the beginning of
the chain
amplitudes of the oscillations θj (t) diminish with growing j. Beginning with a

particular number j ∗∗ , the amplitude of oscillations becomes negligibly small.
Hence, a strict synchronization regime develops for all elements with j ≥ j ∗∗ .
Figure 5.3 shows some time series obtained from numerical simulations for
such a chain, which illustrates this effect of suppressing the nonsynchronous
regime along the chain.
Let us explain a possible mechanism of this suppression effect. When γ is
slightly larger than unity, the rotatory behavior of θ1 is such that its image
point is close to θ1 = π/2 for most of the time. This is because for γ = 1
a periodic motion resulting from merging of the equilibrium states at the
bifurcation point θ1 = π/2 remembers its prehistory and the image point
substantially slows down near θ1 = π/2. Hence, θ2 is an oscillating variable.
As −1 < d1 < 0, the oscillatory motions acquire diminishing amplitudes
in the subsequent elements with growing j. When j −→ ∞, the oscillation
amplitude decreases to zero. Hence, an Andronov–Hopf bifurcation is realized
1.3
(e)
φ20
1.2
1.3
(d)
φ10
1.2
1.3
(c)
φ3
1.2
1.6
(b)
φ2
1.4
1.2
6
(a)
4
φ1
2
0
20000 20020 20040 20060 20080 20100
time
Fig. 5.3. Transition from a nonsynchronous regime to a synchronous one along a
chain (5.15). Time series of phase differences θj for (a) first element, (b) second
element, (c) third element, (d) tenth element, and (e) 20th element. Parameters
are: γ = 1.05, d1 = −0.1
“backward” in space, i.e., a stable oscillatory motion degenerates into a stable

equilibrium state.
Quite another kind of evolution of a nonsynchronous regime along the
chain is observed when γ > 1 and d1 < −1; there is no synchronization at all.
By means of numerical experiments for N = 20 in (5.15), it can be demon-
strated that a gradual complication of the rotatory motion along the chain
happens due to the creation of some new frequencies which are incommensu-
rable with the previous ones. Thus giving rise to the growth of the dimension
of the quasiperiodic motion. This example demonstrates the same scenario as
for developing Landau–Hopf turbulence in space [211].
A number of works [210–217] have been devoted to the investigations of the
nature and the mechanisms of the spatial development of turbulence in dissi-
pative flow systems. In the first studies spatial bifurcations of chaos developing
in a chain of unidirectionally linked generators were found numerically [210].
A periodic state at the beginning of the chain turns into a quasiperiodic
one, as the number of generators is increasing; the latter then changes into
a state of chaotic oscillations. Analytical evidence for this mechanism of the
formation of spatial turbulence is given in [212]. Then a model theory of
spatial turbulence development on the basis of the Feigenbaum universal law

has worked out [212, 213]. Thus, two of several known roots to chaos due to
changes in a control parameter in lumped systems [220] are also realized with
the spatial development of turbulence (here a spatial coordinate along the
chain serves as a control parameter).
It should be noted that the generation of chaos in chains treated in [210,
212–217] is related to the fact that there are chaotic motions in the partial
element affected by its neighbor. The specific character of the chain is that
point elements whose dynamics is described by (5.15), with fixed j (j = 1),
possess a one-dimensional nonautonomous (one variable is time) phase space,
in which chaotic motions are not likely to exist. It is obvious, that due to the
unidirectional coupling the addition of a new coordinate j in system (5.15)
does not change the attractors in phase space of a (j − 1)-dimensional system.
Therefore, taking into account these two facts, one may expect that chaos
does not develop in the chain structure involved.
In order to understand the kinds of motions that may develop in (5.15)
and the way they change with increasing j, we first analyze, how motions vary
in the chain of N elements when the parameters γ or d1 are changing.
Let γ and d1 be such that xj ≡| γ + d1 sin θj−1 |< 1 for j = 1, ..., N and
let d1 be small enough. Then in all the elements of the chain a stationary
synchronization regime sets in. In this regime, in the N -dimensional phase
space (θ1 , . . . , θN ), there exists a single stable equilibrium state O1 (θj ) (with
θj = arcsin γj for all j = 1, ..., N ). Besides O1 , there are (2N − 2) saddle
equilibrium states and one unstable state.
With increasing γ and γ = γ (1) , let xN > 1 for the last element in the chain,
while xj < 1, j = 1, ..., N − 1. Then, along the θN -coordinate there occurs a
merging and vanishing of equilibrium states and the birth of 2N −1 periodic
motions, one of them is stable, and another one unstable, while all the rest
are of a saddle type. In the phase space (θ1 , . . . , θN ) these are straight lines,
which go parallel to the θN -axis through the points having coordinates of
equilibrium states of an (N − 1)-dimensional system. Here the attractor is a
periodic motion. In the last element of the chain, a nonsynchronous rotatory
regime is formed, while in all others, a synchronization regime develops.
As γ = γ (2) > γ (1) is increased further, xN −1 > 1 but all the other xj < 1
for j < N −1. Now along the θN −1 -coordinate, the periodic motions merge and
vanish and a 2N −2 quasiperiodic motions set in. In the phase space (θ1 ,. . . ,θN ),
the attractor is now a two-dimensional torus with a quasiperiodic motion. The
scenario for the formation of a quasiperiodic motion in the (θN −1 ,θN )-plane
with constants θj , j = 1, ..., N − 2 is shown in Fig. 5.4.
Again, we increase γ further. With γ = γ (3) > γ (2) , xN −2 > 1, the two two-
dimensional tori with quasiperiodic motions on them approach each other and
vanish and, instead, 2N −3 three-dimensional tori with quasiperiodic changes
of the θN −2 , θN −1 , θN coordinates are formed. Reasoning along this way, we
can conclude that when γ = γ (k) , k = 4, ..., N in the general case, 2N −k k-
dimensional tori are born. Thus, there is a finite number of k-values across
(a) (b) (c)

2π
θN
0
0 2π 0 2π 0 2π
θN-1 θN-1 θN-1 θN-1
Fig. 5.4. A schematic scenario for the formation of a quasiperiodic motion with an
increase of coupling d1 in the (θN −1 , θN ) plane. (a) There are stable and unstable
periodic motions. (b) There is only one periodic motion. (c) Quasiperiodic motion.
Only several rotations are shown
which, in the transition toward increasing γ, the dimension of the attractor of

the system is increased by unity. Finally, when γ > γ (N ) , one N -dimensional
torus with open winding is the attractor in the N -dimensional phase space
(θ1 , . . . θN ). This means that within the chain in any element of the chain there
appears rotatory motion with a quite different frequency, rationally incom-
mensurate with all the other frequencies.
It is obvious from the analysis performed that the transfer from the
(j − 1)th element to the jth element in the chain of N elements can give
rise to the birth of a new frequency which is incommensurable with the previ-
ous frequencies, i.e., the dimensionality of the quasiperiodic motion increases.
Provided that N −→ ∞, an infinite-dimensional “quasiperiodic motion”
would be the image of the regime developed in such a semiunbounded system.
Therefore, the Landau’s route to turbulence is indeed realized. Historically
this scenario was the first proposed route to turbulence. Ruelle, Takens, and
Newhouse [218, 219] have considered the complex behavior as a chaotic one
and firstly have introduced the notion of strange attractor. They showed, that
a strange attractor can appear in a low-dimensional phase space. According to
the Ruelle–Takens–Newhouse route, the transition from quasiperiodic behav-
ior to chaos appears after the birth of a third rationally incommensurable
frequency.
Numerical modeling of system (5.15) with γ = 2, d1 = −5, N = 5, 10, 20 il-
lustrates the existence of N -dimensional quasiperiodic motions in phase space.
The increasing complexity of the motion along the chain (5.15), as the number
of interacting rotators grows, is readily seen in Fig. 5.5, where the power spec-
tra of trajectories of the chain elements for growing values of j for γ = 2 and
d1 = −5 are shown. As indicated in Fig. 5.5, attractors in the tenth and 20th
elements have practically a continuous power spectra. A detailed analysis of
the spectra shows that in the second element there exists a two-dimensional
quasiperiodic behavior, and in the third element – a three-dimensional quasi-
periodic behavior.
(a) (b) (c) (d) (e)

10.0
5.0
logS(ω)
0.0
5.0
10.0
0 3 60 3 60 3 6 0 3 6 0 3 6
ω ω ω ω ω
Fig. 5.5. Development of instabilities along the chain (5.15). Spectra of the attrac-
tors for several elements. (a) First element, (b) second element, (c) third element,
(d) tenth element, and (e) 20th element. The parameters are γ = 2, d1 = −5, N = 20
It should be noted that our numerical investigations showed also the

existence of quasiperiodic motions whose dimension is lower than that of the
phase space.
5.3 Synchronization Phenomena in a Chain

of Bidirectionally Coupled Phase Oscillators
Now let us turn to a chain of bidirectionally coupled phase systems.

For symmetrically with strength d coupled elements, system (5.8) reads:
φ̇j = ωj + dh(φj−1 − φj ) + dh(φj+1 − φj ) (5.19)
which for the phase difference θj = φj+1 − φj and the frequency mismatch
Δj = ωj+1 − ωj yields:
φ̇1 = ω1 + dh(θ1 ), (5.20)
θ̇j = Δj + dh(θj−1 ) + dh(θj+1 ) − 2dh(θj ), (5.21)
where j = 1, ..., N − 1.
5.3 Synchronization Phenomena 113
The conditions for the existence and the stability of the regime of syn-
chronization of all elements in this ensemble, i.e., a regime of global synchro-
nization, can be studied analytically. This regime corresponds to a stable
stationary state in (5.21), which can be defined from the linear difference
equation with constant coefficients:
Δj
xj−1 − 2xj + xj+1 = − , j = 1, ..., N − 1 (5.22)
d
where xj = h(θ̄j ) , θ̄j is the constant difference between the phases of neigh-
bors, and with the boundary conditions:
x0 = xN = 0. (5.23)
The solution of (5.22) depends strongly on the concrete frequency distribution

ωj . Because h is a 2π-periodic function, there are at least 2N −1 steady states
in (5.21) but, as shown in [4, 206], only one (θ̄1 , ..., θ̄j , ..., θ̄N −1 ), for which
dh/dθj > 0 for all variables, is a stable node. Al the other steady states are
saddles and unstable node.
The frequency of the global synchronization regime for any type of individ-
ual frequency distributions for symmetrically coupled rotators can be easily
defined. After summing up all equations in (5.19), we get:

N
N
φ̇j = ωj . (5.24)
j=1 j=1
Because for global synchronization the frequencies of rotations of all

elements coincide, i.e., Ωj = φ̇j = Ωs for all j = 1, ..., N , (5.24) yields
1
N
Ωs = ωj . (5.25)
N j=1
That means that the frequency of the global synchronization regime Ωs for
symmetrically coupled phase oscillators is always equal to the mean frequency
of the oscillators in the chain.
With an increase of the frequency mismatch or with a decrease of coupling
for a certain oscillator j ∗ , xj ∗ reaches a maximal or minimal possible value
(hmax or hmin ) and steady states disappear via a saddle-node bifurcation.
As a result only one phase difference θj ∗ becomes unbounded, i.e., the variable
θj ∗ rotates, while the rest remains bounded, i.e., they oscillate around the
corresponding states θ̄j , i.e.,
|θj − θ̄j | < Const, for all j = 1, ..., N − 1 and j = j ∗ . (5.26)
Thus, the whole chain is divided into two groups, called clusters, of
mutually synchronized elements. The first cluster consists of the elements
j = 1, ..., j ∗ , which are synchronized with the frequency Ω1 = dtj , whereas

dφ
the elements in the second cluster with l = j ∗ + 1, ..., N are synchronized

with the frequency Ω2 = dφ l
dt . Strongly speaking, the global synchronization
regime breaks up in all oscillators. The motion is firstly quasiperiodic with
dimension two, i.e., in the power spectrum of each oscillator φj there are two
main peaks corresponding to Ω1 and Ω2 . But the components corresponding
to Ω2 for the oscillators of the first cluster (j ∈ [1, j ∗ ]) is weakly pronounced
in comparison to the main component corresponding to Ω1 . Moreover, in long
chains the power spectral component corresponding to Ω2 vanishes for the
first elements of the chain. A similar behavior happens for the oscillators of
the second cluster.
A further increase of the frequency mismatch or decrease of the coupling
leads to another splitting of both clusters, leading to more clusters till a fully
nonsynchronized regime occurs, i.e., all oscillators have different frequencies.
After the first transition to a cluster structure, the mutual synchronization
regime of phase oscillators can be characterized in terms of coincidence of their
mean observed frequencies defined as
Ωj = φ̇j . (5.27)
Then, in order to test for the existence of synchronization regime we can
straightforwardly use the criterium applied for two coupled oscillators from
Chap. 4. We say that two arbitrary (not necessary neighboring) elements i
and j are m : n synchronized (m and n are integers), when
mΩi = nΩj . (5.28)
Another criterium of m : n synchronous state between elements i and j
are the phase locking conditions:
|mφi − nφj | < Const. (5.29)
5.3.1 Synchronization, Clustering and Multistability in Chains

with Linearly Distributed Individual Frequencies
In the next subsections we use the coupling function h in the form of sine.
Then the main model (5.19) is:
φ̇j = ωj + d sin(φj−1 − φj ) + d sin(φj+1 − φj ) , j = 1, ..., N. (5.30)
First, we consider a linear distribution of the individual frequencies ωj (5.13).
Then using the substitutions θj = φj+1 − φj and Δ = ωj+1 − ωj , the model
(5.30) can be rewritten as:
φ̇1 = ω1 + d sin(θ1 ), (5.31)
θ̇j = Δ + d sin(θj−1 ) − 2d sin(θj ) + d sin(θj+1 ), (5.32)
where j = 1, ..., N − 1.
The constant value of Δ allows to solve the linear difference equation (5.22)
analytically. In this case it takes the form:
Δ
xj−1 − 2xj + xj+1 = − , j = 1, ..., N − 1, (5.33)
d
where xj = sin(θ̄j ). From (5.33) the distribution of xj is determined by:
Δ
xj = (N j − j 2 ). (5.34)
2d
Therefore system (5.32) has 2N −1 steady states, whose coordinates θ̄j are
equal to
θ̄j = arcsin[Δ(N j − j 2 )/2d] (5.35)
or
θ̄j = π − arcsin[Δ(N j − j 2 )/2d]. (5.36)
But as was mentioned before, only one steady state with
Δ
θ̄j = arcsin[ (N j − j 2 )]
2d
for all j = 1, ..., N is stable. As follows from (5.34), the steady states disappear
when 2d Δ
(N j −j 2 ) reaches unity. This happens at the critical coupling strength
dcr defined as:
⎧
⎪
⎪ ΔN 2
⎨ 8 , for even N,
dcr = . (5.37)
⎪
⎪
⎩ Δ(N − 1)(N + 1) , for odd N
8
If d becomes smaller than dcr , the break-up of the global synchronization
regime happens, and two clusters of mutually synchronized elements appear.
The size of both clusters is N/2.1 The transitions to (from) global synchro-
nization regime from (to) a fully nonsynchronized regime exhibit interesting
features in dependence on the number of elements and the frequency mis-
match Δ: The main phenomenon is the existence of typical synchronization
trees (Fig. 5.6). Due to the symmetry of the dynamical system, the transition
cascade in these trees is symmetric around the mean individual frequency Ωs .
Similar to two coupled elements, two types of transitions between the cluster
structures are to distinguish: First, a hard transition without intermediate
structures occurs from the state with n clusters to the state with n + 1 or
more clusters or from the state with n + 1 clusters to the state with n or
less clusters. Second, a soft transition happens with a smooth transition of
intermediate structures one into the other. For intermediate structures there
1
We note that here and in all the following chapters we will consider ensembles
with even number of elements.
1.20
1.15
Ωj
1.10
1.05
1.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6
d
Fig. 5.6. Synchronization trees for the chain (5.30) with linearly distributed indi-
vidual frequencies ωj . The parameters are N = 20, ω1 = 1.0, Δ = 0.01. Both
transition types – hard and soft – are presented. The hard transition is more typical
for relatively large coupling, but the soft transition is quite typical for small coupling
exists the regime of multifrequency rotations, when all elements of the chain
(except the edge ones) rotate with different frequencies Ωj . Scaling properties
of the occurring cluster synchronization regime, namely, sizes of clusters, clus-
ter frequencies, etc., will be described in detail in Chap. 6 for weakly coupled
limit-cycle oscillators.
In Fig. 5.8 we plot space–time diagrams of the evolution of sin(φj (t)) (a-d)
for different synchronization regimes. Corresponding mean frequency (Ωj ) dis-
tributions are shown in Fig. 5.9. In all plots in Fig. 5.8 the darker regions
mark higher values of the presented variables. It is clearly visible that with
an increasing coupling, the collective behavior of the elements in the chain
becomes more coherent, i.e., there appear clusters of mutually synchronized
phase oscillators. So for the coupling d = 1 (Fig. 5.8b) there are five clusters
of synchronized elements. For d = 3 (Fig. 5.8c) there are two synchroniza-
tion clusters. At the border of two neighboring clusters there appear phase
difference slips or defects. These are jumps of the phase difference between
neighboring elements of 2π. In Fig. 5.8a–c these events are indicated by a
merging of two white (black) lines. For regular cluster structures in Fig. 5.8b,
c the frequency of the appearance of defects is defined by the mean frequency
mismatch of the neighboring clusters. For d = 4 (Fig. 5.8d) the regime of
global synchronization is reached. Figure 5.8d demonstrates the propagation
of phase waves with a constant phase shift.
0.01
0.00
−0.01
−0.02
Lyapunov exponents
−0.03
−0.04
−0.05
−0.06
−0.07
−0.08
−0.09
−0.10
0.0 0.1 0.2 0.3 0.4 0.5
d
Fig. 5.7. Spectrum of the 20 Lyapunov exponents for a chain (5.30) with linearly
distributed frequencies. Parameters are the same as in Fig. 5.6
In order to give a deeper insight into the synchronization transitions, we

calculate the spectrum of Lyapunov exponents (Fig. 5.7). By inspection of
Figs. 5.6 and 5.7 we can conclude that the soft transitions from one synchro-
nized structure to another one is accomplished by a transition from regular
to chaotic and then again to regular behavior (in Figs. 5.6 and 5.7 see, for
example, coupling intervals around d = 0.05; 0.085; 0.15). That means that for
a low level of ordering, which is typical for a small coupling strength, these
transitions appear through a strong destruction, i.e., through the onset of
disorder. If the coupling is relatively strong, such interstructure transitions are
not observed. As we will show in the next chapters, such transitions are usually
observed in chains of coupled regular as well as chaotic oscillators. We spec-
ulate that this type of evolution is typical for self-organization processes in
nature and engineering.
The evolution of the Lyapunov spectrum exhibits a novel type of chaos–
chaos transition [205]. For small coupling due to the high degree of symmetry
and homogeneity of the system, the spectrum has a sign-symmetric structure.
This leads to a zero phase volume, i.e., the dynamics of the system is quasi-
Hamiltonian. With an increase of the coupling, this sign-symmetry of the
Lyapunov spectrum is violated and a smooth transition to dissipative chaos
is observed. There also exists intervals of coupling for which many attractors
coexist [205]. These intervals mostly appear in small chains at small coupling,
prior the first clustering. The reason for this effects is the reversibility of the
system dynamics.
(a) (b) (c) (d)

5000
4000
3000
2000
1000
0
1 j 50 1 j 50 1 j 50 1 j 50
Fig. 5.8. Space–time plots of evolution sin(φj (t)) for different synchronization
regimes in a chain (5.30) with linear frequency distribution. For the coupling d = 0.5
(a) there are some clusters of synchronization for the elements close to the ends of
the chain and strong incoherent motion for the elements in the middle of the chain.
For d = 1 (b) there are five clusters of synchronized elements. For d = 3 (c) there
are two synchronization clusters. For d = 4 (d) regime of global synchronization is
reached. The parameters are N = 50, ω1 = 1.0, Δ = 0.001
Further, in the evolution of synchronized clusters, structures are formed

which consists of a different number of clusters in dependence on the ini-
tial conditions, i.e., we have the phenomenon of multistability of coexisting
attractors.
1.40
ε=0.5
1.
1.35
2.
4
1.30
1.25
Ωj
1.20
1.15
1.10
1.05
0 10 20 30 40 50
j
Fig. 5.9. Distributions of the mean frequencies Ωj in different synchronization

regimes presented in Fig. 5.8. Parameters are the same as in Fig. 5.8
5.3.2 Synchronization Transitions in Chains with Randomly

Distributed Individual Frequencies
Let us turn to the case of randomly distributed individual frequencies.
Additionally to the chain (5.30), we also consider a two-dimensional lattice of
coupled phase oscillators of first order which are described by:
φ̇i,j = ωi,j +
d sin(φi,j−1 − φi,j ) + d sin(φi,j+1 − φi,j ) (5.38)
d sin(φi−1,j − φi,j ) + d sin(φi+1,j − φi,j ),
where i = 1, ..., M and j = 1, ..., N. N and M are the number of oscillators in
both directions.
Note that here we have to expect synchronization of not only the neigh-
bored oscillators as in the case of linearly distributed frequencies but also
the oscillators whose individual frequencies are nearby. In order to study the
transitions between the synchronous states, we calculate the evolution of the
mean observed frequencies Ωi,j in dependence on the coupling strength as
illustrated in Fig. 5.10 for a chain and in Fig. 5.11 for a lattice (5.38).
Analyzing the synchronization trees for the chain and synchronization
snapshots for the lattice, we see that the main synchronization phenomena
in both cases are similar. Three main types are observed:
1. Two (or more) neighboring rotators having close individual frequencies
become rather fast synchronized. This type of synchronization behavior
is frequent.
1.10
1.05
Ωj
1.00 1,2
d = 0.07
0.75
02 j 0
3,4,5,6,7,8,9
Ωj
0.90
10,11,12,13,14,15,16,17,18
0.92 d = 0.17
0.80 19,20
Ωj 0.18
12,13,15,16,19,20
0.82
02 j 0
0.70
0.00 0.10 0.20 0.30 0.40 0.50
d
Fig. 5.10. Synchronization trees for a chain with randomly distributed individual
frequencies ωj . The parameters are N = 20, ωj randomly distributed in the inter-
val [0.7 : 1.1]. The numbers of mutually synchronized elements are shown; e.g., 1,2
means that the first and second elements are synchronized. In the inserts the distrib-
utions of the mean frequencies for a constant coupling are presented. The top insert
shows a regime of nonlocal synchronization of the elements 12, 13, 15, 16, 19, 20. The
corresponding place in the synchronization tree is marked by an arrow. The bottom
insert shows the transition of the 18th element from one to another synchronization
cluster
2. The formation of nonlocal synchronization where two (or more) non-

neighboring rotators become synchronized, while the rotators in-between
them remain nonsynchronized. But the latter one can quickly brought
by its neighbors to a synchronous regime (for the chain see top insert in
Fig. 5.10).
3. Due to the competition of neighboring clusters, the edge element of one
cluster detaches from this cluster and passes to another cluster (for the
chain see bottom insert in Fig. 5.10).
Therefore the synchronization transitions in ensembles with randomly dis-

tributed frequencies are more complex than in the case of linear frequency
distribution.
5.4 Influence of Non-Uniform Rotations on the Synchronization 121
(g) (h) (i)
(d) (e) (f)
(a) (b) (c)
Fig. 5.11. Synchronization in a lattice (5.38) with individual frequencies ωi,j ran-
domly distributed in the interval [0.9 : 1.1]. The observed frequency distribution Ωi,j
for different coupling strength: (a) d = 0, (b) d = 0.015, (c) d = 0.025, (d) d = 0.045,
(e) d = 0.055, (f) d = 0.065, (g) d = 0.075, (h) d = 0.085, (i) d = 0.095. In
all plots the darker regions mark higher values of the observed frequencies. The
regions marked by the same color correspond to clusters of mutually synchronized
elements. Detailed inspections of these results show the existence of all three types
of synchronization behavior observed for chains (Fig. 5.10). The x-axis corresponds
to j = 1, . . . , N = 50 and y-axis to i = 1, . . . , M = 50
5.4 Influence of Non-Uniform Rotations

on the Synchronization
We now extend the model class (5.30) to systems which include a term
(a sin φj ) in each element, which generates nonuniform rotations, i.e., the
instantaneous frequency of rotations is no more constant. This is the gen-
eral situation for strongly nonlinear oscillators.
The model chain with nonuniformly rotating phase variables is then:
φ̇j = ωj − a sin φj + d sin(φj−1 − φj ) + d sin(φj+1 − φj ),

(5.39)
j = 1, ..., N,
where the parameter a characterizes the degree of nonuniformity of the rota-

tion. It implies mainly that the instantaneous frequency does not coincide
with the individual frequency of periodic rotation (see Sect. 2.5.1). In all our
simulations we take a < ωj for all j, i.e., in all uncoupled systems there exist
rotations. Hence, the frequency of rotations can be defined according to

ω̄j = (ω1 + Δ(j − 1))2 − a2 (5.40)
In our numerical simulations the mean observed frequencies Ωj were com-

puted according to (5.27). Analyzing a chain for ω1 = 1, Δ = 0.001, and
N = 50 (Fig. 5.12), we can conclude that
1. There is a very important difference in the transitions from a fully non-
synchronous state to a fully synchronous state for the elements with uni-
form and nonuniform rotation. For uniformly rotating systems (a = 0,
Fig. 5.12a), this transition appears mostly through the formation of cluster
structures, i.e., via a hard transition, while for rather strong nonuniformly
rotating systems (a = 0.9, Fig. 5.12c) this transitions has a “smooth” char-
acter, i.e., via a soft transition. For intermediate a, both transitions – hard
and soft ones – exist (a = 0.75, Fig. 5.12b). As we will show in the next
chapters, this difference is typically observed in chains of coupled regular
and chaotic oscillators. This is mainly due to the existence of multiple
(a) (b) (c)

1.05 0.74 0.54
ε=0 ε=0
0.09 0.09
0.13 0.17
0.52
0.16 0.2
1.04
0.22 0.34
0.72
0.32 0.37
0.50
1.03
Ωj
0.70 0.48
1.02
0.46 ε=0
0.68 0.1
0.17
1.01
0.44 0.27
0.33
0.37
1.00 0.66 0.42

0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
j j j
Fig. 5.12. Synchronization transitions for nonuniform rotating phase oscillators in
(5.39). (a) a = 0, (b) a = 0.75, and (c) a = 0.9
5.5 Mutual Entrainment in Populations 123
time scales. For uniformly rotating systems (or also, for example, for qua-
siharmonic limit-cycle oscillators) there exist only one time scale while for
nonuniformly rotating systems (or also, e.g., for relaxation type limit-cycle
oscillators) there are several (at least two) time scales, e.g., a slow and a
fast one. But as it is well known (e.g., [79]) the appearance and interaction
of many timescales lead in oscillatory systems to a chaotic behavior. That
is why the typical route to global synchronization regime in ensembles of
nonuniformly rotators is via soft transitions.
2. During the destruction of the global synchronization regime for all a,
two clusters of mutually synchronized elements appear. As our simula-
tion shows, the size j ∗ of the left appearing cluster becomes smaller with
increasing a.
3. The frequency of global synchronization regime increases with the increase
of the parameter a. For a = 0 it is equal to the mean individual frequency
of the rotators, while for a slightly less than unity, it is close to the maximal
individual frequency. The reason for this effect is the following. The phase
evolution in rotators with large a looks as intermittency of relatively long
intervals of weak changes interrupted by relatively short intervals of strong
increase (see Fig. 2.9). The strong change (“firing”) of the phase variable
in the elements close to the right end of the chain provokes an analogous
strong change of the phases in the neighboring element that leads to a
sequential “firing” in all elements in the chain. Here boundary conditions
have to be taken into account.
4. The frequency of the appearing global synchronization regime Ωs with an
increase of the coupling can be well described by:
Ωs = ωN −j ∗ , (5.41)
where ωN −j ∗ is the individual frequency of the (N − j ∗ )th element. N − j ∗

is equal to the length of the right cluster appearing after the destruction
of the global synchronization regime.
5.5 Mutual Entrainment in Populations of Globally

Coupled Phase Oscillators
Population of globally coupled elements are another interesting field of collec-
tive effects in oscillatory networks (e.g., [1, 8]).
First Winfree [221] mathematically reduced the problem of mutual syn-
chronization to that of a collective behavior in an ensemble of coupled phase
oscillators. He showed that such oscillator ensembles could demonstrate a tem-
poral analogue of a thermodynamic phase transition: with the increase of the
coupling a group of oscillators suddenly becomes synchronized.
By using perturbation methods of averaging, Kuramoto [8,222] obtained a

model of weakly globally coupled, nearly identical limit-cycle oscillators. The
governing equations of this model are
d N sin(φ − φ ),
φ̇j = ωj + N k=1 k j
(5.42)
j = 1, ..., N,
where d ≥ 0 is the coupling strength and the individual frequencies ωj are
Gaussian distributed according to some symmetric probability density g(ω),
i.e., g(ω) = g(−ω).
To describe synchronization transitions, Kuramoto introduced the complex
order parameter r
1
N
r exp(iψ) = exp(iφk ) (5.43)
N
k=1
In this formula r and ψ denote an amplitude and a phase of the complex mean
field Z = X + iY = r exp(iψ). The amplitude r of the mean field measures
the degree of coherence of collective behavior. In the extreme cases: r = 0 for
a fully noncoherent behavior and r = 1 for a globally synchronized regime.
Using (5.43), the system (5.42) can be rewritten as
φ̇j = ωj + dr sin(ψ − φj ),
(5.44)
j = 1, ..., N.
Kuramoto analyzed model (5.44) in the limit N → ∞. This allows to get:
1. The exact formula for the critical coupling strength dcr that corresponds
to the transition to the most ordered behavior
2
dcr = . (5.45)
πg(0)
2. The dependence of the order parameter r on the coupling near the critical
point dcr :
16 μ
r≈ , (5.46)
πd3cr −g (0)
where μ is the relative distance to the critical point
d − dcr
μ= . (5.47)
dcr
For the case of a Lorenzian density
γ
g(ω) = , (5.48)
π(γ 2 + ω 2 )
Kuramoto found the analytical expression

dcr
r = 1− , (5.49)
d
which is shown in Fig. 5.13.
0
dcr
d
Fig. 5.13. Typical evolution of the order parameter r according to (5.49)
As a very interesting recent manifestation of synchronization of glob-

ally coupled subjects one can consider the effect observed by the opening
of London’s Wobbly Millennium Bridge in 2000. A spontaneous synchrony
occurred when hundreds of pedestrians caused the bridge to undulate errati-
cally as they unconsciously adjusted their pace to the bridge’s swaying. This
explanation was proposed by Josephson, then experimentally verified and later
numerically calculated [223].
5.6 Synchronization Phenomena in a Chain of Coupled

Pendulum-Like Equations
As mentioned in the beginning of this chapter, a huge variety of physical and
engineering systems can be formulated as coupled pendulum-like systems.
Here, we discuss the collective behavior in a chain of pendulum-like equations
(see Sect. 2.5.2) with a sinusoidal nearest neighbor coupling:
φ̈j + λφ̇j + sin φj = ωj + d sin(φj+1 − φj ) + d sin(φj−1 − φj ),

(5.50)
j = 1, ..., N,
where λ is the friction coefficient, ωj are the external driving forces, and
d is the coupling strength. As in Sect. 5.5, we consider free-end boundary
conditions: φ0 = φ1 and φN +1 = φN .
Equation (5.50) is a discrete version damped forced sine-Gordon equation
studied in many applications (e.g. [198] and [4] and references therein).
Note that for small phase deviations, i.e.,
|φj+1 (t) − φj (t)| 1,
(5.51)
j = 1, ..., N − 1.
(5.50) can be reduced to the damped and forced sine-Gordon equation in the
continuum limit
∂2φ ∂φ 2
∂ φ.
+ λ + sin φ = ω(x) + d (5.52)
∂t2 ∂t ∂x2
This equation has been intensively investigated (e.g., [224–229]) and, there-
fore, our network approach may have special interest in the mentioned appli-
cations.
The individual dynamics of a single element (see Sect. 2.5.2), especially
the existence of a parameter region (region D2 in Fig. 2.10) in which two
attractors – a steady state and periodic motion – exist, provides the existence
of two quite different regimes of global synchronization in (5.50):
1. The first regime is similar to that observed in coupled active rotators

described in Sect. 5.5. It is characterized by a relatively high frequency of
synchronous motion.
2. The other type of a global synchronization regime, the regime with
low frequency, is caused by the existence of traveling single- and mul-
tipulse motions. It was found for the system (5.50). There may be both
pulses traveling to the left (in the direction of decreasing j-solitons)2 and
those traveling to the right (in the direction of increasing j-antisolitons).
The temporal realization φ̇j (t) takes a typical form of soliton (Fig. 5.14).
By that the temporal realization φj (t) is the following: In every element
of the chain there are rotatory periodic motions; for the major part of the
period T0 , the value φj (t) is practically constant. Further in a short time
ΔT0 T0 there occurs a jump of 2π in φj (t).
We demonstrate the low-frequency synchronization regime with numerical

calculations. We fix in (5.50) the friction parameter λ = 0.4, d = 10, and
N = 100, and a linear distribution of the parameters ωj = ω1 Δω(j − 1).
As initial conditions we take:

N
(φj ) = 4 arctan exp(j − )h , φ̇j = 0, j = 1, ..., N.
2
Experiment 1. ω1 = 0.901, Δ = 0.00001. A soliton, having changed its form

slightly, remains the only spatial structure (Fig. 5.14a). The frequency of
the global synchronization regime depends on the velocity of the soliton
propagation.
Experiment 2. ω1 = 0.902, Δ = 0.00001. For some time there are no spatial
structures different from a soliton. Then some local breather-like structure
develops which breaks into a pair: a soliton and an antisoliton. Later on
no new spatial structures arise (Fig. 5.14b). The frequency of the global
synchronization regime is twice larger than in the first experiment.
Experiment 3. ω1 = 0.90205, Δ = 0.00001. In the same way, as in the
previous experiment, the localized structure arising breaks into a soli-
ton and an antisoliton. This is followed by the appearance of one more
2
Here the use of the terms “soliton” and “antisoliton” standing for “pulses” is not
strict; however, so far as this terminology is commonly used in the literature for
the sine-Gordon type equations [224], we consider it expedient to retain them
here.
5.7 Conclusions 127
4.0
3.0 (a)
2.0
.
φj
1.0
0.0
−1.0
4.0
3.0 (b)
2.0
.
φj
1.0
0.0
−1.0
4.0
3.0 (c)
2.0
.
φj
1.0
0.0
−1.0
01 02 03 04 05 06 07 08 09 0 100
j
Fig. 5.14. One (a), two (b) and three (c) solitons in a chain of coupled second-order
phase oscillators in (5.50)
breather-like structure which also breaks into a soliton and an antisoliton.

No new structures arise after this further on (Fig. 5.14c). The frequency
of the global synchronization regime is three times larger than in the first
experiment.
5.7 Conclusions
The properties of networks of different phase oscillators presented in this

chapter are of basic importance for the understanding and description of
synchronization processes in all distributed systems analyzed in the follow-
ing chapters. It is important to note that the further study of synchronization
effects is related to ensembles of coupled regular or chaotic dynamical systems
which have their own characteristic time scales (in the simplest case periodic
oscillators or rotators just the frequencies). Therefore, this problem can be
formulated as the achievement of some relationships between frequencies and
(or) phases. Hence, the observed effects are very common in the networks
studied in the next chapters.
The main effects are:
1. Collective phenomena in a network of weakly connected oscillators can be

analyzed by using of a corresponding phase model.
2. For uni- and bidirectionally coupled phase oscillators with a linear distri-
bution of individual frequencies there is a coupling strength for which a
global synchronization regime appears in a chain of arbitrary length.
3. In a chain with unidirectionally coupled systems, the downstream transi-
tions “synchronous regime to nonsynchronous regime” and “nonsynchro-
nous regime to synchronous regime” are possible.
4. Hard and soft transitions to global synchronization are typical.
5. Neighboring elements can form synchronization clusters.
6. In the case of randomly distributed individual frequencies the effect of the
regime of “nonlocal” synchronization can be observed.
7. Multistability of different synchronous regimes are possible and quite typ-
ical.
8. For nonuniformly rotating elements, the transition to global synchroniza-
tion regime is usually soft.
9. For uniform rotations the frequency of the global synchronization regime is
equal to the mean frequency of the elements in the ensemble. For strong
nonuniformly rotating systems the elements with higher frequency are
dominated in the ensembles.
10. For coupled second-order phase oscillators, there are two types of synchro-
nous regimes with a low and a hard frequency of synchronous motion. First
regime is associated with the existence of soliton-like behavior.
6
Chains of Coupled Limit-Cycle Oscillators
In this chapter collective effects in chains of diffusively coupled limit-cycle

oscillators with different individual frequencies are investigated by asymptotic
and numerical methods. In the case of weak coupling using phase approx-
imation (see Chap. 5), conditions for the onset and existence of a global
synchronization regime are determined. Cluster synchronization regime and
mechanisms of synchronization transitions are studied. Main results are (1)
the existence of mono- and multistable regimes of cluster synchronization and
(2) the soft respective hard transitions between these structures, which are
consisting of a different number of clusters. Synchronization is observed in
a broader range of parameters in a randomly formed chain than in the case
of a regular arrangement of oscillators with monotonically varying individual
frequencies along the chain.
This chapter starts with a short introduction to the studied problem
(Sect. 6.1). Chains of oscillators with a uniform frequency mismatch Δωj =
ωj − ωj−1 = const are investigated in detail in the frame of a discrete ana-
logue of the Ginzburg–Landau equation in Sect. 6.2. We begin with a brief
description of the model (Sect. 6.2.1) and characteristics of the global syn-
chronization regime (Sect. 6.2.2) that are further used for the interpretation
of the properties of cluster synchronization regime (Sect. 6.2.3). Multistable
regimes are described in Sect. 6.2.4. In Sect. 6.3, the formation of clusters sepa-
rated by a region of nonexcited oscillators is interpreted in terms of the effect
of oscillator death [243–246]. The influence of nonlinear distribution of the
individual frequencies is presented in Sect. 6.4. The formation of synchronized
clusters in the presence of regular nonuniformities of the frequency mismatch
is studied in Sect. 6.4.1. In Sect. 6.4.2 it is shown that small-scale nonunifor-
mities, irregular ones inclusive, may expand the region of the parameters in
which synchronization is possible. Nonlinear effects that are beyond the scope
of the Ginzburg–Landau equation are considered in Sect. 6.5 for van der Pol
oscillators. Section 6.6 concludes our study of synchronization phenomena in
chains of coupled limit-cycle oscillators.
130 6 Chains of Coupled Limit-Cycle Oscillators
6.1 Objectives
The investigation of different synchronization phenomena is close to the
problem of the relationship between the properties of temporal dynamics of
oscillations and their changes in spatially extended systems and their chain
analogues. This problem is important for both diagnostics [230, 231] and fea-
sibility of different regimes [232–234]. It is important to emphasize that these
systems have some properties which seem to be surprising and counterintuitive
at the first view: Under certain conditions a time-periodic behavior is real-
ized only in the presence of spatial disorder which is generated either by the
system itself [232] or introduced from outside, e.g., by the dispersion of para-
meters of the elements in the chain [233,234]. In particular, a globally chaotic
regime may be realized in a chain of identical nonlinear regular pendulums
with external forcing under certain conditions. But if there is a dispersion
in their parameters, this chaotic regime is replaced by a periodic one [233].
In an analogous mathematical system, but without external periodic forcing,
it models a parallel chain of current-biased Josephson junctions coupled via
inductors. A small dispersion in the currents gives rise to an enhanced mutual
synchronization of oscillations in these junctions [234].
In this chapter we consider synchronization effects in a chain of coupled
regular oscillators with linearly and randomly distributed individual frequen-
cies (see 5.13). A choice of a completely regular distribution in the form of the
frequency variation that is linear along the chain is motivated by the fact that
it occurs in several situations in engineering and nature as presented next.
6.2 Synchronization Clusters and Multistability

at Linear Variation of Individual Frequencies
Along the Chain
Chains of oscillators with linearly varying individual frequencies are
interesting both conceptually and because they are encountered in various
applications. We mention here two rather illustrative examples:
1. The dynamics of mammalian small intestine. If one isolates them into
sections 1–3 cm long, then each of them is able to oscillate at a definite
frequency and the changes of these frequencies along the intestine can be
regarded to be linear at rather long distances [235].
2. The vortex shedding in a flow past cone-shaped bodies, e.g., supports
or chimney stacks. This can be again described by a chain of coupled
oscillators with linearly varying individual frequencies, if the derivative
with respect to the coordinate along the cone axis is replaced by finite
differences (see, e.g., [236]).
In this type of systems a sufficiently strong coupling between the oscillators
leads to local frequencies of excited collective oscillations that strongly differ
6.2 Synchronization Clusters and Multistability at Linear Variation 131
from the individual frequencies. Besides, steps in the form of well-pronounced

and rather extended plateaus intermitted by a relatively narrow transition
region appear in dependence on their spatial coordinate. As in Chap. 5, we
characterize this effect as cluster synchronization, where under a cluster we
understand a coupled set of oscillators having the same average period T and
the corresponding mean frequency Ω ∼ T −1 , but no demand for a constant
phase difference between the elements which allows for limited variations in
time.
As some control parameter R (e.g., frequency gradient along the chain,
value of coupling, etc.) is changed, the cluster structure is destroyed at R =
Rcr , and then, on passing the critical value, it is regenerated again for R =
Rcr +ΔR but now with a different number of clusters. A chaotic behavior may
appear at R ∈ (Rcr , Rcr + ΔR) and, since the relative share of intervals ΔR
increases with the increase of the frequency gradient and/or the weakening
of coupling, the clusters will eventually disappear in the sea of chaos. A well-
pronounced scenario of the transition to turbulence through cluster fractioning
is also another aspect of high interest in this problem.
Theoretical investigations of cluster synchronization in chains with linear
frequency variations have been carried out for a long time, including modeling
the specific behavior of mammalian small intestine [235, 237–240]. The prob-
lem was formulated and analyzed in a general context in [206] in the frame
of phase equations (see Chap. 5). However, almost simultaneously there were
revealed effects in which amplitude variations are significant [241,242]. A vivid
manifestation of these effects is the formation of clusters of oscillators with
infinitesimal amplitude, even if the conditions of self-excitation in the absence
of coupling are fulfilled for each of them, a phenomenon known as oscillator
death or amplitude death [144, 145, 243–246]. The amplitude effects are also
essential for the formation and restructuring of cluster structures. Therefore,
we employ here equations for slow complex amplitudes, those solution is more
complicated than those obtained from phase equations. Consequently, we have
to use partially solutions obtained numerically.
6.2.1 Model Equations
To demonstrate typical features of such a system, we firstly analyze the

paradigmatic case of a chain of coupled van der Pol oscillators (Sect. 2.2)
having different individual frequencies. Each oscillator is coupled diffusively
with its two nearest neighbors. The model equations read then

d2 xj 2 2 dxj dxj−1 dxj dxj+1
+ (1 + εΔj ) xj − 2ε(p − xj ) = 2εd −2 + .
dt2 dt dt dt dt
(6.1)
Here, j = 1, . . . , N and N is the number of self-oscillators, variable xj
describes the states of ith element, Δj is the frequency mismatch of the oscil-
lators relative to the individual frequency of the first oscillator with ω1 = 1,
ω1 ω2 ω3 ωΝ
Fig. 6.1. Free-ends chain of locally coupled oscillators with different individual
frequencies ωj
and d is the coefficient of coupling between the oscillators. Free-end boundary

conditions (Fig. 6.1) are considered, i.e., x0 (t) = x1 (t), xN +1 (t) = xN (t). The
parameter ε characterizes the smallness of the quantities standing after it when
the asymptotic methods are employed and is omitted in final expressions.
Further consideration, except Sect. 6.5, is carried out in a quasiharmonic
approximation (ε 1) in its traditional interpretation. We introduce two
different time scales, the fast ξ = ωt with ω = 1 + εω (1) + . . .) and the slow
one η = εt. Hence, (6.1) is written, to an accuracy of the terms ∼ ε2 , in the
form
d2 xj d2 xj dxj
(1 + 2εω (1) ) 2 + 2ε + xj − 2ε(p − x2j ) =
dξ dξdη dξ

dxj−1 dxj dxj+1
−2ε Δj xj + d −2 + . (6.2)
dξ dξ dξ
(0) (1)
The solution is sought as an expansion xj + εxj + 0(ε2 ) to an accuracy of
the terms 0(ε2 ) leading to
(0)
d2 xj (0)
+ xj = 0, (6.3)
dξ 2
(1) (0) (0) (0)

d2 xj (1) (0) 2 dxj d2 xj d2 xj
+ x = −2(p − (x ) ) − 2 − 2ω 1 −
dξ 2 j j
dξ dξdη dξ 2
(0) (0) (0)
dxj−1 dxj dxj+1
2(Δj xj + d( −2 + )). (6.4)
dξ dξ dξ
This gives
(0)
xj = zj (η) exp(iξ) + zj∗ (η) exp(−iξ), (6.5)
where the complex amplitudes zj (η) are determined from the resolvability con-
dition for the system (6.3) and (6.4) that reduces in this case to the require-
ment of the absence of resonance terms ∼ exp(±iξ) in the right-hand side of
(6.4). If this is fulfilled, we obtain
dzj
= iΔj zj + (p − zj2 )zj + d(zj+1 − 2zj + zj−1 ); (6.6)
dt
which is a discrete in space analog of the well-known Ginzburg–Landau equa-
tion (for overview, see [247]). Without loss of generality, we can suppose that
ω (1) = 0 because the solutions in the taken approximation are invariant to

ω (1) if their final form is written in the initial variables.
By passing to real amplitudes and phases, zj = ρj exp(iφj ), we obtain a
set of equations consisting of two groups that can be referred to as amplitude
ρj respectively phase φj :
ρ̇j = (p − ρ2j )ρj + d(ρj+1 cos θj+1 − 2ρj + ρj−1 cos θj−1 ), (6.7)
j = 1, ..., N

ρj+2 ρj ρj+1 ρj−1
θ̇j = Δ̄j + d sin θj+1 − + sin θj + sin θj−1 . (6.8)
ρj+1 ρj+1 ρj ρj
j = 1, ..., N − 1
ρ2
φ̇1 = Δ1 + d sin θ1 . (6.9)
ρ1
Here, θj = φj+1 − φj is the phase difference and Δ̄j = Δj+1 − Δj is the
frequency mismatch. We consider the following free-end boundary conditions:
ρ0 = ρ1 ; ρN +1 = ρN ; φ0 = φ1 ; φN +1 = φN . (6.10)
For the case of a linear distribution of individual frequencies along the

chain we have Δ̄j = Δ for all j.
The synchronization conditions are the coincidence of the observed (mean)
frequencies Ωj (third frequency definition in Sect. 2.3)
φj (T ) − φj (0)
Ωj = φ̇j = lim . (6.11)
T →∞ T
The frequencies Ωj can be also calculated as the 2πnj (T )/T ratio, where
nj (T ) is the number of typical features of the time series (e.g., the maxima
exceeding certain values) in the time interval T . Next, typical solutions of this
model will be presented.
A qualitative picture of the spatiotemporal structure of oscillations is
obtained by plotting shadow-graphs of zj (t) on the (j, t)-plane. Chains of
different number of elements at p = 0.5 were investigated in our numerical
experiments.
6.2.2 Global Synchronization in an Assembly, Stationary Phase

Distributions, Synchronization area
The simplest solution of (6.7) and (6.8) is a stable equilibrium state (ρ̇j = 0,
θ̇j = 0) which corresponds to the regime of global synchronization in the
chain, i.e., all oscillators of the chain are in the regime of full synchronization.
In this regime the amplitude equations in a zero approximation (coupling is
weak) give the same oscillation amplitudes for all elements of the ensemble.
Then one can use only phase equations (see Sect. 5.3.1) and by taking into
account the condition Δ̄j = Δ, the system of equations for the stationary
phase differences θ̄j is rewritten in the form (see Chap. 5)
Δ + d(sin θ̄2 − 2 sin θ̄1 ) = 0; (6.12)
Δ + d(sin θ̄j−1 − 2 sin θ̄j + sin θ̄j+1 ) = 0, j = 2, ..., N − 2; (6.13)

Δ + d(sin θ̄N −2 − 2 sin θ̄N −1 ) = 0. (6.14)
The distribution of θ̄j is determined by (see (4.33) and (4.34)):
Δ
sin θ̄j = (N j − j 2 ). (6.15)
2d
It follows from (6.15) that the system (6.13) has 2N −1 steady states (see,
e.g., [4, 242]) and only one of them (for −π/2 < θ̄j < π/2) is stable. As the
frequency mismatch Δ is increased, the condition of synchronization regime
for all elements:
Δ
(N j − j 2 ) < 1, (6.16)
2d
that coincides with the condition for the existence of equilibrium states [4],
is violated first for j = N/2 at even N , i.e., for the middle element of the
chain. Thus, the condition of global synchronization regime in the chain (or
the respective synchronization area) under the suppositions made above for
the values of the parameters of the initial system is given by the inequality

ΔN 2

8d < 1. (6.17)
The correction to the frequency of the synchronized oscillations, Δωc , can be

determined from the equation for the phase φ1 :
φ̇1 = d sin θ̄1 + Δ, (6.18)
such that
Δωc = Δ(N − 1)/2, (6.19)
i.e., the frequency Ωs of the global synchronization regime is equal to the mean
individual frequency in the chain. Moreover (see also Chap. 5), this is true for
all other distributions of the individual frequencies in the quasiharmonic case.
For Δ = 8d/N 2 , we have θ̄N/2 = π/2. In this case, the stable and unsta-
ble equilibrium states merge and a rotatory (with infinite growth of phase
differences θN/2 ) periodic motion is born in the phase space of the system
of equations for the phase difference. Because all the elements of the chain
are coupled, the appearance in the middle element of the regime of unlimited
advance of phase differences between this element and its neighbors leads to
a transition from a stationary regime of global synchronization to the regime

of oscillations θj (t) near a certain constant value of θj , with the amplitude
depending on j in all the elements of the chain except the middle one. The
closer the elements are to the ends of the chain, the smaller the amplitude
of the oscillations is. In the case of a long chain, the current values of θj
are nearly constant (or constant) for the edge elements, i.e., a regime of syn-
chronization occurs. Thus, the chain is divided into two clusters of equal size
(N/2) that consist of mutually synchronized elements at different average fre-
quencies. A more detailed description of cluster synchronization regime will
be presented next.
6.2.3 Regimes of Cluster Synchronization
Depending on the specific values of the parameters, two principal regimes

are realized, as Δ/d is increased. The first of them is the regime of multi-
frequency generation, when most elements of the chain (except, perhaps, the
edge ones) generate different frequencies like in Fig. 6.2a, b. The second one
is the regime of cluster synchronization, when all the oscillators are divided
into a few groups inside each of which all the elements oscillate at the same
(a) (b) (c)

0.2 0.2 0.2
0.15 0.15 0.15

Ωj
Ωj
Ωj
0.1 0.1 0.1
0.05 0.05 0.05
0 0 0
0 100 0 100 0 100
j j j
0.04 0.04 0.04
0.03 0.03 0.03
0.02 0.02 0.02

Δ Ωj
Δ Ωj
Δ Ωj
0.01 0.01 0.01
0 0 0
− 0.01 − 0.01 − 0.01

0 100 0 100 0 100
j j j
Fig. 6.2. Formation of synchronization clusters. Averaged frequencies Ωj and their
difference ΔΩj = Ωj+1 − Ωj for perfect (b) (d = 1) and intermediate (a) (d = 1.2),
(c) (d = 1.45) cluster structures for Δ = 2 × 10−3
0.3
d=0.00
d=0.55
d=1.20
0.25 d=3.80
0.2
Ωj
0.15
0.1
0.05
0
0 50 100
j
Fig. 6.3. Perfect cluster synchronization. Averaged frequencies Ωj for different
values of coupling d for Δ = 2 × 10−3
average frequency (Figs. 6.2b and 6.3). The values of the frequency for each
cluster (except the edge ones) are close to those obtained by averaging the
individual frequencies over all the elements forming the cluster. In the case
of a linear dependence of the frequency on j in Fig. 6.3, this corresponds to
the intersection of the lines Ω = Ωj = φ̇j and Ω = jΔ exactly in the middle
of the cluster. These cluster structures are periodic in time – the frequency
differences between the clusters in such structures coincide and are equal to
the lowest cluster frequency in terms of the amplitude equations (6.6):
Ωn = Δ(N − 1)/(n + 1). (6.20)
The size of the clusters Nn for small Δ may be approximated, to an accuracy

of ±1 element, by the relations for the middle clusters:
N −1
Nn = , (6.21)
n+1
and for the edge clusters:
3N −1
Nn = . (6.22)
2 n+1
Here, N (= 100) is the number of all elements and n(= 2, . . .) is the number of
clusters. The sizes of middle clusters Nn at the instant they break are plotted
in Fig. 6.4. The scaling of Nn in dependence on the parameters Δ and d is
similar to the one that specifies, in the constant amplitude approximation
(|zj | = |z0 |), the limiting size of the chain with free ends in which a global
synchronization regime (6.17) occurs:
1.6
1.4
ln(d / d 0) 1.2
0.8
ln (Δ/Δ 0)
0.6
0.4
0.2
0
0.4 0.5 0.6 0.7 0.8 0.9 1
ln Δ Nn
Fig. 6.4. Scaling properties of cluster structures. Critical values of the frequency
gradient (times) in the range Δ ≈ (0.5 − 17) × 10−3 for d = 1 and the coupling
coefficient (plus) in the range d ≈ 0.3 − 3.8 for Δ = 2 × 10−3 , at which the n-cluster
structure breaks prior to the transition from the n to the (n + 1) cluster depending
on the size of the middle cluster Nn . The scale is logarithmic to an accuracy of
arbitrarily chosen origin; the straight lines correspond to the dependence (6.20)
1/2
8d
Nn ∼ . (6.23)
Δ
The spatiotemporal behavior of cluster structures is illustrated in Fig. 6.5,

where the darker regions mark the higher values of intensities of |zj |2
(Fig. 6.5a) and real parts Re(zj ) (Fig. 6.5b) of the complex amplitudes of
oscillations. Time series of intensities for the middle elements of the chain are
shown in Fig. 6.6. A detailed comparison of the data given in these figures as
well as in Fig. 6.2 leads to the conclusion that indeed perfect cluster structures
are formed here (for d = 0.8; 1.2; 1.8 in Figs. 6.2 and 6.5).
The intensities of both middle elements |z49 |2 and |z50 |2 decays periodi-
cally almost to zero at the cluster boundaries (Fig. 6.6c). With an increasing
distance from the boundary of the clusters, the magnitude of intensity drops
d=0.8 0.9 1.0 1.1 1.2 1.3 1.45 1.6 1.8 1.8
(a) (b)
Fig. 6.5. Space–time diagrams: (a) intensities of |zj |2 and (b) real parts Re(zj )
for Δ = 0.002 and different values of coupling coefficients. The defects, which are
clearly seen as minima (white regions) of the local amplitude, appear regularly at
certain positions on the chain in the case of perfect cluster structures. The spatial
coordinate j = 1, . . . , 100 is plotted on the abscissa axis, and time t ∈ [0, 4000] on
the ordinate axis
decreasing so that the change of the real part of the complex amplitudes zj in
the (j, t)-plane (Fig. 6.5b) represents correctly the phase of zj . It is important
to emphasize that the formation of a defect in the spatiotemporal pattern of
the phase (or Re(zj )), that is visualized as a singularity of the intensity field
of |zj |2 , corresponds to the transition between the clusters.
Since the number of the defects, nD , formed in one period of a perfect
cluster structure is a unity less than the number of clusters n and their repe-
tition rate is T = 2πΩn−1 , the number of defects per unit time ρD is equal to
0.5
|z51(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
0.5
|z50(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
0.5
|z49(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
0.5
|z48(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
(a)
0.5
|z51(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
0.5
|z50(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
0.5
|z49(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
0.5
|z48(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
(b)
0.5
|z51(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
0.5
|z50(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
0.5
|z49(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
0.5
|z48(t)|2
0.4
0.3
0.2
0.1
0
50000 51000 52000 53000 54000 55000 56000
(c)
Fig. 6.6. Intensity time series for the middle elements of the chain for Δ = 0.002:
(a) d = 1.45, (b) d = 1.65, and (c) d = 1.8
nD Δ(N − 1) n − 1
ρD = = . (6.24)
T 2π n+1
Estimates by these formulas are in very good accordance with the data
obtained directly from numerical solutions. In particular, the number of the
defects is equal to 44, 40, and 39 for the case shown in Fig. 6.5a at d=0.8, 1.2,
and 1.8, and to 45, 42, and 38, respectively, when calculated by the formula
(6.24). Note that, when the transitions between the structures with n and n+1
clusters are caused by changes of the coupling coefficient d, the average defect
density changes only slightly at n ≥ 4. At the same time, their relative position
in the (j, t)-plane alters significantly. For example, in Fig. 6.5a it changes from
completely ordered at d = 1.2 to irregular at d = 1.45, and then again to a
regular one but now with a different symmetry at d = 1.8. The time series
undergo corresponding changes too (see Fig. 6.6).
Now, we discuss some consequences of the scalings of the maximal size of
clusters (6.23) and of the number of defects (6.24) for two limiting transitions
to infinitely long chains N → ∞ at a constant interval of oscillator frequencies
Δω = Δ(N −1) = const. In the first of these two transitions, the thermody-
namic one, the coupling coefficient d remains constant dN = D = const. In the
second one, the “continuous,” the coupling coefficient dN = DN 2 , so that the
corresponding second-order difference in (6.1) and (6.6) tends to the second
derivative with respect to the spatial coordinate. As follows from (6.23), the
maximal size of the clusters in the thermodynamic limit changes as
1/2
8DN
Nn ∼ ,
Δω
so that their relative size N̄n /N and the interval of variations of individual
frequencies along the cluster length, N̄n (Δω )/N , tends to zero as the num-
ber of elements is increased. At the same time, the quantities Nn /N and
N̄n (Δω )/N change approximately as [(8N D)/Δω ]1/2 in the continuous
limit. As a result, the regime of global synchronization N1 = N will inevitably
be established as N → ∞. The mean density of defects in the (j, t)-plane, as
is seen from (6.24), will remain constant in either case, of course if we speak
about the range of the parameters in which the number of clusters is much
larger than unity.
Both the picture of synchronization presented above and its description
in a rather general form on the basis of numerical solutions are possible due
to the high degree of symmetry and homogeneity of the problem in a qua-
siharmonic approximation at small frequency gradients and coupling coeffi-
cients. Actually, the meaningful quantity in this approximation is not the
frequency itself but the frequency difference Δj . Consequently, the system
may be regarded to be a homogeneous one at Δj = Δ = const if the edge
effects are neglected. The picture is becoming more complicated, as Δ and d
are increased; that makes the effects of multistability and the changes of the
amplitudes of oscillations along the chain essential.
For applications it is important to note that the observed cluster synchro-
nization regime is sufficiently stable under the influence of fluctuations [249].
6.2.4 Multistability
The most vivid manifestation of the phenomenon of multistability is the for-

mation of structures containing a different number of clusters depending on
initial conditions. To uncover such coexisting structures in numerical experi-
ments, we have adapted the initial conditions to small variations of the para-
meters. The adaptation procedure is as follows: The frequency mismatch Δ
was varied successively by +5 × 10−4 or by −5 × 10−4 . The values from the
steady-state solution obtained in the previous variant were taken as the ini-
tial conditions for zi (t). Although this procedure does not guarantee that
all possible regimes will be found, it enables to reveal qualitatively different
transitions in the domain, where the states possessing a different number of
clusters coexist, i.e., in the region of multistability and in the region of para-
meters where it is absent. The existing domains of the structures which have
a definite number of clusters are shown in Fig. 6.7. We have again to types of
transitions:
– A hard transition, i.e., a transition without intermediate structures occurs

from a state with four clusters to a state with five clusters (Fig. 6.8a).
At that Δ increases by less than 5 × 10−4 .
– A soft transition, i.e., a smooth transition of intermediate structures one
into another, occurs at a much greater interval of variations Δ ≈ 2.2×10−3
(Fig. 6.8b).
Note that a nonmonotonic dependence of the number of clusters on the

magnitude of the frequency mismatch (Fig. 6.9) is observed in the region of
multistability when solutions with the same initial conditions are sought (in
particular, xj (0) = −2 for even j, xj (0) = 2 for odd j, and yj (0) = 0 for
all j). This is evidently due to an intricate structure of the basins of the
corresponding attractors in phase space; the deformation of which leads to
the alternating initial conditions in each of them. This is the basic mechanism
for the formation of the structures of mutually synchronized elements (see
Sect. 6.2.1).
The sophisticated structure of the phase portrait of the system considered
does not exclude that multistable regimes of other types, when the structure
of the clusters rather than their number is changed, may also be observed.
For the verification of this hypothesis we conducted a series of experiments
in which the amplitude and phase distributions formed earlier in the clusters
but now with a different number of elements were taken as initial conditions
(usually, the number of elements in the cluster is changed by Nn = ±1, 2).
(a)
10
5
n
0
0 0.005 0.01 0.015 0.02
(b)
10
5
n
0
0 0.005 0.01 0.015 0.02 0.025 0.03
(c)
10
5
n
0
0 0.01 0.02 0.03 0.04
Δ
Fig. 6.7. The ranges of frequency gradients at which structures with n perfect
clusters of the type shown in Fig. 6.2b are given for coupling coefficients d = 1 (a);
d = 2 (b); d = 5 (c). Multistability is clearly seen for d = 5 there exist the intervals of
Δ where coexisting (i) 4 and 5 ([∼ 0.005; ∼ 0.007]), (ii) 5 and 6 ([∼ 0.008; ∼ 0.0135]),
(iii) 6 and 7 ([∼ 0.0135; ∼ 0.022]) cluster structures
Δ=0.008
(a) Δ=0.0064
(b)
0.7 0.6
Ωj
Ωj
0 0
0 20 40 60 80 100 0 20 40 60 80 100
j j
Δ=0.0075 Δ=0.006
1.2 1
Ωj
Ωj
0.6 0.5
0 0
0 20 40 60 80 100 0 20 40 60 80 100
j j
Δ=0.007 Δ=0.0055
1.2 0.5
Ωj
Ωj
0.6
0 0
0 20 40 60 80 100 0 20 40 60 80 100
j j
Δ=0.0065 Δ=0.005
1 0.6
0.4
Ωj
Ωj
0.5 0.2
0 0
0 20 40 60 80 100 0 20 40 60 80 100
j j
Δ=0.006 Δ=0.0042
1 0.4
Ωj
Ωj
0.5 0.2
0 0
0 20 40 60 80 100 0 20 40 60 80 100
j j
Fig. 6.8. Averaged frequencies Ωj in the transitions from n to n + 1 clusters:

(a) hard transition for d = 5; (b) soft transition for d = 1. The corresponding
regions of parameters are indicated in Fig. 6.7a, c by “◦”
6.3 Oscillation Death 143
1.6
Δ = 0.003
Δ = 0.006
Δ = 0.007
1.4 Δ = 0.008
Δ = 0.009
Δ = 0.01
Δ = 0.013
1.2
1
Ωj
0.8
0.6
0.4
0.2
0
0 20 40 60 80 100
j
Fig. 6.9. Nonmonotonic sequence of the number of clusters (4,5,4,5,6,5,6 upward)
at monotonic variation of the frequency gradient Δ and identical initial conditions
(d = 5). The corresponding values of the parameters are marked in Fig. 6.7 by
“times”
We find that the same cluster structure was always established in the region
of the parameters of interest with such variations of initial conditions.
6.3 Oscillation Death
A distinguishing feature of a cluster synchronization regime with a still further

increase of coupling d and mismatch Δ is the formation of clusters separated
by a region of unexcited oscillators [241, 242]. The formation of such regions
may be interpreted as manifestation of oscillator death [144, 243–246] which
was described for two coupled van der Pol oscillators in Sect. 4.1.2.
This effect is observed in an assembly of coupled oscillators. By virtue of
collective effects, they may be realized either globally, i.e., in all elements, or
locally, i.e., in clusters of neighboring elements.
For a better understanding of the transitions in long chains, the following
interpretation of oscillator death is useful. Consider a linearized equation (6.6)
for one of the oscillators:
żj = iΔzj + pzj + d(zj−1 − 2zj + zj+1 ).

Here, the term containing the oscillation amplitude of the (j − 1)th and
(j + 1)th oscillators, at the breaking of the synchronization regime, may be
regarded as a nonresonant external force, and the term −2dzj that depends
on the magnitude of coupling exerts the same effect as additional losses. If the
losses exceed the amplification (2d > p), there remain only forced oscillations
with amplitude ρj = 2d/|iΔ + p − 2d| that decrease as the mismatch Δ is
increased. Therefore, oscillation death can be observed in the jth element. An
essential aspect in the case of chains is that, with the increase of mismatch Δ,
the synchronization conditions are violated not throughout the whole chain
at once but locally. Namely, in the neighborhood of the weakest element of
the chain, i.e., at the site where the regime of global synchronization breaks
earliest, i.e., in the middle of the chain (see Sect. 6.2.1). In this case, for suffi-
ciently large mismatches, i.e., when the influence of the neighbors is no longer
a resonant one, the coupling acts as effective damping, and for 2d > p the
corresponding element becomes unexcited. As Δ is increased, the region of
oscillation death is expanding so fast that there remain only two clusters that
are not fractioned due to the local desynchronization any longer because the
increase of the parameter Δ/d, that usually leads to a breaking of synchroniza-
tion regime, is compensated by the decrease of the size of the clusters [243].
This is illustrated in Fig. 6.10, where two clusters of synchronized elements
at the edges of the chain are separated by an area of unexcited oscillations.
Therefore, the oscillation death as well as the synchronization is the reason of
7
6 (b)
5
4
Ωj
3
2
1
0
0 20 40 60 80 100
j
0.8
0.7 (a)
0.6
0.5
ρj2
0.4
0.3
0.2
0.1
0
0 20 40 60 80 100
j
Fig. 6.10. Oscillation death. (a) Squared mean amplitudes |ρj |2 and (b) averaged
frequencies Ωj in the case of oscillator death in a chain for d = 5 and Δ = 0.06.
There are two clusters each consisting of 20 mutually synchronized elements at the
ends of the chain
6.4 Effects of Nonuniformity of the Frequency Mismatch Gradient 145
pattern formation in such oscillatory networks. In Sect. 6.4 we try to answer

how such patterns, namely, structures of the clusters of the synchronization
can be controlled.
6.4 Effects of Nonuniformity of the Frequency Mismatch

Gradient in the Formation of Synchronized Clusters
In engineering and physiological applications the differences of the parameters
are not so simple as a linear increase; they are typically more inhomogeneous.
This leads to a more complicated behavior but it also offers new options for
the control of synchronous behavior.
6.4.1 Sensitivity of the Structures to Regular Nonuniformities
We can distinguish at least two mechanisms controlling the spatial structure

of the system of interest. In both additional periodic inhomogeneities of the
frequency mismatch Δj along the chain are introduced. One of them is asso-
ciated with the transformation of an attractor (or attractors), the other one is
attributed to changes only of attraction basins (see, e.g., [233, 234]). It can be
expected that the second mechanism is readily realized in the above-mentioned
case of nontrivial dependence of the number of clusters on the magnitude of
the frequency mismatch gradient. Indeed, for the parameter values and initial
conditions like in Fig. 6.7 (“*” in (c)), a relatively small, periodic along j
correction to the individual frequencies:

2π · n∗ (j − 1)
ωj = Δ(j − 1) + α sin (6.25)
N
leads to a change of the number of clusters formed. Particularly, for Δ = 0.009

and n∗ = 5, the number of clusters is n = 6 if α = 0, and n = 5 if α = 0.0001.
By introducing a perturbation with a much larger amplitude α ≥ 0.002, one
can influence not only the process of cluster formation in the regime of tran-
sition but also the structures that have already been formed. For example,
the transition from six to five clusters occurs for α = 0.002, Δ = 0.009 (see
Fig. 6.11a). This case corresponds to the first mechanism of forcing, namely,
destruction of one of the multistable states, in this case, the structure of six
clusters. Another possible manifestation of this mechanism is the formation
of synchronization clusters from a nonsynchronized state, when the transition
from a chaotic state in space and time to a state where five clusters take place
(Fig. 6.11b).
Note that the expression (6.25) does not give a clear picture of the rela-
tionship between the magnitudes of the uniform and nonuniform components
of the frequency mismatch. Only a comparison either of their gradients or
of their changes over the modulation period along the chain is physically
t=5000 t=1500
0.2 0.2
Δ Ωj
Δ Ωj
0.1 0.1
0 0
− 0.1 −0.1
0 20 40 60 80 100 0 20 40 60 80 100
t=3000 t=1000
0.2 0.2
Δ Ωj
0.1
Δ Ωj
0.1
0 0
− 0.1 − 0.1
0 20 40 60 80 100 0 20 40 60 80 100
t=2000 t=500
0.2 0.2
Δ Ωj
0.1
Δ Ωj
0 0.1
− 0.1 0
0 20 40 60 80 100 −0.1
0 20 40 60 80 100
t=0 t=0
0.2
0.2
Δ Ωj
0.1
Δ Ωj
0 0.1
− 0.1 0
0 20 40 60 80 100 − 0.1
0 20 40 60 80 100
j j
(a) (b)
Fig. 6.11. Controlling cluster structures. Restructuring of a five cluster structure

from a six cluster structure (a) and chaotic state (b) under the action of periodic
inhomogeneity (6.25). The parameters are (a) d = 5, Δ = 0.009, α = 0.002; (b)
d = 2, Δ = 0.005, α = 0.005. The frequencies were determined by averaging over
time Δt = 1, 000 prior to subsequent record of Ωj distribution. The calculations
were performed for the values of the parameters marked by asterisk in Fig. 6.7
meaningful. By comparing these quantities, we see that the perturbation is

determined by the parameter (α/Δ)(2πn/N ) that did not exceed 1/3 in all
the cases considered.
6.4.2 The Effect of Random Dispersion of Individual Frequencies

on Cluster Synchronization
We restrict our consideration to one aspect of the effect exerted by spatially

irregular parameter variations on the formation of the synchronized structures.
Namely, we investigate the dependence of spatial structures on the magnitude
of the random distribution of the mismatch relative to some mean at a con-
stant complete range of frequency variations. The critical values of d averaged
over 20 samples of random numbers, at which the transition between cluster
structures occurs, are presented in Fig. 6.12. This yields that regime of global
synchronization is established at much smaller values of coupling in the case
of a random frequency distribution as compared to a linear frequency distrib-
ution. This effect is actually observed for each realization of random frequency
distribution.
This phenomenon has the following qualitative explanation. In the case of
a linear frequency distribution, the left-hand neighbor of a certain element j
is, on the average, phase lagged (θj > 0); whereas the right-hand neighbor is,
on the average, phase advanced (θj+1 < 0). In other words, they “pull” the
element in opposite directions and, in this sense, their actions are compen-
sated. A possible situation for a random frequency distribution is when both
neighbors either advance a definite oscillator in phase or lag behind it, i.e.,
they “pull” it in the same direction. As a result, the neighboring frequencies
of all the three elements come closer together giving rise to a synchronization
cluster. Apparently, the clusters may emerge at an arbitrary site of the chain.
6.5 Synchronization in a Chain of van der Pol Oscillators 147
60
50
40
30 1
d
20
2
10
3
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Δ ω*
Fig. 6.12. Average critical values of d at which a transition between different cluster
structures occurs. Region 1 corresponds to the regime of global synchronization.
Two and three clusters of mutually synchronized elements exist in regions 2 and 3,
respectively. The averaging was made over 20 sample random individual frequencies.
Δ = 0.002
6.5 Synchronization in a Chain of van der Pol Oscillators
We have already mentioned that the picture of synchronization described

above and simple scalings at linear variation of individual frequencies along the
chain are due to the high degree of symmetry and homogeneity of the problem.
This is true, on the one hand, when the chain is sufficiently long so that the
edge effects do not introduce significant distortions into the clusters. On the
other hand, the quasiharmonic approximation should be valid, ε 1. If this
is not the case the gradient of the individual frequencies is the essential para-
meter. The situation changes here cardinally with the increase of ε because
nonlinear distortions of the shape of oscillations become pronounced. Oscilla-
tions become then of the relaxation type (see Fig. 2.1b). Since these distortions
are greater in the low-frequency region than in the high-frequency one, the
symmetry of the problem is violated. In particular, the amplitude distribution
becomes essentially nonsymmetric (Fig. 6.13) even at a stable synchronization
regime for small Δ, when the amplitude modulation is a few percent. As a
result, the distribution of stationary phase differences of the neighbors that
matches the amplitude distribution is no longer symmetric. Consequently,
the regime of global synchronization is broken in the elements close to the
beginning of the chain rather than in the middle elements (Fig. 6.14). In spite
of the strong dependence of the shape of oscillations on the nonlinearity ε,
one can observe in the interval of its intermediate values ε ≈ 1 an almost
1.44
ε=0.3
ε=0.5
ε=0.7
1.43 ε=1.0
1.42
Aj
1.41
1.4
1.39
1.38
0 50 100
j

Fig. 6.13. Mean amplitudes Aj = x2j + ẋ2j for different values of ε for d = 5 and
Δ = 0.00025
1.008
1.006
(b)
Ωj
1.004
1.002
1
0 20 40 60 80 100
j
1.5
1.4
(a)
Aj
1.3
1.2
1.1
0 20 40 60 80 100
j

Fig. 6.14. (a) Mean amplitudes Aj = x2j + ẋ2j and (b) averaged frequencies Ωj
at different values of ε and Δ for d = 5. The symbols are: “plus” for ε = 0.02
and Δ = 0.0006; “times” for ε = 0.1 and Δ = 0.0006; “stars” for ε = 0.3 and
Δ = 0.00065; and “squares” for ε = 0.5 and Δ = 0.0007. The values of Δ are slightly
larger than the critical values Δcr at which the regime of global synchronization is
disturbed
6.5 Synchronization in a Chain of van der Pol Oscillators 149
linear relationship between ε and the critical value of mismatch Δ∗ , starting

from which the regime of global synchronization breaks. A linear relationship
exists also between ε and the squared size of the smallest of the two clusters
formed as the global synchronization regime breaks. Examples of such rela-
tionships are given in Fig. 6.15. Another sequence of increasing the asymmetry
is a change of the frequency Ωs of the global synchronization regime. With
an increase of the nonlinearity strength ε, the frequency Ωs increases and for
rather large ε is close to the maximal frequency in the chain ωmax = Δ(N −1).
This means that the fastest oscillator dominates in the chain.
Another important consequence follows from Fig. 6.15b. For fixed coupling,
the appearance of global synchronization for a strong nonlinearity takes place
for a larger frequency mismatch than for a weak nonlinearity.
From an application point of view a transient time to synchronous regime
is a very important characteristic. Here also chains of quasiharmonic oscilla-
tors and chains of relaxation oscillators demonstrate quite different proper-
ties: Relaxation oscillators are relatively fast if compared with quasiharmonic
oscillators (Sect. 2.2). The former ones need only a few cycles to synchro-
nize [153, 248].
0.0012
0.001
0.0008
(b)
Δ∗
0.0006
0.0004
0.0002
0 0.5 1 1.5 2 2.5 3
ε
3000
2500
2000
(N* )2
1500
(a)
1000
500
0
0 1 2 3
ε
Fig. 6.15. (a) The dependence on ε of the critical value of mismatch Δ∗ starting
from which the regime of global synchronization is not realized; (b) the dependence
on ε of the squared quantity N ∗ = 50 − Nc , where Nc is the length of the smaller of
the two synchronization clusters that are formed after breaking the regime of global
synchronization, d = 5
6.6 Conclusions
In this chapter the collective behavior of a chain of diffusively coupled non-

identical van der Pol oscillators at weak and relatively strong nonlinearity has
been treated.
Asymptotic and numerical methods to obtain synchronization features for
linearly and randomly distributed along the chain individual frequencies have
been used.
The main findings are:
– Typical features for the onset and existence of regimes of global (all-to-all)
and cluster (partial) synchronization have been explored.
– Two scenarios, soft and hard, of the transitions between the structures con-
sisting of a different number of synchronization clusters have been revealed.
In the first case, a gradual tuning of the spatial distribution of averaged
frequencies is observed, while in the second one, the transition from the
structure of n synchronized clusters to the structure of n+1 clusters occurs
in a stepwise fashion as a consequence of multistability.
– The effect of different types of individual frequency distributions on syn-
chronization in a constant range of frequency variations is investigated. It
is revealed that the characteristics of the synchronization are strengthened
when an irregular distribution of individual frequencies is used.
– The oscillation death effect has been observed. This effect can also lead to
the creation of synchronized cluster structures.
– Possible mechanisms to control synchronization structures based on the
application of nonhomogeneous forces have been proposed.
– The difference of the transitions to global synchronization regime for weak
and strong nonlinearity has been clarified. For a strong nonlinearity in
the chain with a fixed frequency dispersion, global synchronization regime
occurs at smaller coupling than for weak nonlinearity.
The discrete Ginzburg–Landau equation may be regarded as a model of an

arbitrary nonequilibrium medium near a critical point (Andronov–Hopf bifur-
cation). Consequently, the presented results may be extended to a broad class
of discrete media in different fields. Moreover, many of the effects observed in
chains of coupled periodic oscillators also occur in regimes of chaotic phase
synchronization in a chain of coupled Rössler oscillators (Chap. 7), in a chain
of coupled intermittent oscillators (Chap. 8), in a chain of coupled circle maps
(Chap. 9), and in oscillatory networks with complex topology (Chap. 14).
7
Ensembles of Chaotic Oscillators
with a Periodic-Doubling Route
to Chaos, Rössler Oscillators
Synchronization in ensembles of coupled chaotic oscillators is a topic of great

interest, due to its high theoretical significance and many applications in
a variety of fields including neural networks [250–256], electronic circuits
[257–259], optics [260–264], chemistry [67, 265, 266], etc.
In this chapter we extend our studies to chains of continuous in time
chaotic oscillators. We are interested in whether the phenomena typically
encountered in networks of periodic oscillators (see Chaps. 5 and 6) can be
observed for chaotic systems as well. The main effects discussed here are
(1) the development of global phase synchronization regime, i.e., all elements
of the chain are synchronized in phase and (2) the formation of several clusters
of phase synchronized oscillators. We also present properties of the collective
behavior inherent in chaotic networks.
The chapter is organized as follows. Section 7.1 is devoted to the study
of synchronization effects in a chain of identical Rössler oscillators, which is
then extended to nonidentical oscillators and we start with a phase-coherent
chaotic attractor. After describing the model in Sect. 7.2, phase and frequency
definitions as well as criteria for synchronous behavior in a chaotic oscillator
networks are discussed. In Sects. 7.3 and 7.4 we present synchronization effects
in chains of coupled Rössler oscillators with a linear and a random distribution
of natural frequencies. Section 7.5 is devoted to synchronization of Rössler
oscillators with a nonphase-coherent attractor, the funnel attractor. Effect of
anomalous collective behavior of coupled chaotic oscillators is discussed in
Sect. 7.6. In Sect. 7.7 we summarize the chapter.
7.1 Synchronization Effects in a Homogeneous Chain

of Rössler Oscillators
In homogeneous networks of Rössler or other continuous in time chaotic sys-

tems complete synchronization is the main interesting phenomenon. In this
152 7 Phase Synchronization in Ensembles of Chaotic Oscillators
section, following [267–269] we consider N identical diffusively coupled arbi-

trary oscillators with periodic boundary conditions:
u̇j = f (uj ) + dE(uj+1 − 2uj + uj−1 ) (7.1)
where j = 0, 1, . . . , N − 1, uj ∈ R , the function f : R → R is nonlinear, d
n n n
is a scalar coupling parameter, and E = diag(e1 , . . . en ) is a constant diagonal

diffusion matrix with 0 ≤ ei ≤ 1. The regime of global synchronization corre-
sponds to a spatial homogeneous state u0 = u1 = ... = uN −1 , which defines a
n-dimensional invariant manifold M . In M the dynamics is governed by the
equation of the uncoupled oscillator ṡ = f (s). The stability of the synchronous
state is determined by linearizing (7.1) around s(t). This leads to
ξ˙j = Jf (s)ξj + dE(ξj+1 − 2ξj + ξj−1 ), (7.2)
where ξj = uj − s, Jf (s) is the Jacobian of f at s(t). The linear stability
equations (7.2) can be diagonalized by expanding into spatial Fourier modes,
−1
√ N
ξj = (1/ N ) ηk exp(−2πijk/N ). (7.3)
k=0
Carrying this out yields linearized equations for the Fourier amplitudes ηk
η̇k = [Df (s) − 4d sin2 (πk/N )E]ηk , (7.4)
where k = 0, 1, . . . , N −1. Solving (7.4) for a special function f (s) and a special
solution s(t), we obtain all Lyapunov exponents (LEs) which determine the
stability of the synchronization manifold M . In [267,268] it was found that for
coupled Rössler systems increasing coupling d can lead to destabilizing of the
synchronous state. This phenomenon is called a short-wavelength bifurcation.
In addition to global and cluster synchronization regimes in ensembles of
identical systems, the phenomena of antiphase and in-phase–antiphase syn-
chronization were observed [270–274]. These types of synchronization are
defined by the existence of stable linear transversal invariant manifolds.
Such antiphase synchronization is observed in a system of coupled oscillators
where all corresponding variables of oscillators are equal with opposite sign.
In in-phase–antiphase synchronization, one set of the corresponding variables
is equal, whereas the other one is equal with opposite signs. Pattern forma-
tion and synchronized chaos was also studied in two-dimensional lattices of
identical Rössler oscillators [275].
7.2 Basic Model of a Nonhomogeneous Chain, Phase

and Frequency Definitions, and Criteria of Phase
Synchronization
Our basic model is now a chain of nonidentical Rössler oscillators with nearest-
neighbor diffusive coupling, i.e., with the same coupling type as in the previous
chapters. The dynamical systems are [165]:
7.2 Basic Model of a Nonhomogeneous Chain 153
ẋj = −ωj yj − zj ,
ẏj = ωj xj + ayj + d(yj+1 − 2yj + yj−1 ), (7.5)
żj = 0.4 + (xj − 8.5)zj ,
j = 1, . . . , N.
Here N is the number of the oscillators in the chain and d is the coupling
coefficient. The parameter ωj corresponds to the natural frequency of the
individual oscillator. Like in the previous chapters, we treat two cases (1) a
linear distribution of natural frequencies ωj = ω1 + Δ(j − 1), where Δ is the
frequency mismatch between neighboring elements and (2) a random distrib-
ution of natural frequencies in the range [ω1 , ω1 + Δ(N − 1)].
We assume again free-end boundary conditions:
y0 (t) = y1 (t) ; yN +1 (t) = yN (t) . (7.6)
Because the Rössler system typically has periodic windows as the parameter
ωj is varied, we choose ω1 and Δ in such a way that at least large periodic
windows are avoided.
In order to study phase synchronization effects here, one needs to have
appropriate phase and frequency definition. In Sect. 2.3 we have proposed a
rather general definition of phase which works well for both phase-coherent
and funnel attractors. Correspondingly, we define for each element in the chain
the phase as:
ẏj
φj = arctan . (7.7)
ẋj
As the phase is well defined, one can straightforwardly calculate the phase
difference between neighboring oscillators θij = φi − φj . This enabled to
analyze synchronization phenomena in chains as follows. If the phase dif-
ference does not grow with time, but remains bounded, i.e.,
|θij (t)| < Const (7.8)
we have a 1:1 phase locking between the ith and the jth oscillators in the
chain. Note that these oscillators are not necessary the neighbors. A weaker
condition of synchronous motion is the coincidence of the averaged partial
frequencies:
Ωi = Ωj , (7.9)
which can be calculated as shown in Sect. 2.3.
If
Ωj = Ωs , j = 1, . . . , N (7.10)
a global synchronization regime occurs. In (7.10) Ωs is the common mean
frequency for all elements in the chain, i.e., the frequency of global synchro-
nization regime.
7.3 Phase Synchronization in a Chain with a Linear

Distribution of Natural Frequencies, Phase-Coherent
Rössler Oscillators
As was discussed in Sect. 4.2.1 for two coupled Rössler oscillators, the transi-
tions to chaotic phase synchronization (CPS) regime strongly depend on the
topology of chaotic attractor, namely, whether the chaotic attractor is phase
coherent or funnel? Here we consider both cases separately.
In order to underline the similarity of synchronization phenomena in net-
works of periodic and chaotic systems, we start with a chain of phase-coherent
Rössler oscillators.
7.3.1 Theoretical Study
The transition to CPS can be better understood by transforming the original

system to cylindric coordinates. By introducing
φj =
arctan(yj /xj ),
ρj = x2j + yj2 , (7.11)
zj = zj ,
we obtain
ρ̇j = −zj cos φj + aρj sin2 φj +

d sin φj (ρj+1 sin φj+1 − 2ρj sin φj + ρj−1 sin φj−1 ),
φ̇j = ωj + a cos φj sin φj + zj sin φj /ρj +
d cos φj (ρj+1 sin φj+1 − 2ρj sin φj + ρj−1 sin φj−1 )/ρj , (7.12)
żj = 0.4 + (ρj cos φj − 8.5)zj ,
j = 1, . . . , N.
Introducing the phase difference variable θj = ψj+1 − ψj for the slow

phases ψj = φj − ω0 t, averaging (7.12) (see also Sect. 4.2.1), and taking into
account that in the zero approximation of coupling and frequency mismatch,
the averaged ρj is Const for all j = 1, . . . , N , one gets:
d
θ̇j = Δj + (sin θj−1 − 2 sin θj + sin θj+1 ), (7.13)
2
j = 1, . . . , N − 1,
where Δj = ωj+1 − ωj . Hence, we obtain analogous phase difference equations

as already studied for coupled phase systems (Chap. 5) resp. coupled limit-
cycle oscillators (Chap. 6).
From (7.13) and for a linear frequency distribution, i.e., for Δj = Δ =
const, we find the distribution of the averaged stationary phase differences θ̄j :
7.3 Phase Synchronization in a Chain with a Linear Distribution 155
Δ
sin θ̄j = (N j − j 2 ). (7.14)
2d
and the condition for the onset of global CPS:

ΔN 2

8d < 1, (7.15)
Our computer simulations show very good agreement between theoretical and
numerical results. This supports the idea that synchronization of chaotic oscil-
lators with a rather simple topology, especially, phase-coherent attractors, is
very similar to the case of synchronization of periodic oscillators (Chaps. 5 and
6). However, there is the important difference that the phase difference θj now
is not a constant, but chaotically fluctuates driven by the chaotic behavior of
the amplitudes.
7.3.2 Numerical Results
We perform numerical simulations for chains of N = 20 − 50 oscillators for

different values of the parameters Δ, ω1 , d. The main quantities calculated are
the observed frequencies Ωj and the spectra of LEs. Generally, by increas-
ing the coupling d, all these frequencies Ωj become equal; i.e., global phase
synchronization regime sets in. As in the case of coupled periodic oscillators
and coupled chaotic circle maps, the regime of global synchronization in the
chain (7.5) can again appear in two ways, depending on the relatively fre-
quency mismatch Δ/ω1 . Below these two scenarios, referred to as soft and
hard transitions, are described.
Small Frequency Mismatch: Soft Transition to a Synchronous

State
We first consider a relatively small frequency mismatch Δ/ω1 1. With

increase of coupling, mutual synchronization occurs at both ends of the chain,
i.e., first only elements with small resp. large j are synchronized (see Fig. 7.1).
With a further increase of d, an increasing number of oscillators join the syn-
chronous regions, so they become more and more extended (Fig. 7.1). Finally,
for d exceeding a critical value d∗ , all oscillators are synchronized, i.e., all
mean frequencies Ωj coincide. This smooth transition to global synchroniza-
tion regime is a soft transition similar to that one obtained for chains of
periodic oscillators. But due to the high level of chaoticity the distribution
of mean frequencies is not symmetric and the frequency of global synchro-
nization regime is not equal to the averaged observed frequency of oscillators.
With the transition to global synchronization regime, the amplitudes of the
oscillators remain chaotic, which is clearly marked by the Lyapunov spectrum
shown in Fig. 7.2. The number of positive LEs remains N even in the regime
of global synchronization at d > d∗ . As has been mentioned in Sect. 4.2, the
1.039
d=0
0.003
1.038 0.006
0.009
1.037
Ωj
1.036
1.035
1.034
0 4 8 12 16 20
j
Fig. 7.1. Soft transition to global synchronization regime in a chain of Rössler
oscillators (7.5). Averaged frequencies Ωj for different values of coupling d. The
parameters are: N = 20, Δ = 0.0002, and ω1 = 1. The frequency of global synchro-
nization regime is less than the mean individual frequency
0.08
d = 0.003
0.006
0.009
0.06
Lyapunov exponents
0.04
0.02
0.00
−0.02
0 10 20 30 40
k
Fig. 7.2. Forty largest LEs for different coupling for the regimes reported in Fig. 7.1
appearance of PS in a system of two coupled oscillators manifests itself in

the Lyapunov spectrum, namely, one of the zero exponents becomes negative,
while the two largest ones remain positive. For the chain, the N largest LEs
remain positive, while only one zero exponent survives, and N − 1 become
negative. Then, a high-dimensional synchronized chaos is observed where the
number of positive LEs is equal to the number of the coupled elements, i.e.,
it is clearly a hyperchaos–hyperchaos transition.
The regime of global phase synchronization is rather sensitive to external
noise. We have simulated the dynamics of the chain (7.5) with noisy Gaussian
terms added to the right-hand side of the equations for xj and yj . Even
small noise leads to a nonconstant distribution of the observed frequencies;
so, strictly speaking, global synchronization regime is not observed. Noise
destroys full phase synchronization regime in the chain in the same way as it
destroys it in a system of two coupled oscillators [98], i.e., noise-induced phase
slips occur.
Large Frequency Mismatch: Disorder Reduction and Clustering
For a relatively large frequency mismatch Δ/ω1 , we first observe a reduction of

chaos. As the coupling d grows, the number of positive LEs decreases. Before
any synchronization effects set in, only a few LEs remain positive (Fig. 7.3).
Synchronization regime now is created in the form of clusters (see Fig. 7.4):
the oscillators are divided into groups having the same observed frequency,
0.04
d = 0.03
0.06
0.18
0.02 0.7
Lyapunov exponents
0.00
−0.02
−0.04
0 10 20 30 40
k
Fig. 7.3. Forty largest LEs in a chain of Rössler oscillators (7.5). The parameters
are N = 50, Δ = 0.009
1.5
1.4 d = 0.0
0.18
0.7
1.3
3.0
5.0
Ωj
1.2
1.1
1
0 10 20 30 40 50
j
Fig. 7.4. Hard transition to global synchronization regime in a chain of Rössler
oscillators (7.5). Averaged frequencies Ωj for different values of coupling d. The
parameters are N = 50, Δ = 0.009, and ω1 = 1
with relatively large frequency difference between the groups. As in the case
of coupled periodic oscillators with a relatively strong frequencies mismatch,
the transition to global synchronization regime and transitions between clus-
ter structures occurs through jumps in the observed frequency distributions.
Hence, we have hard transitions. As a rule, with an increase of d, the clusters
of mutually synchronized oscillators appear rather abruptly. With a further
increase of coupling, the width of the clusters grows in parallel, i.e., the num-
ber of clusters decreases, and, finally, only one cluster remains, and global
synchronization regime is formed. The cluster formation is clearly visible in a
space–time diagram (Fig. 7.5). In all panels a gray scale is used with minimal
values being represented by white and maximal by black. The left panel shows
sin(φj ) = yj /Aj (for system (7.11)); hence the white stripes correspond to the
phase ≈ −π/2 and the black stripes to the phase ≈ π/2. The right panel
shows the amplitudes Aj of the oscillators. To characterize the instantaneous
phase difference between neighboring oscillators, we plot in the center panel
the quantity
φj+1 (t) − φj (t)
sj = sin2 ( ) (7.16)
2
which is zero if the phases are equal and one if they differ by π.
This presentation exhibits the defects, which are clearly seen as maxima
(black regions) of sj and minima (white regions) of the local amplitude. They
appear regularly at certain positions in the chain. Hence, the border between
the clusters is sharp (see Fig. 7.5). Obviously, the frequency difference between
Fig. 7.5. Observed frequencies Ωj and space–time structures in a chain of 50 coupled

oscillators (7.5) with Δ = 0.009 and different couplings. (a) d = 0.03: no clusters are
observed, although relatively large regions of phase coherence are seen. (b) d = 0.06:
first clusters appear, but the defects are extended in time. (c) d = 0.18: a regular
train of defects is observed. (d) d = 0.6: at the transition from seven to six clusters
an irregular sequence of defects is seen near the right edge of the chain. (e) d = 0.7:
a regular train of well-localized defects
the clusters (beat frequency) is equal to the frequency of the appearance of

defects. Near a transition at which the number of clusters changes, the defects
appear irregularly in both space and time (see Fig. 7.5d), and the border
between the clusters becomes smeared.
The difference in the chain dynamics for small and large frequency mis-
matches directly corresponds to the properties of two interacting systems dis-
cussed in Sect. 4.2.1. First let us mention that a larger frequency mismatch
requires a larger coupling for synchronization regime to occur. We have shown
in Sect. 4.2.1 that for small couplings the frequency difference can be arbi-
trary small; therefore, with an increase of coupling a smooth transition to
synchronization regime is observed in a chain. Contrary to this, for large cou-
plings the frequency difference is either zero or finite; therefore, synchronous
clusters are formed with jumps between them.
Oscillation Death
If the coupling between the elements is not very small, the interaction can
lead not only to synchronization, but also to a suppression of oscillations.
This effect, known as oscillation death, is observed both for pairs and chains
of periodic oscillators (see Sects. 4.1.2 and 6.3). This loss of self-excitation
of two chaotic oscillators due to interaction has been discussed in Sect. 4.2.1.
Here we demonstrate that this effect also appears in a chain of coupled chaotic
oscillators.
To explain it, we rewrite the second equation of (7.5) as
ẏj = ωj xj + (a − 2d)yj + d(yj+1 + yj−1 ). (7.17)
Then it becomes clear that the influence of the coupling can be considered as
some additional damping acting on the system. For large enough frequency
mismatch, the force from the neighboring oscillators is not resonant and does
not compensate the increased losses. As a result, if 2d > a the oscillator can
leave off the self-excited regime, and the oscillations decay or “die out.”
This effect can occur locally in chains of chaotic oscillators, and can go
hand in hand with synchronization. This is illustrated in Fig. 7.6, where a
state with two synchronous clusters near the ends of the lattice is separated
by nonoscillating elements. Note that oscillation death in all oscillators is
also possible.
7.4 Synchronization in a Chain with Randomly

Distributed Natural Frequencies
Now we describe effects of coupling on a chain with randomly distributed
natural frequencies (cf. [278, 279]). As in the case of linear increasing frequen-
cies, the regime of global synchronization arises via the formation of clusters
(hard transition) (Fig. 7.7).
As for regular oscillators (see Chaps. 5 and 6) the essential difference is that
for the same mismatch between the largest and the smallest partial frequen-
cies ωj , global synchronization regime appears for considerably lower values
7.4 Synchronization in a Chain with Randomly Distributed Natural Frequencies 161
Fig. 7.6. Oscillation death in a chain of Rössler oscillators (7.5). Space–time dia-
grams of the evolution of yj , sj , and Aj . In the middle of the chain (j=23–38) the
oscillations are suppressed due to the interaction, i.e., oscillation death is observed.
The parameters are N = 50, Δ = 0.015, ω1 = 1, and d = 0.75
of coupling than for a linear distribution. Qualitatively, this is easy to explain.

For the case of linearly distributed frequencies, the left neighbor of an element
is in average behind in phase, and the right neighbor is, respectively, ahead.
Hence they “pull” the oscillator in different directions, and in this sense their
actions are compensated. For the random case it is possible that both neigh-
bors are behind (or ahead) in phase, and, respectively, both speed the element
down (or up). As a result, their frequencies tend to each other, and these ele-
ments form a synchronous cluster. Such clusters can arise in arbitrary places
in the chain and coexist with the oscillators that do not belong to any cluster.
With an increase of coupling, the clusters are firstly formed at the location
of elements with a smaller frequency mismatch. We note that distributions of
mean frequencies do not depend on initial conditions, i.e., for each random
distribution of partial frequencies in (7.5) there exists only one attractor.
Figure 7.7 demonstrates the effect of nonlocal synchronization (see Chaps. 5
and 6), where an oscillator or a cluster of oscillators is synchronized not to
a nearest oscillator or clusters of oscillators, but to some remote oscillator or
cluster in the chain.
1.10
1.08
1.06
1.04
1.02
1.10
1.08
1.06
1.04
1.02
1.10
1.08
1.06
1.04
1.02
1.10
1.08
1.06
1.04
1.02
1.10
1.08
Ωj
1.06
1.04
1.02
0 10 20 30 40 50
j
Fig. 7.7. Averaged frequencies Ωj in a chain of Rössler oscillators with ran-
domly distributed natural frequencies ωj in the interval [1, 1.05]. The number of
elements N = 50, ω1 = 1.0. From bottom to top different coupling strengths
d = 0, 0.01, 0.02, 0.05, 0.2. Effect of nonlocal synchronization is clearly seen for
d = 0.05
7.5 Phase Synchronization of Rössler Oscillators

with the Funnel Attractor
Now we go to the more complicated case of nonphase-coherent oscillators,
especially to the oscillators with the funnel attractor (see Sect. 2.3). Such an
attractor has two different time scales, which prevent global synchronization
in large ensembles: The competition between a large number (2N ) of different
7.5 Phase Synchronization of Rössler Oscillators with the Funnel Attractor 163
1.06
d=0
0.1
0.5
1.04
5
10
Ωj / Ω1
1.02
1.00
0.98
0 10 20 30 40 50
j
Fig. 7.8. Averaged frequency ratios Ωj /Ω1 in a chain of nonphase-coherent Rössler

oscillators with linearly distributed natural frequencies ωj . N = 50, a = 0.23,
ω1 = 0.98, Δ = 0.001
time scales makes the achievement of cluster synchronization rather difficult

and sometimes even impossible (see Sect. 4.2). For example, for linear increas-
ing ωj we observe in Fig. 7.8 neither global and even not cluster phase syn-
chronization (besides small clusters at the ends of the chain). However, there
are some couplings for which averaged frequencies for neighboring elements
locating around the middle of the chain are very close (d = 5, Fig. 7.8). A
detailed analysis shows that these frequencies are slightly different and syn-
chronization regime does not occur here at all.
In order to show that some ordering of cooperative behavior for funnel
attractors can nevertheless be observed we study the chain consisting of iden-
tical oscillators in the funnel regime. As the coupling increases, relatively
large regions of coherent rotation of the oscillators are formed (Fig. 7.9).
These regions are separated by defects that appear spontaneously due to
locally irregular phase slips. It is interesting that their lifetime is relatively
large compared to the cluster regime of the phase-coherent attractors. This
is because after a slip the phase difference between neighboring oscillators is
≈π, which roughly corresponds to the unstable but stationary configuration
of the phases. The relaxation to the stable phase difference ≈0 is therefore
slow.
To describe the synchronization processes of the funnel attractors quanti-
tatively, we characterize the instantaneous phase difference by sj (7.16), where
the phases are defined according to (7.7). The average of sj then quantifies
the degree of synchronization.
Fig. 7.9. Ordering in a chain of identical Rössler oscillators with funnel attractors
(a = 0.23). Space–time diagrams of evolution of yj /Aj , sj , and Aj for different
couplings: (a) d = 0.02 and (b) d = 0.05. The values of yj (t) normalized to the
amplitude are depicted in order to make the phase dynamics visible [yj /Aj = sin φj ]
In Fig. 7.10 the dependence of s (for independent oscillators it is obviously

0.5) on the coupling is shown for different parameter values of the Rössler
attractor. While for the phase-coherent case global synchronization regime
appears for very small couplings, in the funnel case a rather slow decay of s
is observed. Therefore the tendency to more ordered (not necessary synchro-
nous) behavior is clearly seen. The same situation takes place for randomly
7.6 Anomalous Collective Behavior of Coupled Chaotic Oscillators 165
0.5
a = 0.19
0.21
0.4 0.23
0.25
0.27
0.3
s
0.2
0.1
0.0
0.00 0.05 0.10 0.15 0.20
d
Fig. 7.10. Averaged value of the phase difference s = sj vs. coupling d for dif-
ferent values of the parameter a in a chain of identical Rössler systems with funnel
attractors
distributed ωj . Therefore for funnel attractors GS and CPS are coming in

parallel.
We will note that by the study of synchronization phenomena in ensem-
bles of rather complex oscillators other technique, especially recurrence plot
technique [280], can be effectively used.
7.6 Anomalous Collective Behavior of Coupled

Chaotic Oscillators
As we have shown before for coupled Rössler systems, the tendency to get a
more coherent behavior with an increase of coupling is well pronounced. How-
ever, this tendency is not general. In order to measure the degree of synchro-
nization in a chain of interacting oscillators, we define the frequency disorder
as the standard deviation σ(d) = (Ωj − Ωj )2 of all oscillator frequencies Ωj
depending on the coupling d. For phase-coherent Rössler systems we expect
a decrease with d, as shown for 500 coupled Rössler oscillators (6.5) with
periodic boundary conditions and randomly distributed ωj . Then we analyze
analogous a chain of coupled foodweb oscillators modeled by [31, 281]:
x˙j = a (xj − x0 ) − α1 xj yj
y˙j = −bj (yj − y0 ) + α1 xj yj − α2 yj zj + d
N i (yi − yj ) (7.18)
z˙j = −c (zj − z0 ) + α2 yj zj .
It describes a three-level “vertical” food chain, where the vegetation x

is consumed by herbivores y which themselves are preyed upon by the top
predator z. In the absence of interspecific interactions, the dynamics is linearly
expanded around the steady state (x0 , y0 , z0 ) with coefficients a, b, and c that
represent the respective net growth and death rates of each species. Predator–
prey interactions are introduced via Lotka–Volterra terms with strength α1
and α2 .
Despite their minimal structure, the equations are able to capture com-
plex dynamics which matches real data for example in the Canadian hare–lynx
cycle [31,281–283]. In some parameter range this model shows phase-coherent
chaotic dynamics, where the trajectory rotates with a nearly constant fre-
quency in the (x, y)-plane but with chaotic dynamics that appear as irregular
spikes in the top predator z (Fig. 7.11). This behavior of the foodweb model
Roessler Foodweb
20
30
10
20
0
y
−10 10
−20 0
−20 −10 0 10 20 0 10 20 30
x x
2.0 1.5
N=500 N=500
1.5
1.0
σ (%)
σ (%)
1.0
0.5
0.5
0.0 0.0
0.0 0.1 0.2 0.0 0.1 0.2
d d
Fig. 7.11. Comparison of the transition to synchronous regime in a chain of 500

locally coupled Rössler systems 7.5 (left) and foodweb models (7.18) (right). Oscil-
lators have been coupled in the y-variable with strength d to next neighbors in a ring
with periodic boundary conditions. Initial values were set randomly. Top: attractor
projection of the uncoupled system in the (x, y)-plane. Bottom: standard deviation
of frequencies, σ(d), as a function of coupling strength. Parameter values: Rössler
system a = 0.15; Foodweb model x0 = 1.5, y0 = 0, z0 = 0.1, α1 = 0.1, α2 = 0.6,
a = 1, c = 10. Parameters bj and ωj are taken in both systems as uniformly distrib-
uted random numbers in the range 0.97 ± 0.025
7.7 Conclusions 167
is reminiscent to the Rössler system and therefore one might expect similar
synchronization properties in both systems. To explore this in more detail,
we compare the transition to synchronous regime in coupled chains of Rössler
and foodweb systems. Quenched disorder is introduced by taking bj and ωj
for each oscillator from the same statistical distribution. Despite the fact that
both systems have a very similar attractor topology, we find fundamental
differences in their response to the interaction. For the ensemble of Rössler
systems, the onset of synchronization regime is as expected and σ(d) decreases
monotonically with increasing coupling strength, in accordance to the above
theory. In contrast, the ensemble of foodweb models shows a totally different
behavior. Here, with increasing coupling the frequency disorder is first ampli-
fied leading to a maximal decoherence for intermediate levels of coupling. Only
for much larger coupling strength, the frequency disorder is reduced again and
global synchronization regime sets in. This unusual increase of disorder with
coupling strength is called as anomalous phase synchronization [284]. This
effect emerges because the interaction may perturb the oscillators away from
their attractors. This brings the nonisochronicity – amplitude dependence of
frequency – of the oscillation into play. Disorder enlargement occurs if the
nonisochronicity has positive covariance with the natural frequency of the
oscillation. For the foodweb model this is exactly the case, whereas the oscil-
lations in the phase-coherent Rössler system are practically isochronous.
7.7 Conclusions
In this chapter we have analyzed phase synchronization effects in a chain of
diffusively coupled chaotic Rössler oscillators.
– When the individual attractor is phase coherent, the dynamics of a chain
is similar to that of a chain of regular oscillators (Chaps. 5 and 6). In a
nonhomogeneous chain, synchronization regime appears when the coupling
exceeds some threshold.
– We have again found two scenarios of synchronization transitions: in the
soft one a gradual adjustment of the mean frequencies is observed, while in
the hard one intermediate clustered states occur. The borders of clusters
appear in the space–time diagrams as positions where phase defects take
place. We have demonstrated that these defects can be both periodic and
irregular. These two scenarios directly correspond to the synchronization
properties of two interacting systems (see Chap. 4): for small coupling the
mean frequencies are adjusted gradually, while for large couplings a virtual
jump is observed.
– In the soft transition global synchronization regime is accompanied by a
high chaoticity of the behavior: the number of the positive LEs is equal to
the number of coupled elements, while in the hard transition the disorder
is strongly reduced: only a few LEs remain positive.
– The phenomenon of chaotic oscillation death occurs in the chain of Rössler

oscillators with linearly distributed natural frequencies.
– If the dynamics of the individual system is more complex, like the funnel
attractor in the Rössler oscillator, phase synchronization regime is rather
difficult or even impossible to achieve. But the tendency to a more coherent
behavior is well expressed with an increase of coupling.
– In the systems with nonisochronous oscillations anomalous collective
behavior is observed.
A work similar to the study presented above was done in [276, 277], where
one- and two-dimensional lattices of identical Rössler oscillators have been
considered. Observed in [276], the effect of the appearance of a macroscopic
mean field for very small couplings can be interpreted in our terms as the
appearance of a phase-synchronous state. In the case of nonidentical oscillators
the transition is, however, nontrivial, as we have shown in this chapter.
To conclude, the presented phase synchronization effects in chains of cou-
pled chaotic elements support the idea that phase synchronization is a univer-
sal phenomenon of coupled chaotic systems and is similar to synchronization
in networks of periodic oscillators.
8
Synchronization of Intermittent-Like
Oscillations in Chains of Coupled Maps
In the previous chapters synchronization phenomena in ensembles of

time-continuous oscillators are treated. Many systems in nature and tech-
nology and their corresponding mathematical models are, however, discrete
in time, e.g., population dynamics [285, 286], asteroidal motion [287], many
systems under external force [288, 289], laser dynamics [290]. Because of that
it is important to consider synchronization effects in ensembles of coupled
time-discrete elements.
This chapter is devoted to the study of synchronization phenomena in
chains of coupled maps demonstrated chaotic intermittent behavior.
As was mentioned in the previous chapters, the study of chaotic phase
synchronization requires the knowledge of equations for the evolution of the
phase variables (as it is for coupled Rösller oscillators or circle maps) or at
least an appropriate definition of phases (see Chap. 2). Although, there are so
far no unambiguous methods to obtain such equations resp. phase definitions,
in some cases specific properties of the chaotic attractors allow to retrieve the
phases of chaotic oscillations in a rather straightforward way. Additionally
to the class of phase-coherent oscillators (cf. Chap. 7), where chaos appears
through a period doubling cascade, a suitable phase can be also easily intro-
duced for systems with intermittent-like behavior, especially for those with
typical chaotic oscillations with different types of intermittency, spiking, and
bursting neurons. In this chapter we investigate the collective dynamics in
chains of such discrete in time chaotic systems – chaotic maps. Such chains
are important for the understanding of mechanisms behind the transition
from low-dimensional chaos (which may correspond to synchronized chaotic
systems) to developed (spatiotemporal) turbulence that often looks like inter-
mittent chaotic behavior, in particular spatiotemporal intermittency (STI).
It is one of the most fascinating phenomena appearing in a wide range of
extended systems in several experimental situations, such as chemical reac-
tions [292], Rayleigh–Benard convection [293, 294], planar Couette flow [295],
fluid flows between rotating electrical cylinders [296], Taylor–Couette flows
[297], ensembles of spiking–bursting neurons, as well as in theoretical models,
170 8 Synchronization of Intermittent-Like Oscillations
as coupled map lattices [298] or partial differential equations (e.g., complex

Ginzburg–Landau equation) [299, 300].
The chapter is organized as follows. In Sect. 8.1 we shortly describe the
behavior of the quadratic map generating chaotic type-I intermittent behav-
ior, remind definitions of the phase and the frequency of oscillations, and
give criteria for the existence of synchronization regime in chains of coupled
maps. Synchronization phenomena as well as transitions from synchronous to
nonsynchronous state with a linear resp. a random distribution of the con-
trol parameter are discussed in Sects. 8.2 and 8.3. In the rest of this chap-
ter we study collective phenomena in ensembles of another type oscillators
with intermittent-like behavior: spiking and bursting maps. In Sect. 8.4 we
discuss chaotic phase synchronization in a chain of coupled spiking maps. In
Sects. 8.5 and 8.6 effects of mutual and external chaotic phase synchronization
in ensembles of bursting oscillators are treated. The results are summarized
in Sect. 8.6.
8.1 Model of Coupled Intermittent Maps, Phase

and Frequency, Synchronization Criteria
Here, we focus on synchronization phenomena in chains of coupled noniden-

tical maps with an intrinsic type-I intermittent chaotic behavior.
First we remind (details see in Sect. 2.4) how to define frequency and phase
of intermittent oscillations. A chaotic intermittent dynamics has a distinct
characteristic time scale (CTS) Tc = τ + T , where T is duration of turbulent
stage and τ is the length of the laminar stage. The average length of laminar
stage (ALLS) is defined for a single element as [71]:
1
τ0 ∝ √ , (8.1)
ε − εcr
where ε is the bifurcation parameter and εcr the critical value for chaos onset.
For coupled maps studied below, the CTS Tc can be calculated numerically as:
1
M
Tc = lim (kl+1 − kl ), (8.2)
M →∞ M
l=1
where kl is the moment when the lth laminar stage sets in or in other words
when the lth firing occurs. We note, that in the studied maps because of
τ /T 1 the time of a full cycle is Tc = τ + T , i.e., the time between the
beginning of two sequential laminar stages is practically equal to τ . Therefore,
the coincidence of averaged τ in such a chain leads to the coincidence of
averaged Tc . One can also introduce a phase of the intermittency, attributing
to each interval between the starts of the laminar stage (or in other words
between two firings) a 2π phase increase:
8.2 Linearly Distributed Control Parameters 171
k − kl
φk = 2π + 2πl, kl ≤ k < kl+1 , (8.3)
kl+1 − kl
where k is discrete time.
The presence of a CTS and a suitable phase allows to formulate the pro-
blem of chaotic phase synchronization in ensembles of coupled units with
intermittent behavior. So, if < τj > or the corresponding frequencies
Ωj = 2π/ < τj > (8.4)
of all units become equal, this manifests their global 1:1 frequency entrain-
ment. If the conditions
|φkl − φkm | < Const (8.5)
are fulfilled for all k, one can speak about a 1:1 phase locking between the lth
and the mth units.
We demonstrate mutual phase synchronization of chaotic intermittent
oscillations for a chain of diffusively locally coupled nonidentical quadratic
1D maps:
xk+1
j = fj (xkj ) + d(xkj−1 − 2xkj + xkj+1 ),
(8.6)
j = 1, . . . , N,
where fj (x) has a form (see Sect.2.4):

⎧
⎨ εj + x + x2 , if x ≤ 0.2,
fj (x) = (8.7)
⎩
g(x − 0.2) − εj − 0.24, if x > 0.2
For the parameter g < 5 the laminar stage duration is distributed in

a rather narrow band, i.e., the chaotic behavior is highly coherent. We use
g = 2 (see Fig. 2.8).
The parameter εj defines the CTS in the individual jth oscillator. As
before we treat two cases (1) a linear distribution of the parameter εj : εj =
ε1 + Δε(j − 1), where Δε is the parameter mismatch between neighboring
elements and (2) a random uniform distribution of the εj in the range [ε1 , ε1 +
Δε(N − 1)]. We assume free-end boundary conditions:
xk0 = xk1 ; xkN +1 = xkN . (8.8)
8.2 Linearly Distributed Control Parameters, Soft

Transition to Global Synchronization Regime
First, a chain with a linear distribution of the parameters εj is explored.
The observed frequencies Ωj in dependence on the coupling are presented in
0.0025
(a) (b) (c)
0.002
0.0015
Ωj
0.001
0.0005
0
0 0.01 0.02 0.03 0 0.01 0.02 0.03 0 0.01 0.02 0.03
d d d
Fig. 8.1. The dependence of Ωj (8.4) on the coupling for ε = 0.000001 and
for three different values of Δε in a chain of 50 coupled maps (8.6) and (8.7).
(a) Δε = 0.000001; (b) Δε = 0.000005; (c) Δε = 0.00001
Fig. 8.1. In all diagrams with an increase of the coupling the tendency for the
formation of a more coherent behavior is clearly seen. Then the parameter
mismatch Δε controls whether a global synchronization regime is created
(Fig. 8.1a) or not (Fig. 8.1b, c). But in all cases the increase of coupling ends
up in a fully incoherent behavior, i.e., a nonsynchronous state. The detailed
analysis of the frequency distribution Ωj vs. coupling (see Fig. 8.2) shows
that the transition to global synchronization regime is smooth, i.e., a gradual
adjustment of frequencies is observed. The reason of such a soft route to a
global synchronization regime is the existence of two quite different timescales:
the slow laminar stage and the fast firing stage. It is well known (see, e.g., [79])
that the appearance and interaction of several timescales (at least two) can
lead in oscillatory systems to a chaotic behavior. Another consequence of the
mixed slow–fast motion is a large value of the frequency in the regime of global
synchronization. It is close to the maximal individual frequency [301]. The rea-
son for this effect is the following: For a sufficiently large coupling, the strong
change (firing) of the dynamical variable in the elements close to the right
end of the chain is faster than in the other elements. This provokes analogous
strong change of the dynamical variable in the neighboring element which also
provokes his neighbor and so on. This process leads to a sequential firing in
all elements in the chain.
8.3 Randomly Distributed Control Parameter 173
0.0008 0.0009
(a) (b)
0.0006
0.0008
Ωj
0.0004
d=0
0.0005 0.0007
0.001
0.0002 0.0015
d=0.01
0.002
0.02
0.0025
0.022
0.0035
0.023
0.0000 0.0006
0 10 20 30 40 50 0 10 20 30 40 50
j j
Fig. 8.2. The dependence of the observed frequencies Ωj on j in (8.6) and
(8.7) for different couplings d for (a) transition “nonsynchronous state–synchronous
state” and (b) transition “nonsynchronous state–synchronous state.” ε = 0.000001,
Δε = 0.0000001, and N = 50
Desynchronization appears also through a soft change of the observed fre-

quencies (Fig. 8.2b). A detailed analysis of the transition from a synchronous
to a nonsynchronous regime (synchronization–desynchronization) is presented
for randomly distributed parameter εj in Sect. 8.3.
8.3 Randomly Distributed Control Parameter,

Transition to Spatiotemporal Intermittency
For randomly distributed εj , the evolution of the observed frequency distri-
bution is shown in Fig. 8.3. We observe here three types of transitions to global
synchronization regime:
1. Two adjacent elements (clusters) with close frequencies can be easily
synchronized and a new and larger cluster appears
2. Nonlocal synchronization regime can occur, i.e., an element (a cluster of
elements) becomes synchronized not to a nearest-neighbor element (clus-
ter), but to some other element (cluster) having a close rotation number.
At that the observed frequencies of the elements (clusters) in between are
considerably different
0,0014
(e)
Ωj
0,0013
0,0012
0,00135
(d)
Ωj
0,0013
0,00125
0,0014
(c)
Ωj
0,0012
0,0014 (b)
Ωj
0,0012
0,0015
(a)
Ωj
0,001
0 10 20 30 40 50
j
Fig. 8.3. The dependence of the observed frequencies Ωj on j in (8.6) and (8.7) for
different couplings (a) d = 0, (b) d = 0.0005, (c) d = 0.001, (d) d = 0.0015, and (e)
d = 0.0025. ε = 0.000001, Δε = 0.0000001, and N = 50
3. One element (group of elements) at the edge of one cluster can go to

another neighboring cluster
Similar to the case of linearly distributed parameters εj , in the case of

a random distribution of εj the regime of global synchronization can disap-
pear with an increase of the coupling; at the critical value d∗cr this synchro-
nous regime becomes unstable. In the chain a triangular embedding is now
formed. The onset of such embedding in some place in the chain leads to the
propagation of firing processes in one or, more typically, in both directions.
Propagating firing fronts are usually unstable and a new triangular embed-
ding appears and this process repeats. Therefore, the domains with a large
laminar stage duration (and synchronized intermittency) are alternating with
domains of complex spatiotemporal behavior, which in the presented context
we call spatially turbulent regime. This spatially turbulent regime appears
suddenly and extends to the whole chain. Then it suddenly disappears and
in the whole chain the laminar regime (synchronized intermittency) is again
realized. With an increase of coupling the duration of the spatially turbulent
regime grows and correspondingly the duration of the laminar regime becomes
shorter. During the laminar regime the elements remain synchronized. Beyond
the critical value d∗∗
cr , the laminar regime is no longer observed and the regime
8.3 Randomly Distributed Control Parameter 175
15000
(a) (b) (c) (d) (e)
1
1 j 50 1 j 50 1 j 50 1 j 50 1 j 50
Fig. 8.4. Space–time plots of xj for synchronous (b) and nonsynchronous

regimes (a, c, d, e) in (8.6) and (8.7) for εj randomly distributed in the inter-
val [0.000005; 0.000015]. N = 50, d = 0.001 (a), d = 0.04 (b), d = 0.0056 (c),
d = 0.07 (d), d = 0.15 (e)
of fully developed STI sets in. This rich spatiotemporal dynamics in the syn-
chronous and nonsynchronous regimes is illustrated in Fig. 8.4. The left panel
corresponds to a nonsynchronous behavior (small values of coupling), but
there are already several clusters of mutually synchronized elements. Only
panel (b) corresponds to a synchronous regime. In all plots the darker regions
mark higher values of the presented variables.
Next, we analyze these processes observed by using our phase definition
(8.3). This way the regimes of perfect (Fig. 8.4b) and imperfect (Fig. 8.4c)
chaotic phase synchronization are characterized by a phase distribution φj ,
which is a sequence of intervals with constant phase, separated by ±2π-kinks.
The position of the kinks at constant time corresponds to phase slips. In the
synchronous regimes the phase slips appear with the frequency of synchro-
nous motion. In the nonsynchronous regimes phase slips appear randomly
and rather fast.
In the presented model STI appears due to the relatively strong interaction
of many units. The specific property in our observation consists in the exis-
tence of a transition from a fully coherent (synchronous laminar) to a fully
noncoherent (nonsynchronous turbulent) behavior. In order to demonstrate
this transition, we plot in Fig. 8.5 the ratio D of the number of laminar stages
corresponding to the synchronization regime and the full number of laminar
stages. It is clearly seen that (1) for d ≥ d∗cr the turbulent stages appear very
rarely, whereas (2) for d ≤ d∗∗
cr there are very short intervals of laminar stages.
In our numerical study we also examined chains of different sizes and
different boundary conditions and find that qualitatively all described above
effects are the same.
0,8
0,6
D
0,4
0,2
0
0,05 0,06 0,07
d
Fig. 8.5. The dependence of the ratio D on the coupling for a chain of 50 elements
(8.6) and (8.7) with εj randomly distributed in the interval [0.000005; 0.000015].
d∗cr ≈ 0.049 and d∗∗
cr ≈ 0.067
8.4 Collective Oscillations in a Chain of Spiking Maps 177
8.4 Collective Oscillations in a Chain of Spiking Maps

There is a similar type of behavior often observed in neurobiological systems
that resembles intermittency and is usually called “spiking.” The rich col-
lective dynamics of coupled intermittent systems urges analogous studies of
neural ensembles. In simulations we next study a chain of locally coupled
nonidentical model maps (replicating neural spiking activity) [291]:
1
xk+1
j = f (xkj , xk−1
j , yjk ) + d(xkj+1 − 2xkj + xkj−1 ),
2
1
k+1
yj = yj − μ(xj + 1) + μσj + μ d(xkj+1 − 2xkj + xkj−1 ),
k k
(8.9)
2
j = 1, . . . , N,
where xj and yj are the fast and slow variables, respectively. μ = 10−3 and σj
are the parameters of the individual map and d is the coupling. The function
f (·, ·, ·) has the form:
⎧
⎪
⎪ α/(1 − xk ) + y k , if xk ≤ 0,
⎪
⎪
⎪
⎪
⎪
⎪
⎨ α + y k , if 0 < xk <
k
f (x , x , y ) = α + y k and xk−1 ≤ 0,
k−1 k
(8.10)
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ −1, if xk ≥ α + y k
⎩
or xk−1 > 0
This model is a modification of the model presented in Sect. 2.6. In our

study we take α = 3.5. In dependence on the parameters, the individual
dynamics of the map (in (8.9) d = 0) is ranging from a regular spiking to a
chaotic spiking or bursting behavior and can, therefore, be used for an effec-
tive modeling of neuron-like elements. Several typical spatiotemporal regimes
(including pulse and spiral wave propagation) for networks of identical maps
(8.9) and (8.10) were presented in [291].
Here, we show synchronization phenomena in a chain of locally coupled
nonidentical maps. As for maps with a type-I intermittent behavior, the phase
of spikes can be defined by (8.3), implying a 2π increase between subse-
quent spikes. The frequency can be analogous defined by Ωj = 2π/ < τj >,
where τj is the time interval between two spikes. Simulations show that as
the coupling increases, three different kinds of spatiotemporal dynamics are
generated (1) similar to the case of a chain of intermittent maps at small
coupling neurons are spiking asynchronously (Fig. 8.6a), (2) at a medium cou-
pling they synchronize (Fig. 8.6b), (3) but at large coupling the synchroniza-
tion regime gets destroyed and spatiotemporal chaos sets on (Fig. 8.6c, d).
However, the formation of spatiotemporal chaos is different here compared
to intermittent maps (8.6) and (8.7): Initially rare spikes appear and act as
phase slips or defects; further they evolve into synchronized in-phase chaotic
Fig. 8.6. Space–time plots of xj in (8.9) and (8.10) for synchronous (b) and non-
synchronous regimes (a, c, d) for σj randomly distributed in the interval [0.15; 0.16].
N = 100, d = 0.005 (a), d = 0.05 (b), d = 0.09 (c), d = 0.2 (d) and (e). Panel (e)
is an enlargement of panel (d)
bursts (Fig. 8.6d, e). Note, that spikes forming these bursts are correlated in
space, as they appear as triangular embedding with a fractal-like spatiotempo-
ral structure. The transition observed shows how spiking maps can produce
bursting behavior if they form a spatially extended system. Why collective
complex behavior differs for intermittent and spiking maps? This is due to
the interplay between fast and slow dynamics that produces spiking behavior
in system (8.9) and (8.10). The slow variable regulates the threshold value and
when the threshold gets too high, it forces spike events to stop propagating
along the chain and the burst ends. Until the fastest neuron is recovered, no
spiking is observed in the chain and that separates bursts clearly. Quite on the
contrary, there is no slow variable in the intermittent map that would regulate
turbulent outbursts and they multiply freely in the regime of spatiotemporal
chaos.
8.5 Synchronization in Ensembles of Globally

Coupled Bursting Oscillators
Different aspects of controlled collective behavior in ensembles of neuronal
elements is in the focus in many recent studies. For instance, effects of the
imposed periodic signal on the dynamics of an isolated neuron or a small
group of neurons have been studied theoretically and experimentally. They
8.5 Synchronization in Ensembles of Globally Coupled Bursting Oscillators 179
show an impressive variety ranging from giving rise to either quasiperiodic

or chaotic temporal patterns to external phase locking phenomenon [303].
At the same time the problem of controlling the dynamics of medium- and
large-size neural ensembles by means of external synchronization is far less
analyzed. Basing on the known possibility of synchronization of an ensemble
of first-order phase oscillators by a common external periodic force [304], one
may conjecture that if each bursting oscillator in a large ensemble is forced
by a common periodic signal, global phase locking should occur. However,
arranging such common driving may turn out to be quite problematic in real
neuron ensembles. Thus elaborating techniques of controlling the dynamics of
a neural ensemble by a locally imposed (i.e., applied to a single element) signal
looks far more advantageous for real applications.
In this section we first show the effect of the onset of mutual CPS in
an ensemble of chaotically bursting oscillators. Then we demonstrate that
an external periodic driving applied to only one arbitrary taken oscillator
can lock the phases and the frequencies of all elements in the ensemble. The
autonomous ensemble is modeled by a set of the two-dimensional maps cou-
pled via the mean field (for individual map see Sect. 2.6):
⎧
⎨ αj ε
N
x(j, n + 1) = 1+x(j,n) 2 + y(j, n) + N x(i, n),
(8.11)
⎩ i=1
y(j, n + 1) = y(j, n) − σx(j, n) − β.
The parameters αj are taken randomly and uniformly distributed in

[4.1, 4.9] so that each uncoupled map produces either chaotic bursts or con-
tinuous chaotic oscillations. If the mean-field coupling is quite strong, initially
uncorrelated, timescale-different neuronal bursts become synchronized [91].
A common rhythm of bursting also appears in the elements, which produces
continuous chaotic spiking when isolated. We would like to stress that in this
kind of ensemble the onset of coherent bursting cannot be described in terms
of CPS, because no CTS of bursting can be found in a continuously spiking
neuron.
In contrast, we set αj randomly and uniformly distributed in [4.1, 4.4],
β = σ = 0.001 obtaining thus only bursting neurons [91]. In Fig. 8.7a typical
realizations of the fast dynamics (the evolution of the x variable) of isolated
oscillator are shown. As outlined in Sect. 2.6, the existence of a CTS allows to
introduce a phase and a frequency of bursting in each oscillator. The phase of
the bursting oscillations (strongly speaking, the phase of bursting events) in
the jth oscillator φ(j, n) increases linearly between the moments nk at which
bursts start (k being a number of a burst) and gain a 2π growth over each
time interval nk+1 − nk : φ(j, n) = 2πk + 2π(n − nk )/(nk+1 − nk ) (see also
(8.3)). The frequency of the bursts is an average speed of the phase increase:
Ωj = lim (φ(j, n) − φ(j, 0))/n or can be defined as Ωj = 2π/< τj >, where
n→∞
τj is the time interval between two bursts.
8.5.1 Mutual Synchronization
First we show that the transition to mutual phase synchronization takes place
on the bursting timescale of the globally coupled oscillators, while on the
spiking timescale they behave asynchronously. The onset of mutually syn-
chronized bursting in the studied ensemble has much in common with the
classical example of global phase entrainment of phase oscillators (Chap. 5).
The transition to synchronized bursting is observed as the coupling between
the oscillators is increased (Fig. 8.7b). A nonzero mean field is also formed and
its oscillations make neurons develop a common rhythm. Remarkably, only the
slow timescale (i.e., bursting) dynamics becomes coherent. The spikes remain
uncorrelated and do not substantially contribute to the mean-field (close to
periodic) dynamics. The frequencies Ωi of the bursting oscillators calculated
for different coupling illustrate the appearance of one synchronized cluster and
its gradual increase in size (Fig. 8.8). The asymptotic (very large n) behav-

N
ior of the order parameter r = | eiφ(j,n) |/N indicates a second-order phase
j=1
transition to coherence (Fig. 8.9), which is typical for mean-field coupled phase
oscillators [8, 305] (see Sect. 5.5).
mean field
mean field
α=4.4 α=4.4
α=4.1 α=4.1
(a) (b)
0 1000 2000 0 1000 2000
n n
Fig. 8.7. Time series x(j, n) of two neurons (8.11) and the mean field from
(a) an uncoupled ensemble (ε = 0) and (b) a coupled ensemble (ε = 0.04), where
synchronization regime of the bursts is achieved, in the absence of an external signal
(A = 0). Random values of αj are implemented, σ = β = 0.001, N =1,000
0.03
ε=0.025
Ωj 0.02
0.01
0.016 0.018 0.02 0.022 0.024 0.026 0.028 0.03 0.032
0.024
ε=0.035
Ωj
0.022
0.02
0.016 0.018 0.02 0.022 0.024 0.026 0.028 0.03 0.032
0.0215
ε=0.045
Ωj
0.021
0.016 0.018 0.02 0.022 0.024 0.026 0.028 0.03 0.032
Ω0j
Fig. 8.8. Frequencies of bursting in the mean-field coupled ensemble (8.11) vs. those
at the zero mean-field coupling show a growth of the synchronization cluster as the
coupling coefficient ε is gradually increased. The external signal is absent (A = 0),
N =1,000
0.9
0.8
0.7
0.6
0.5
r
0.4
0.3
0.2
0.1
0
0 0.01 0.02 0.03 0.04 0.05 0.06
ε
Fig. 8.9. The order parameter r vs. the mean-field coupling coefficient ε indicates
a second-order phase transition to CPS of bursting in (8.11), N =1,000
8.5.2 External Synchronization
Now we demonstrate the effect of external chaotic phase synchronization of

the bursting behavior in the studied ensemble by a periodic driving applied
to only one arbitrary taken neuron. Once all the neurons have developed a
common timescale of bursting behavior, it may be attempted to synchronize
this internal rhythm to an external periodic driving. We fix the mean-field
coupling coefficient ε = 0.1, so that neurons stay mutually phase synchronized.
Then we subject one arbitrarily taken neuron j ∗ to a harmonic signal at the
x variable
ε
N
∗ αj ∗ ∗
x(j , n + 1) = + y(j , n) + x(i, n) + A sin ωn (8.12)
1 + x(j ∗ , n)2 N i=1
The equations for the y variable of this neuron and those for the other neu-
rons remain unchanged. The number of bursting oscillators in the simulated
ensemble is N = 50. By applying a frequency of the driving signal ω which
differs a bit from those of the mean field, we obtain external phase locking
(i.e., external CPS) of bursts (Fig. 8.10) in dependence on the forcing ampli-
tude A. This transition to external CPS in this ensemble is characterized by
the following main properties:
x 10−3
2.5
2 A=0.05
A=0.09
1.5
A=0.15
1
0.5
Ωj − ω
−0.5
−1
−1.5
−2
−2.5
0.014 0.015 0.016 0.017 0.018
ω
Fig. 8.10. The difference between frequencies of bursting in oscillators and the
driving frequency vs. the driving frequency for three values of the driving amplitude
demonstrates external CPS in the ensemble of bursting oscillators (8.11) and (8.12).
The fixed mean-field coupling ε = 0.1 ensures mutual phase synchronization regime
between oscillators in the absence of the driving signal, N = 50
1. At relatively large amplitudes of the driving outside the synchronization

region the driven oscillator switches to a frequency different from those of
the driving signal and of the other mutually synchronized oscillators. At
smaller amplitudes of driving this effect is far less pronounced.
2. The synchronization plateau enlarges basically in the direction of higher
frequencies of the driving as its amplitude increases.
3. For A > ε a further increase of the amplitude of the driving does not
enlarge the synchronization plateau significantly.
In Fig. 8.11 we illustrate the dynamics of the phases of the oscillators

inside (the phases are locked) and outside (the phases are drifting) the syn-
chronization region. These computations confirm that the synchronous motion
persists whatever neuron is driven. Hence, this phenomenon provides also a
very efficient control strategy for a variety of experiments and applications
especially in Neuroscience. This effect is also obtained for all generated real-
izations of the randomly distributed parameters αi , for which the unforced
ensemble demonstrates mutually synchronized bursting.
The observed phenomena can be explained by the influence of the slow
variable y on the fast x. This results in triggering (terminating) a burst in an
isolated oscillator, when y exceeds (gets below) the correspondent thresholds
[91]. An imposed periodic signal precipitates a burst of a neuron into the
450
400
350 ω=0.0158
300
250
nω − φj
200 ω=0.0156
150
100
50 ω=0.0153
0
−50
0 2 4 6 8 10
n x 105
Fig. 8.11. Drifting (ω = 0.0158 and ω = 0.0156) and locked (ω = 0.0153) phases
φj bursting oscillators illustrate the transition to CPS in a neural ensemble (8.11)
and (8.12). For each frequency ω the differences nω − φj for all j = 1, . . . , N are
presented. Some lines practically coincide, N = 50
quiescent regime, when positive, and delay it, when negative, which tends to
synchronize the driven neuron. The mean-field coupling term reflects averaged
individual dynamics of the neurons. Suppose, that the external frequency
exceeds that of the mutually synchronized autonomous ensemble. Then the
periodic signal will fasten the oscillations of the driven neuron. When the
whole ensemble is about to start (stop) bursting, the global dynamics becomes
very sensitive to changes of the amplitudes of the individual oscillators. If the
fastened neuron starts (stops) bursting (passing ahead of the others), the
abrupt change of its amplitude increases (decreases) the mean-field value,
pushing the other neurons toward bursting (silence). Thus higher frequencies
win the competition with the lower ones in the neural ensemble. Quite on the
opposite, should the frequency of the driving signal be smaller than that of
the autonomous ensemble, only tiny synchronization effects can be expected.
From this the frequency asymmetry of the synchronization plateaus results.
We would like to stress that a local driving can result in external CPS of the
whole ensemble only when the oscillations in the autonomous ensemble are
mutually synchronized. Oscillators that are not bursting coherently with the
driven oscillator are not susceptible to the driving signal. The contribution of
the driven oscillator to the mean field is proportional to ε/N and does not
depend upon the amplitude of the driving A. That explains why the increase
of the synchronization region is limited when A is increased and ε is fixed. It
x 10−3
1.2
1
Ndr =4
0.8
Ndr =2
Δω
0.6
0.4
Ndr =1
0.2
0
0.005 0.01 0.015 0.02
1/N
Fig. 8.12. The average size of the synchronization plateau Δω (see details in the
text) for the fixed mean-field coupling ε = 0.2 and the driving amplitude A = 0.15
vs. the inverse number of oscillators in the ensemble 1/N for different numbers of
driving neurons Ndr
8.6 Conclusions 185
also follows that the synchronization region should decrease as the number of
oscillators in the ensemble grows. To analyze this dependence, we calculate
the size of the synchronization plateau Δω for a fixed mean-field coupling
ε = 0.2 and driving amplitude A = 0.15 in ensembles with a different number
of oscillators N . For each size value N we generate 100 realizations of random
αj , j = 1, . . . , N and average the obtained sizes of synchronization regions. In
Fig. 8.12 we observe that the synchronization region size scales as: Δω ∝ 1/N .
For control purposes in very large ensembles one can apply the same driving
signal not to one but to several arbitrary taken neurons Ndr (see Fig. 8.12 for
two and four driving neurons). The increase of the number of drivers leads to
the increase of synchronization plateaus.
8.6 Conclusions
In this chapter we have analyzed phase synchronization phenomena in chains
of coupled chaotic intermittent maps.
The main properties are:
1. For chains of locally coupled maps

– There are global and cluster phase synchronization effects in a chain
of nonidentical chaotic oscillators with a type-I intermittent behavior.
– A very important feature is that an increase of the coupling strength
can also lead to desynchronization phenomena, i.e., regimes of global
or cluster synchronization is changed by a regime where synchronous
motion is intermittent with an incoherent state. Then a regime of fully
incoherent nonsynchronous state, STI, is formed.
– Analogous synchronization phenomena, especially synchronization–
desynchronization transition with the increase of the coupling have
been observed in a chain of locally coupled nonidentical maps demon-
strating spiking activity.
It is important to note that the appearing chaotic traveling spikes (forming
triangular embedding), which correspond to fully developed turbulence,
construct nothing but space–time fractal bursting. Our results show that
the transition to STI is quite typical for intermittent discrete in time
and space systems (see also [298]), which are often used for modeling
of dynamical processes in oscillatory media. Presented findings elucidate
complex and intriguing collective dynamics of intermittent and spiking
spatially extended systems, and may be used in various applications like
developed (spatiotemporal) turbulence and complex behavior in neuro-
biological networks. Another challenge are experimental studies on these
results in various fields, where type-I intermittency has been reported
so far.
2. For chains of globally coupled bursting maps

– We have analyzed the role of CPS in the onset of mutually synchro-
nized dynamics in a globally coupled ensemble of bursting oscilla-
tors. This transition occurs on the timescale of bursting, while on the
timescale of spiking synchronization does not appear. Features typical
of a second-order phase transition to synchronization regime have been
observed.
– We have also demonstrated the effect of the onset of external CPS of
the ensemble of bursting (mutually phase locked when autonomous)
by a periodic driving applied to only one arbitrary taken oscillator.
We also propose an explanation of the mechanism behind this effect.
– Our findings show that the studied ensemble can be effectively syn-
chronized by the frequency of the driving signal that is higher than
that of the autonomous ensemble, while lower driving frequencies are
practically unable to cause such synchronous state.
– This phenomenon offers a new way for an efficient controlling of the
bursting activity in neural ensembles.
One can also anticipate that this mechanism of regulating neural behavior
may be identified in natural neurobiological systems.
9
Regular and Chaotic Phase Synchronization
of Coupled Circle Maps
In this chapter we study conditions for an onset of regular and chaotic phase
synchronization (PS) regimes in ensembles of coupled circle maps (CMs) [183].
For networks of coupled maps different problems of synchronization, pat-
tern formation, and spatiotemporal chaos have been investigated [306–311].
In most studies, however, identical coupled maps have been analyzed. It
is evident that for networks of coupled identical elements the investigation
of PS makes no sense, because the individual frequencies of the uncoupled
elements already coincide. Therefore we consider here mainly coupled non-
identical maps, i.e., the more realistic case that usually arises in nature and
engineering where subsystems are never identical. In contrast to other maps,
CMs have the strong advantage that the phase variables exist which allows
immediately to apply criteria to test for of synchronization similar to those
used for the detection of PS in time-continuous systems (see Chaps. 5–7).
As for other phase systems, synchronization in ensembles of coupled CMs
has found important practical applications in electronics, radioengineering
or communications, in particular, in networks of digital phase-locked loops
(DPLLs) [4, 80, 81, 312, 313]. Ensembles of coupled CMs can be used as rather
simple but paradigmatic models to investigate processes of mutual synchro-
nization in ensembles of relaxation time-continuous dynamical systems (see
Chap. 6). In this case, each value of the phase variable can be interpreted as
an onset of a new impulse (or firing event) [1, 314, 315]. The systems treated
in this chapter belong to the broad class of “pulse-coupled” systems arising
in many branches of science and engineering; e.g., pulse-coupled systems have
been investigated as models of neural networks [187, 316–323], cardiac pace-
maker cells [324–326], or in communication [327].
The chapter is organized as follows. In Sect. 9.1 we present a model of
chains of coupled CMs. Then we consider synchronization effects in ensembles
of identical circle maps (Sect. 9.2). This will be helpful for the study of collec-
tive behavior in nonhomogeneous chains that has been done in the rest of this
chapter. In Sect. 9.3 we state the problem and discuss criteria of synchronous
188 9 Regular and Chaotic Phase Synchronization of Coupled Circle Maps
behavior. Sections 9.4 and 9.5 are devoted to the synchronization of regular
and chaotic CMs, respectively. The results are summarized in Sect. 9.6.
9.1 Common Model for a Chain of Coupled Circle Maps

The main model of this chapter is a chain of nonidentical CMs which are
locally coupled
φk+1
j = ωj + φkj − F (φkj ) +
(9.1)
d1 sin (φkj−1 − φkj ) + d2 sin (φkj+1 − φkj ).
Here j = 1, . . . , N corresponds to the number of individual CMs and d1 and d2

are the coupling strengths. The function F (φkj ) is as in Sect. 2.5.4 a piece-wise
linear 2π-periodic function of the form
φkj
F (φkj ) = c (9.2)
π
defined in the interval (−π, π] and c is the control parameter. In the following
we use for the ensembles basic properties of a single circle map (Sect. 2.5.4)
and synchronization effects for two coupled CMs (Sect. 4.3).
The parameters ωj characterize the individual frequencies. For simplicity,
we call ωj as frequency.
We assume that the system is subjected to:
1. Periodic boundary conditions:
φkj = φkj+N (9.3)
2. Free-end boundary conditions:
φk0 = φk1 ; φkN +1 = φkN . (9.4)
Let us briefly discuss our choice of coupling. System (9.1) with the
nonlinear coupling can be regarded as a model of a multichannel chain of
partial DPLLs connected in parallel by phase-mismatching signals [329]. To
realize these connections in a chain in its simplest variant, it is necessary
to compare the output signals of two neighboring DPLLs with the help of a
separate phase discriminator (PDs) and then to apply the obtained phase-
mismatching signal for the frequency control of both generators (Fig. 9.1).
Some similar one- and two-dimensional space models of coupled identical CMs
have been studied in [4].
9.2 Synchronization in a Chain of Identical Circle Maps 189
DPLLj−1 DPLLj DPLLj+1
PDj−1 PDj
Fig. 9.1. Digital phase-locked loops (DPPLs) coupled via phase discriminators
(PDs)
We analyze the nonlinear coupling between the partial elements in the form
of sine of the phase differences because this kind of coupling naturally arises
in models of ensembles of weakly locally diffusively coupled time-continuous
oscillators (see Chap. 5). In contrast to often used types of diffusive coupling
like linear phase difference between the neighbors:
d1 (φkj−1 − φkj ) + d2 (φkj+1 − φkj ), (9.5)
or through the same nonlinear functions as individual functions for each ele-
ment:
d1 (F [φkj−1 ] − F [φkj ]) + d2 (F [φkj+1 ] − F [φkj ]), (9.6)
the sine type coupling exhibits some special properties of the dynamics of
populations of time discrete elements. The most important advantage of the
sine coupling (9.1) is that it generates mutually phase synchronous rotations
already for a very small coupling d compared to the cases (9.5) and (9.6).
9.2 Synchronization in a Chain of Identical Circle Maps
In this section we study the case of identical maps, i.e., ωj = ω, in the chaotic
regime, i.e., c < 0. In homogeneous chains of chaotic oscillators the regime
of complete chaotic synchronization is possible. This regime is realized if the
state
φk1 = φk2 = . . . = φkN = φ̄k (9.7)
is stable.
The specific type of the considered function F (φ) (9.2) allows to find
analytically the stability conditions of the synchronous state.
We will consider symmetrically and asymmetrically coupled maps.
9.2.1 Symmetrically Coupled Maps
First we study the case of symmetrically coupled maps, i.e., d1 = d2 = d.

Linearizing around the uniform solution (9.7), the system (9.1) for devia-
tions ξjk = φkj − φ̄k is:

ξjk+1 = ξjk + F (φ̄k )(1 − 2d)ξjk + d(ξj−1
k k
+ ξj+1 ),
(9.8)
j = 1, . . . , N,
or in the vector form:

ξ k+1 = Dξ k , (9.9)
k
where ξ = [ξ1k , ξ2k , . . . , ξN
k T
] , and the Jacobian matrix D is
⎛ ⎞
ρ0 (1 − 2d) d 0 ··· δd
⎜ d ρ0 (1 − 2d) d ··· 0 ⎟
D=⎜ ⎟,
⎝ ······ ······ ······ ······ ······ ⎠
δd ··· 0 d ρ0 (1 − 2d)
where δ = 1 for periodic boundary conditions and δ = 0 for free-end boundary

conditions.
Because ρ0 is constant, the matrix D consists of constant coefficients, i.e.,
it does not depend on k. This allows us to find the eigenvalues of the solution
(9.7). One of them is always equal to the eigenvalue of the individual map ρ0 .
We remind that ρ0 > 1 as c < 0. While ρ0 manifests chaotic rotations, the
stability of the synchronized state is determined by the other N −1 eigenvalues.
If |ρl | < 1 for all l = 1, 2, . . . , N − 1, then the state (9.7) is stable.
Let us consider separately both boundary conditions.
Periodic Boundary Conditions
For periodic boundary conditions the eigenvalues are:
ρl = ρ0 − 2d(1 − cos ψl )
(9.10)
l = 0, 1, . . . , N − 1,
where ψl = 2πl/N . If d = 0, i.e., the maps are uncoupled, hence all N eigen-
values are equal to ρ0 . With increasing d, all eigenvalues except ρ0 become
smaller and go into the interval (−1; 1). The largest eigenvalue among them
are ρ1 = ρN −1 . The smallest eigenvalues are ρN/2 for even N and ρ(N ±1)/2 for
odd N . Therefore, in order to obtain the stability region of the synchronous
state, it is sufficient to look on the eigenvalues ρ1 and ρN/2 for even N , and
ρ1 and ρ(N +1)/2 for odd N . Then the synchronization region in (9.1) can be
determined from the inequalities:
ρ1 < 1; ρN/2 > −1 for even N,

(9.11)
ρ1 < 1; ρ(N +1)/2 > −1 for odd N,
which provides the interval of the coupling parameter d that corresponds to

the synchronous regime:
1. For even N
1
d∗∗ ≡ c
2π(1−cos(2π/N )) <d< 2 − c
4π ≡ d∗ (9.12)
2. For odd N
2π−c
d∗∗ ≡ c
2π(1−cos(2π/N )) <d< 4π(1−cos(π[N +1]/N )) ≡ d∗ . (9.13)
These conditions show that the stability of the synchronous state is com-
pletely determined by (a) the eigenvalue of the single map and (b) by the
number of coupled maps in the ensemble. In Fig. 9.2 the dependencies of d∗∗
and d∗ on the parameter c for different N is shown. One can see that with
increasing N the synchronization region Sp become smaller.
Assuming a chain of fixed length N , one can calculate the maximal value of
ccr , such that at c ≤ ccr a synchronous regime does not exist for any coupling.
0.8 0.8 0.8

(a) (b) (c)
0.6 0.6 0.6
Sp Sf Sf Sf
Sp Sp
0.4 0.4 0.4
d
0.2 0.2 0.2
0 0 0
− 0.8 − 0.6 − 0.4 − 0.2 0 − 0.2 − 0.15 − 0.1 − 0.05 0 − 0.03 − 0.02 − 0.01 0
c c c
Fig. 9.2. Synchronization regions in the chain (9.1) with periodic boundary condi-
tions (region Sp bounded by dashed lines) and free-end boundary conditions (region
Sf bounded by solid lines) for N = 10 (a), N = 20 (b), N = 50 (c). Note different
scales for the horizontal axis
10
periodic BC
free-ends BC
1
−ccr
0.1
0.01
0.001
0 10 20 30 40 50 60 70 80 90 100
N
Fig. 9.3. Dependence of the critical parameter value −ccr on the number of elements
N in the chain (9.1) with periodic and free-end boundary conditions. Synchroniza-
tion takes place in regions under the curves
This value can be easily found from 9.12 and 9.13. For example, for even N
it reads (Fig. 9.3):
cos(2π/N ) − 1
ccr = 2π (9.14)
cos(2π/N ) + 1
With increasing coupling the last unstable eigenmode to be observed when
approaching synchronization region from below for d a bit less than d∗∗ has
the wave number ψ1 = 2π/N and the spatial period equal to N . Hence, a
long-wave bifurcation takes place. The distribution of φkj for fixed k will be
sinusoidal: a sin(2πj/N ) + b, where a and b depend on k.
For even N the first mode which becomes unstable at the transition from
a synchronous to a nonsynchronous regime is the eigenmode corresponding to
l = N/2, ψ1 = 2π/N , and spatial period equal to 2. Therefore for d a bit larger
than d∗ the “antiphase” (“saw”) distribution occurs, at which the values φk
for all k are identical for all even elements, i.e., φk2j = φ̄keven as well as for all
odd elements, i.e., φk2j−1 = φ̄kodd . Hence, a short-wave bifurcation takes place.
These and other distributions φj in the chain (9.1) with periodic boundary
conditions at N = 256, c = −0.0001 for different couplings d after 10,000,000
iterations are shown in Fig. 9.4. For chosen parameters with increasing d
the transition “nonsynchronous regime–synchronous regime” occurs at d∗∗ ≈
0.2113. The transition “synchronous regime–nonsynchronous regime” takes
place at d∗ ≈ 0.5. Sinusoidal and “antiphase” distributions of φj are presented
φj − 0.8136 (a)
− 0.8137
− 0.8138
− 0.6
(b)
− 0.8
φj
−1
(c)
− 0.06186
φj
− 0.06188
2
(d)
0
φj
−2
−4
2
(e)
0
φj
−2
0 64 128 192 256
j
Fig. 9.4. Distributions φj in the chain (9.1) with periodic boundary conditions
for N = 256, c = −0.0001 at different coupling strength d after 10,000,000 iter-
ations. d = 0.01 (solid line), d = 0.04 (dotted line corresponds to distribution
φj = a sin(2πj/N ) + b, d = 0.07 (dashed line corresponds to homogeneous syn-
chronous state) (a), d = 0.508 (“antiphase” distribution) (b), d = 0.748 (c) (values
φj only for even j are shown), d = 0.88 (d), d = 0.91 (e)
in Fig. 9.4a, b, correspondingly. With further increasing d a modulation of φj

takes place. (Fig. 9.4c). Then more and more eigenvalues become less than
−1, i.e., more and more corresponding modes become unstable, which leads
to more complex distribution of φj (Fig. 9.4d, e).
This analysis allows to identify only local stability of synchronous states.
Due to the homogeneity of the chain other regimes are possible too, i.e., the
system (9.1) is multistable. So there exist stable splay-state distribution with
linear distribution φkj : φkj = a(k)+2πj/N for all k. For this state the difference
θjk ≡ φkj+1 − φkj = 2π/N does not depend on j. Therefore in the system for θjk
there exists the spatially homogeneous state θ1k = θ2k = . . . = θN k k
−1 = θ̄ . Its
stability can be analyzed in the same manner as the stability of the state ϕ̄k .
θ̄k loses its stability with increasing of coupling and “antiphase” distribution
of θjk is realized (Fig. 9.5).
Free-End Boundary Conditions

Now we study free-end boundary conditions: ξ0k = ξ1k and ξN
k k
+1 = ξN in (9.8).
The eigenvalues in this case are:
ρl = ρ0 − 2d(1 − cos ψl )
(9.15)
l = 0, 1, . . . , N − 1,
3
d=0.504
d=0.509
2
1
φj
−1
−2
−3
0 20 40 60 80 100
j
Fig. 9.5. Distribution φj in the chain (9.1) with periodic boundary conditions at
N = 100, c = −0.0001 for different couplings d after 10,000,000 iterations. At
d = 0.504 the distribution φj is linear: φj = a + 2πj/N . At d = 0.505 “antiphase”
regime is stable
where ψl = πl/N . As for periodic boundary conditions the largest eigenvalue

after ρ0 is ρ1 , but the smallest one is now ρN −1 . Looking on these ρl , one can
calculate the stability region Sf of the synchronous state
2π−c
d∗∗ ≡ c
2π(cos(π/N )−1) <d< 2π(1+cos(π/N )) ≡ d∗ (9.16)
Such regions Sf for different N are presented in Fig. 9.2 which shows that for
chains of fixed length the stability region of the synchronous state in chains
with periodic boundary conditions Sp is essentially larger compared to those
with free-end boundary conditions Sf .
The critical value ccr is:
1
ccr = π(1 − ) (9.17)
cos(π/N )
This curve lies below the curve ccr (9.14) for periodic boundary conditions
(Fig. 9.3), i.e., we find again that for chains closed in a ring the synchronous
regime can exist for larger size (larger N ) than in chains with free ends.
In Fig. 9.6 the distributions of φj in the chain (9.1) with free-end boundary
conditions for N = 256, c = −0.0001 for different d after 10,000,000 iterations
are shown. At the transition to a synchronous regime at d a bit less than d∗∗ ,
the last stable eigenmode corresponding to l = 1 has the wave number π/N
− 0.8267 (a)
− 0.8268
− 0.8269
(b)
0.590509
0.590508
− 0.76
(c)
− 0.8
− 0.84
− 0.88
2
(d)
0
−2
2
(e)
0
φj
−2
0 64 128 192 256
j
Fig. 9.6. Distributions φj in the chain (9.1) with free ends for N = 256, c = −0.0001
for different coupling strength d after 10,000,000 iterations. d = 0.01 (a), d = 0.211
(b) (distribution φj = a sin(πj/N ) + b, d = 0.501 (c), d = 0.864 (d), d = 0.99 (e)
and the spatial period 2N . This mode is sinusoidal a + b sin(πj/N ) (where a, b

depend on k) (Fig. 9.6b). In long chains at the transition to a nonsynchronous
regime at d a bit larger than d∗ the first stable eigenmode corresponding to
l = N − 1 is close to the “antiphase” (ψN −1 ≈ π (Fig. 9.6c). With increasing
d > d∗ , the distributions of φj becomes more complex.
Therefore for both boundary conditions the transition from a synchronous
state to nonsynchronous one occurs through long- and short-wave instabilities
[328].
9.2.2 Effect of Asymmetry of Coupling
In the case of asymmetrically coupled maps (9.1) the stability conditions can
be also found analytically.
For periodic boundary conditions the eigenvalues are complex:
ρl = ρ0 − (d1 + d2 )(1 − cos ψl ) + i(d1 − d2 ) sin ψl

(9.18)
l = 0, 1, . . . , N − 1,
where ψl = 2πl/N . The synchronous state is stable if all eigenvalues ρl ,
l = 1, . . . , N − 1 are located inside the circle with radius 1. Because the
maximal absolute values are given by the eigenvalues ρ1 and ρN/2 for even N ,
and ρ1 and ρ(N ±1)/2 for odd N . Hence, the boundaries of the stability regions
can be obtained from the equalities:
|ρ1 | = 1; |ρN/2 | = 1 for even N,

(9.19)
|ρ1 | = 1; |ρ(N +1)/2 | = 1 for odd N,
which define the stability region for even N in the coupling parameter plane
(d1 , d2 ) by the curves:
2π
ρ20 − 2(1 − cos )∗
N
2π
∗[d∗∗ ∗∗ ∗∗ 2 ∗∗ 2 ∗∗ ∗∗
1 + d2 + (d1 ) + (d2 ) − 2d1 d2 cos ]=1 (9.20)
N
ρ0 + 1
d∗1 + d∗2 = .
2
For free-end boundary conditions the eigenvalues are:
√
ρl = ρ0 − (d1 + d2 ) + 2 d1 d2 cos ψl
(9.21)
l = 0, 1, . . . , N − 1,
where ψl = πl/N . The stability region of the synchronous state defined from:
ρ1 = 1; ρN −1 = −1 (9.22)
is then bounded by the curves:

π
d∗∗ ∗∗ ∗∗ ∗∗
1 + d2 − 2 d1 d2 cos + 1 = ρ0
N
(9.23)
π
d∗1 + d∗2 + 2 d∗1 d∗2 cos − 1 = ρ0 .
N
In Figs. 9.7 and 9.8 the synchronization regions for periodic and free-end
boundary conditions are presented. In the first case this region is located
around the line d1 = d2 . But in the second case synchronization region con-
tains two parts located close to axes (d1 = 0, d2 = 0) and (d2 = 0, d1 = 0). One
can also see that (1) with increase of complexity of motion (larger c) the syn-
chronization regions become smaller and (2) for periodic boundary conditions
the asymmetry of coupling leads to decrease of the synchronization region,
but for free-end boundary conditions the asymmetry causes enhancement of
synchronization. In contrast to the case of symmetrical coupling the transi-
tion from nonsynchronous regime to synchronous one and from synchronous
regime to nonsynchronous one occurs via long-wavelength bifurcation. For
example (Fig. 9.7a, b), this bifurcation is observed for small values d1 and on
2 2 2
(a) (b) (c)
1.5 1.5 1.5
S
d2
1 1 1
0.5 0.5 0.5

S S
0 0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2
d1 d1 d1
Fig. 9.7. Synchronization regions S in the chain (9.1) of asymmetrically coupled

CMs with periodic boundary conditions for c = −0.07 (a), c = −0.17 (b), and
c = −0.5 (c), N = 10
2 2 2
(a) (b) (c)
1.5 1.5 1.5
d2
1 1 1
S S
0.5 0.5 0.5
S
S S
0 0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2
d1 d1 d1
Fig. 9.8. Synchronization regions S in the chain (9.1) of asymmetrically coupled

CMs with free-end boundary conditions for c = −0.06 (a), c = −0.16 (b), and
c = −0.3 (c), N = 10
increasing values d2 . For free-end boundary conditions the transitions between

synchronous and nonsynchronous regimes can also be different for symmet-
rical and asymmetrical coupling. So, the transition from synchronous regime
to nonsynchronous one and from nonsynchronous regime to synchronous one
can be accomplished by long-wave bifurcation.
9.2.3 Synchronization in Lattices of Coupled Maps
Let us show instabilities of synchronous states in a lattice of symmetrically

coupled circle maps
n,m = ω + φn,m − F (φn,m ) + d[sin (φn+1,m − φn,m )

φk+1 k k k k
+ sin (φkn−1,m − φkn,m ) + sin (φkn,m+1 − φkn,m ) (9.24)
+ sin (φkn,m−1 − φkn,m )].

Here n = 1, . . . , N and m = 1, . . . , M correspond to the number of individual

CMs and d is the coupling strength.
The multipliers for periodic boundary conditions are the following:
ρl1 ,l2 = ρ0,0 − 2d(2 − cos ψl1 − cos ψl2 ))
(9.25)
l1 = 0, 1, . . . , N − 1, l2 = 0, 1, . . . , M − 1
where ρ0,0 is the multiplier of individual map, ψl1 = 2πl1 /N and ψl2 =
2πl2 /M .
Let us focus on the case of even N and M . Then stability region of synchro-
nous regime can be found from the conditions ρ1,1 = 1 and ρN/2,M/2 = −1.
It is bounded by:
1. The transition from nonsynchronous state to synchronous one (through

the long-wave bifurcation):
ρ0,0 − 1
d∗∗ ≡ (9.26)
2(2 − cos 2π 2π
N − cos M )
2. The transition from synchronous state to nonsynchronous one takes

(through the short-wave bifurcation):
ρ0,0 + 1
d∗ ≡ (9.27)
8
For free-end boundary conditions the stability interval of synchronous state
can be defined in a similar way. It yields:
ρ0,0 − 1
d∗∗ ≡ 2π < d <
2(2 − cos 2π
N − cos M )
(9.28)
ρ0,0 + 1 ∗
2π ≡ d .
2(2 + cos 2π
N + cos M )
Spatial structures observed in numerical experiments (see Fig. 9.9 for

examples) are consistent with the instabilities described above.
We would like to note that the paper [330] formulates a general approach
to assessing stability of synchronization in case of symmetrical coupling and
elucidates synchronization of chaotic maps in networks with complex network
topologies.
9.3 Ensembles of Coupled Nonidentical Circle Maps 199
Fig. 9.9. Spatial structures in the lattice (9.24) of coupled CMs with periodic
boundary conditions for ω = 0.5, c = −0.001, N = M = 20, and d = 10−6 (a),
d = 4.2 · 10−5 (b), d = 0.28 (c) and d = 0.5 (d)
9.3 Ensembles of Coupled Nonidentical Circle Maps

and Criteria of Synchronization
Now let us turn to nonhomogeneous chains, i.e., we take different ωj . We have
distributions of them as in the previous chapters:
1. A linear distribution of the individual frequencies:
ωj = ω1 + Δ(j − 1), j = 1, . . . , N , where Δ is the uniform frequency
mismatch
2. A random distribution of the individual frequencies:
ωj = ω1 + Δξj , j = 1, . . . , N, where Δ = const and ξj are uniformly
distributed random numbers in the interval [−0.5; 0.5].
As shown in previous sections, the qualitative behavior in large chains
with periodic boundary conditions with an even or odd number of elements
and in chains with free-end boundary conditions is very similar. Therefore,

we assume in the following sections only free-end boundary conditions.
We use two criteria to test for m : n synchronous rotations, where m and
n are integers. A regime of m : n phase synchronization of regular rotations
as well as chaotic regimes between the ith and the lth elements can be defined
as phase entrainment or locking
| m φki − n φkl | < Const, (9.29)
for all k = 1, 2, . . .. This condition of phase difference boundness is a direct

extension of the phase synchronization regime conditions for continuous in
time systems (see Chaps. 5–7) and have been used in Chap. 8. A weaker cri-
terion for the analysis of both types of synchronous motion is based on their
rotation numbers:
1 φM − φ1i
ρi = lim i , (9.30)
2π M →∞ M
i.e., we test for
m ρi = n ρl . (9.31)
If the condition (9.31) is fulfilled, we infer frequency locking between the ith
and the lth elements. Because of its simplicity of calculation and convenience
of presentation, we often use in the following the criterion (9.31). By the
study of synchronization effects in a chain of coupled CMs, the fulfillment of
the conditions (9.29) and (9.31) for all j = 1, . . . , N − 1 means the existence
of a global synchronization regime. If these conditions are satisfied only for
several neighboring elements, we have a regime of cluster synchronization.
In dependence on the coherence parameter c, we study three different cases
which correspond to the three types of behavior of a single CM (1) strongly
coherent (uniformous) regular rotations (c = 0), (2) noncoherent regular rota-
tions (c > 0) (Sect. 9.3), and (3) chaotic rotations (c < 0) (Sect. 9.4).
9.4 Synchronization and Clustering in a Chain

of Regular CMs
First we will focus on synchronization phenomena in a chain with a linear

distribution of the individual frequencies and then for random distributions.
9.4.1 Linear Distribution of Individual Frequencies
Studying synchronization of regular maps, we distinguish the cases of strongly

coherent (uniformous) and noncoherent rotations:
9.4 Synchronization and Clustering in a Chain of Regular CMs 201
Synchronization of Strongly Coherent CMs
Theory
System (9.1) for c = 0 and a linear distribution of the partial frequencies ωj

can be rewritten as:
φk+1
1 = ω1 + φk1 + d sin θ1k ,
(9.32)
θjk+1 = Δ + θjk + d (sin θj+1

k
− 2 sin θjk + sin θj−1
k
),
j = 1, . . . , N − 1 (9.33)
with Δ = ωj+1 − ωj , θjk = φkj+1 − φkj and the boundary conditions: θ0k =
k
θN = 0.
The conditions for the onset of a global synchronization regime and the
frequency of globally synchronized rotations can be found analytically in anal-
ogy with the case of time-continuous uniformly rotating phase oscillators (see
Chap. 5). The stable fixed point θjk+1 = θjk = θ̄j for each j = 1, . . . , N − 1 in
system (9.33) corresponds to a regime of global synchronization in the chain.
Then the system of equations for the stationary phase differences θ̄n can be
written as (compare with 5.19 and 6.13–6.15):
Δ + d (sin θ̄2 − 2 sin θ̄1 ) = 0,
Δ + d (sin θ̄j+1 − 2 sin θ̄j + sin θ̄j−1 ) = 0,

(9.34)
j = 2, . . . , N − 2,
Δ + d (sin θ̄N − 2 sin θ̄N −1 ) = 0
As follows from Chap. 5 (see (5.33) and (5.34)), the distribution of θ̄j is
Δ
sin θ̄j = (N j − j 2 ). (9.35)
2d
It follows from (9.35) that the system (9.33) can have 2N −1 fixed points, but
only one of them (θ¯j ∈ [−π/2; π/2] for all j = 1, . . . , N − 1) is stable. As the
frequency mismatch Δ is increased, the condition for the existence of a stable
fixed point is
Δ
| (N j − j 2 )| < 1. (9.36)
2d
This is violated firstly for j = N/2 at even N , i.e., for the middle element
in the chain. Thus, the condition for the existence of a fixed point in the
N -element chain is again given by the inequality
ΔN 2
| | < 1. (9.37)
8d
This condition for the existence of a global synchronization regime in system
(9.1) coincides with the results of our numerical experiments. The rotation
number of synchronous rotations ρs can be determined from (9.32):
φk+1
1 = ω1 + φk1 + d sin θ̄1k =
(9.38)
Δ(N −1)
ω1 + φk1 + 2
and is equal to the mean rotation number over all elements in the ensemble
Δ(N − 1)
ρs = ω1 + . (9.39)
2
With increasing of the frequency mismatch Δ, a loss of global synchronization
regime takes place. For a long chain a two cluster synchronization occurs, i.e.,
the chain is divided into two clusters of equal sizes that consist of mutually
synchronized CMs of different rotation numbers.
Numerical Results
Numerical simulations allow to analyze in detail all possible regimes in the

chain with a linear distribution of individual frequencies. For each CM the
rotation number ρj has been calculated. If all the rotation numbers are equal,
global synchronization regime takes place. If the rotation numbers are only
equal for some groups of neighboring elements, cluster synchronization is
formed. The dependence of all rotation numbers ρj for j = 1, . . . , n on cou-
pling d is presented in Fig. 9.10 and is very similar to the synchronization
trees for coupled continuous in time systems (see previous chapters). A care-
ful investigation of Fig. 9.10 shows that the transition trees of synchronization
generated by the system (9.1) does not qualitatively depend on the frequency
mismatch Δ. Indeed looking on the change of the ρj distribution with an
increase (or decrease) of coupling d, we find that the appearance and disap-
pearance of clusters of synchronization happens according the same scenario:
with a decrease of d after the loss of global synchronization regime, two clus-
ters of synchronization appear, then the two-cluster structure is changed into a
four-cluster structure, and exists till a six-cluster structure appears, etc. Clus-
ter synchronization structures are intermittent with nonsynchronous states.
For relatively strong coupling, the transition from a n-cluster to a m-cluster is
hard. Two main routes correspond to this transition (1) through the splitting
of all clusters (1 to 2, 2 to 4) or splitting only of some clusters in the ensemble
Fig. 9.10. Transition trees of synchronization in system (9.1) for the rotation num-
bers ρj vs. the coupling d for different frequency mismatches Δ and for uniform
rotations c = 0, ω1 = 0.1. (a) Δ = 0.001, (b) Δ = 0.0001, and (c) Δ = 0.00001,
N = 100
(e.g., 4 to 6) and (2) through a merging of two clusters into one (e.g., 6 to 3).
The values of the rotation number for each cluster (except the edge ones) are
close to those obtained by averaging the individual rotation numbers over all
elements forming the cluster. For a weak coupling this interstructural transi-
tion is soft: A gradual adjustment of the rotation numbers takes place. For
such soft transitions most elements of the chain (except the edge ones) rotate
with different rotation numbers.
It should be noted that the transition n-cluster structure − > nonsynchro-
nized state − > again n-cluster structure is quite typical. Usually at such
transitions the number of elements in the clusters is changed. Transition trees
of synchronization are very similar not only qualitatively but they also, as
found numerically, exhibit well-expressed scaling properties. So, if in the N -
element chain with the frequency mismatch Δ1 some cluster structure appears
(disappears) at the coupling dcr
Δ1 , then the same cluster structure appears (dis-
appears) in the same chain but with another Δ2 at the coupling:
Δ1
dcr cr
Δ 2 = dΔ 1 . (9.40)
Δ2
Therefore, knowing the evolution of the rotation number distribution for some
Δ1 , one can easily calculate a similar evolution for any other value of frequency
mismatch Δ2 .
As in the case of two coupled CMs, the global synchronization regime can
disappear with the increase of coupling. The stable fixed point of system (9.34)
corresponding to the global synchronization regime at some critical coupling
d, loses its stability through a period doubling. But if all θj which become
oscillating around
Δ
θ̄j = arcsin (N j − j 2 ) (9.41)
2d
remain bounded, global synchronization regime still exists. Figure 9.10 does
not show this change, because there the averaged rotation numbers are plot-
ted. But when coupling becomes larger than some coupling d∗ , the oscillatory
behavior of θj no longer exists and the global synchronization regime disap-
pears. Note that not only the regime of global synchronization can be changed
into the regime of global nonsynchronization. Figure 9.10a presents an exam-
ple of a nonsuccessful transition to global synchronization regime in a chain
with a relatively large frequency mismatch Δ. The two-cluster structure is
immediately transformed into fully incoherent rotations.
Effects of Noncoherent Rotation on Synchronization

of Regular Maps
Now we are going to elucidate how the increase of noncoherence (c = 0)

influences the synchronization in chains. For a linear individual frequency
distribution Fig. 9.11 presents distributions of the rotation numbers ρj in the
transition to global synchronization regime in strongly coherent (uniform rota-
tions) (Fig. 9.11a) and noncoherent (Fig. 9.11b) regimes. The main finding
is that noncoherence can destroy the clusters of synchronization which are
excited at coherent rotations. As our numerical experiments show for a lin-
ear distribution of individual frequencies ωj even for very weak noncoherence
(c = 0.0001), i.e., all elements rotate slightly nonuniformly, the transition
to globally synchronized rotations is usually soft, and only the transition
from a two-cluster structure to a one-cluster structure is hard (Fig. 9.11b).
The boundaries of clusters existing in the noncoherent case become firstly
slightly smooth if the parameter c increases (Fig. 9.11b). If the noncoher-
ence increases (c ≈ 0.001), all clusters except the edge ones are completely
destroyed. Since there still exists a possibility of synchronization at relatively
large c of uncoupled CMs, synchronization cluster structures can appear with
further increase of c.
Two opposite effects are observed under the influence of noncoherence on
the global synchronization in a chain. As in the case of two-element systems, a
rather small increase of noncoherence of rotations practically does not change
the size of the global synchronization region. But with a further increase of
noncoherence, even in the case of the absence of coupling, some neighboring
(a) (b) (c) (d)

0.098 0.098 0.096 0.096
ρj
0.097 0.097 0.095 0.095
0.098 0.098 0.097 0.097
ρj
0.097 0.097 0.095 0.095
0.099 0.099 0.097 0.097
ρj
0.096 0.096 0.095 0.095
0.099 0.099 0.097 0.097
ρj
0.096 0.096 0.094 0.094
0.099 0.099 0.097 0.097
ρj
0.096 0.096 0.093 0.093
0.100 0.100 0.098 0.098
ρj
0.095 0.095 0.093 0.093
0 25 50 0 25 50 0 25 50 0 25 50
i i i i
Fig. 9.11. Regular synchronization in system (9.1). Rotation number ρj distribution

in the transition to global synchronization regime in strongly coherent (uniform rota-
tions) (a,c) and noncoherent (b,d) regimes. For a linear (a,b) individual frequency
distribution ωj = ω1 + Δ(j − 1) we take ω1 = 0.6, Δ = 0.005, c = 0 (a), c = 0.00005
(b) and coupling from bottom to top d = 0, 0.03, 0.042, 0.08, 0.106, 0.158. In the
coherent case only clustered structures of synchronization are presented. For a ran-
dom (c,d) individual frequency distribution ωj = ω1 + Δξj we take ω1 = 0.6,
Δ = 0.025, c = 0 (c), c = 0.05 (d) and coupling from bottom to top d =
0, 0.004, 0.008, 0.016, 0.02, 0.052. ξj are uniformly distributed in the interval [-0.5;0.5]
elements can have the same rotation numbers, i.e., they belong to one interval
of equal rotation numbers (see Fig. 2.12). Thus, there are clusters for which
the rotation numbers coincidence of rotation numbers without coupling. If the
difference between the rotation numbers of the elements is small enough, the
occurrence of a global synchronization regime happens for smaller coupling
than in the case of weak noncoherence. There, the common rotation num-
ber coincides with the rotation number of the elements in the largest cluster.
Usually this situation is observed for intermediate values of c. For strong non-
coherence, the rotation number difference can become very large. Then in spite
of the existence of clusters with coincident ρj in an uncoupled chain, global
synchronization regime can be observed only for stronger coupling (Fig. 9.12).
0.8
0.7
0.6
0.5
d+
0.4 b1=1.0
0.3 1.2
0.2 1.6
1.8
0.1
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4
c
Fig. 9.12. Critical value of coupling d+ corresponding to the transition to global

synchronization regime vs. c for different values ω1 and fixed N = 20 and frequency
mismatch Δ = 0.01 in (9.1)
Moreover, the transition to globally synchronized rotations in a chain of cou-

pled regular CMs is rather complex in dependence on c (Fig. 9.12).
9.4.2 Random Distribution of Individual Frequencies
The rotation number of global synchronization for any type of an individual

frequency distribution for uniformous rotations can be easily calculated. After
summarizing of all equations in (9.1), we have

N
N
(φk+1
j − φkj ) = ωj . (9.42)
j=1 j=1
Because in the global synchronization regime the rotation numbers of all ele-
ments coincide, i.e., ρ1 = . . . = ρj = . . . = ρN = ρs , (9.42) yields
1
N
ρs = ωj . (9.43)
N j=1
This means that the rotation number of the global synchronization regime is
equal to the mean rotation number of all elements in the ensemble (as in the
case of coupled phase (Chap. 5) and limit-cycle (Chap. 6) oscillators).
For randomly distributed frequencies ωj , the evolution of the rotation num-
ber distribution is qualitatively quite similar to the case of continuous-time
active rotators (Chap. 5). The following three types of transitions can then be
observed in the synchronization trees (1) two adjacent elements (clusters) with
close frequencies can be easily synchronized and a new cluster appears; (2)
a regime of nonlocal synchronization can occur, i.e., an element (a cluster of
elements) becomes synchronized not to a nearest-neighbor element (cluster),
but to some other element (cluster) having a close rotation number. At that
the rotation numbers of the elements (clusters) in between are considerably
9.5 Chaotic Phase Synchronization 207
different; and (3) one element (group of elements) at the edge of one cluster
can go to another neighboring cluster.
Synchronization clusters are more stable (Fig. 9.11c, d) with respect to
the increase of noncoherence of rotations. Small c does not practically change
the number of clusters, the number of elements in the clusters, and mean
rotation numbers. With a further increase of the noncoherence, a transition to
the global synchronization regime through the appearance of well-pronounced
clusters is still observed. Only the structure of intermediate clustered states
can be different.
9.5 Chaotic Phase Synchronization
Phase synchronization in ensembles of locally coupled chaotic elements was

firstly studied in chains of weakly diffusively coupled chaotic Rössler oscil-
lators (see Chap. 7). Many phenomena already observed in a population of
periodic oscillators (see Chap. 6) are found there too, especially to mention
the existence of several clusters of mutually synchronized elements and global
synchronization regime. The collective behavior in a chain of coupled chaotic
(c < 0) CMs (9.1) exhibits similar properties.
Chains with a linear individual frequency distributions are explored now.
Both hard (Fig. 9.13a) and soft (Fig. 9.13b) transitions to synchronized behav-
ior are observed. The first route takes place for rather coherent rotations
(small −c). But the second route is typical for strong noncoherent cases.
Even the transition to the global synchronization regime is accompanied by a
smooth adjustment of the rotation numbers.
The rich spatiotemporal dynamics of the noncluster and cluster synchro-
nization structures is illustrated in Fig. 9.14. In all these space–time plots the
darker regions mark higher values of the presented variables. The two left
panels show the quantity sin(xkn ), so that the white stripes correspond to the
phase ≈ 3π/2 and the black stripes to the phase ≈ π/2. The right panel shows
the quantity (see Chap. 7)
φkj+1 − φkj
sj = sin2 ( ) (9.44)
2
which characterizes the instantaneous phase difference between neighboring
oscillators. We have then that sj = 0 if the phases are equal and sj = 1 if
they differ by π. The space–time behavior of the boundaries between clusters
corresponds to the positions where phase difference slips or defects occur.
These defects are clearly seen as maxima (black regions) of sj . They can
follow regularly in time at certain positions in the chain; this case corresponds
to the existence of strong jumps between the clusters (Fig. 9.14c). If cluster
structures do not exist or the borders between them are smooth, then defects
appear irregularly in both space and time (Fig. 9.14d).
1.22 1.22
1.20 (a) 1.20 (b)

d=0.0 d=0.0
1.18 0.06 1.18 0.06
0.14 0.14
1.16 0.29 1.16 0.29
0.39 0.39
0.45 0.45
1.14 1.14
0.66 0.66
1.12 1.12
ρj /ρ1
1.10 1.10
1.08 1.08
1.06 1.06
1.04 1.04
1.02 1.02
1.00 1.00
0.98 0.98
0 10 20 30 40 50 0 10 20 30 40 50
j j
Fig. 9.13. Hard (a) and soft (b) transitions to regime of global chaotic phase syn-
chronization in system (9.1). Relative rotation numbers ρj /ρ1 for different coupling
coefficients d for linear distribution of individual frequencies and for ω1 = 0.6, fre-
quency mismatch Δ = 0.0002, c = −0.002 (a) and c = −0.4 (b)
The critical values of coupling corresponding to the onset of the global syn-
chronization regime in a chain of 50 chaotic CMs with a linear distribution
of individual frequencies ωj for ω1 = 0.6, different values of frequency mis-
match Δ and different values of c are presented in Fig. 9.15. With an increase
of the parameter −c the value d+ that slightly increases first and then can
decrease and increase again, and finally increases. After some critical value
−c∗ a synchronization is impossible due to the very high noncoherence of
rotations.
9.6 Conclusions
In this chapter a rich variety of phenomena in the formation of regular and

chaotic phase synchronization in ensembles of identical and nonidentical circle
maps has been presented.
9.6 Conclusions 209
(a) (b) (c) (d)

2000 4000
1600 3200
1200 2400
800 1600
400 800
k=0 k=0
1 j 50 1 j 50 1 j 50 1 j 50
Fig. 9.14. Space–time plots of evolution of (a, b) sin(φkj ) and sj (9.44) (c, d) by
hard (a, c) and soft (b, d) transitions to a regime of global chaotic phase synchroniza-
tion for a linear distribution of individual frequencies in (9.1). The parameters are
N = 50, ω1 = 0.6, frequency mismatch Δ = 0.002, coupling d = 0.39 and c = −0.002
(a, c) and c = −0.4 (b, d)
1. Identical maps
– The stability conditions depend only on the eigenvalue ρo of the indi-
vidual map, the number N of maps in the ensemble, and on the asym-
metry of coupling. For fixed ρ0 , there is a critical size Ncr of the chain
such that for N > Ncr a homogeneous synchronization regime cannot
occur for any coupling strength. For a fixed size N there is a critical
0.9
Δ=0.0005
0.8 0.001
0.002
0.7
0.6
d+
0.5
0.4
0.3
0.2
0.1
− 0.8 − 0.7 − 0.6 − 0.5 − 0.4 − 0.3 − 0.2 − 0.1 0.0
c
Fig. 9.15. Critical value of coupling corresponding to the transition to global syn-
chronization regime vs. −c for different values of the frequency mismatch Δ in system
(9.1)
value of the eigenvalue ρcr cr

0 such that for ρ0 > ρ0 a synchronization
regime cannot be realized.
– In symmetric chains with increasing coupling (1) the transition from a
nonsynchronous state to a synchronous one occurs through long-wave
bifurcation: the sinusoidal distribution of phase variable is replaced by
homogeneous distribution and (2) the transition from a synchronous
state to a nonsynchronous one occurs through short-wave instabilities:
the homogeneous distribution is replaced by a saw distribution. In
asymmetrical case the last transition can be accompanied by long-
wave bifurcation.
– For chains consisting of the maps, the synchronous regime exists in
longer chains for periodic boundary conditions than for free-end ones.
Actually, adding of only one connection (passing from a free-end chain
to a ring) can sufficiently increase the synchronization region. This is
intuitively clear because the largest distance between nodes in a ring
is twice shorter.
– For periodic boundary conditions the asymmetry of coupling leads to
decrease of synchronization region.
– For free-end boundary conditions the asymmetry is a reason of the
enhancement of synchronization.
9.6 Conclusions 211
2. Nonidentical maps
– For chains of coupled circle maps, the typical features are the onset
and the existence of global (all-to-all) and cluster (partial) synchro-
nization regimes. These regimes have been observed both for regular
and chaotic maps.
– For high coherent rotations (c = 0) and for any type of frequency
distribution the rotation number of the global synchronization regime
is equal to the mean rotation number of elements in the ensemble.
– As well as for identical maps, increase of coupling strength can lead to
desynchronization phenomena, i.e., global or cluster synchronization
regime is changed by fully incoherent nonsynchronous state.
– As for chains of periodic and chaotic continuous in time oscillators,
for coupled maps two scenarios of transition to global synchronization
regime or transition between cluster structures have been found (1) a
gradual adjustment of the rotation numbers is observed and (2) the
transition occurs through the appearance of synchronized clusters.
– Hard transition between the cluster structures is more typical for
highly coherent rotations, while a soft transition is more often observed
for noncoherent rotations.
All presented properties especially the result that a synchronization regime
can be destroyed through increasing of the coupling strength is of special
importance for the design of DPLLs in order to realize stable synchronization
in engineering applications.
10
Controlling Phase Synchronization
in Oscillatory Networks
In this chapter we present an automatic control method of phase locking of

regular and chaotic nonidentical oscillations, when all subsystems interact
by a feedback [332]. This method is basing on the well-known principle of
feedback control which takes place in nature and is successfully used in
engineering. Considering the models of coupled systems in biology, neuro-
science, and ecology one can see that in many of them the coupling between
interacting elements is nonlinear, and usually has the form of quadratic func-
tions of the subsystem variables. Such a coupling serves as the basis of an
internal self-organization mechanism leading to a balanced motion in these
systems. Synaptically coupled neurons [342, 343], phase transitions in human
hand movement [344], ecological systems [339], or spinal generators of locomo-
tion [345], are only some well-known examples of balanced cooperative oscil-
latory motion, caused by such a nonlinear coupling. In engineering, nonlinear
coupling, is used, for example, in coupled lasers [333, 334] or phase-locked
loops (PLL) [57] (see also Sect. 2.5).
In contrast to unidirectional (Fig. 10.1a) and bidirectional (Fig. 10.1b)
coupling, the approach presented here supposes the existence of a special
controller, which allows to change the parameters of the controlled systems
(Fig. 10.1c). First we discuss general principles of automatic phase synchro-
nization (PS) for arbitrary coupled systems with a controller whose input is
given by a special quadratic form of coordinates of the individual systems
and its output is a result of the application of a linear differential operator.
We demonstrate the effectiveness of our approach for controlled PS on several
examples (1) two coupled regular oscillators, (2) coupled regular and chaotic
oscillators, (3) two coupled chaotic Rössler oscillators, (4) coupled chaotic
Rössler and Lorenz oscillators, (5) ensembles of locally coupled regular oscil-
lators, (6) ensembles of locally coupled chaotic oscillators, and (7) ensembles
of globally coupled chaotic oscillators.
214 10 Controlling Phase Synchronization in Oscillatory Networks
(a)
OSCILLATOR 1 OSCILLATOR 2
(b)
(c)
α1 α2
X1 X2
CONTROLLER
Fig. 10.1. Three main schemes of interelement coupling. Unidirectional (a),

bidirectional (b), and by a feedback loop (c)
10.1 General Principles of Automatic Synchronization

First, we describe automatic phase locking for the case of two arbitrary regular
or chaotic oscillators given by the system:
ẋ1,2 = F1,2 (x1,2 , ω1,2 ), (10.1)
where x1,2 and F1,2 are n-dimensional vectors, ω1,2 are parameters defining the
time dependence rate (in some cases, frequencies) of oscillators x1,2 (t).1 Our
purpose is to synchronize two such oscillators by using a feedback control of
the timescales of coupled oscillators in such a way that the new characteristic
−1
time scales T1,2 ∼ Ω1,2 become identical. Here Ω1,2 are the mean observed
frequencies of the oscillators being controlled. In order to synchronize coupled
subsystems, we apply a feedback control in the following form (Fig. 10.1c):
ẋ1,2 = F1,2 (x1,2 , ω1,2 (1 + α1,2 u)),

(10.2)
Lu = Q(x1 , x2 ),
where L is a linear operator
dk dk−1 d
L = γk k
+ γk−1 k−1
+ ... + γ1 + a0 (10.3)
dt dt dt
1
Often, the time dependence rates (or frequencies) can be expressed in terms of
multipliers of the right-hand parts: ẋ1,2 = ω1,2 F1,2 (x1,2 ).
10.1 General Principles of Automatic Synchronization 215
acting as a low-pass filter, all γk are nonnegative constants; Q(x1 , x2 ) is a

quadratic form
Q = xT1 Hx2 , (10.4)
where H is a n × n matrix H; α1,2 are feedback controlling coefficients; u(t) is
the control variable added in (10.1) in such a way that it is able to change the
characteristic time scales of the interacting subsystems. The scheme modeled
by (10.2)–(10.4)) works in the following simple manner:
1. The two signals x1 and x2 taken from both interacting systems go to a mul-
tiplier (first part of the “Controller” shown in Fig. 10.1c) which generates
the product u = Q(x1 , x2 ) of these signals. The spectrum of the oscilla-
tions Q consists of a “low” part defined by the difference Ω2 − Ω1 and a
“high” part defined by the sum Ω2 + Ω1 .
2. The signal u is conducted through the low-pass filter (second part of the
“Controller” shown in Fig. 10.1c), which damps the “high” frequency part
due to a specially designed transfer function. Hence, the control variable
u(t) becomes a slow-varying time function, whose spectral band goes to
zero. After the filtering, u(t) is added to both interacting systems (10.2) in
such a way that it may change their characteristic time scales. The main
goal is that this procedure provides a balance between the new timescales,
i.e., Ω2 = Ω1 .
In addition to the comparison of the observed frequencies of the controlled
systems, we are also interested in the evolution of their phase difference, which
is typically used in the study of PS.
This principle of getting synchronization is effectively used in applications
of PLL in a large number of radio- and telecommunication devices, radio-
location [57], coupled lasers [333,334], etc. It also takes place in a huge variety
of examples in nature, where the interaction of some oscillatory objects leads
to their balanced behavior. This balanced behavior is achieved by a nonlinear
interaction of the elements [8, 335–340]. Usually this coupling has the form of
a quadratic function of the interacting elements [346]. This type of coupling
is able to minimize efficiently the oscillator’s phase difference and, therefore,
it leads to synchronization.
It is important to emphasize that this principle can be applied not only
to coupled self-oscillatory systems but to other systems as well. For example,
let us consider a controlled linear oscillator:
ẋ = y,
ẏ = (ω 2 + αu)x + λy, (10.5)
τ u̇ = −u + βxy.
As a result of the control (α = 0 in (10.5)), a stable limit cycle appears.

So the balance of phase and amplitude is achieved by the same mechanisms
of frequency control as in (10.2). Note that at τ = 0, i.e., u = βxy, the

system (10.5) becomes the classical van der Pol equation related to a natural
self-exited generator. This observation may be interpreted as self-control (self-
organization) leading to a synchronization of voltage (x) and current (y) in
the generator circuit.
10.2 Two Coupled Poincaré Systems

As the simplest case, we consider feedback control of PS in two unidirectionally
(drive-response) coupled Poincaré systems:
ẋ1 = −ω1 y1 − λ(x21 + y12 − p2 )x1 ,

ẏ1 = ω1 x1 − λ(x21 + y12 − p2 )y1 ,
ẋ2 = −(u + ω2 )y2 − λ(x22 + y22 − p2 )x2 , (10.6)
ẏ2 = (u + ω2 )x2 − λ(x22 + y22 − p2 )y2 ,
u̇ = −γu + βx1 x2 ,
where ω1,2 are the frequencies, p is the amplitude of oscillations, λ > 0 is a

damping parameter of both oscillators, u is the control variable, and α and β
are the parameters of the controller. The quadratic form Q has the simplest
form Q(x1 , x2 ) = βx1 x2 . In this configuration the first oscillator is the drive
system and the second one is the response, which has to be locked via the
feedback control governed by the variable u. Here the linear operator L in
d
(10.2) has the simple form L = dt + γ. The first oscillator has the solution:
x1 = p cos(ω1 t + ψ1 ). From this and using polar coordinates for the response
system: x2 = ρ cos φ, y2 = ρ sin φ, we rewrite (10.6) in the form
ρ̇ = λρ(p2 − ρ2 ),
φ̇ = (u + ω2 ), (10.7)
u̇ = −γu + βp cos(ω1 t + ψ1 )ρ cos φ.
Let ω2 = ω1 + Δ. Then introducing the new phase variable θ = φ − ω1 t − π/2

and averaging system (10.7) we obtain
θ̈ + γ θ̇ + γΔ + βp2 sin(θ − ψ1 )/2 = 0. (10.8)
This pendulum-like equation describes the synchronization regime of the drive

and response systems. The existence of this regime is defined by a stable steady
¯
state in (10.8) with the coordinates (θ̄ = ψ1 +arcsin(2γΔ/(βp2 )); θ̇ = 0), which
does exist in the range
|Δ/p2 | < β/2γ. (10.9)
In the contrast to the linear diffusive coupling, for example by the terms
β(x2,1 − x1,2 ), the condition (10.9) depends on the values of the amplitudes of
10.3 Coupled van der Pol and Rössler Oscillators 217
oscillations p. Larger p provides the appearance of a synchronization regime

at smaller values of coupling β.
10.3 Coupled van der Pol and Rössler Oscillators

Now two structurally different oscillators are coupled: a regular van der
Pol oscillator and a chaotic Rössler oscillator (see Chap. 2). The equations
describing the control scheme (same L and Q as in (10.2)) for PS of such
oscillators are:
ẋ1 = −ω1 (1 + α1 u)y1 − z1 ,

ẏ1 = ω1 (1 + α1 u)x1 + ay1 ,
ż1 = b − cz1 + x1 z1 , (10.10)
ẋ2 = y2 ,
ẏ2 = −(ω2 (1 + α2 u))2 x2 + ε(p2 − x22 )y2 ,
u̇ = −γu + βx1 x2 ,
where x1 , y1 , z1 are the variables of the Rössler oscillator and x2 , y2 are those
of the van der Pol oscillator. u is again the control variable added to both
subsystems, α1,2 , β and γ are the control parameters. We set β = γ = 1. For
the van der Pol oscillator we choose the following set of parameters: ω2 = 1,
ε = 0.01, and p = 4. The parameters of the Rössler oscillator will be chosen
as: a ∈ [0.15 : 0.2], b = 0.1, c = 8.5, and ω1 = 1.2 For these values the
topology of the chaotic Rössler attractor is rather simple, i.e., phase coherent
(see Sect. 2.3), and one can introduce the phase in the form
φ1 = arctan(y1 /x1 ). (10.11)
Because for the chosen ε the phase trajectory of the van der Pol oscillator
regularly monotonously oscillates around the origin, we can use a similar def-
inition of the phase
φ2 = − arctan(y2 /x2 ). (10.12)
In order to test for the existence of PS between Rössler and van der Pol
oscillators, we use as in the previous chapters the two criteria:
1. PS sets in if the mean frequencies of both coupled subsystems become
equal (frequency locking):
Ω2 = Ω 1 , (10.13)
where the frequencies are defined as
φ1,2 (T ) − φ1,2 (0)
Ω1,2 = lim . (10.14)
T →∞ T
2
For ω1 = 1 in the chaotic Rössler oscillator the mean in time frequency does not
equal to 1 and therefore, uncoupled oscillators have some frequency mismatch.
2. The second criterion we use is the bounded phase difference

|φ2 − φ1 | ≤ const. (10.15)
We consider two types of unidirectional (drive-response) feedback coupling:
(a) We control the characteristic time of the Rössler oscillator (α2 = 0 in
(10.10)) (Fig. 10.2)
(b) We control the characteristic time of the van der Pol oscillator (α1 = 0
in (10.10)) (Fig. 10.3).
In both cases there are critical values of the feedback control parameters
∗
α1,2 corresponding to the onset of synchronization regime:
1. First, we study the case where the van der Pol oscillator is the drive
system and the Rössler oscillator is the controlled response system. We
set a = 0.15, so that the chaotic attractor of the Rössler oscillator is phase
coherent. To illustrate the transition to PS, we plot the mean frequency
difference and the three largest Lyapunov exponents (LEs) vs. the control
parameter α1 (Fig. 10.2a), as well as a bifurcation diagram (Fig. 10.2b).
We find that PS occurs at α1∗ = 0.00123. There the behavior of the Rössler
oscillator remains chaotic but with the mean observed frequency Ω1 equal
to the frequency ω2 of the van der Pol oscillator. A similar situation of
chaotic frequency locking was described in Sect. 2.2, where the effects of PS
were observed in a chaotic system forced by an external periodic signal. In
contrast to this, our interacting subsystems are autonomous and therefore
without coupling two zero LEs exist. In this case the transition to phase
synchronization regime can be analyzed by means of the LEs spectrum. As
it can be seen from Fig. 10.2a the frequency locking occurs approximately
(shortly after) at the same value of α1 for which one of zero LEs becomes
negative. Our numerical experiments (for other values of a) show that
usually the behavior of the controlled Rössler oscillator remains chaotic.
But there are also intervals of α1 ([0.0024; 0.0033]) where the behavior
of the Rössler oscillator becomes periodic (Fig. 10.2). Thus, very small
coupling allows to control chaotic systems in such a way that (1) we can
govern the mean frequency of oscillations and (2) we can get periodic
oscillations too.
2. In the second case of unidirectional feedback coupling, the Rössler oscil-
lator is the drive system and the van der Pol oscillator is the controlled
response system. Here we analyze not only the phase-coherent chaotic
attractor (a = 0.16) but also the funnel attractor (a = 0.24). In the latter
case the topology of attractor is rather complicated and the phase cannot
be defined as in (10.11). Thus we apply another phase definition (see
Sect. 2.3):
φ1 = arctan(ẏ1 /ẋ1 ), (10.16)
and use the same two criteria (10.13) and (10.15) as in the previous case.
We plot in Fig. 10.3a the difference of the mean observed frequencies
10.3 Coupled van der Pol and Rössler Oscillators 219
λ, Ω1−ω2 0.1
−0.1
−0.3
−0.5
0.000 0.001 0.002 0.003 0.004
18.0
14.0
x1max
10.0
6.0
2.0
0.000 0.001 0.002 0.003 0.004
α1
Fig. 10.2. Transition to phase synchronization regime for unidirectionally (α1 = 0

in (10.10)) feedback coupled Rössler and van der Pol oscillators. The van der Pol
oscillator is the drive system and the Rössler oscillator is the controlled response
system. The mean observed frequency of the Rössler oscillator Ω1 at α1∗ = 0.00123
becomes equal to the frequency of the van der Pol oscillator ω2 . (a) The three largest
Lyapunov exponents, one of which is always zero, and the difference of the mean
observed frequencies Ω1 − ω2 (circles) vs. the control parameter α1 . (b) Maxima of
x1 vs. α1
Ω1 − Ω2 vs. the feedback control parameter α1 for different values of a.

In all cases PS occurs at some critical values α2∗ , but with increasing of a
(i.e., increasing the complexity of the Rössler attractor) a larger value α2∗
is needed to achieve the locking. The onset of PS is well manifested in the
bifurcation diagrams (Fig. 10.3b–d). One can see that with increasing of
α2 the interval l of possible maximum values of x2 becomes larger at first.
But at the transition point to synchronous motion a strong shrinking
of the interval l is observed. That means that the variables x2 and y2
become localized in a relatively small area.
We have also performed numerical simulations where the van der Pol and
the Rössler oscillator are mutually coupled by feedback (α1,2 = 0). The effect
of both regular and chaotic PS has been observed there as well.
Fig. 10.3. Transition to phase synchronization regime for unidirectionally (α2 = 0 in

(10.10)) feedback coupled Rössler and van der Pol oscillators. The Rössler oscillator
is the drive system and the van der Pol oscillator is the controlled response system.
The observed frequency of the van der Pol oscillator Ω2 after some α2∗ becomes equal
to the mean frequency of the Rössler oscillator Ω1 . (a) The difference of the mean
observed frequencies Ω1 − Ω2 (circles) vs. control parameter α2 . (b–d) Maximal
values of x2 vs. α2 . The parameters are: (b) a = 0.24, (c) a = 0.22 (for a = 0.22
attractor in Rössler oscillator is periodic), and (d) a = 0.16
10.4 Two Coupled Rössler Oscillators
Now we demonstrate feedback control of chaotic phase synchronization for

two coupled Rössler oscillators:
ẋ1,2 = −ω1,2 (1 + α1,2 u)y1,2 − z1,2 ,

ẏ1,2 = ω1,2 (1 + α1,2 u)x1,2 + ay1,2 , (10.17)
ż1,2 = b − cz1,2 + x1,2 z1,2 ,
u̇ = −γu + βx1 x2 ,
where x1,2 , y1,2 , z1,2 are the variables of the first and second Rössler oscillator,
respectively. We set a = 0.15, b = 0.1, c = 8.5, ω1 = 0.98, and ω2 = 1.02.
Hence, for both oscillators the phase definitions (10.11) can be used. The
10.4 Two Coupled Rössler Oscillators 221
existence of PS between Rössler oscillators is tested again by the criteria

(10.13) and (10.15).
We computed the LEs spectrum (Fig. 10.4a), the mean frequency difference
(Fig. 10.4a), and the evolution of the phase difference (Fig. 10.4b). PS sets in at
the essentially small coupling α1∗ = −α2∗ ≈ 0.000415. Note that shortly before
one of the zero LEs becomes negative. With increasing of α1 , the frequency
difference decreases smoothly (without jump), i.e., a soft transition to PS takes
place. This is manifested in the evolution of the phase difference, namely for
the control parameters close to the critical value α1∗ phase locking at large
time intervals is observed (Fig. 10.4b).
Let us compare the effectiveness of the proposed coupling scheme with
respect to the diffusive coupling usually considered. In the last case the equa-
tions of motions of two diffusively coupled Rössler oscillators are
ẋ1,2 = −ω1,2 y1,2 − z1,2 ,

ẏ1,2 = ω1,2 x1,2 + ay1,2 + β(y2,1 − y1,2 ), (10.18)
ż1,2 = b − cz1,2 + x1,2 z1,2 .
0.2
λ,Ω1−Ω2
0.1
0.0
−0.1
0.0000 0.0002 α1 0.0004 0.0006
250
200
α1=0.0004
150
φ2−φ1
100 α1=0.000405
50 α1=0.00041
α1=0.000415
0
−50
0 20000 40000 60000 80000 100000
time
Fig. 10.4. Synchronization of two coupled Rössler oscillators (10.17). The parame-
ters are a = 0.15, b = 0.1, c = 8.5, ω1 = 0.98, ω2 = 1.02, α1 = −α2 , and β = γ = 1.
(a) The four largest LEs and the difference of the mean observed frequencies Ω1 −Ω2
(circles) vs. the control parameter α1 . (b) Difference of the phases φ2 − φ1 for
nonsynchronous (α1 = 0.0004; 0.000405; 0.00041) and synchronous (α1 = 0.000415)
regimes
In order to estimate the critical coupling strength corresponding to the

appearance of synchronization regime, we convert into cylindric coordinates:
x1,2 = ρ1,2 cos φ1,2 and y1,2 = ρ1,2 sin φ1,2 . Then for the feedback coupled
oscillators (10.17), the averaged equation for the difference θ = ψ2 − ψ1 of
slow phases ψ1,2 = φ1,2 − ω0 t is
θ̈ + αθ̇ − αβρ1 ρ2 sin θ = αΔ. (10.19)
If we neglect the fluctuations of the amplitude, (10.19) has the stationary

solution.
θ̄ = − arcsin(Δ/(βρ1 ρ2 )). (10.20)
This state exists and is stable if
Δ
| | < β. (10.21)
ρ1 ρ2
The equation for the phase difference of (10.18) is (for details see Sect. 4.2.1)
β ρ21 + ρ22
θ̇ − sin θ = Δ (10.22)
2 ρ1 ρ2
The stable stationary state
θ̄ = arcsin(2Δρ1 ρ2 )/(β(ρ21 + ρ22 )) (10.23)
exists in the range

β > (|2Δρ1 ρ2 )|/(ρ21 + ρ22 ). (10.24)
If we take not strongly different oscillators (i.e., ρ1 ≈ ρ2 ) this range is
reduced to
β > |Δ|. (10.25)
Therefore, by equivalent parameters of the interacting oscillators the syn-
chronization range for the feedback coupling (10.17) is p2 = ρ1 ρ2 times larger
than for the diffusive coupling (10.18). This estimation is in very good agree-
ment with our numerical results.
We have also analyzed the synchronization transitions for the simplest
case of a linear operator L. For γ 1 the filtered control variable u can be
expressed by sin(φ2 − φ1 ), where the phases φ1,2 are introduced by (10.11).
Then (10.17) can be rewritten as
ẋ1,2 = −ω1,2 (1 + α1,2 sin(φ2,1 − φ1,2 ))y1,2 − z1,2 ,

ẏ1,2 = ω1,2 (1 + α1,2 sin(φ2,1 − φ1,2 ))x1,2 + ay1,2 , (10.26)
ż1,2 = b − cz1,2 + x1,2 z1,2 .
10.5 Coupled Rössler and Lorenz Oscillators 223
1.05
1.04
a = 0.15
1.03
Ω2/Ω1
a = 0.16
1.02 a = 0.17
a = 0.18
1.01
1.00
0.99
0.00 0.02 0.04 0.06
α1
Fig. 10.5. Synchronization of two coupled Rössler oscillators (10.17). The ratio of
the mean observed frequencies Ω2 /Ω1 vs. the control parameter α1 . The parameters
of the individual oscillators are the same as in Fig.10.4
The dependency of the mean frequencies ratio Ω2 /Ω1 on the parameter α1 =

−α2 for different a shows the onset of PS again for a very small coupling
strength (Fig. 10.5).
10.5 Coupled Rössler and Lorenz Oscillators

Now we will apply the automatic phase synchronization to the coupled Rössler
and Lorenz oscillators, i.e., chaotic oscillators with a well-pronounced differ-
ence in topology (see Sects. 2.3 and 2.4). The model is:
ẋ1 = τ (−ω(1 + α1 u)y1 − z1 ),

ẏ1 = τ (ω(1 + α1 u)x1 + ay1 ),
ż1 = b − cz1 + x1 z1 ),
ẋ2 = σ(y2 − x2 ), (10.27)
ẏ2 = rx2 − y2 − x2 z2 ,
ż2 = (1 + α2 u)(−bz2 + x2 y2 ),
u̇ = −γu + x1 z2 ,
where x1,2 , y1,2 , z1,2 are the variables of the Rössler and Lorenz oscillators,
respectively. The parameters a, b, c and phase of the Rössler oscillator are the
same as in the previous case; ω = 0.98, τ = 8.3, σ = 10, r = 28, and b = 8/3
and the phase of the Rössler oscillator is measured as before. The phase of
the Lorenz oscillator is calculated as θ = arctan((z2 − 27)/( x22 + y22 − 12))
(see Sect. 2.4). In Fig. 10.6 we present results of the transition to chaotic PS
between Rössler and Lorenz oscillators. One can see an interval of α where
PS occurs. Therefore, using the proposed scheme we are also able to achieve
chaotic PS between oscillators with a strong difference in their topology.
0.00
− 0.05
Ω1−Ω2
− 0.10
− 0.15 γ = 0.6
0.8
− 0.20 1.0
1.2
− 0.25
0.0 0.0005 0.001 0.0015 0.002
α
Fig. 10.6. Synchronization of coupled Rössler and Lorenz oscillators (10.27). The
parameters are a = 0.15, b = 0.1, c = 8.5, ω = 0.98, σ = 10, r = 28, b = 8/3, α1 =
−α2 = α, and β = γ = 1. The difference of the mean observed frequencies Ω1 − Ω2
vs. the control parameter α
10.6 Principles of Automatic Synchronization

in Networks of Coupled Oscillators
We can straightforward extent this control scheme proposed for two coupled
oscillators to a network of oscillators. Let us consider an ensemble of arbitrary
regular or chaotic oscillators given by:
ẋj = Fj (xj , ωj ), (10.28)

j = 1, . . . , N
where xj and Fj are n-vectors, ωj are parameters defining the time dependence
rate of the oscillations xj (t), and N is a number of oscillators.
In order to synchronize these systems, we generalize the feedback control
(10.2) in the following form:
ẋj = Fj (xj , ωj + αj uj ),
Luj = Qj (x1 , . . . , xN ), (10.29)
j = 1, . . . , N,
where L is again a linear operator acting as a low-pass filter; the function

Qj (x1 , . . . , xN ) is:

N
Qj (x1 , . . . , xN ) = Qk (xj , xk ), (10.30)
k=1,k=j
where Qk is a quadratic form Qk = xTj Hxk which characterizes the coupling

between the jth and the kth oscillators.
10.7 Synchronization of Locally Coupled Regular Oscillators 225
OSCILLATOR OSCILLATOR OSCILLATOR

j−1 j j+1
CONTROLLER CONTROLLER
j−1 j
Fig. 10.7. Local feedback coupling in a chain of oscillators
Now we study whether the control variable uj (t) added to each oscillator
can provide a synchronous behavior between interacting elements. Figure 10.7
presents a simple scheme which roughly describes the proposed coupling tech-
nique (cf. also the scheme of a multichannel chain of partial digital phase-
locked loops connected in parallel by phase-mismatching signals, Fig. 9.1).
We demonstrate the method of feedback control for PS for ensembles of
(1) locally coupled regular oscillators (Sect. 10.7), (2) locally coupled chaotic
oscillators (Sect. 10.8), and (3) globally coupled chaotic oscillators (Sect. 10.9).
10.7 Synchronization of Locally Coupled Regular

Oscillators
As the simplest case we analyze feedback control of PS in an ensemble of
locally mutually coupled Poincaré systems:
ẋj = −(αj uj + ωj )yj − λ(x2j + yj2 − p2 )xj ,

ẏj = (αj uj + ωj )xj − λ(x2j + yj2 − p2 )yj , (10.31)
u̇j = −uj + βj+1 xj yj+1 + βj−1 xj yj−1 ,
where j = 1, ..., N , ωj are the frequencies, p is the amplitude of oscillations,
λ > 0 is a damping parameter of the oscillators, uj is the control variable,
and αj and βj are the parameters of the jth controller. We assume free-end
boundary conditions: β0 = βN +1 = 0. For the quadratic form Qj we take the
simplest form of coupling with nearest neighbors
Qj = βj+1 xj yj+1 + βj−1 xj yj−1 . (10.32)

d
In this example we set the linear operator L in the form L = dt + 1. Using
polar coordinates xj = ρj cos φj , yj = ρj sin φj , we rewrite (10.31) in the form
ρ̇j = λρj (p2 − ρ2j ),

φ̇j = αj uj + ωj ,
u̇j = −uj + βj+1 ρj ρj+1 cos φj sin φj+1 (10.33)

+βj−1 ρj ρj−1 cos φj sin φj−1 ,
j = 1, . . . , N
As in the previous chapters, we start with a linear increasing distribution

of individual frequencies ωj = ω1 + Δ(j − 1), and αj = α, βj = β. Then
by introducing the phase difference variable θj = φj − φj+1 , α̂ = αβ/2 and
averaging the system (10.33), we obtain:
φ̈1 + φ̇1 = ω1 + α̂p sin θ1 , (10.34)
θ̈j + θ̇j = Δ + α̂p (sin θj+1 − 2 sin θj + sin θj−1 ),

(10.35)
j = 1, . . . , N − 1
with the boundary conditions: θ0 = θN = 0. These equations describe the
synchronization regime in an ensemble of coupled Poincaré systems. The exis-
tence of a PS regime is defined by a stable steady state in (10.35). This state
(θ̄1 , . . . , θ̄j , . . . , θ̄N −1 ) in system (10.35) corresponds to a regime of global syn-
chronization in the chain. Hence, the system of equations for the stationary
phase differences θ̄n can be written as (see Chaps. 5–7)
Δ + α̂p(sin θ̄2 − 2 sin θ̄1 ) = 0,

Δ + α̂p (sin θ̄j+1 − 2 sin θ̄j + sin θ̄j−1 ) = 0, (10.36)
j = 2, . . . , N − 2,
Δ + α̂ p(sin θ̄N − 2 sin θ̄N −1 ) = 0
The distribution of θ̄j is (see Chap. 5):

Δ
sin θ̄j = (N j − j 2 ). (10.37)
2α̂p
It follows from (10.37) that the system (10.35) can have 2N −1 steady states.
But only one of them (θ̄j ∈ [−π/2; π/2] for all j = 1, . . . , N − 1) can be stable.
As the frequency mismatch Δ is increased, the condition for the existence of
steady states
Δ
| (N j − j 2 )| < 1 (10.38)
2α̂p
is violated first for j = N/2 at even N , i.e., for the middle element in the
chain. Thus, the condition for the existence of a stable steady state in the
N -element chain is given by the inequality
ΔN 2
| | < 1. (10.39)
8α̂p
The frequency of globally synchronized oscillations Ωs may be determined
from the (10.35), such that
10.7 Synchronization of Locally Coupled Regular Oscillators 227
Δ
Ωs = ω1 + (N − 1) (10.40)
2
Then the frequencies for all elements are equal to the mean frequency of
the elements in the ensemble. With an increase of the frequency mismatch Δ
(or decrease of the coupling α), a loss of global synchronization regime takes
place. For a long chain the two synchronization clusters occur, i.e., the chain
is divided into two clusters each of size (N/2) both consisting of mutually syn-
chronized oscillators. Further increase of Δ (decrease of α) leads to a sequence
of destruction of the one cluster structure of the synchronized elements and
to the appearance of another structure. This sequence obtained in numerical
experiments is shown in Fig. 10.8. From this figure we recognize two types of
transitions between cluster structures. In the first type a hard transition with-
out intermediate structures occurs from the state with n (n + 1) clusters to
the state with n + 1 (n) clusters (see, for example, the interval [0.019 : 0.023]).
In the second type, a soft transition happens with a smooth transition of
intermediate structures one into the other. As follows from Fig. 10.8, the the-
oretically and numerically obtained conditions of the global synchronization
regime and the frequency of the global synchronization regime are in very
good agreement.
1.020
1.018
1.016
1.014
1.012
1.010
Ωj
1.008
1.006
1.004
1.002
1.000
0.998
0.00 0.01 0.02 0.03 0.04 0.05 0.06
α
Fig. 10.8. Observed frequencies Ωj in a chain of Poincaré systems (10.31) with
linear distribution of individual frequencies vs. α. N = 20, p = 1, ω1 = 0.98,
Δ = 0.001
10.8 Synchronization of Locally Coupled Chaotic

Oscillators
Now we will demonstrate feedback control of chaotic PS in ensembles of locally
coupled Rössler oscillators
ẋj = −(ωj + αj uj )yj − zj ,

ẏj = (ωj + αj uj )xj + ayj ,
żj = b − czj + xj zj , (10.41)
u̇j = −γj uj + βj+1 xj yj+1 + βj−1 xj yj−1 ,
j = 1, . . . , N
We set a = 0.15, b = 0.1, c = 8.5, αj = α, γj = βj = 1. As in the Poincaré

systems, we introduce a gradient distribution of natural frequencies ωj = ω1 +
Δ(j −1) with ω1 = 0.98, and Δ = 0.0001. Another variant considered below is
a random distribution of natural frequencies in the range [ω1 , ω1 + Δ(N − 1)].
We again assume free-end boundary conditions: β0 = βN +1 = 0. The control
d
scheme Qj is the same as in Sect. 10.7 and L = dt + γj . As a condition of
synchronization regime, we again consider the coincidence of the observed
partial frequencies defined according to (10.14).
We have performed numerical simulations with a chain of 100 elements
with both, a linear and a random distribution of the individual frequencies.
For each element for different α the frequency Ωj has been calculated. We
find that in both cases by increasing of coupling α all frequencies Ωj become
equal, i.e., global chaotic PS sets in.
We have also analyzed synchronization transitions for the simplest case
of a linear operator L. For γj 1 the filtered control variable uj can be
expressed as
uj = sin(φj+1 − φj ) + sin(φj−1 − φj ), (10.42)
where the phases φj are introduced by (10.11). Then (10.41) can be rewrit-
ten as
ẋj = −ωj yj − zj
−ωj α(sin(φj+1 − φj ) + sin(φj−1 − φj ))yj ,
ẏj = ωj xj + ayj (10.43)
+ωj α(sin(φj+1 − φj ) + sin(φj−1 − φj ))xj ,
żj = b − czj + xj zj ,
j = 1, . . . , N.
The dependencies of the mean frequencies Ωj on the parameter α with linear

(Fig. 10.9) resp. random (Fig. 10.10) distributions of the individual frequencies
exhibit the onset of PS for a very small coupling term.
10.8 Synchronization of Locally Coupled Chaotic Oscillators 229
1.04
1.035
Ωj
1.03
α = 0.0
0.04
0.1
0.16
0 20 40 60 80 100
j
Fig. 10.9. Mean frequencies Ωj in a chain of Rössler oscillators (10.41) with a

linear distribution of individual frequencies for different α. The number of elements
N = 100, ω1 = 0.98, Δ = 0.0001
1.06
α=0.0
0.01
1.055 0.02
0.06
1.05
1.045
Ωj
1.04
1.035
1.03
1.025
0 20 40 60 80 100
j
Fig. 10.10. Mean frequencies Ωj in a chain of Rössler oscillators (10.41) with

randomly distributed frequencies in the interval [0.98,1] for different α. The number
of elements N = 100
10.9 Synchronization of Globally Coupled Chaotic

Oscillators
Finally we study the potential of the presented method for globally coupled
Rössler oscillators
ẋj = −(ωj + αj uj )yj − zj ,
ẏj = (ωj + αj uj )xj + ayj ,
żj = b − czj + xj zj , (10.44)
N
u̇j = −γj uj + xj k=1,k=j βk yk ,
j = 1, ..., N.
We take the same parameters of individual elements as in Sect. 10.8 and ran-
domly distributed frequencies ωj . Let us fix again all γj 1. Then the filtered
control variable uj can be described in the form:

N
uj = sin(φj − φk ) (10.45)
k=1,k=j
and (10.44) can be rewritten as
1.0
0.8
0.6
<R>
0.4
0.2
0.0
0.0 0.000005 0.00001 0.000015 0.00002
α
Fig. 10.11. Frequency entrainment in the ensemble of globally feedback coupled
Rössler oscillators (10.46) with randomly distributed frequencies in the interval
[0.98,1]. Order parameter R (10.47) vs. coupling α. The number of elements
N = 100
10.10 Conclusions 231
ẋj = −ωj yj − zj
N
−ωj α k=1,k=j sin(φj − φk )yj ,
ẏj = ωj xj + ayj
N (10.46)
+ωj α k=1,k=j sin(φj − φk )xj ,
żj = b − czj + xj zj ,
j = 1, ..., N.
Following [8], we characterize the degree of synchronization by means of the

order parameter:
Nl
R = lim , (10.47)
N →∞ N
where Nl is the size of the largest cluster of synchronized oscillators. This

frequency order parameter is for fully incoherent oscillators R = 0, and reaches
for globally synchronized behavior the maximum R = 1. The order parameter
R averaged over ten samples of randomly distributed frequencies ωj is shown
in Fig. 10.11. We see that there exists a critical value α∗ such that all oscillators
become synchronized. This transition from a fully incoherent behavior to a
fully coherent (synchronized one) has been typically observed in ensembles of
globally coupled elements.
10.10 Conclusions
In this chapter we have described a feedback control method for automatic
phase locking of regular and chaotic oscillators.
The main advantages of this method are the following:
– The effect of the amplitudes of the interacting subsystems on the difference

of their phases provides a high efficiency of this approach.
– The proposed method can be used for automatic synchronization of
oscillators of different nature (regular and chaotic), and different topol-
ogy (e.g., coupled Rössler and Lorenz oscillators) and complexity (e.g.,
chaotic and hyperchaotic Rössler oscillators).
– A phase synchronization regime sets already in at very small values of
control parameters, which is very important from an energetical point of
view.
– The method can be used to synchronize elements coupled in small (two
units) and large (chains and lattices) ensembles. In the latter case the
coupling can be local or global.
This presented approach can be helpful (1) for the understanding of self-
organization mechanisms in many systems in nature and possibly also for
their manipulation and (2) for the design of different schemes of automatic
synchronization and could be applied to communication, engineering, ecology,
and medicine.
Close to the problem of controlled synchronization is related the problem

of chaos control which consists (1) in suppression of chaos and transition
to regular behavior or (2) in control of properties (e.g., spectrum power) of
chaotic behavior. First this problem was formulated in [347, 348] and after
the paper [349] became very attractive. Nowadays there are a lot of different
controlling schemes most of them described in [350].
11
Chains of Limit-Cycle Oscillators
Previous chapters were mainly devoted to the investigation of synchronization

effects inside one chain or network of coupled oscillators and rotators. In this
chapter, phenomena of collective oscillations in a system of two coupled chains
of identical limit-cycle oscillators are investigated [331, 351–353]. It is shown
that a pair of coupled chains with different collective frequencies exhibit stable
fronts between two possible asymptotic states: synchronous oscillations and
oscillation death. The inhomogeneous states formed by the fronts persist at
weak coupling between the oscillators in each chain due to the discrete space
variable, thus providing conditions for the existence of localized structures. At
stronger coupling, the interface between both the regions of the chains may
propagate. Different examples of synchronization patterns and their dynamics
are presented including nontrivial effects such as (1) synchronized clusters
induced by disorder and (2) transitions from nonpropagation to propagation
of fronts via intermittency.
The chapter is organized as follows. We start (Sect. 11.1) with a short
introduction to the problem and the presentation of the studied model. Then
(Sect. 11.2) mechanisms of synchronized cluster formations are presented.
Chains of oscillators coupled only through the dissipative terms are discussed
in Sect. 11.3. Propagation versus nonpropagation transitions and the influence
of noise on chains of dissipatively and conservatively (non-zero “dispersion”)
coupled oscillators are treated in Sect. 11.4.
11.1 Introduction and Model

The investigation of interaction of large systems, each being an ensemble con-
taining many coupled oscillators [354], has growing interest from theoreti-
cal and practical points of view. In particular, weakly coupled chains model
the interaction between two parallel, long Josephson junctions [355, 356], the
boundary between two semi-infinite chains of atoms absorbed on the surface
of metals [357,358], or the interaction of excitations in elastic ferroelectric and
234 11 Chains of Limit-Cycle Oscillators
ferromagnetic systems [359,360]. The main phenomena usually studied in this

context are (1) synchronization between the elements inside each ensemble
and (2) synchronization between the elements belonging to different ensem-
bles. Phase synchronization in coupled continuous space–time systems, i.e.,
systems modeled by partial differential equations have been studied for cou-
pled complex Ginzburg–Landau equations (GLEs). For unidirectional [361]
and bidirectional [362, 363] coupling between GLEs and for an externally
forced GLE [364], transitions to phase and frequency locking were studied.
The motivation that GLE has attracted a big attention of research is the
following. The GLE describes the modulation of a spatially extended system
which has undergone an Andronov–Hopf bifurcation. It is thus very general,
as shown from the analysis of many physical, biological, and chemical systems
(see [145]).
Here we consider two coupled chains: A and B of equal length (see
Fig. 11.1) of oscillators whose dynamics in a quasiharmonic approximation
is described by the following equations for the slowly varying complex ampli-
tudes aj and bj , i.e., we consider two coupled discrete GLEs (synchronization
phenomena in a single discrete GLE have been studied in Chap. 6):
Chain A:
a˙j = (p + iΔ)aj − (1 + iα)|aj |2 aj +

(d1 + id2 )(aj+1 − 2aj + aj−1 ) + c(bj − aj ), (11.1)
Chain B :
b˙j = pbj − |bj |2 bj

(11.2)
+d1 (bj+1 − 2bj + bj−1 ) + c(aj − bj ),
for j = 1, . . . , N , with the boundary conditions a0 = a1 , aN +1 = aN , b0 = b1 ,
and bN +1 = bN (free end). Here Δ and α describe the linear and nonlinear
frequency mismatches of the oscillators, p is the growth rate, d1 and d2 are
coefficients of active and reactive coupling between the elements inside such
chains, respectively, and c is the coefficient of coupling between both chains.
This is a fairly generic system, which probably does not directly describe
experimental situations, but is an appropriate model to study phenomena
that have relevance in nature and engineering.
A
a1 a2 a3 aN
B
b1 b2 b3 bN
Fig. 11.1. Two coupled chains (11.1) and (11.2)

11.3 Dissipative Coupling (Zero “Dispersion”) 235
11.2 Mechanism of Localized Structure Formation
We mainly focus on localized structures with partially or completely syn-

chronized oscillators. One basic mechanism responsible for the formation of
such structures (adequately referred to as coherent structures) is associated
with the oscillator death (see Chap. 6). Due to this effect, the trivial equilib-
rium state of (11.1) and (11.2) is stable for sufficiently strong coupling c and
large linear mismatch Δ. At the same time, if the linear mismatch can be
compensated by the nonlinear one (α|ajk |2 ≈ Δ), an equilibrium state with
finite amplitudes is also stable within a certain range of parameters. For a
homogeneous state, we obtain
|aj |2 = |bj |2 = Δα − 4(c − p)+

#1/2 −1
2 1 2 1 2 2 2 1 2
[Δα−4(c−p)] −16 1+ α Δ −c −(c−p) 4 1+ α ,
4 4 4
(11.3)
with
2 −1
2 2 2 2 2 1 1 2
4c − 4(c − p) < Δ < 4c − 4(c − p) + Δα − 2(c − p) 1+ α ,
2 4
(11.4)
Δα − 4(c − p) > 0.
One can naturally expect that, for certain limitations on the value of the
coupling between the elements, states with aj = bj = 0 and aj = bj = 0 can
coexist in both chains and thus form stationary fronts and localized structures.
Depending on the initial and boundary conditions and on external actions,
these states may form patterns of various configurations and topologies as
discussed in the following.
11.3 Dissipative Coupling (Zero “Dispersion”)

Here we consider dissipatively (d2 = 0) coupled oscillators in each chain (11.1)
and (11.2) and start with a desynchronization phenomenon.
11.3.1 Desynchronization of Front Propagation
Let us define the initial amplitude distribution as a step function:

2 2 0.0 j = 1, . . . , 50
aj = bj =
0.72 j = 51, . . . , 128
(here Δ = 2; α = 5.75; p = 0.5; c = 0.51).

0.8
V 0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
d1
Fig. 11.2. Velocity v of the front of the transition from the excited to the non-excited
state in (11.1) as a function of the coefficient of coupling between the cells inside
the chain (Δ = 2.0, α = 5.75, c = 0.51). “+” correspond √ to numerical simulations.
Dashed line corresponds to the dependence v ≈ 0.92 d1 − 0.09
At weak coupling between the elements inside the array (d1 dcr 1 = 0.09),
a motionless (stationary) front that separates the oscillators into excited states
and non-excited ones is formed. At strong coupling (d1 > dcr 1 ), the regime of
oscillator death prevails, and the region in which this is realized broadens and
embraces all the array. The transition process and the influence of it on the
initial conditions occur simultaneously in a complicated way with a special
complication of the structures formed near the critical value dcr 1 ≈ 0.09. But
the velocity v of the front propagation of the oscillator death may still be
determined for some realizations, i.e., at a definite choice of initial conditions.
Numerical simulation shows that the velocity remains almost unchanged in
the range of the chain coupling c = [0.5, 1.0] and has a root dependence on
cr
√ parameter at d1 > d1 as it is typical for critical phenomena:
the coupling
v ≈ 0.92 d1 − 0.09 (Fig. 11.2). It is worth to note that a deviation from the
root dependence and the jump-wise transition from zero to nonzero velocities
at nonzero coupling coefficients d inside the chains is observed only in discrete
models. Partial differential equations for complex amplitudes predict a smooth
transition starting from dcr1 = 0.
A more detailed analysis of the transition processes at the interface
between both regions having two different states gives a direct confirmation
that the transition to an unexcited state is actually determined by desyn-
chronization. This is illustrated in Fig. 11.3 by time series of the real and
imaginary parts of the complex amplitudes aj and bj on one arbitrary chosen
cell (j = 100) in each chain at the moving front. Another useful characteristic
is the difference of phases ϕa and ϕb in both chains. The damping of the oscil-
lations is preceded by an abrupt change in the phase difference ϕa100 − ϕb100
with a subsequent formation of oscillations having different frequencies which
1.0
|a100|2 (a)
0.8 s
0.6
0.4
0.2
0.0
100 150 200 250 300
1.0
Re(a100)
(b)
0.5 Re(b100)
0.0
− 0.5
−1.0
100 150 200 250 300
t
Fig. 11.3. Desynchronization in (11.1) and (11.2). Plot of the oscillation intensity
ϕa −ϕb
(|a100 |2 ) and of s = sin2 ( 100 2 100 ) characterizing the phase difference (a) and real
parts of complex amplitudes a100 and b100 (b) of the 100th oscillator at the transition
to the trivial equilibrium state aj = bj = 0 (Δ = 2.0, α = 5.75, p = 0.5, c = 0.51,
d1 = 0.15, N = 128)
are close to the corresponding intrinsic frequencies (Ωa ∼ Δ, Ωb ∼ 0) of the

individual chains.
11.3.2 Localized Synchronization Structures
Another typical feature in the system (11.1) and (11.2) is the existence of
localized synchronization structures. The bistability of the stationary regimes
allows us to chose initial conditions in such a way that these structures can
be formed in any place of the chain. Fig. 11.4 presents the structures that are
formed when localized oscillations aj = bj = |a| with amplitudes determined
according to (11.3) are specified at the initial instant of time. The distributions
of initial and established amplitudes almost coincide at weak coupling (d1
dcr
1 ) between the oscillators inside the chains. As the internal coupling d is
increased, the influence of the coupling with unexcited oscillators becomes
significant. As a result the amplitude |aj | and, consequently, the nonlinear
phase shift determined by the term α|aj |2 aj strongly depend on their position
in the chain. At d1 ≈ dcr 1 , this shift increases so that a local breakdown of
0.5
d1=0.001
0.05
0.4 0.1
0.3
|aj|2
0.2
0.1
0.0
0 16 32 48 64 80 96 112 128
j
Fig. 11.4. Stationary structures of system (11.1) and (11.2) with parameters Δ =
2.0, α = 5.75, p = 0.5, c = 0.51, and d = 0.001, 0.05, 0.1 that are realized under the
initial conditions a2j = b2j = 0 for j = 1, . . . , 50, a2j = b2j = 0.434 for j = 51, . . . , 78,
and a2j = b2j = 0 for j = 79, . . . , 128
the synchronization regime may occur and smaller structures are formed. The
resulting amplitude distribution near this critical value depends significantly
on the coupling and the initial amplitude.
11.3.3 Nonlocal Synchronization in Nonhomogeneous Chains
Now we explain an interesting mutual synchronization of elements in two

isolated clusters. We take here a nonhomogeneous distribution of the natu-
ral frequencies, namely a staircase distribution, and initial conditions which
allows us to form three different clusters which we call I (elements j = 13, 14),
II (elements j = 16, 17), and III (elements j = 19, 20). At relatively large
coupling strength, oscillations in clusters I and III become completely syn-
chronized, while the cluster II remains unsynchronized. Our detailed study
shows that the phase difference ϕI − ϕIII of the oscillations of these clus-
ters (see Fig. 11.5) tends to the same value Δϕ as t → ∞ irrespective of
the initial value. This fact and the value of Δϕ = 0 itself are independent
of the absence or presence of cluster II at j = 16, 17, if its frequency dif-
fers markedly or if it is incommensurate with the frequency of the mutually
synchronized clusters I and III (e.g., because of local inhomogeneity in the
chain). This resembles the effect of indirect or nonlocal synchronization which
was observed for three coupled oscillators [354] and in many other examples
0.5
φ14−φ19
−0.5
1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000
t
Fig. 11.5. Nonlocal synchronization regime in (11.1) and (11.2). Onset of in-phase
synchronization regime at different initial conditions in two isolated clusters (I for
j = 13, 14 and III for j = 19, 20) separated by cluster II for j = 16, 17 for the chains
with parameters N = 32, p = 1.0, α = 8.0, c = 2.0, d = 0.15, and Δj = 6.0 for
j = 1, . . . , 15, and j = 18, . . . , 32 and Δj = 5.9 for j = 16, 17
of oscillatory ensembles (see Chaps. 5–9, and [365]). A possible explanation of

the synchronization of clusters I and III is that a non-resonant interaction of
clusters I and III with cluster II is inefficient at small d because of a relatively
large frequency difference. Whereas the effect of coupling even between more
remote but resonantly (due to the closeness of their frequencies) interacting
clusters I and III (ωI = ωIII ) is significant and sufficient for synchronization,
the phase difference ϕI − ϕIII in the established regime does not depend on
its initial value. Investigation of conservation of the synchronization between
the cells and between groups of cells in spatially extended complex structures
is interesting in the context of modeling processes of recognition and storage
of spatial images in the cortex [354, 366].
11.3.4 Fully Incoherent (Turbulent-Like) Oscillations
In [367] it was shown that a single GLE can exhibit spatiotemporally

disordered regimes, especially amplitude and defect turbulence. In some cases
the disordered regimes are reviewed and argued to be of the spatiotemporal
intermittency type [368]. Two coupled discrete GLEs demonstrate a very
similar intermittent behavior. Here we consider the case of an asymmetric
coupling between the chains (for its description c1 and c2 must be substituted
for c in (11.1) and (11.2), respectively). Supposing that the oscillatory struc-
ture is homogeneous (aj = a, bj = b), stable regimes of pulsations and
nonsynchronized oscillations are possible besides the regimes of synchroniza-
tion and oscillator death. Clusters of oscillations in one of these regimes,
depending on the initial conditions, are formed at sufficiently weak coupling
d1 . When one of the competing regimes is synchronized and the other one
is not, significantly nonstationary structures may be formed in the system,
which leads to a specific turbulent regime called spatiotemporal intermittency
(Fig. 11.6a) (see also transition to spatiotemporal intermittency in chains of
(a) (b)
1000
800
600
400
200
t=0
1 50 100 1 50 100
j j
Fig. 11.6. Amplitude turbulence in coupled chains. Spatiotemporal distribution of

oscillation intensity in the course of pattern formation in system (11.1) and (11.2) at
asymmetric coupling (c1 = c2 ) for weak d1 = 0.05 (a) and strong d1 = 0.3 diffusion
(b) in the chain. Parameters are: p = 0.5, Δ = 2, α = 5.75, c1 = 0.51, c2 = 5.0.
Initial conditions are: a2j = b2j = 0.01 for j = 1, . . . , 50 and a2j = b2j = 0.5 for
j = 51, . . . , 100
11.4 Nonscalar (Dissipative and Conservative) Coupling 241
coupled intermittent chaotic maps described in Chap. 8). In this case, the
front propagation becomes chaotic due to a non-stationary dynamics of the
domains, giving rise to a process similar to directional percolation [369].
As the coupling d1 between the cells is increased, only those structures
which consist of a small amount of oscillators with strong oscillations survive,
like in the case of symmetric coupling between the chains. At a small degree of
asymmetry of coupling between the chains, intensive oscillations are excited
in isolated groups of oscillators and occur as short trains. The position of such
oscillations is random in space and time. As the nonsymmetry of coupling is
increased, the average density and repetition rate of these trains grow and
there occurs a transition from oscillations intermittent in space and time to
developed turbulence – spatiotemporal intermittency (STI) (Fig. 11.6b).
11.4 Nonscalar (Dissipative and Conservative) Coupling

More intriguing effects occur in the collective dynamics of the chain when there
is a nonzero conservative (nonzero “dispersion”) coupling. First, we consider
the phenomenon of the formation of spontaneous bursting structures.
11.4.1 Bursting Structures
Suppose that a localized excitation with amplitudes close to the stationary

ones – as described by (11.3) – is defined at the initial moment of time.
Then, at d2 = 0, either localized structures are formed (if d1 < dcr 1 ) or the
excitations are damped out (if d1 > dcr 1 ). The additional presence of a reactive
component in the coupling coefficient (d2 = 0) in one of the chains induces
localized structures in the latter case too. The arising features significantly
depend on the sign of the product αd2 . If αd2 < 0, i.e., when the corresponding
Schrödinger equation describes self-compression of the localized perturbation
in the limit α → ∞, d2 → ∞, the damping is accelerated. In the opposite
case (αd2 > 0), the occurring structures are the result of a competition of
two effects: the expansion of the region of oscillator death described above
and the spreading, in the opposite direction, of a localized excitation due to
the combined action of the nonlinearity αa|a|2 and the dispersion d2 . Let us
consider possible structure formations as a results of such a competition for
an example of the evolution of a localized perturbation with a large initial
amplitude for p = 0.5; d1 = 0.3; c = 0.51; Δ = 2.0; α = 5.75; and N = 100.
As the dispersion d2 is increasing at αd2 > 0, the propagation velocity
of the fronts forming the localized structure decreases. At a certain critical
value d1 ≈ 0.3, the effect of oscillator death is balanced by a nonlinear self-
expansion of the perturbation, and a stable structure is formed. Its size grows
as the dispersion d increases and, eventually, when the second critical value
d2 ≈ 0.9 is attained, the effect of self-expansion becomes predominant. This
leads to delocalization of the excitation which finally embraces all the array.
The most interesting phenomena occur when the values of d2 approach

d2 . Within the framework of a rigorously symmetric arrangement (aj ≡
a100−j , bj ≡ b100−j , i = j, . . . , 50), an almost stationary amplitude distrib-
ution |aj |, |bj | is first formed from the homogeneous localized excitation for d2
slightly smaller than the critical value. This amplitude distribution is retained
at large intervals of dimensionless time (∼ 103 –104 ). After that, for much
shorter time scales (≈ 10), the structure (regions of high-intensity oscillations)
abruptly expands symmetrically and then rapidly recovers its initial quasista-
tionary amplitude distribution. Still further, this process is repeated. As d2 is
increased, the average repetition rate of such bursts increases. The time inter-
vals between these events are random, even in the absence of noise. Bursting
structures in this case look like in the case of nonsymmetric structures located
at the boundary of the chains (see Fig. 11.7).
Additional forcing or perturbation of the initial conditions that breaks the
symmetry with respect to the center of the chain does not change significantly
the structure in the time intervals between the bursts. However, the structures
shift in one or another direction at each burst. These shifts occur randomly,
both at noise forcing and without it, when the initial conditions are taken to
be asymmetric. A random walk of the structure eventually leads it to the end
of the chain. Thus, we naturally encounter the problem of a boundary layer
in the considered chain of self-excited oscillators.
11.4.2 Nonpropagation to Propagation Transition

via Intermittency
Boundary layer structures is formed as a result of the above mentioned random

walk-like behavior. From a physical point of view, the following interpretation
can be useful. Consider two semi-infinite coupled chains, with a boundary
layer structure formed by some initial perturbation of several elements near
the boundary at t = 0. In this case for a vanishing reactive coupling (d2 = 0),
this initial excitation remains localized for any value of d1 . Front propagation
is possible only for sufficiently large values of the reactive coupling d2 >

d2 (d2 ≈ 0.9 for p = 0.5; d = 0.3; c = 0.51; Δ = 2.0; α = 5.75). In certain
domains of d1 (in particular for d1 ≈ 0.3, which is used here) the transition
from nonpropagation to propagation occurs via intermittency.
An example of such a situation is given in Fig. 11.7. Again, we observe
intermittency in the dynamics of amplitudes of oscillators, when long laminar
phases of constant amplitudes are interrupted by chaotic bursts. The number
of bursts during the time interval Δt = 50, 000 vs. the value of reactive cou-
pling d2 is plotted in Fig. 11.8 for active coupling d1 = 0.302. This exhibits two
regions of intermittent behavior. The transition to intermittency in the left
and right regions occurs after the loss of stability of a periodically oscillating
structure with periodically modulated amplitudes of a and b (quasistation-
ary mode) and of a stationary structure with constant amplitudes of a and
25000
t=0
j=1 100
Fig. 11.7. Bursting structure in (11.1) and (11.2). Dependence of the oscillation
intensity |aj (t)|2 on the element’s position in the array j and time t (shading corre-
sponds to the intensity)
b (stationary mode), respectively. Time series of the intensity of oscillations

of one of the elements for d2 from both intermittent regions are presented in
Fig. 11.9.
To identify the bifurcations which lead to this chaotic intermittent behav-
ior, we employ two commonly used tools [79]. First, we have constructed a
one-dimensional return map by taking the local maxima from the observed
time series |a95 (t)|2 . Figure 11.10 shows the return maps for different values
of the coupling parameter d2 for the transition from the quasistationary mode
70
60
number of bursts
50
40
30
20
10
0
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95
d2
Fig. 11.8. Dependence of the number of bursts on the reactive coupling d2 at

d1 = 0.302 in (11.1) and (11.2). For d2 > 0.93, there are no intermittent bursts and
a front propagation is observed
0.4 (a)
|a(90)|2
0.3
0.2
0.1
0
0 5000 10000 15000 20000
time
0.4
|a(90)|2
0.3 (b)
0.2
0.1
0
0 5000 10000 15000 20000
time
Fig. 11.9. Examples of time series showing intermittency in (11.1) and (11.2).
Intervals of laminar behavior are randomly interrupted by short bursts: (a) d2 = 0.56
and (b) d2 = 0.72
to intermittency. Return maps have a quadratic-like form, which corresponds

to the simplest map exhibiting type-I intermittency (according to the classifi-
cation by Pomeau and Manneville [72]) and which can be written in the form
Aj+1 = ε + Aj + A2j , (11.5)
where ε is the distance from the bifurcation point that depends on d2 .

We have also estimated the statistical distribution of the laminar intervals
by a histogram (Fig. 11.11). It shows that this distribution has two peaks. This
property of the distribution is the second important criterion for the existence
of type-I intermittency. The peak in the region of long–laminar intervals is less
sharp than the one in the region of short–laminar intervals. Such a distribution
0.190
0.185
d2=0.55
Ai+1 d2=0.547
d2=0.546
0.180
0.175
0.175 0.180 0.185 0.190
Ai
Fig. 11.10. The first return map constructed from the maxima of the time series
of the intensity of oscillations of the 95th element in (11.1) for three different values
of d2 : d2 = 0.546 – before the bifurcation (stable attracting point), d2 = 0.547
and d2 = 0.55 – after the bifurcation from the quasistationary mode (quadratic-like
map); d1 = 0.3
250
200
150
N
100
50
0
200.0 1200.0 2200.0
L
Fig. 11.11. Histogram for the distribution of laminar intervals length for the 95th
element in (11.1) after the bifurcation from the quasistationary mode. Parameters:
d1 = 0.3, d2 = 0.55
of laminar interval lengths is usually observed for type-I intermittency in

the presence of noise [375]. In our 4N -dimensional system, the role of such
noise can be presumably played by some nonresonant modes, which become
slightly excited above the bifurcation threshold d∗2 , but do not grow during
intermittency.
The behavior is quite different in the region where intermittency is born

from a stationary mode. The corresponding return maps obtained from local
maxima of the time series of intensity and sample histogram are presented
in Figs. 11.12 and 11.13, respectively. One can see that the return map is not
0.210
0.190
Ai+1
d2=0.75
d2=0.72
0.170
0.150
0.150 0.170 0.190 0.210
Ai
Fig. 11.12. The first return map constructed from the maxima of the time series
of the intensity of oscillations of the 95th element in (11.1) for two different values
of d2 after bifurcation from the stationary mode, d1 = 0.302
125
100
75
N
50
25
0
0.0 2000.0 4000.0 6000.0
L
Fig. 11.13. Example of a histogram for the distribution of laminar interval length
for the 95th element in (11.1) after the bifurcation from the stationary mode. Para-
meters: d1 = 0.302, d2 = 0.75
strictly one dimensional. The examination of the bifurcation suggests that the
bifurcation may correspond to type-I intermittency (i.e., the Floquet multi-
plier crosses the unit circle at +1), but the return map has a significant high-
order (for example, cubic) term, i.e., this term is essential near the bifurcation
point (a similar map with an essential cubic term was considered in [376]).
11.4.3 Noise Influence
Finally, we consider the action of external noise on the intermittent front

dynamics treated above. We introduce an additive noise into our system (11.1)
and (11.2) in the following manner:
Chain A:
a˙j = (p + iΔ)aj − (1 + iα)|aj |2 aj

+(d1 + id2 )(aj+1 − 2aj + aj−1 ) + c(bj − aj ) + δ(ξ1 + iξ2 ), (11.6)
Chain B :
b˙j = pbj − |bj |2 bj

(11.7)
+d1 (bj+1 − 2bj + bj−1 ) + c(aj − bj ) + δ(ξ3 + iξ4 ),
where ξk , k = 1, 2, 3, 4 are independent random numbers uniformly distributed
in [−0.5; 0.5], and δ is the noise intensity. The results of numerical simulations
are presented in Fig. 11.14 (cf. Fig. 11.8). Quite naturally small-intensity noise
slightly alters the frequencies of bursting. But when the strength of the noise
δ is increased, an unexpected noise-induced phenomenon is observed. First,
the frequency of the chaotic bursting increases for some values of d2 , and for
140
120
number of bursts
100
80
60
40
20
0
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95
d2
Fig. 11.14. The number of bursts per 5 × 104 times units in the system with noise
in (11.6) and (11.7) vs. d2 , d1 = 0.302. The amplitude of noise δ = 0 (no noise, +),
0.0001 (×), 0.0005 (∗), 0.001 (), 0.002 (), and 0.003 (◦). The curves break at the
values of δ, where intermittency is replaced by pure propagation
the noise intensity δ = 0.002 and 0.003, a chaotic bursting arises for such
values of d2 , where bursting is observed. The most important observation is,
however, that the number of chaotic bursts decreases and even vanishes for
certain domains of d2 . Thus, noise can suppress chaotic intermittent bursting
and provides a way to control the intermittent front dynamics. For a possible
explanation of this nontrivial effect, the multistability of different regimes
should be taken into account. Under the noise influence, the bursting regimes
become unstable and only the trivial state (aj = bj = 0) remains attracting.
11.5 Conclusions
In this chapter, effects of collective oscillations in a system of two coupled
chains of identical limit-cycle oscillators, namely coupled discrete Ginzburg–
Landau equations, are investigated. The main findings are the following:
– A pair of coupled chains with different collective frequencies exhibit stable
fronts between two possible asymptotic states: synchronous oscillations
and oscillation death
– The inhomogeneous states formed by the fronts persist at weak coupling
between the oscillators in each chain due to discrete space variable; thus
providing conditions for the existence of localized synchronization struc-
tures
– At stronger coupling, the interface between the two regions of the chains
may propagate
– Synchronized clusters can be induced by disorder
– Transitions from nonpropagating to propagating fronts between synchro-
nized and nonexcited oscillators occur via intermittency
– Noise plays a constructive role in regularization of collective behavior of
oscillators
On the Basis of these finding we would like to underline the following: The
mechanisms of the formation of localized structures considered above have
a strong impact on a better understanding of the origin of low-dimensional
chaos in multidimensional and extended systems. In particular, weakly cou-
pled clusters with m < N cells are frequently formed in large ensembles of
N oscillators, including the ones having different parameters. As to expect
from the above analysis, only clusters with synchronized (at least partially)
cells may survive in a definite region of the parameters. As a result, the
effective number of degrees of freedom sufficient for the description of the
dynamics of the system decreases substantially. Our analysis confirms that,
at certain conditions, this mechanism is also effective at chaotic synchroniza-
tion (see Chaps. 7–9) and leads to the formation of localized structures with
low-dimensional chaotic dynamics. On the other hand, the formation of vari-
ous complex patterns observed at relatively simple but inhomogeneous initial
conditions indicates that the studied phenomenon may also be responsible
for the formation of spatial disorder in the propagation of fronts in chains of

oscillators.
We would like to emphasize that the simplicity and generality of the mech-
anism of oscillator death allows us to extend the above conclusions to the case
of coupled multidimensional arrays in the presence of linear and nonlinear fre-
quency mismatch. Besides, interpreting two coupled chains as a limiting case
of a strongly inhomogeneous two-dimensional array and proceeding from the
above analysis, it is to be expected that inhomogeneity (for example, fre-
quency mismatch) in two-dimensional systems along one of the coordinates
will lead to the appearance of states localized along both the coordinates.
We conclude this chapter with a few comments on the role of spatial and
temporal irregularities in networks of dynamical elements. A traditional view
is that chaos and/or disorder acts in a destructive way. Many examples are
known presently which provide evidence of the opposite. We have already
mentioned that disorder introduced into networks of oscillators can enhance
synchronization and can make chaotic dynamics more regular [370–374]. Noise
introduced into nonlinear dynamical systems can lead to nontrivial effects
too. A well-known example is stochastic resonance [377–380]. Spatially uncor-
related noise can enhance stochastic resonance effects in an array of cou-
pled bistable elements [381, 382], facilitate signal propagation in arrays of
bistable systems [383, 384], sustain traveling waves in subexcitable chemical
media [385], sustain patterns (including the spiral ones) [386–389], induce
pattern transitions [390] and fronts [391], etc. Here we have demonstrated
how noise can suppress chaotic bursting in systems with oscillatory death.
This effect is a new example for a nontrivial action of noise in arrays of cou-
pled oscillators. Other aspects of noise influence on the synchronization and
synchronization-like processes in oscillatory and excitable networks are dis-
cussed in Chap. 13.
12
Chains and Lattices of Excitable
Luo–Rudy Systems
Previous chapters are devoted to synchronization phenomena in ensembles of

self-oscillatory systems. In this chapter we apply the synchronization theory
presented before to describe in detail possible responses of one- and two-
dimensional excitable media on an external periodic force [392]. This study is
especially motivated by the important finding in cardiology that ventricular
fibrillation might arise from spiral wave chaos [393]. Ventricular fibrillation is
potentially lethal arrhythmia that can result in sudden cardiac death, which
is, for example, responsible for 250,000 deaths in the US each year [394].
Our objective is therefore to investigate wave interactions in excitable
media and to explore the feasibility of using overdrive pacing to suppress such
spiral wave chaos. This work is based on the fact that in excitable media,
propagating waves with the highest excitation frequency eventually overtake
all other waves. We analyze effects of low-amplitude, high-frequency pacing
in one-dimensional and two-dimensional networks of coupled, excitable cells
governed by the Luo–Rudy systems [94] used as a model of cardiac tissue.
First it is shown that in the one-dimensional cardiac model there are narrow
high-frequency regions of 1:1 synchronous response (synchrony) between the
input stimulus and the system’s response. The frequencies in this region are
higher than the intrinsic spiral wave frequency of cardiac tissue. When we pace
the two-dimensional cardiac model with frequencies from this region, we find
that spiral wave chaos can, in some cases, be suppressed. When we, however,
couple the overdrive pacing by reducing some parameters that are responsi-
ble for the calcium channel conductance, we find that spiral wave chaos can
be suppressed in all cases. These results suggest that low-amplitude, high-
frequency overdrive pacing, in combination with calcium channel inhibitors
(e.g., class II or class IV antiarrhythmic drugs), may be useful for eliminating
such life-threatening fibrillation. This technique allows us to reduce the mag-
nitude of electrical force on the cardiac tissue several times in comparison to
other techniques, which are used in implanted cardioverter defibrillators [395].
The chapter is organized as follows. In Sect. 12.1, we discuss the objec-
tives of our study. Then (Sect. 12.2) we describe the dynamical system used
252 12 Chains and Lattices of Excitable Luo–Rudy Systems
as a model of cardiac tissue. The theoretical basis of our method is dis-

cussed in Sect. 12.3. In Sect. 12.4, we present computational findings for one-
dimensional ensembles, which indicate that it may be possible to defibrillate
cardiac tissue using low-amplitude, high-frequency pacing. We systematically
explore the parameter space governing the pacing stimulus and present a map
of the most promising stimulus waveforms. In addition, we show that the pac-
ing technique can be optimized if it is used in conjunction with antiarrhythmic
drugs, specifically those which block calcium channels. In Sect. 12.5, we use
the results found in the one-dimensional study to explore the elimination of
spiral waves in two dimension. Finally, we discuss how the presented novel
defibrillation approach could be realized in a clinical setting by implanting
both a drug pump and a pacemaker, and programming them so that they are
simultaneously activated at the onset of fibrillation.
12.1 Objectives
Forced by the sinus rhythm pacemaker, waves of electrical activity propagate

throughout the heart, eliciting a simultaneous contraction of the ventricles.
However, in diseased heart tissue, tachycardia can develop when excitatory
spiral waves locally reexcite tissue prior to the next stimulus from the sinoa-
trial node. Such life-threatening ventricular fibrillation might arise if these
spiral waves break up into spiral wave chaos [396–399].
To overcome this serious regime, defibrillation techniques have been devel-
oped. Current ventricular defibrillation techniques rely on the application of
a very strong voltage shock (5 kV) to the heart. Then in most cases this shock
halts all electrical activity within the heart and prevents a local reexcitation
of the tissue. Once the heart cells repolarize in synchrony, electrical waves
from the sinoatrial node take over and a sinus rhythm resumes. However, the
energy necessary for a successful defibrillation using this technique is quite
painful and often strong enough to damage the tissue [400].
Besides other spiral wave suppression techniques [401–408], a number of
experiments have explored the use of low-amplitude, high-frequency pacing
(i.e., overdrive pacing) as an alternative defibrillation technique [409–414]. In
each of these experiments, the pacing had only local effects, resulting in only
small areas of organized electrical activity. Once the pacing was suspended, the
local region of capture was reinvaded by the surrounding electrical activity,
and the tissue remained in a state of fibrillation. Thus, in all cases, over-
drive pacing was only marginally effective to eliminate the arrhythmias. It is
important to note that the stimulation parameters, including stimulus wave-
form and input frequency, for each of these studies were similar. Specifically,
each of these experiments utilized stimuli consisting of either monophasic
(amplitude fluctuation from a baseline in one direction) square-wave pulses of
2 ms duration or symmetric biphasic (amplitude fluctuation from baseline in
12.2 Cardiac Model 253
two directions) square-wave pulses of 2 ms duration, and a pacing frequency

slightly higher or lower than the average excitation frequency of the fibrillating
tissue.
Extensive research has been recently done on a periodic pacing of spiral
waves in excitable media [415–418]. These studies have shown that a periodic
train of pulses can eliminate spiral waves in an excitable medium when the
frequency of the pacing wave is greater than that of the spiral wave. In effect,
the high-frequency wave pushes the low-frequency wave beyond the edges of
the medium, thereby suppressing its wavefronts. Sparse spiral waves can also
be annihilated using a lower frequency pacing wave when the pacing source is
close to the core of the spiral wave [419]. It is important to note that each of
these studies primarily explore such effects where a frequency of the periodic
pacing exists on the spiral waves.
In this chapter, we propose another approach and explore the feasibility
of using overdrive pacing – external driving with relatively high frequency
– to eliminate spiral waves and spiral wave chaos in cardiac tissue. Our in
numero experiments are based on theoretical principles underlying the physics
of interacting waves in excitable media. We consider a stimulation parame-
ter space that is considerably larger than that utilized in the aforementioned
experimental and theoretical studies. In addition, we examine the possibility
of coupling overdrive pacing with calcium channel blockers. Although cal-
cium channel antagonists are often considered to be proarrhythmic [420, 421],
they have also been shown to transform spiral wave chaos into quasiperiodic,
meandering spiral wave activity [421]. We explore, in a series of computational
experiments, the possibility of exploiting this effect to enhance the effective-
ness of overdrive pacing to eliminate arrhythmias.
12.2 Cardiac Model
We conduct all our computational experiments with a monophasic description

of ventricular myocardium. The model is given by the expression (for details
of the single cell model and its dynamics see Sect. 2.7.3)
2
∂t = −Iion − Istimulus + Cm D∇ V,
Cm ∂V
(12.1)
dgi g∞ −gi
dt = τgi ,
where V is the membrane voltage and V = V (x, t) for a one-dimensional

(1D) medium and V = V (x, y, t) for a two-dimensional (2D) medium, Cm =
1 μF cm−2 is the membrane capacitance, D = 0.001 cm2 ms−2 is the diffusion
2 2 2
coefficient, ∇2 = ∂∂2 x for 1D and ∇2 = ∂∂2 x + ∂∂2 y for 2D, Istimulus is the input
stimulus, and Iion is the sum of six ionic currents:
Iion = Ina + Isi + Ik + Ik1 + Ikp + Ib . (12.2)

The forms of all these currents are given in Sect. 2.7.3. gi (V, t) is the product
of one or more gating variables, g∞ = αgi /[αgi + βgi ] is the steady-state value,
τgi = 1/[αgi + βgi ] is the time constant, and the αs and βs are the functions
of the membrane voltage. The conductances and reversal potentials used in
the simulations are listed in Sect. 2.7.3.
12.3 Methods: Theoretical Basis

The behavior of interacting waves in general homogeneous excitable media is
governed by the following four fundamental properties:
1. The wave with the highest frequency will eventually overtake all the other
waves [415,418,422]. This is due to the fact that slower waves are progressively
invaded by faster ones.
2. A given medium supports interacting spiral waves of a single fre-
quency [415, 423]. When spiral waves of different frequencies interact in a
single medium, the wave with the highest frequency will dominate according
to (1). Thus, only waves of a single frequency can eventually coexist in a
given medium. This characteristic arises from general properties of the sys-
tem’s action potential, in particular, its refractory period. The spiral wave
frequency varies from medium to medium even though a unique frequency
exists for every medium. In human cardiac tissue, the intrinsic spiral wave
frequency is approximately 6.3 Hz [424].
3. Planar wavefronts travel faster than convex wavefronts [425]. In traveling
waves with a significant positive curvature, every successive time step requires
more cells to be excited. The additional transverse current load needed to
support this expanding wavefront causes the curved wave to travel slower
than a planar wave.
4. The time to suppress colliding periodic waves depends inversely on two
factors: (a) the frequency difference between the waves and (b) the velocity of
the wave with the highest frequency [422, 426]. Thus, a slower wave is more
quickly invaded by a faster one as the frequency and/or velocity of the faster
wave increases.
We use properties (1)–(4) to guide our in numero experiments. Especially,
on the basis of the first and second properties, we explore pacing frequencies
greater than the intrinsic spiral wave frequency of cardiac tissue. In addition,
on the basis of the third property, we generate planar wavefronts by incor-
porating into the model the equivalent of a strip electrode. Finally, on the
basis of the fourth property, we utilize the highest frequencies which could be
produced in the medium.
To find the highest frequencies supported by the medium, we characterize
the frequency response of the system by examining the synchrony between
the applied stimulation and the resulting action potentials of the medium.
Note that high-frequency stimuli in excitable media typically do not result in
a 1:1 synchrony between the input (pulse stimulus) and the response (action
12.4 Computational Results 255
potential). This effect is due to the system’s refractory period. Instead at

times, the cells in the medium may respond to every other pulse or every third
pulse depending upon the refractory state resulting in 2:1 or 3:1 synchrony;
more complicated dynamics is also possible [427, 428].
12.4 Computational Results

We perform computer simulations first in one dimension using a chain of 30
cells, and then in two dimensions using a network of 300×300 cells. The one-
dimensional simulations were conducted to investigate the frequency response
of the cardiac model to various stimuli. We then use the high-frequency regions
of a 1:1 synchrony found in the one-dimensional studies as the basis for explor-
ing the elimination of spiral waves and spiral wave chaos in large ensembles
in two dimensions.
The one-dimensional simulations are performed by stimulating the first
cell in the 30-cell chain for 5.0 s and by averaging the action potential fre-
quency of the last cell in the chain over the final 1.0 s of the computational
trial. In the one-dimensional case, the output frequency (i.e., the action poten-
tial frequency of the last cell) is well defined because the excitatory stimulus
propagates in only one direction through the chain. However, in the two-
dimensional case, the output frequency is not clearly defined because exci-
tatory stimuli can flow into a cell from four directions, often causing action
potentials to be produced before the cell returns to its resting potential. In
our simulations, we neglect changes in the action potential morphology over
time, and we group all K:K synchrony regions, where K is a positive integer,
into the 1:1 synchrony region. We similarly group all 2K:K synchrony regions
into the 2:1 synchrony region.
We use two integration methods for the simulations: the forward Euler
(FE) method and the operator splitting (OS) method [429, 430]. The FE
approach was implemented with a fixed time step of 0.01 ms, and the OS
method was implemented with an adaptive time step ranging from 0.05 to
0.005 ms. We simulate isotropic media with a spatial step of 0.028 cm, and we
use the following no-flux boundary conditions for all simulations:

∂V ∂V ∂V ∂V
= = = =0
∂x ∂x ∂y ∂y
x=0 x=L y=0 y=L
where L is the tissue length.
12.4.1 One-Dimensional Simulations
In our first series of computational experiments, we periodically pace the one-

dimensional chain of initially non-excited 30 cells with a square-wave stimulus
with the amplitude = 100 μA cm−2 . As our experiments shown, this length is
15
1:1
10
Output Freq. (Hz)
2:1
0
0 5 10 15 20
Input Freq. (Hz)
Fig. 12.1. Output frequency vs. input frequency for a driven one-dimensional chain
of 30 cells. The stimulus consists of a square wave with an amplitude of 100 μA cm−2
and a duty cycle of 50%. The input–output frequency characterization for non-
excited cells, initial conditions are marked by plus and for adapted initial conditions
by circle. The dashed horizontal line at 6.3 Hz represents the spiral wave frequency
of cardiac tissue
enough to observe the main phenomena in 1D. The input–output frequency

characterization for this system is shown in Fig. 12.1. The spiral wave fre-
quency for the cardiac tissue is shown as a horizontal dashed line at 6.3 Hz.
Thus, an output frequency greater than this line could potentially result in the
suppression of spiral waves. The initial straight line represents 1:1 synchrony,
or multiples thereof, in which the system responds to every input stimulus.
Beginning at an input frequency of 7.3 Hz, the system responds to every other
pulse resulting in 2:1 synchrony. The system returns to 1:1 synchrony, inter-
mittent within the 2:1 synchrony region. This narrow 1:1 intermittency region
centered around an input frequency of 12.7 Hz (Fig. 12.1) appears to be a
promising region for suppression. In the same figure we plot results of another
series of computations with other initial conditions. We adapt the initial con-
ditions to small parameter variations: By changing the input frequency, the
solutions from the previous simulations were taken as initial conditions for the
new simulations. This procedure enables us to reveal the region of bistability
of synchronous responses (Fig. 12.1). Corresponding input–output lines are
marked by “o.” Therefore, the response of excitable media to external stim-
ulation strongly depends on the conditions where this stimulation has been
110
90
Amplitude 70
50
30
10
−10
0 T 2T 3T 4T 5T
time
Fig. 12.2. External force as a sequence of rectangle pulses with a 50% (T /2) duty
cycle
applied. It is important to note that bi-stability in the model of externally

forced single cell was observed in Sect. 3.6.
An important point in this system is to study the effects of pulse duration
on the system’s input–output frequency response by varying the percentage
of cycle time (the duty cycle) where the square-wave pulse was greater than
zero. The sequence of pulses with duty cycle duration of 50% is presented in
Fig. 12.2. Results for a range of duty cycles are shown in Fig. 12.3. It can
be shown that for the duty cycles included in Fig. 12.3, the 1:1 intermit-
tency region only appears for the 50% value. Moreover, we found that the 2:1
synchrony region breaks up as the duty cycle decreases.
We next examine the effects of the stimulus amplitude on the 1:1 and 2:1
synchrony regions. The results for various amplitudes, as a function of the
input frequency, are shown in Fig. 12.4. The 1:1 intermittency region shown
in Fig. 12.1 curves toward lower frequencies and breaks up for smaller stimulus
amplitudes. It is important to point out that the 2:1 synchrony regions with
output frequencies greater than 6.3 Hz and thus input frequencies greater than
12.6 Hz could potentially be useful for suppressing spiral waves as well. Thus,
portions of the 2:1 synchrony region to the right of the 1:1 intermittency region
in Fig. 12.4 may be effective for eliminating spiral waves in two-dimensional
media.
Given the promising results for a duty cycle of 50% (Fig. 12.4), we further
investigate the effects of the stimulus amplitude on the 1:1 and 2:1 synchrony
regions for duty cycles ranging from 40 to 60%. Figure 12.5 indicates that
as the duty cycle is decreased from 50 to 40%, the 1:1 intermittency region
breaks up, while the 2:1 synchrony region remains stable. Figure 12.5 also
shows that, as the duty cycle is increased beyond 50%, the 1:1 intermittency
(a) fixed 2 msec duty cycle (b) 12.5% duty cycle

15 15
Output Freq. (Hz)
10 10
5 5
0 0
0 10 20 30 0 10 20 30
(c) 25.0% duty cycle (d) 37.5% duty cycle

15 15
Output Freq. (Hz)
10 10
5 5
0 0
0 10 20 30 0 10 20 30
(e) 50.0% duty cycle (f) 62.5% duty cycle

15 15
Output Freq. (Hz)
10 10
5 5
0 0
0 10 20 30 0 10 20 30
Input Freq. (Hz) Input Freq. (Hz)
of 30 cells. The stimulus consists of a square wave (see Fig. 12.2) with an amplitude
of 100 μA cm−2 and a duty cycle of: (a) a fixed, 2 ms duration; (b) 12.5%; (c) 25.0%;
(d) 37.5%; (e) 50.0%; and (f ) 62.5%. As in Fig. 12.1, the dashed horizontal lines
represent the intrinsic spiral wave frequency of 6.3 Hz. Note that the part of Fig. 12.1
has been included as (e) for reference. Initial conditions: non-excited cells
region is invaded by surrounding 2:1 synchrony regions. This clarifies why the
1:1 intermittency region quickly disappears if the duty cycle departs from a
value of 50%.
We also examine effects of stimulus waveform on the 1:1 and 2:1 synchrony
regions. Figure 12.6 presents the results for six different waveforms typically
used for cardiac pacing [395]:
1. Monophasic square-wave pulses
2. Symmetric biphasic square-wave pulses of 2 ms duration
3. Sinusoidal force
4. Pulses with ramp wavefronts and vertical waveback
5. Pulses with vertical wavefronts and ramp waveback
6. Pulses with ramp wavefronts and waveback
1:1 synchrony
2:1 synchrony
100
Stimulus Ampl. (μAmp/cm2)
80
60
40
20
6 8 10 12 14 16 18 20
Input Freq. (Hz)
Fig. 12.4. Regions of 1:1 plus and 2:1 circle synchrony, as a function of stimulus
amplitude and input frequency, for a driven one-dimensional chain of 30 cells. The
stimulus consists of a square wave with a duty cycle of 50%. Initial conditions: non-
excited cells. Middle region marked by plus is 1:1 intermittency synchrony region
It comes that monophasic square-wave pulses of 2 ms duration (Fig. 12.6a)

rarely result in 1:1 synchrony (only four data points over the explored para-
meter space resulted in 1:1 synchrony). Similar results are obtained if the
medium is paced with symmetric biphasic square-wave pulses of 2 ms dura-
tion (Fig. 12.6b). We also find that sinusoidal inputs (Fig. 12.6c) do not lead
to 1:1 synchrony at high frequencies, whereas stimuli with ramp wavefronts
and vertical wavebacks (Fig. 12.6d) generate large regions of 1:1 synchrony,
including a 1:1 intermittency region. For stimuli with vertical wavefronts and
ramp wavebacks (Fig. 12.6e), the branch of the initial 1:1 synchrony region
disappears and the 1:1 intermittency region becomes scattered. The 1:1 inter-
mittency region disappears entirely when a triangle waveform with a ramp
wavefront and waveback (Fig. 12.6f) is used to excite the system.
It is interesting to note that the two most effective waveforms, the square
waveform and a waveform with a ramp wavefront and a vertical waveback, pos-
sess a vertical waveback. It is possible that a vertical waveback enables cells to
recover more quickly following an excitation, making them more responsive to
high-frequency inputs. In contrast, a slowly decaying waveback, which injects
current into a cell during its recovery, may prolong the refractory period by
1:1 synchrony
2:1 synchrony
40.0% duty cycle 45.0% duty cycle

100 (a) 100 (b)
60 60
20 20
10 15 20 10 15 20
50.0% duty cycle

100 (c)
60
20
10 15 20
55.0% duty cycle 60.0% duty cycle

100 (d) 100 (e)
60 60
20 20
10 15 20 10 15 20
Fig. 12.5. 1:1 plus and 2:1 circle synchrony regions, as a function of stimulus
stimulus consists of a square wave with a duty cycle of: (a) 40%, (b) 45%, (c) 50%,
(d) 55%, and (e) 60%. Note that Fig. 12.4 has been included as c for reference.
Initial conditions: non-excited cells
keeping the membrane voltage elevated. This would make a cell less responsive
to high-frequency inputs.
We also investigate the effects of the calcium channel antagonists on our
one-dimensional cardiac model. Therefore, we eliminate transmembrane cal-
cium fluxes by setting Gsi = 0 throughout each in numero experiment. The
input–output frequency response for various duty cycles are shown in Fig. 12.7.
The main results are:
1. The elimination of slow inward calcium currents serves to increase the
intrinsic spiral wave frequency from 6.3 to 25.0 Hz.
2. The calcium channel antagonists eliminate the intermittency regions found
in the original model (Fig. 12.7). Hence, we can conclude that it may
only be possible to eliminate spiral waves using a combined action of
calcium channel blockers and overdrive pacing with a narrow band of
input frequencies around 27 Hz at duty cycles less than or equal to 37.5%.
Our simulation shows the absence of bistability.
1:1 synchrony
2:1 synchrony
100 (a) 100 (b)

60 60
20 20
10 15 20 10 15 20
Stimulus Ampl. (μAmp / cm2)
100 (c) 100 (d)

60 60
20 20
10 15 20 10 15 20
100 (e) 100 (f)

60 60
20 20
10 15 20 10 15 20
Fig. 12.6. 1:1 plus and 2:1 circle synchrony regions, as a function of stimulus
stimulus consists of the following waveforms (shown to the right of each plot): (a) a
monophasic square-wave pulse of 2 ms duration, (b) symmetric biphasic square-wave
pulses of 2 ms duration, (c) a sine wave, (d) a waveform with a ramp wavefront and
a vertical waveback, (e) a waveform with a vertical wavefront and a ramp waveback,
and (f ) a triangle wave. Initial conditions: non-excited cells
12.4.2 Two-Dimensional Simulations
We use the results from the one-dimensional simulations to explore the possi-
ble suppression of spiral waves and spiral wave chaos in the practically much
more interesting case of two dimensions. The analysis of the one-dimensional
chain in Fig. 12.4 suggests especially that overdrive pacing of the system at a
frequency of 12.4 Hz with a square-wave stimulus of amplitude 50 μA cm−2
and duty cycle 50.0% should be sufficient for suppression of spiral waves
in a two-dimensional medium. First, we present sequential snapshots of the
two-dimensional cardiac model subject to such stimulation (Fig. 12.8). The
excitation was applied, starting at t = 0.00 s, with a model equivalent to a
strip electrode at the top of the medium to ensure a fast conduction velocity.
Figure 12.8 also shows that the initial spiral wave chaos (t = 0.00 s) is entirely
eliminated after 5.00 s of pacing. By continuing the pacing beyond this point,
we see that more-or-less planar wavefronts travel through the medium for at
(a) fixed 2 msec duty cycle (b) 10.0% duty cycle

30 30
Output Freq. (Hz)
20 20
10 10
20 30 40 20 30 40
(c) 12.5% duty cycle (d) 25.0% duty cycle

Output Freq. (Hz)
30 30
20 20
10 10
20 30 40 20 30 40
(e) 37.5% duty cycle (f) 50.0% duty cycle

30 30
Output Freq. (Hz)
20 20
10 10
20 30 40 20 30 40
of 30 cells. The calcium current is eliminated in the model by setting Gsi = 0
throughout each in numero experiment. The stimulus consists of a square wave with
an amplitude of 100 μA cm−2 and a duty cycle of: (a) a fixed, 2 ms duration; (b)
10.0%; (c) 12.5%; (d) 25.0%; (e) 37.5%; and (f ) 50.0%. The dashed horizontal lines
represent the intrinsic spiral wave frequency of 25.0 Hz
least 1 s (from t = 5.00 s to t = 6.00 s). Shortly after, however, spiral wave
chaos is reinitiated, as can be seen at t = 6.25 s. Thus, there exists a window,
in this case of approximately 1.0 s, in which the medium is free of any spiral
wave chaos. If the pacing is halted during this window, the medium will be
left in a resting state. This finding has important consequences for cardiology
and shows especially that controlled intervals of overdrive pacing can be used
to suppress spiral wave chaos.
In general, however, we get limited success for suppressing spiral wave
chaos in two-dimensional media using overdrive pacing alone. The high-
frequency stimuli predicted from the one-dimensional studies did not always
result in planar wavebacks during capture. In many cases, subsequent wave-
fronts approach the wavebacks of prior stimuli, encountering cells in various
refractory states. This leads, in some instances, to the reinitiation of spiral
wave chaos after local capture. The main reason for this effect is the bistability
of the synchronous response of the media to the external stimulation.
t = 0.00 sec t = 2.00 sec t = 4.00 sec

40 mV
20
0
− 20
− 40
− 60
− 80
t = 5.00 sec t = 5.25 sec t = 5.50 sec
t = 5.75 sec t = 6.00 sec t = 6.25 sec
Fig. 12.8. Sequential time images of a successful suppression of spiral wave chaos,
using overdrive pacing, in a network of 300×300 cells. The system is excited with the
model equivalent of a strip electrode of 2×300 cells located at the top border of the
medium. The stimulus consists of a square wave with an amplitude of 50 μA cm−2
and a duty cycle of 50%. The input frequency is 12.4 Hz. The membrane voltage
(mV) of the cells in each image is color coded as indicated in the bar located to the
right of the top row of panels
Therefore, we next analyze effects due to coupling of overdrive pacing

with calcium channel blockers. The inward calcium current in our cardiac
model serves to keep the membrane voltage elevated after the initial, sodium-
dependent depolarization. By inhibiting this current (Gsi = 0), the absolute
refractory period of each cell is effectively reduced. When a stimulus is then
applied to the model, the waveform encounters fewer non-excitable cells and
thus it has a higher chance of entraining the spiral waves. As noted earlier,
the reduced refractory period also leads to an increase in the intrinsic spiral
wave frequency of the medium. We, therefore, have to increase the frequency
of our overdrive pacing to keep the input frequency higher than that of the
spiral wave activity within the medium.
We inhibit the calcium flux in the model by setting the maximum cal-
cium conductance, Gsi , to zero after the first time step in the simulations.
Figure 12.9 presents results of a computational experiment where the top two
rows of cells were paced at a frequency of 27.9 Hz with a square-wave stim-
ulus of amplitude 50 μA cm−2 and duty cycle 12.5%. Inhibiting the calcium
flux causes the initial spiral wave chaos to conform to a more regular, quasi-
periodic meander, which is consistent with previously reported results [431].
This dynamic effect allows the applied stimulation to entrain the medium
progressively. After 4.0 s, the applied stimulation has already annihilated all
spiral wave activity within the medium (Fig. 12.9). When the stimulation is
subsequently halted, the medium returns to its resting state within 270 ms.
As was mentioned for the one-dimensional case, the synchronous response of
t = 0.0 sec t = 0.5 sec t = 1.0 sec

40 mV
20
0
− 20
− 40
− 60
− 80
t = 1.5 sec t = 2.0 sec t = 2.5 sec
t = 3.0 sec t = 3.5 sec t = 4.0 sec
Fig. 12.9. Sequential time images of a successful suppression of spiral wave chaos,
using overdrive pacing in conjunction with calcium channel antagonists, in a network
of 300×300 cells. The system is excited with the model equivalent of a strip electrode
of 2×300 cells located at the top border of the medium. The stimulus consists of a
square wave with an amplitude of 50 μA cm−2 and a duty cycle of 12.5%. The input
frequency is 27.9 Hz. The calcium channels in the model are blocked at t = 0.0 s
by setting Gsi = 0. The membrane voltage (mV) of the cells in each image is color
coded as indicated in the bar located to the right of the top row of panels
the medium to the external stimulation is unique, i.e., it does not depend on
the current state of the cells in the media.
Next, we investigate effects of the electrode geometry on the suppression
of spiral waves by using a point source to generate convex waves. As it may be
impossible to generate a completely planar wavefront outside of the computa-
tional realm, it is important to investigate various electrode geometries. The
experimental conditions are similar to those in Fig. 12.9. We find that spiral
wave chaos could be suppressed; however, the time to get suppression is now
longer. We further explore the sensitivity of the spiral wave suppression to the
magnitude of the calcium current by repeating the above computational exper-
iment with modified values of the maximum calcium conductance, Gsi . Then,
spiral wave chaos could be suppressed similarly to that shown in Fig. 12.9
with reduced, nonzero levels of inward calcium current (Gsi greater than 0
but less than the nominal 0.07). However, the time to get suppression of spi-
ral wave chaos increases as the level of inward calcium current is increased.
In general, we yield more episodes of a successful suppression of spiral wave
chaos using the combined action of overdrive pacing and calcium channel
antagonists, than in using overdrive pacing alone. In fact, we find that spi-
ral wave chaos could be suppressed in all cases when the overdrive pacing is
coupled with calcium channel blockers.
12.5 Conclusions
An important challenge in cardiology is to find rather weak external actions

which induce a suppression of fibrillation in the cardiac tissue. Recent findings
indicate that ventricular fibrillation might arise from spiral wave chaos. In this
chapter, we have therefore first characterized the frequency response of a one-
dimensional cardiac model to provide promising frequencies and waveforms
of external forces for suppressing spiral waves and spiral wave chaos in two
dimensions. We have used these results to guide two-dimensional simulations
and have shown that it is possible to apply overdrive pacing to eliminate
spiral waves and spiral wave chaos in two-dimensional media. We have also
demonstrated that calcium channel blockers could be used to enhance the
effectiveness of overdrive pacing in eliminating arrhythmias.
The main findings of this chapter are the following:
– The highest frequencies supported by the medium occur, in general, in the
1:1 intermittency region that arises for a duty cycle of 50%.
– If the duty cycle is decreased, the intermittency region becomes irregular,
resulting in lower output frequencies. If the duty cycle is increased, the 1:1
intermittency region disappears.
– Squared waveforms (of sufficient duration) and waveforms with ramp wave-
fronts and vertical wavebacks of the external pulses give rise to robust,
high-frequency 1:1 intermittency regions.
– There occur regimes of bistability of a synchronous response of the media

to an external stimulation.
– Some, but not all, of the stimuli suggested by the one-dimensional sim-
ulations could be used to suppress spiral wave chaos in two-dimensional
media.
– The elimination of slow inward calcium currents causes spiral wave chaos to
conform to a more regular, quasiperiodic meandering spiral wave behavior,
a pattern characteristic of ventricular tachycardia.
– If the cardiac model is additionally subjected to overdrive pacing, the
meandering activity could be suppressed.
These theoretical finding can be applied for defibrillation in the follow-

ing way: In a clinical setting, the calcium current can be inhibited by using
class II or class IV antiarrhythmic drugs, which selectively block the inward
calcium flux through the cell membrane. For example, Esmolol, a class II
agent, and Verapamil, a class IV agent [432], could be delivered intravenously
to inhibit the calcium flux selectively through the L-type calcium channel.
Esmolol has a 1–2 min activation time with a half-life of 10 min, and Ver-
apamil, in intravenous form, has an immediate onset of activation with a
half-life of 4 min [433]. These short-lived, fast-acting agents could be used to
inactivate the calcium channels of the heart for short periods of time. Our sim-
ulations suggest that defibrillation could be attained if the drugs are delivered
quickly in conjunction with overdrive pacing. In a clinical setting, this could
be realized by implanting both a drug pump and a pacemaker, and program-
ming them so that they are simultaneously activated at the onset of fibrilla-
tion. Alternatively, long-term low-dose oral calcium channel blockers could be
administered in conjunction with an implanted pacemaker. Finally, we would
like to note that one of the fundamental challenges in using implanted defib-
rillators is to find early signs of ventricular fibrillation. Recently proposed
recurrence-plot-based method [434] allows us to do such detection. Therefore,
the application of the proposed defibrillation scheme in combination with a
fibrillation detection method seems to be very promising for the fibrillation
suppression.
Part III
Synchronization in Complex Networks

and Influence of Noise
13
Noise-Induced Synchronization
in Ensembles of Oscillatory
and Excitable Systems
In the previous chapters, we have demonstrated various synchronization

phenomena in arrays of coupled deterministic oscillators. However, noise is
inevitably present in experimental and natural systems. Therefore, it is of
interest and importance to explore the effects of noise on the robustness of
the synchronization process.
Noise may influence synchronization in different ways. Usually, it has a
degrading effect, for example, inducing phase slips of phase-locked oscillators
or resulting in an intermittent loss of synchronization in complete synchro-
nization (CS) of coupled chaotic systems.
On the other hand, noise may play a constructive role in enhancing syn-
chronization or, more generally, in inducing more order. This has been firstly
demonstrated in the context of stochastic resonance [435]. By stochastic reso-
nance, noise can optimize the response of a nonlinear system to a weak exter-
nal signal. Stochastic resonance has also been studied from the viewpoint of
noise-enhanced phase synchronization (PS) [436] of the switching events to
the external signal, because noise controls the average switching rate of the
system and the response is optimal when it is close to that of the external
signal.
Noise-enhanced synchronization may have meaningful implications in
various fields. For example, the circumstance that different systems are not
coupled or only weakly coupled but subjected to a common random forcing
is of great relevance in life sciences, especially in neuroscience or ecology. In
neural systems, different neurons, commonly connected to another group of
neurons, receive a common input signal which often approaches a Gaussian
distribution as a result of integration of many independent synaptic cur-
rents [437]. We would emphasize the experimental observation of the spiking
behavior of animal cortical neurons in response to external stimulus. When
the stimulus is a constant input current, the neurons generate different spike
sequences in repeated experiments with the same driving signal. Remarkably,
when an amount of colored Gaussian noise is added in addition to the con-
stant input to mimic the synaptic input in actual neural systems, the neurons
270 13 Noise-Induced Synchronization
produce repetitive spike sequences in repeated experiments with the same

fluctuating driving signal. This reliability of spike timing is of significance for
understanding signal encoding by spike timing of neurons. In ecology, food
webs [438, 439] and forest ecosystems [440] over a large geographical region
are affected by similar environmental fluctuations. Actually, observations
have shown synchronous oscillations of populations [438–440], and the com-
mon environmental fluctuations may play an important role in this collective
behavior in ecology. It is important to emphasize that, in these systems, both
the coupling and the common random forcing are often rather weak among
the elements which are generically nonidentical.
This chapter is devoted to effects of noise on collective dynamics of
oscillatory and excitable ensembles. We begin with a brief demonstration of
degrading effects of noise on synchronization of phase oscillators and limit-
cycle oscillators (Sect. 13.1). Then the following typical noise-induced synchro-
nization effects are presented. Noise-induced CS of uncoupled identical chaotic
oscillators (Sect. 13.2.1) and noise-induced PS of uncoupled nonidentical
chaotic oscillators are discussed in Sect. 13.2.2. Then we study noise-enhanced
PS of a chaotic laser due to periodic forcing (Sect. 13.3.1), noise-enhanced
PS in two weakly coupled nonidentical oscillators (Sect. 13.3.2), arrays of
globally coupled nonidentical Rössler oscillators (Sect. 13.3.3), and experi-
ments with electrochemical oscillators (Sect. 13.3.4). In Sect. 13.4, we show
noise-enhanced synchrony among locally coupled excitable cells, and noise-
enhanced synchrony of locally coupled excitable cells to weak external forcing,
and noise-sustained resonant pattern formation in a two-dimensional (2D)
lattice of excitable cells.
13.1 Degrading Effects of Noise: Noise-Induced

Phase Slips
We start with synchronization of two coupled periodic oscillators, where the

synchronization regime is defined as locking of the phases φ1,2 , |φ1 − φ2 | <
const (see Sect. 4.1). The frequencies Ωj = φ̇j are also locked, i.e., Ω1 = Ω2 ,
while the amplitudes can be quite different, so that nonidentical periodic
oscillators can be synchronized with each other by a rather weak coupling.
Hence, synchronization of weakly coupled periodic oscillators can be described
by the dynamics of the phase difference θ = φ2 − φ1 , i.e.,
θ̇ = Δω − ε sin θ, (13.1)
where Δω = ω2 − ω1 is the difference between natural frequencies of the oscil-

lators, and ε is the coupling strength. Equation (13.1) can also be expressed as
d
θ̇ = − V (θ), (13.2)
dθ
13.1 Degrading Effects of Noise: Noise-Induced Phase Slips 271
where V (θ) = −θΔω − ε cos θ is a washboard potential. The synchronization

regime is achieved when the parameters satisfy
Δω
| | < 1, (13.3)
ε
which forms the synchronization region (see Chap. 4 for a more detailed dis-
cussion). In this region, the system has a stable steady state θ̄ = arcsin Δω
ε
which corresponds to a minimum of the washboard potential V (θ). In gen-
eral, the synchronization regime can be destroyed when the oscillators are
disturbed by noise. Then the dynamics of the phase difference reads as
θ̇ = Δω − ε sin θ + Dξ(t), (13.4)
where D is the intensity of the noise. A detailed analytical description of

this system is possible using the Fokker–Planck equation if we assume the
noise ξ(t) to be Gaussian delta-correlated (for details we refer to [110]). In
general, noise makes the phase difference fluctuate around the minimum of
the washboard potential V (θ), and climb over the energy barrier occasionally
to move into the neighboring minima, as illustrated in Fig. 13.1a. As a result,
we can observe noise-induced 2π slips in the phase difference (Fig. 13.1b).
Now we show how noise induces different responses to periodic and chaotic
oscillators. In coupled periodic oscillators, the synchronization behavior can
be predicted from that of the phase oscillators just discussed. Let us consider
two coupled van der Pol oscillators (see Sect. 4.1.2)
ẍ1,2 − (1 − x21,2 )ẋ1,2 + ω1,2

2
x1,2 = ε(ẋ2,1 − ẋ1,2 ) + Dξ1,2 . (13.5)
The ξ1 and ξ2 in the two oscillators are independent Gaussian noises with
intensity D. Introducing the phase φj by
x˙j
φj = − arctan , (13.6)
xj
0 60
(a) (b)
50
−1 40
V(θ)
30
θ
−2 20
10
−3 0
0 10 20 30 0 5000 10000
θ time
Fig. 13.1. (a) Systematic plot of the washboard potential V (θ) = −θΔω − ε cos θ
for the system (13.1). (b) Noise makes the phase difference θ fluctuate and induces
phase slips
0.01 (a) 0.01 (c)
ΔΩ D=0.015 D=0.1
0.00 0.00
0.01 0.01 (d)
(b)
ΔΩ
D=0.05 D=0.3
0.00 0.00
0.017 0.018 0.019 0.020 0.021 0.017 0.018 0.019 0.020 0.021
ε ε
Fig. 13.2. Frequency difference ΔΩ vs. coupling strength ε. Solid lines: noise-free
case D = 0; circles: independent noise; Left panel (a,b): two coupled periodic van der
Pol oscillators (13.5); right panel (c,d): two coupled chaotic Rössler oscillators (13.7).
The natural frequencies are ω1 = 0.99 and ω2 = 0.97
we can compute the frequencies Ωj = φ̇j of the oscillations. Figure 13.2 (left
panel) shows the difference of the mean frequency ΔΩ = |Ω1 − Ω2 | as a
function of the coupling strength ε for various noise levels. Without noise,
D = 0, synchronization is achieved when ε > εps = Δω = 0.02, as predicted
by (13.3) (see the solid lines in Fig. 13.2a,b). When a small noise is included,
D = 0.015, we can see that the transition to synchronous regime is smeared
(Fig. 13.2). The frequency difference ΔΩ is not vanishing when the coupling
strength is slightly beyond the threshold εps , because noise induces phase slips.
These phase slips occur less frequently when the coupling becomes stronger,
so that the synchronization region is restored at larger coupling strength.
When the noise intensity is higher, phase slips happen more frequently, so that
effective synchronization with small ΔΩ ≈ 0 would require a much stronger
coupling strength, as seen in Fig. 13.2. Hence noise leads only to degrading
effects in periodic oscillators.
A different response to noise occurs in two coupled chaotic Rössler
oscillators
ẋ1,2 = −ω1,2 y1,2 − z1,2 + ε(x2,1 − x1,2 ),

ẏ1,2 = ω1,2 x1,2 + 0.15y1,2 + Dξ(t), (13.7)
ż1,2 = 0.4 + (x1,2 − 8.5)z1,2 ,
with ω1 = 0.99 and ω2 = 0.97. As seen in Fig. 13.2c,d by the solid line, the
transition point to the perfect PS regime in the absence of noise, εps = 0.0208,
is somewhat higher than that of the limit-cycle oscillators at εps = 0.020. In
the presence of noise, phase slips occur in the originally phase synchronized
regime, so that ΔΩ is not vanishing for ε slightly beyond the threshold. Com-
pared to the periodic oscillators in the left panel, it is important to emphasize
that noise makes the frequency difference ΔΩ smaller than that of the noise-
free oscillators for ε < εps . A smaller frequency difference indicates that phase
13.2 Noise-Induced CS and PS in Uncoupled Chaotic Oscillators 273
slips are reduced and a degree of synchrony is enhanced due to the noise, i.e.,
noise induces more order here.
This comparison shows that, in chaotic systems, noise may play a con-
structive role to induce and enhance synchronization. In the following, we
will show that noise can enhance synchrony of chaotic systems in different
ways and with different mechanisms. Similar effects we have also discussed in
Chap. 11.
13.2 Noise-Induced CS and PS in Uncoupled

Chaotic Oscillators
13.2.1 Noise-Induced CS of Identical Chaotic Oscillators
Synchrony of nonlinear systems which are not coupled directly, but influ-
enced by a common random force, has drawn considerable attention and has
been a topic of a long-standing controversy. Firstly, it was shown that noise
can induce synchrony in periodic oscillations [441–443]. More interesting, the
phenomenon of noise-induced order was reported on a chaotic map which
is directly connected to the Belousov–Zhabotinsky chemical reaction [444].
There, a small amount of noise may change a chaotic trajectory of the sys-
tem into a state similar to a periodic orbit smeared with noise [444], which
makes the largest Lyapunov exponent (LE) (λ1 ) negative, and leads to a slower
decay of correlations and an improvement of the state predictability [445,446].
A negative largest LE means that, in an ensemble of systems with identical
laws of motion and common noise, such as the motion of floating particles on
a surface of an incompressible fluid [447], the states in the phase space shrink
into a single point [447, 448], i.e., noise induces CS in chaotic systems.
However, whether in general common noise can induce synchrony of
chaotic systems has been a topic of long-standing dispute. In [449], numer-
ical simulations show that a white noise can induce synchrony of the logis-
tic map, and in the Lorenz system, CS was observed for uniform noise in
[0, W ] with large enough W values, but not for unbiased one. In [450] it was
pointed out that the largest LE of the noisy logistic map is positive. The
CS observed in [449] is actually an artifact of finite precision in numerical
simulations [450, 451]. Several other authors, on the other hand, restudied
the problem by examining the properties of the noise applied to the systems.
The largest LE of the noisy Lorenz system is found to be the same as that
of the system driven constantly only by the mean value W/2 of the noise,
indicating that the bias of the noise plays the crucial role in synchroniza-
tion processes [452, 453]. Recently, in [454] authors analyzed CS of chaotic
systems by noise in an experiment with Chua’s circuit, drawing the general
conclusion that CS might be achieved only by a biased noise, but not by an
unbiased one, and synchrony is only a result of noise-induced order because
the biased noise drives the system into a noise-smeared periodic orbit, similar
as in [444]. In fact, the nonzero-mean value of noise can be separated and

viewed as an additional parameter which may move the system into another
dynamical regime. Considering the system with only a constant bias of noise
as a new system which can still be chaotic, the controversy can be focused
to the question whether unbiased noise can induce synchrony in originally
chaotic systems. The reason has not been clearly addressed, although there
are numerical and experimental examples of noise-induced synchrony.
However, it has been demonstrated that common unbiased noise can
indeed induce synchrony [455, 456]. This controversy has been recently clar-
ified [455, 457], by showing that the bias of the noise is not the key point,
while the existence of a contraction region inherent in the system state space
is crucial for noise-induced synchronization-like phenomena.
In this section, we first demonstrate the importance of such a contraction
region for noise-induced synchrony in time–discrete chaotic maps. Then this
idea of a contraction region is extended to time-continuous systems. Finally,
it is pointed out that, in nonidentical systems, a common noise can induce a
degree of PS even if it is not strong enough or impossible to induce CS [457].
Time–Discrete One-Dimensional Chaotic Maps
For two uncoupled identical maps linked by a common additive noise ξ,
xn+1 = f (xn ) + Dξn , (13.8)

yn+1 = f (yn ) + Dξn , (13.9)
the synchronous behavior is determined by the linearized dynamics of the

synchronization error δxn = yn − xn ,
δxn+1 = f (xn )δxn . (13.10)
A stable synchronous regime requires that the LE
1
N
λ = lim ln |f (xn )| (13.11)
N →∞ N
n=1
becomes negative. Note that the Jacobian f (x) also governs the linearized
dynamics in the noise-free case. In this sense, the LE of stochastic systems is
well defined as in the deterministic systems [458]. The difference is that the
attractor in the phase space is modified by the noise. The noise-free map is
chaotic, i.e., λ > 0. To make λ < 0, we need a significant contribution from
regions where |f (xn )| < 1. Such regions are called contraction regions Cf of
the map, defined as

Cf = {x|f (x)| < 1}. (13.12)
In the contraction regions, nearby trajectories converge to each other tem-

porally, and the synchronization error δxn in (13.10) shrinks. Note that the
1.0 1.0
0.5
f(x) 0.5 0.0
−0.5
(a) (b)
0.0 −1.0
0.0 0.5 1.0 −2.0 −1.0 0.0 1.0 2.0
x x
0.8 0.8
0.6 0.6
0.4 0.4
λ
0.2 (c) 0.2 (d)

0.0 0.0
−0.2 −0.2
0.40 0.40
E
0.20 0.20
(e) (f)
0.00 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
noise intensity D noise intensity D
Fig. 13.3. Contraction regions and synchrony in chaotic maps. Left panel (a,c,e):
Logistic map f (x) = 4x(1 − x). Right panel (b,d,f ): A chaotic neuron model f (x) =
tanh(A1 x) − B tanh(A2 x) with A1 = 20, A2 = 2, and B = 1.5. The contraction
regions where |f (x)| < 1 are highlighted by thick solid lines in (a) and (b). (c) and
(d) give the LE λ, and (e) and (f ) the synchronization error E as a function of the
noise intensity D
contraction regions are intrinsic in the map f (x), but independent of the
forcing Dξ. A nonzero-mean value of the forcing D ξ , however, may shift the
new dynamical system xn+1 = f (xn ) + D ξ to different dynamical regimes,
e.g., to a periodic orbit from the previously chaotic regime.
Figure 13.3 depicts two maps which have quite different structures of con-
traction regions Cf . The well-known logistic map f (x) = 4x(1 − x) has only
a small contraction region around x = 0.5, which is manifested by the thick
line in Fig. 13.3a. The other chaotic map f (x) = tanh(A1 x) − B tanh(A2 x)
in Fig. 13.3b is a chaotic neuron model [459], which has broad contraction
regions at large |x|. In fact, these two maps have very different synchrony
properties in the presence of noise ξ uniformly distributed in [−1, 1]. As seen
in Fig. 13.3c, λ is always positive for the logistic map so that a common unbi-
ased noise cannot synchronize two identical logistic maps. The time average
of the synchrony error E = |xn − yn | displays a similar pattern (Fig. 13.3e),
but it is well above zero. Note that around D = 0.5, both λ and E reach a
minimum. Here noise makes the trajectory spend a lot of time in the con-
traction region, so that the synchrony error can reach rather small values in
some time intervals. In the chaotic neuron model, λ undergoes a transition to
negative values at D = 0.87, as shown in Fig. 13.3d. Accordingly, beyond the
threshold, the synchrony error E decays to zero after a transient time. A more
detailed study has shown that, in the presence of strong enough noise, the sto-
chastic trajectories shift to spend a lot of time in the contraction regions [455]
where they converge to each other. When the convergence becomes dominant,
λ becomes negative, and the local instability in the regions |f (x)| > 1 cannot
keep the map chaotic. Due to the coexistence of contraction and expansion
in the phase space, close to the transition point, the separation of the trajec-
tories exhibits a very intermittent behavior [447], and the probability density
of the separation satisfies a scaling law [448]. This intermittent behavior has
been shown [460] to be on–off intermittency [461].
Time-Continuous Chaotic Systems: Rössler vs. Lorenz Systems
In two continuous systems subjected to a common noise
ẋ1 = f (x1 ) + Dξ(t), (13.13)

ẋ2 = f (x2 ) + Dξ(t), (13.14)
where (x1 , x2 ∈ RM ), the small initial difference δx(t) = x2 (t) − x1 (t) evolves
according to the linearized dynamics
δ ẋ = Jf (x1 )δx, (13.15)
where Jf (x) stands for the Jacobian matrix. This linear equation is the same
as the noise-free case (D = 0) where the system is chaotic, i.e., the maximal LE

1 T |δx(t)|
λ1 = lim ln (13.16)
T →∞ T 0 |δx(0)|
is positive. In the presence of noise, however, the trajectory x is different from
that in the noise-free system, and may explore some regions in the phase space
which are not reachable by the original chaotic system. As in the maps, for CS
to occur, λ1 should become negative, which is possible only when there exist
well-exposed contraction regions. In general, contraction regions are defined as
the region in the phase space where nearby trajectories converge. A necessary
condition for the contraction is

Cf = {x Re[Λi (x)] < 0 (i = 1, . . . , M )}, (13.17)
namely, all the M eigenvalues Λi (x) of the Jacobian matrix Jf (x) have a
negative real part.
There is different synchronization-like behavior [457] in the Rössler systems
ẋ1,2 = −ωy1,2 − z1,2 ,

ẏ1,2 = ωx1,2 + 0.15y1,2 + Dξ(t), (13.18)
ż1,2 = 0.4 + z(x1,2 − 8.5),
with ω = 0.97, and the Lorenz systems

ẋ1,2 = σ(y1,2 − x1,2 ),
ẏ1,2 = ρx1,2 − y1,2 − x1,2 z1,2 + Dξ(t), (13.19)
ż1,2 = −bz1,2 + x1,2 y1,2 ,
with the parameters σ = 10, ρ = 28, and b = 8/3. The noise ξ(t) is Gaussian
white, ξ(t)ξ(t − τ ) = δ(τ ).
The results of λ1 and the synchrony error E = |x1 − x2 | for the Rössler
system and the Lorenz system are shown in Fig. 13.4. In the Rössler system,
λ1 stays positive till the systems become unstable for D > 4. CS regime is
never observed here. This behavior is similar if the noise is applied to the x
variable, or to both x and y variables. Noise is not applied to the z variable,
because rather small noise already makes the system unstable. However, in the
Lorenz system, λ1 becomes negative for D > 33.3, and two identical Lorenz
systems with common noise achieve CS, as is also expressed by the vanishing
synchrony error E beyond the transition point.
It is important to note that, in the Lorenz system, even for rather strong
noise, the basic “butterfly” structure is preserved (Fig. 13.5). The systems
0.075 1
0.000 0
λ1
(a) (b)
−0.075 −1
10 10
5 5
E
(c) (d)
0 0
0.0 1.0 2.0 3.0 4.0 0 10 20 30 40
Fig. 13.4. The largest LE λ1 of the Rössler system (a) and the Lorenz system
(b). The average synchronization error E of the Rössler system (c) and the Lorenz
system (d)
80
40
z
(a) (b) (c)

0
−25 0 25 −25 0 25 −25 0 25
x x x
Fig. 13.5. Trajectories in the phase space of the Lorenz system at different noise
intensity: (a) D = 0, (b) D = 20, and (c) D = 40. The dotted background shows the
contraction region in the plane y = 0
now explore a larger region of the phase space with increasing noise intensity
D. In the CS regime D > 33.3, the trajectory is much more complex than a
smeared periodic orbit and is quite different from the external noise.
In the Lorenz system with the parameters considered here, the origin
(0, 0, 0) is a saddle point “embedded” in the chaotic attractor (i.e., the chaotic
trajectory goes into the vicinity of this point). Hence the chaotic trajectories
leaving the neighborhood of this saddle point will come back to its neigh-
borhood. Due to this homoclinic return of chaotic orbits, there exists a large
contraction region close to the stable manifold of the saddle point (Fig. 13.5).
In the presence of noise, the trajectories cannot come as close to the saddle
point as in the noise-free system; instead they explore deep into the con-
traction region. The modification of the attractor in the phase space changes
the competition between contraction and expansion. When the contraction
dominates over the expansion, the largest LE λ1 becomes negative and CS
occurs.
In the Rössler system, the trajectories spiral outward following the
guidance of the two-dimensional unstable manifold of the focus and are
fold back by the nonlinearity. A contraction region with all three Re(Λi ) < 0
does exist, but the contraction is very weak because the largest Re(Λi ) is
close to zero. In addition, in the presence of noise the system still spends
only a small portion of time in the contraction region. The contraction is not
sufficient to induce CS. There are also regions in the phase space where all
Re(Λi ) > 0, and strong enough noise (here D > 4) makes the system access
to such regions and breaks the system down easily.
To summarize, the controversy on the mechanism of noise-induced CS
of chaotic systems can be resolved in the following manner. A significant
contraction region plays a decisive role, as in the one-dimensional chaotic
maps. Noise may change the balance between contraction and expansion, and
a synchronous regime occurs when contraction becomes dominant. Whether
the noise is biased or unbiased is not the key point. If there does not exist a
contraction region, CS cannot occur with any additive common driving signal.
This understanding from the Lorenz system is very instructive to observe
noise-induced synchronization in other systems. In general, it is expected that,
in a system possessing the structure of a saddle point embedded in a chaotic
attractor and homoclinic return of orbits (homoclinic chaos), the generic exis-
tence of a large enough contraction region and the sensitivity of the system
to noisy perturbation near the saddle point will result in noise-induced syn-
chrony. Note that such a structure is typical for the spiking behavior of many
neural, chemical, and laser systems, and noise-induced synchrony may play
an important functional role there. We present two examples in the following.
Chaotic Hodgkin–Huxley Neuron Model
The first spiking system is the Hodgkin–Huxley model [462] of thermally sen-
sitive neurons which was proposed in [463, 464] to mimic spike train patterns
observed in electroreceptors from dogfish and catfish, and from facial cold
receptors and hypothalamic neurons of the rat. The model equations read
dV
CM = −Il − Id − Ir − Isd − Isr + Dξ(t)
dt
ar φ(T )(ar∞ − ar )
= ,
dt τr
asd φ(T )(asd∞ − asd )
= ,
dt τsd
asr φ(T )(−ηIsd − θasr )
= , (13.20)
dt τsr
with Id = ρ(T )gd ad∞ (V − Vd ) and Ik = ρ(T )gk ak (V − Vk ), (k = r, sd), where
(T −T0 )/10 (T −T0 )/10
ak∞ = [1+exp(−sk (V −V0k ))]−1 and ρ(T ) = A1 , φ(T ) = A2 ,
ξ(t) is the Gaussian noise with intensity D. Here V is the membrane potential,
Il is the leakage current, and Id and Ir are fast currents representing Na and
K channels (see also Sect. 2.7.1).
In the chaotic spiking regime, the system possesses an unstable steady state
which has both stable and unstable two-dimensional manifolds. Therefore, it
is a saddle point S. A trajectory starting close to the stable manifold will
approach the neighborhood of S and leave it along the unstable directions.
This saddle point S is embedded in the chaotic attractor, i.e., a typical chaotic
trajectory may have very close recurrence to S after a sequence of spikes. The
eigenvalues corresponding to the stable manifold are real (λ3 < 0, λ4 < 0), but
the eigenvalues corresponding to the unstable manifold are complex (λ1,2 =
μ ± iν) and −λ3,4 > μ [465], and the chaotic dynamics resulting from this
Shilnikov condition [466] is a typical mechanism of chaotic spiking in neuron
models [467].
Close to the stable manifold, a large contraction region exists in the phase
space where nearby trajectories converge, while close to the unstable mani-
fold nearby trajectories diverge. The contraction and expansion in this high-
dimensional system can be manifested, by the largest local LE λτ ,
1 |δx(t)|
λτ (t) = ln , (13.21)
τ |δx(t − τ )|
where |δx(t)| is a small distance between two trajectories in the phase space at
time t, and τ is a finite time. λτ measures the average expansion or contraction
rate during the finite interval τ . Here τ = 50 ms is about half of the duration
of a single spike, but is much smaller than the average interspike interval
T ∼ 400 ms. With this τ value, quick and large changes of local stability
along the orbits during a spike have been smoothed out considerably and we
can see clearly the changes of the stability close to the saddle point where the
orbits slow down.
A typical chaotic spike sequence of the neuron at T = 10 (without noise
input) is shown in Fig. 13.6a, along with the largest local LE λτ in Fig. 13.6b.
(a)
0
V (mv)
S
− 40
− 80
(b) 120
Λτ (1/s)
0
S
− 120
2.0 3.0 4.0 5.0 6.0
time (s)
Fig. 13.6. Spike train (a) and local LE (13.21) (b) of the chaotic neuron (13.20)
at T = 10, without noise D = 0. The arrows indicate the saddle point S
The saddle point S is indicated by arrows in Fig. 13.6 close to t = 2.5 and
t = 4.5. The neuron generates a sequence of spikes between successive returns
to the neighborhood of S. After a a few spikes away from S in the phase
space, the orbit approaches S, and is guided by the stable manifold as seen
by negative λτ ; then it departs from S following the guidance of the unstable
manifold for a long time, as manifested by small positive λτ . The average
contraction rate is stronger than the expansion rate since the eigenvalues
satisfy −Λ > μ. A single spike (e.g., close to t = 2.8 and t = 5.3 in Fig. 13.6)
follows this recurrence to S. Such a single spike is well separated from the
next burst, because the orbit cycles S and slows down around S. For each
spike, λτ is positive during the activation phase, while it is negative during
the relaxation phase. In the absence of noise, the spike sequence is chaotic so
that the largest LE is positive, i.e., λ1 = λτ > 0.
Note that, in this system, the dynamics is sensitive to the noisy input close
to the saddle point, while the spikes are not affected much. This property is of
importance for biological information processing using spike trains which are
well defined even in the presence of a fluctuating input. The synchronization
behavior in this system is determined by the competition between contraction
and expansion in the phase space. Since the contraction rate is larger than
the expansion rate around S, in the presence of a suitable level of noise, the
orbits can still approach S and experience contraction, while they cannot
follow the guidance of the weakly unstable manifold for a long time. There
are two consequences of this behavior:
1. Those long intervals in the noise-free case resulting from following the
unstable directions of S have been reduced considerably.
2. The expansion degree is reduced correspondingly compared to the noise-
free case. As a result, the largest LE λ1 = λτ becomes negative at
a threshold noise level Dc ≈ 3.5 pA cm−2 . Beyond the threshold, CS is
achieved for two identical neurons forced by a common Gaussian noise, as
2
(a)
λ1(1/s)
1
1
1.0
(b)
E
0.5
0.0
0.0 1.0 2.0 3.0 4.0 5.0
noise intensity D (pA / cm2)

Fig. 13.7. (a) The largest LE λ and (b) the relative synchrony error E (filled
circles for two identical neurons and squares for two nonidentical neurons with a
5% difference of the leakage parameter gl ) vs. the noise intensity D
seen by the average relative synchrony error E = |V1 − V2 | /σV between

the two neurons (Fig. 13.7). Here σV2 is the variance of V (t) over time,
which increases with D because a larger input generates a more frequent
spiking.
In summary, noise-induced CS occurs due to contraction region generi-
cally present along the stable manifold of a saddle point in the realistic neu-
ron model. This mechanism thus interprets the experimental observations of
repetitive spiking in [437] from the viewpoint of dynamical systems, which
should be meaningful for understanding neuron encoding [468].
Homoclinic Chaotic Laser
Very similar noise-induced CS of spike trains due to a saddle point S has been
observed in an experiment of homoclinic chaotic lasers [469].
The experimental setup consists of a CO2 laser with an intracavity loss
modulator (parts 1–4 in Fig. 13.8), driven by a feedback signal which is pro-
portional to the laser output intensity (parts 6–8 in Fig. 13.8). This system
is operating in a homoclinic chaotic regime where the laser output consists
of a chaotic sequence of spikes. To investigate the role of external noise, a
Gaussian noise generator with a frequency cut-off at 50 kHz is inserted into
the feedback loop.
Before presenting the results of the experiments, we analyze the model of
this laser. The dynamical equations describing appropriately the experimental
system read:
ẋ1 = k0 x1 (x2 − 1 − k1 sin2 x6 ),

To digital scope
1 1
4
2 6
3 3 7
R
9 8
B0
Noise source
Registered noise signal
Real Time PC Board Input Output
Fig. 13.8. Experimental setup of the laser system
ẋ2 = −γ1 x2 − 2k0 x1 x2 + gx3 + x4 + p0 ,

ẋ3 = −γ1 x3 + gx2 + x5 + p0 ,
ẋ4 = −γ2 x4 + zx2 + gx5 + zp0 ,
ẋ5 = −γ2 x5 + zx3 + gx4 + zp0 ,
rx1
ẋ6 = −β(x6 − b0 + ) + Dξ(t). (13.22)
1 + αx1
Here, x1 represents the laser output intensity and x6 the feedback voltage
signal which controls the cavity losses. The other variables denote the internal
states of the system. More details of the model are given in [470]. With the
following parameters k0 = 28.5714, k1 = 4.5556, γ1 = 10.0643, γ2 = 1.0643,
g = 0.05, p0 = 0.016, z = 10, β = 0.4286, α = 32.8767, r = 160, and
b0 = 0.1031, the model reproduces very well the regime of homoclinic chaos
observed experimentally.
To study synchronization-like phenomena of this system [469], it is crucial
to take into account the small intrinsic noise present in the experimental
system. The level of the intrinsic noise level (≈ 7 mV) measured in the feedback
variable (x6 ) is about 0.14% of the feedback signal x6 in the experimental
system. In the model, D = 0.0005 is equivalent to the intrinsic noise amplitude
in x6 .
Without added noise, the laser output x1 displays large spikes, followed
by a fast damped train of a few oscillations toward a saddle point S and a
successive longer train of growing oscillations spiraling out from S, as seen
from simulated time series of the model in Fig. 13.9a. The damped oscillation
manifests a strong contraction along the stable manifold in the phase space.
The contraction along the stable manifold is illustrated in Fig. 13.9b by the
convergence of two nearby trajectories. The growing oscillations manifest a
(b) S
(a)
T
x1
0.0 0.5 1.0

time (ms)
Fig. 13.9. (a) Laser output intensity x1 of the noise-free model (13.22).
(b) Projection of the trajectory in the 3D phase space (x1 , x2 , x6 )
weak expansion along the unstable manifold, which can be described approx-
imately as
X(t) ∼ X0 exp[Λu (t − t0 )] cos ω(t − t0 ), (13.23)
where Λu ± iω are the leading eigenvalues of the unstable manifold of S,
and X0 is the distance from S at any reinjection time t0 . Thus, the smaller
X0 is, the longer is the time taken to spiral out. In the presence of noise,
the trajectory on average cannot come closer to S than the noise level, and
perform those oscillations very close to S, while the dynamics along the stable
manifold is not much affected by the noise due to the strong contraction.
With a larger X0 , the system spends a shorter time following the guidance
of the unstable manifold, resulting in a reduced degree of expansion. This
highly nonuniform sensitivity to noise perturbation changes the competition
between contraction and expansion, and contraction may become dominant
at large enough D. The largest LE λ1 in the model changes as a function
of the noise intensity D. As seen in Fig. 13.10a, λ1 undergoes a transition
from a positive to a negative value at Dc ≈ 0.0031. Beyond Dc , two identical
laser models x and y with different initial conditions but with the same noisy
driving Dξ(t) achieve CS regime after a transient, quantified by the vanishing
|x1 −y1 |
normalized synchrony error E = |x 1 −x1 |
. At larger noise intensities, the
expansion becomes again significant, and CS is lost when λ1 becomes positive
for D > 0.052. Notice that, even when λ1 < 0, the trajectories still have access
to the expansion region where small distances between them grow temporally.
As a result, when the systems are subjected to additional perturbations, CS is
lost intermittently, especially for D close to the critical values. Actually, in the
experimental laser system, there always exists a small intrinsic noise source.
To take into account this intrinsic noise into the model, we introduce into
the equations x6 an equivalent amount of independent noise (with amplitude
D1 = 0.0005) in addition to the common one Dξ(t). By comparison, it is
evident that the sharp transition to CS in fully identical model systems is
smeared out.
10 1.7 1.7
E E
λ1 (1/ms)
0 0.0 0.0
(a) (b)
10 −1.7 −1.7
0.00 0.02 0.04 0.06 0 50 100 150
noise intensity D noise intensity D (mV)
Fig. 13.10. Noise-induced CS in the model (13.22) (a) and the experiment (b)
(Fig. 13.8). (a) Dotted line: the largest LE λ1 ; solid line: normalized error of syn-
chrony E between two fully identical laser models x and y; diamonds: E between
two lasers with a small independent noise (intensity D1 = 0.0005) equivalent to the
intrinsic noise in the experiments
This noise-induced CS is also observed in the experiment. As consistent

with numerical results with small independent noise, E does not reach zero
due to the intrinsic noise, and it increases slightly at large D, as depicted in
Fig. 13.10b.
It is important to emphasize that noise-induced CS is achieved for rather
small noise. In particular, the onset of CS occurs at an experimental added
noise (D ≈ 20 mV) of about 0.42% of the feedback signal x6 . This tiny amount
of noise only affects the system’s behavior close to the saddle S, while it does
not change the main geometrical feature of the spike. This feature is similar to
neuron spiking in the presence of a fluctuating signal where the shape of the
spikes is preserved, while the interspike intervals are altered. This is important
for biological information processing. Such a feature is a consequence of the
highly nonuniform sensitivity to noise perturbations in these systems, which
is different from the Lorenz system where noise has much stronger changed
the attractor to achieve CS (see Fig. 13.5).
Figure 13.11 shows time series of the laser intensity from the model and
the experiment under three different noise levels. The corresponding noise
levels relative to the feedback signal x6 in the model and experiment are
equivalent. Without a common noise, the two identical lasers x1 and y1 are
not synchronized; the interspike interval T is large on average and fluctu-
ates (Fig. 13.11a,d). When subject to a common noise with a level about
1–2% of the feedback signal x6 , x1 and y1 become almost completely synchro-
nized even in the presence of the intrinsic noise (Fig. 13.11b,e). In parallel the
spike sequences become more coherent with a smaller average T , because the
escaping time from the vicinity of the saddle S has been reduced; yet, the
main geometry of the spikes is preserved. Finally, at a larger noise, synchro-
nous states are intermingled with short epochs of nonsynchronous motions
(Fig. 13.11c,f). All features of noise-induced synchrony in the model and in
the experimental system are in good agreement (cf. [471] for more details).
D=0 (a) (d) D=0
D=0.013 (b) (e) D=66mV
D=0.03 (c) (f) D=133mV
5.0 5.5 6.0 6.5 7.0 8.0 9.0 10.0

time (ms) time (ms)
Fig. 13.11. Time series of output intensities (arbitrary units) of two lasers x1
(solid lines) and y1 (dotted lines) with a common noise. Left panel (a–c): model
systems (13.22) including independent noise (amplitude D1 = 0.0005 to intrinsic
noise level). Right panel (d–f ): experimental system (Fig. 13.8)
Synchronization of nonlinear systems by a common driving signal has also

been understood as the consistency in the response of the system to complex
driving signals [472].
13.2.2 Noise-Induced PS of Nonidentical Uncoupled

Chaotic Systems
In Sect. 13.2.1, we have shown that under certain conditions noise may induce
CS in chaotic systems, depending on the structure of the phase space. It is
interesting to study whether noise can also induce weaker degrees of synchro-
nization, similar to temporal locking of phases as in PS, when it is not strong
enough or not able to induce CS.
In Chap. 4, we have shown that in many systems there is the regime of PS
associated with the transition of one of zero LE to negative values. In time-
continuous autonomous chaotic systems ẋ = f (x) (f : Rn → Rn ), there is a
zero LE λ0 corresponding to perturbations of the motion along the trajectory,
which, in phase-coherent chaotic systems, can be connected to the phase. In a
system subjected to a weak noise, ẋ = f (x)+ξ, most time it is |f (x)| |ξ| and
one can also roughly speak about a motion along the trajectory and connect
the original zero LE to it and link it to the phase dynamics. Numerical results
show that λ0 becomes negative for strong enough noise, and in parallel, the
phases of the two slightly nonidentical systems coupled only by common noise
become statistically correlated [457].
To study PS due to noise, two systems with a small parameter mismatch
are considered: particularly, ω1 = 0.97 and ω2 = 0.99 in the Rössler sys-
tem (13.18) and σ1 = 10, ρ1 = 28 and σ2 = 10.2, ρ2 = 28.5 in the Lorenz
system (13.19). This slight parameter difference does not change the LE spec-
tra of the systems much.
In noisy chaotic systems, the phase of the dynamics is no longer coherent,
i.e., a linearly increasing function of time. A phase linked to λ0 now cannot
be rigorously defined as in the noise-free case. Nevertheless, as discussed in
Chaps. 2–4, we can practically calculate a phase variable as in the determin-
istic system by the method of Poincaré section, e.g.,
t − τk
φ(t) = 2π[k + ], τk < t < τk+1 , (13.24)
τk+1 − τk
where τk and τk+1 are two successive crossings of a Poincaré section after
cycling a reference point (unstable fixed point of the noise-free system). In
stochastic systems, it is impossible to observe perfect synchrony of phases
φ1 and φ2 , i.e., |φ1 − φ2 | < const. An appropriate approach to study PS
in stochastic systems is to regard the phases in a stochastic manner, i.e., to
compute the distribution of the cyclic phase difference, P (Δφ), on [−π, π] [473,
474]. A peak in P (Δφ) manifests a preferred phase difference between the
systems. Preferred phase differences are expected to occur at least when λ0
becomes appreciably negative.
Without noise the two nonidentical Rössler systems are not phase synchro-
nized and the phase difference decreases almost linearly (Fig. 13.12a, D = 0).
Accordingly, the distribution of the phase differences Δφ on [−π, π] is almost
uniform (Fig. 13.12b). While for strong enough noise, where λ0 < 0, one
observes many plateaus in the phase difference, i.e., many phase locking epochs
(Fig. 13.12a, D = 3.0), and this is manifested by a pronounced peak around
Δφ = 0 in the distribution P (Δφ) (Fig. 13.12d); this demonstrates a noise-
induced imperfect PS. For D close to the transition of λ0 , the peak is not as
pronounced and is not located around Δφ = 0 (Fig. 13.12c). Similar properties
are also observed in the Lorenz system.
The degree of noise-induced imperfect PS can be measured quantitatively
by the mutual information between the cyclic phase dynamics on [−π, π] of
the two systems
p(i, j)
M1 = p(i, j) ln , (13.25)
i,j
p 1 (i)p2 (j)
where p1 (i) and p2 (j) are the probabilities when the phases φ1 and φ2 are in
the ith and jth bins, respectively, and p(i, j) is the joint probability that φ1
−70
(a) D=3
−100
Δφ
D=0
−130
5000 5500 6000
time
0.02 (b) (c) D=0.7

D=0 (d) D=3
P(Δφ)
0.01
0.00−π π −π π −π π
0 0 0
Δφ Δφ Δφ
Fig. 13.12. (a) Time series of the phase difference Δφ of two nonidentical Rössler
systems (ω1 = 0.97 and ω2 = 0.99), at noise intensity D = 0 and D = 3. In this
presentation, the phases are unwrapped. The distribution of the phase difference is
modulated into [−π, π] for D = 0 (b), D = 0.7 (c), and D = 3 (d)
1
0.00 0
(a)
λ0
−1 (b)
−0.05 −2
0
−2
10−1
10 10−2
−3 10−3
M
10 (d)
10
−4 (c) 10−4
−5 10 −5
10 10 0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 5 10 15 20
Fig. 13.13. The second LE λ0 of the Rössler system (a) and the Lorenz system
(b). Normalized mutual information M of the Rössler system (c) and the Lorenz
system (d)
is in the ith bin and φ2 in the jth bin. The number of bins of [−π, π] in our
simulations is N = 100. M1 is normalized into [0, 1] as M = M1 /Sm , where
Sm = ln N is the Shannon entropy of the uniform distribution of p1 and p2 .
Numerical results for M are shown in Fig. 13.13 along with the second LE λ0
as a function of the noise intensity D. Due to the incoherence of the phases,
an exact correspondence between the transition of λ0 and phase synchrony
would not be expected. Nevertheless, when λ0 becomes appreciably negative,
M increases rapidly, indicating an increasing degree of PS.
0.015 D=0 D=0.1 D=0.4

(a) (b) (c)
D1=0.15
D1=0.05 D1=0.1
P(θ)
0.000
−π 0 π −π 0 π −π 0 π
θ θ θ
Fig. 13.14. Distribution P (θ) in the zero-order approximation of the phase

model (13.28) with Δω = 0.02, at different D and D1 values
To understand PS induced by common noise, one can examine an approx-

imate dynamics of the phase difference θ = Δφ in the Rössler system, i.e.,
φ 1 + φ2 θ
θ̇ = Δω + 2Dξ sin sin + F (A), (13.26)
2 2
where F (A) denotes the fluctuations coming from the chaotic amplitudes and
Δω = ω1 − ω2 . In a simplified version by averaging over the fast phases φ1
and φ2 , we get
√ θ
θ̇ = Δω + 2Dξ sin + D1 η, (13.27)
2
where F (A) is described
√ here as independent Gaussian noise√ η. Theθ com-
mon noise term 2Dξ sin θ2 has a nonzero-mean value 2Dξ sin 2 ξ =
D2
4 sin θ ξ [475], which gives rise to a systematic contribution to the average
dynamics (with respective to ξ) of the system. A zero-order approximation is
then
D2
θ̇ = Δω + sin θ + D1 η, (13.28)
4
which does not include the remaining zero-mean stochastic contribution from
ξ. The analysis shows that when two oscillators are forced by a common
noise, their phases establish a relationship which is equivalent to the case that
they are coupled and subjected to perturbations. This zero-order approxima-
tion yields qualitatively the same features observed in the Rössler oscillators
(Fig. 13.14).
13.3 Noise-Enhanced PS in Weakly Coupled

Chaotic Oscillators
In Sect. 13.2.2, we have shown that a common noise can induce a degree of
PS or CS of uncoupled oscillators. In this section, we treat effects of noise on
synchronization of chaotic oscillators which are weakly coupled to a periodic
forcing, or to other chaotic oscillators. Note that the coupling considered here
is too weak to induce perfect PS (subthreshold coupling).
13.3 Noise-Enhanced PS in Weakly Coupled Chaotic Oscillators 289
13.3.1 Noise-Enhanced PS of a Chaotic Laser Due to Periodic

Forcing
We start with the chaotic laser model (13.22). We have demonstrated in

Fig. 13.11 that noise reduces the average value and the variability of the inter-
spike interval T . It is interesting to know whether these noise-induced changes
in the time scale of the system can also enhance the synchronization of the
system to an external periodic forcing. For this purpose, a periodic modulation
of the pumping parameter p0 , with a relative amplitude A and frequency fe , is
introduced in the system (13.22) (part 9 in the experimental setup Fig. 13.8),
namely
p(t) = p0 [1 + A sin(2πfe t)]. (13.29)
Without noise (D = 0) and driving signal (A = 0), the orbit approaches
the saddle S via a few quickly decaying oscillations (stable manifold) and
leaves S via a series of slowly growing ones (unstable manifold). It may have
a different number of oscillations before generating a large spike, depending on
the distance from S at the previous reinjection. As a result, the model displays
a broad range of time scales, and there are many peaks in the distribution
P (T ) of the interspike interval T (Fig. 13.15a). With a small noise (D =
0.0005), the orbits can no longer perform some oscillations very close to S,
resulting in a clear change in the time scales: P (T ) is now characterized by
a dominant peak followed by a few exponentially decaying ones (Fig. 13.15c).
This distribution of T is typical for small D in the range D = 0.00005–0.002.
The experimental system with only intrinsic noise (equivalent to D = 0.0005 in
the model) has a very similar distribution P (T ). At larger intensity D = 0.01,
0.2
(a) (c) (e)
P(T)
0.1
0.0
0.2
(b) (d) (f)
P(T)
0.1
0.0
0.0 0.5 1.0 0.0 0.5 1.0 0.0 0.5 1.0
T (ms) T (ms) T (ms)
Fig. 13.15. Probability density of interspike intervals of the noise-free (D = 0,

a, b) and the noisy (D = 0.0005, c, d; D = 0.01, e, f ) laser model (13.22). Upper
panel : without external forcing, A = 0; lower panel : with forcing, A = 0.01. The
signal period Te in (b), (d), and (f ) corresponds to the average interspike interval
T0 (D) (vertical dotted lines) of the unforced model in (a), (c), and (e), respectively
100
D=0
70
θ(t) 40 D = 0.01
10
D = 0.0005
−20
0 100 200 300 400 500
time (ms)
Fig. 13.16. Phase difference between the laser model and the driving signal at
various noise intensities. The forcing amplitude is A = 0.01, and the signal period
Te is the same as in Fig. 13.15b,d,f for D = 0, D = 0.0005, and D = 0.01, respectively
noise eliminates most of the oscillations around S (see Fig. 13.11b for a similar
behavior); the fine structure of the peaks is smeared out and P (T ) becomes a
unimodal peak with a lower height (Fig. 13.15e). Note that the average value
T0 (D) of T decreases with D. Consequently, the average spiking rate of the
unforced laser model, f0 (D) = 1/T0 (D), is an increasing function of D.
As a result of these noise-induced changes in the time scales, the model
displays a quite different response to a weak signal (A = 0.01) with a frequency
fe = f0 (D) = 1/T0 (D), i.e., equal to the average spiking rate of the unforced
model. At D = 0, P (T ) of the forced model still has many peaks (Fig. 13.15b),
while at D = 0.0005, T is sharply distributed around the signal period Te =
T0 (D) (Fig. 13.15d). However, at larger intensity D = 0.01, P (T ) becomes
lower and broader again (Fig. 13.15f). To examine phase synchronization due
to the driving signal, the phase difference θ(t) = φ(t) − 2πfe t is computed.
Here the phase φ(t) of the laser spike sequence is simply defined by (13.24),
where τk now is the spiking time of the kth spike. Figure 13.16 displays the
phase difference θ corresponding to Fig. 13.15b,d,f. At D = 0, the phase of the
laser model is not locked by the external forcing. On the contrary, with a small
noise D = 0.0005, phase slips occur very rarely and phase locking becomes
almost perfect when noise generates a pronounced time scale in the system. At
stronger intensity D = 0.01, noise becomes dominant over the signal around
the saddle S, and it induces many random-like phase slips. The behavior is
similar for driving frequencies close to f0 (D).
Now we study the synchronization region (1:1 response) of the laser model
systematically in the parameter space of the driving amplitude A and the
relative initial frequency difference Δω = (fe − f0 (D))/f0 (D). The actual
relative frequency difference in the presence of the signal is calculated as
ΔΩ = (f − fe )/f0 (D), where f = 1/ T t is the average spiking rate of the
model subject to the pumping modulations. The results are shown in Fig. 13.17
for three different noise intensities. In the noise-free model (Fig. 13.17a), the
0.015
0.010
A
0.005
(a) (b) (c)
0.000
−0.3 0.0 0.3 −0.3 0.0 0.3 −0.3 0.0 0.3
Δω Δω Δω
Fig. 13.17. Synchronization region of the laser model at various noise intensities.
A dot is plotted when |ΔΩ| ≤ 0.003: (a) D = 0, (b) D = 0.0005, and (c) D = 0.005
synchronization behavior is quite complicated and featureless: at weak ampli-

tudes (about A < 0.012), there does not exist a tongue-like region similar
to the Arnold tongue in phase-coherent oscillators; for a fixed A, ΔΩ is not
a monotonous function of Δω and it vanishes only at some specific signal
frequencies; at stronger driving amplitudes (about A > 0.012), the system
becomes periodic at a large frequency range. The addition of a small noise,
D = 0.0005, drastically changes the response: a tongue-like region, where
effective frequency locking (|ΔΩ| ≤ 0.003) occurs, can be observed similar
to that in usual noisy phase-coherent oscillators (Fig. 13.17b). The synchro-
nization region shrinks for stronger noise intensities D = 0.005 (Fig. 13.17c).
A more detailed analysis can be found in [476]; there it has also been shown
that noise can induce n : m locking of the laser to the modulation signal.
It is meaningful to mention that the homoclinic chaotic laser system is
not phase-coherent because of the wide fluctuations of the return time T .
In phase-coherent chaotic oscillators, such as the Rössler oscillators, T fluctu-
ates slightly around the mean value, namely (see also Fig. 2.3),
|T / T − 1| < 1. (13.30)
As a result, in phase-coherent oscillators, there exists a coordinate change

of the chaotic attractor into the phase φ and amplitudes A [111] (e.g., see
Chap. 4). In the presence of weak perturbations by an external periodic signal
or by another phase-coherent oscillator, if the response does not show a signi-
ficant dependence on the amplitudes A, then the phase can be locked similarly
in periodic oscillators. Phase locking in such a case can be well described by
the phase oscillators, probably with a small amount of noise.
A common property of the Lorenz system, the chaotic neuron model, and
the chaotic laser system is that there is a saddle point embedded in the chaotic
attractor. In the neighborhood of this point, the trajectory slows down con-
siderably. The return time is very broad and the condition (13.30) cannot be
satisfied. A change of the coordinate to phase thus does not exist for the whole
attractor, while it may exist in some part of the attractor away from the sad-
dle point [111]. Even though a phase may also be defined heuristically as in
the spiking sequence of the laser, the response of the phase to perturbations
by the signal is strongly correlated with the amplitude A. In fact, in Chap. 3,
it has been shown that PS of the Lorenz system to periodic signal is imperfect:
there is no region as in the Arnold tongue of phase-coherent oscillators where
the phase is locked forever [477].
Our analysis of the laser system has shown that, when the system is kicked
away from the neighborhood of the fixed point by noise, the degree of phase-
coherence is increased so that the system shows an enhanced synchronization
to weak periodic signal. The locking tongues are now similar to phase-coherent
oscillators in the presence of noise. Such a property seems to be general also in
excitable systems where noise kicks the system to escape the stable fixed point
to generate a sustained oscillation. Noise-enhanced synchrony of excitable
systems will be shown in Sect. 13.4.
13.3.2 Noise-Enhanced PS of Two Coupled Rössler Oscillators
Now we consider two mutually coupled chaotic Rössler oscillators influenced

by a noise:
ẋ1,2 = −ω1,2 y1,2 − z1,2 + ε(x2,1 − x1,2 ),

ẏ1,2 = ω1,2 x1,2 + 0.15y1,2 + Dξ1,2 (t), (13.31)
ż1,2 = 0.4 + (x1,2 − 8.5)z1,2 .
Here ω1 = 0.97 and ω2 = 0.99. The added noise Dξi (t) (i = 1, 2) is now
composed of a common part e(t) and an independent part ηi (t), satisfying
√ √
ξi (t) = Re(t) + 1 − Rηi (t). (13.32)
Both e(t) and ηi (t) are assumed to be Gaussian noise and δ-correlated in time.
This description allows independent variations of the correlation R between
ξ1 and ξ2 and the noise intensity D.
Without noise, the two oscillators achieve PS for the coupling strength
ε > εps = 0.0208, where εps is the transition point to PS. In the weak cou-
pling region ε < εps , the phases are not yet fully synchronized, while when ε
approaches εps , appreciable PS epochs can be observed between phase slips.
For example, at ε = 0.0205, the phase difference Δφ = φ1 − φ2 decreases con-
tinuously. But there are many epochs of PS interrupted by phase slips; and
typically the PS epochs last for about 300 cycles of oscillations (Fig. 13.18).
Adding a proper amount of noise to the two oscillators (e.g., D = 0.1) can
prolongate remarkably the duration of the synchronization epochs: the two
oscillators maintain phase synchronization for a period of about 3,000 cycles
of oscillations. However, for stronger noise (e.g., D = 0.3), phase slips occur
more frequently again.
This phenomenon of noise-enhanced phase synchronization can be quanti-
fied by the duration τ of the PS epochs. It is found that the average duration
104
0 a b
D=0.1
50
103
<τ>
Δφ
D=0.3
ε=0.0205
100 ε=0.0200
ε=0.0195
D=0 102
ε=0.0190
150
0 25000 50000 0.0 0.1 0.2 0.3 0.4
time noise intensity D
Fig. 13.18. Noise-enhanced PS in two weakly coupled Rössler chaotic oscillators

(ε = 0.0205). (a) Phase difference vs. time for different noise intensity D. (b) Average
duration of PS epochs vs. noise intensity D for different coupling strength ε. Error
bars are shown for ε = 0.0205
τ increases with the noise intensity D, reaches a maximal value, and

decreases for larger D for all coupling strengths analyzed (Fig. 13.18b). Thus
there is an optimal amount of noise, which enhances PS most significantly.
Such a resonant-like behavior is typical in the interplay between noise and
nonlinearity in dynamical systems [435].
The results are almost the same for different noise correlation R, but at
large noise intensity D, τ takes a slightly smaller value if the noise is inde-
pendent (R = 0), compared to that for a common noise (R = 1). Unlike in
Sect. 13.2.2, here the enhanced PS is not simply due to the correlation of noise.
The underlying mechanism can be understood [478] in terms of unstable peri-
odic orbits (UPOs) embedded in the chaotic attractors [479], as discussed in
Chaps. 3 and 4. Let us recall this mechanisms. Chaotic trajectories can have
rather close recurrence to these orbits, but noise generally prevents the system
from following the UPOs closely for a long time. The PS regime of coupled
chaotic oscillators can be viewed as phase locking of all pairs of UPOs. Since
different UPOs have slightly different periods, the Arnold locking tongues are
also slightly different, and the overlapping region of the locking tongues of
each pair corresponds to the full synchronization region of the chaotic trajec-
tories. When a parameter crosses the border and moves out of this region,
some pairs of the UPOs become unlocked while others remain locked. Phase
slips of a chaotic oscillation now become possible, but only when the system
comes to follow one of the unlocked pairs for a time long enough for a phase
slip to occur. Figure 13.19 confirms that phase slips are indeed associated
with UPOs, which is seen especially clearly for the coupling strength ε close
to the threshold εps where only a few pairs of UPOs become unlocked. Peri-
odic orbits are manifested by almost vanishing ΔXk = |Xn+k − Xn | which is
the difference between the x variable after every k successive returns to the
Poincaré section y = 0, x < 0, with a return time Tk . So an orbit identified in
this way is not an exact UPO, but a close approximation of it when the chaotic
0 0
(a) (d)
Δφ
Δφ
−20 −20
9 6.4
(b) (e)
6 6.3
T4/4
ΔX4
3 6.2
0 6.1
6 6.4
4 (c) (f)
6.3
ΔX2
T4/4
2 6.2
0 6.1
2000 4000 6000 2000 4000 6000
t t
Fig. 13.19. Illustration of phase slips (a, d) induced by unlocked UPOs (b, e,
period-4 for oscillator 1; c, f, period-2 for oscillator 2) at ε = 0.0205. The insets
in (b) and (c) show the unlocked UPOs around t = 2, 500. T4 /4 in (e) and (f ) is
the average return time calculated by every four returns to the Poincaré section in
system (13.31)
trajectory is very close to it. It is seen that phase slips occur between a period-
4 UPO in oscillator 1 and a period-2 UPO in oscillator 2 which are followed
closely by the systems for a fairly long time (∼ 30 cycles). While most orbits
are locked with return times fluctuating around a common value (T = 6.24),
these UPOs have clearly much smaller and larger return times (Fig. 13.19e,f),
thus remain unlocked by the coupling. With a noise of D = 0.1, such a long
time staying close to UPOs is rarely observed, and meanwhile most of the
phase slips are eliminated (Fig. 13.18). At stronger noise, e.g., D = 0.3, phase
slips develop quickly when the oscillators come to some orbits with quite large
differences in the return times, which cannot follow UPOs closely.
Noise-enhanced PS thus can be explained as follows. Noise exerts two
effects (1) it prevents the system from staying close to the unlocked UPOs
for long enough times to allow a phase slip to occur and (2) it generates fluc-
tuations in the return times and may induce phase slips of locked orbits, as
it does in coupled periodic orbits. The degree of PS is enhanced when (1)
dominates over (2) at a weak noise level, while it is degraded again when (2)
becomes dominant at large noise. There thus exists an optimal noise inten-
sity, yielding the maximal enhancement as a result of the competition between
these two influences and it is a resonance-like phenomenon. At a smaller cou-
pling strength ε, more orbits become unlocked, and phase slips may develop
already during a shorter time τsl when the oscillators approach some unlocked
orbits. When noise prevents a phase slip, the trajectories may approach other
unlocked orbits quickly, thus the enhancement of PS becomes less pronounced

(Fig. 13.18). For ε well below εps , phase slips occur frequently and are not
always clearly associated to UPOs. Beyond εps , only (2) is active, and perfect
PS is interrupted by noise-induced phase slips. Increasing correlation R of
noise can slightly enhance PS further.
13.3.3 Noise-Enhanced PS in Arrays of Globally Coupled

Rössler Oscillators
Now let us consider an ensemble of N globally coupled nonidentical Rössler

oscillators
ẋi = −ωi yi − zi + ε(X − xi ),

ẏi = ωi xi + 0.15yi + Dξi , (13.33)
żi = 0.4 + zi (xi − 8.5),
N
where X = 1/N i=1 xi is the mean field. The parameters ωi are randomly
and uniformly distributed in [ω0 − δ, ω0 + δ] with ω0 = 1.0 and δ = 0.025. The
noise is again constructed via (13.32).
For a small ensemble, e.g., N = 5, a cascade of clustering of frequen-
cies is formed with an increase of the coupling strength until a global
PS regime is achieved, where all the oscillators are locked to a common
frequency (Fig. 13.20a); i.e., we have a hard transition as discussed in pre-
vious chapters. Close to the transition point of the global synchronization
regime, one can also observe a connection between phase slips and unlocked
UPOs (Fig. 13.20b,c). Although independent noise prevents the system from
1.06 20
5 (a)
1.05
Δφ1,5
40
1.04 4 (b)
60
Ωi
1.03
3
1.02 4 (c)
ΔX1
2
1.01 2
1
1.00 0
0.00 0.05 0.10 2000 3000 4000 5000
ε time
Fig. 13.20. (a) Transition to PS in the ensemble of N = 5 globally coupled

chaotic Rössler oscillators (13.33) without noise (D = 0). A cascade of clustering
of frequencies can be seen clearly. (b) Phase slips between oscillators 1 and 5 are
generated by period-1 UPOs of oscillator 1, as seen by ΔX1 in (c); ε = 0.08 is close
to the threshold of global PS where the oscillators 2–5 have formed a cluster
staying close to unlocked UPOs, it also induces phase slips among clustered
oscillators. As a whole, there is no appreciable enhancement of PS of the
ensemble by independent noise. In the following, we consider a global noise
R = 1, which is common to all the oscillators in the ensemble.
In the globally coupled ensemble, the transition to PS can be quantitatively
described by the mean order parameter r(t) , where

r(t) = | Pj (t)|/ |Pj (t)| (13.34)
j j
quantifies the degree of clustering of the state vectors Pj (t) = (xj , yj ) in the
phase space at time instant t. This definition of the order parameter based
on state points in the phase space has the same meaning as that defined in
(5.43) based on the phase of the oscillators. For unsynchronized oscillators,
the state vectors are scattered in the phase space and r is close to zero; while
r > 0 indicates a degree of clustering of the state vectors. Due to clustering,
a collective oscillation emerges in the ensemble of the oscillators. So the order
parameter r(t) is closely related to the oscillation amplitude of the mean field
X, which is quantified by the variance var(X) of X over time. The frequency
locking behavior can be quantified by σΩ which is defined as the standard
deviation of the mean frequency Ωj = φ˙j t of the oscillators in the ensemble.
In the absence of noise, all the oscillators are locked to a common frequency
and σΩ becomes zero at large enough coupling ε. However, a smaller σΩ may
not always indicate a stronger degree of synchronization, especially in the
frequency clustering region where σΩ may not be small but the degree of
synchronization is actually high.
As illustrated in Fig. 13.21, in the noise-free ensemble consisting of 1,000
elements, all the measures vary slowly until ε ≈ 0.043 where they start to
change quickly, and finally global PS is achieved at εps ≈ 0.093. In the
crossover region 0.043 < ε < 0.093, many oscillators are locked into clus-
ters with a common frequency within a cluster and different frequencies for
different clusters. A collective oscillation emerges when a significantly large
cluster is formed, and its amplitudes increase with ε when more oscillators are
included into the leading cluster. Unlike in the case of two coupled oscillators,
when independent noise (R = 0) is added to the ensemble, the PS regime is
degraded, as indicated by smaller var(X) and r , because noise degrades the
synchronization regime of those locked oscillators significantly. In the cluster-
ing region, σΩ becomes smaller than that of the noise-free ensemble, which,
however, does not mean a higher degree of PS. In contrast, a global noise
(R = 1) which is common to all the oscillators can enhance PS significantly
in the weak coupling regime: σΩ is considerably smaller, and var(X) and r
are clearly larger than those in the absence of noise.
For a fixed coupling strength ε < εps , there is an optimal noise inten-
sity D which enhances synchronization the most, and the mean field displays
the most coherent oscillation. For smaller D, the effective coupling due to
the global noise does not enhance synchronization, and for a too large noise
50 D=0
40 D=0.7, R=0
var(X)
30 D=0.7, R=1
20
10 (a)
0
0.02
σΩ
0.01
(b)
0.00
1.0
<r>
0.5
(c)
0.0
0.00 0.05 0.10 0.15
ε
Fig. 13.21. Transition to PS in an ensemble of N = 1, 000 globally coupled chaotic
Rössler oscillators with or without noise. (a) The variance of the mean field X.
(b) Standard deviation of the distribution of the mean frequency Ωj . (c) The order
parameter (13.34)
intensity, the oscillations become fairly incoherent and the synchronization

regime is degraded again. A typical behavior is shown in Fig. 13.22.
13.3.4 Experimental Observation of Noise-Enhanced PS
Noise-enhanced PS has been observed in experiments with coupled electro-

chemical oscillators. The oscillations result from electrodissolution of nickel
in sulfuric acid solution [480]. Depending on the system parameters, the
oscillation can be periodic or chaotic. The electrodes (oscillators) can be glob-
ally coupled to the potentiostat through individual parallel resistors, Rind ,
and through one collective series resistors, Rcoll (see Fig. 13.23). The coupling
strength is characterized by the fraction of collective resistance, ε = Rcoll /Rtot
where Rtot = Rcoll + Rind /N .
PS of two and an array of globally coupled oscillators without added noise
has been reported [480]. Here noise also leads to interesting phenomena [478,
481]. A Gaussian noise ξ with zero mean and a standard deviation of 1.0 is
applied to the potential of all electrodes, V = V0 + Dξ where V0 is an offset
potential, and D is the noise intensity. Thus in the experiments the applied
noise is common, i.e., R = 1. In the experiments, the time series of the current
i(t) of each oscillator is recorded for the analysis of synchronization.
Fig. 13.22. Different behavior of synchrony in an ensemble of N = 1, 000 globally

coupled (ε = 0.045) chaotic Rössler oscillators with a global noise of various inten-
sities. Upper panel : space–time plot of yi in gray scales (with white representing
maximal and black representing minimal values); middle panel : the corresponding
mean field X vs. time; lower panel : the corresponding distributions of the mean
frequency Ωj of the oscillators
Two Oscillators
We start with two coupled oscillators. The two electrodes have a small fre-
quency mismatch (Δω = 14 mHZ) due to surface heterogeneity. With an
added weak coupling, a PS regime occurs at about εps = 0.08; i.e., the fre-
quencies of both oscillators become equal.
Similar to two coupled Rössler oscillators in Sect. 13.3.2, just before the
threshold, one can observe clearly even in this experiment the correspondence
between phase slips and UPOs. For example, at ε = 0.06, the observed fre-
quency mismatch (ΔΩ = 5 mHz) is smaller than that seen without coupling,
however, the coupling is not strong enough for PS. During the time of the
experiment, one phase slip was observed (Fig. 13.24a). The analysis of the time
series of the two oscillators shows that the phase slip occurs when both oscilla-
tors approach the neighborhood of an unlocked period-3 UPO. In Fig. 13.24b,
Counter Electrode
Reference Electrode
V
Working Electrode
Potentiostat
Rind
Rcoll
Fig. 13.23. Schematic diagram of the experimental setup: globally coupled elec-
trochemical oscillators
3 (a)
0
Δφ
3
6
0.10 (b)
ΔX3/mA
0.05
0.00
0 50 100 150
t /s
Fig. 13.24. Two coupled (ε = 0.06) chaotic electrochemical oscillators just below
εP S = 0.08. (a) Phase difference between both oscillators vs. time. (b) The difference
between the next return values of the current maxima (ΔX3 = |Xn − Xn−3 |, where
Xn is the nth maximum) of the two oscillators (solid : oscillator 1, dashed : oscillator
2) as a function of time
the ΔX3 values are shown as a function of time for both oscillators. The coin-
cidence of the approach of the unlocked UPOs and the phase slip confirms the
numerical predictions about the dynamics close to but below εps .
10
15 (a) (d)
10 0
Δφ
5
0 −10
0.10 (b) 0.10 (e)
ΔX4/mA
0.05 0.05
0.00 0.00
0.10 (c) 0.10 (f)
ΔX4/mA
0.05 0.05
0.00 0.00
0 50 100 150 0 50 100 150
t /s t /s
Fig. 13.25. Two coupled (ε = 0.04) chaotic oscillators without (left panel, a–c)
and with (right panel, d–f ) small amounts of common, zero-mean Gaussian noise
(standard deviation of 3 × 10−4 V, measured at 200 Hz). Top row : phase difference
between the oscillators vs. time; middle and bottom row : the difference between the
next return values of the maxima of the oscillators, ΔX4 = |Xn − Xn−4 | (middle:
oscillator 1, bottom: oscillator 2), vs. time. Rtot = 500 Ω, V0 = 1.350 V
Noise-enhanced PS is investigated at a somewhat lower coupling strength,

ε = 0.04. During the time of the experiment (about 200 oscillations), there
are two phase slips between the oscillators (Fig. 13.25a) corresponding to a
frequency difference ΔΩ = 12 mHz. As can be seen in Fig. 13.25a–c, the first
phase slip can be attributed to the unlocked period-4 UPOs. It is seen that the
phase slips occur more frequently and develop more quickly than in Fig. 13.24
at the stronger coupling strengths. Moreover, the second phase slip cannot
be clearly linked to UPOs. These observations are also in agreement with the
numerical calculations in Sect. 13.3.3. Adding a small amount of zero-mean
Gaussian noise with the intensity D = 3 × 10−4 V (measured at 200 Hz) to
the (common) potential of the electrodes results in a qualitatively different
synchronization behavior. With this small noise, the deterministic nature of
the electrodissolution process is still dominant; the reconstructed attractors
resemble those without noise. However, the phase slips are eliminated and the
phase difference fluctuates around zero (Fig. 13.25d). The oscillators do not
have as long time of residence close to UPOs (Fig. 13.25e,f) as in the noise-free
case. The absence of phase slips during the 200 oscillations of the experiment
is consistent with the model calculations which predict lengths of the phase
synchronized epochs on the order of a thousand oscillations.
Array of 64 Oscillators
Now we experiment with a population of 64 chaotic oscillators [481] with a

very weak (ε = 0.014) global coupling, and explore the effect of noise on the
dynamics by changing the noise intensity D. In Fig. 13.26a, the time series of
one element in the array without noise is shown with its power spectrum. They
are similar to those of a single element. The space–time plot (Fig. 13.26c) of the
elements does not show any obvious sign of synchronous motion. The elements
in the array exhibit some inhomogeneity; there is a rather broad distribution
of frequencies (Fig. 13.26d) and phases (Fig. 13.26e). Hence, the variations of
the mean field time series, i.e., mean current, h(t) = 1/N k ik (t), are small
(Fig. 13.26b). Thus the coupling is not strong enough for PS.
Now we show the effect of added noise. If the noise intensity is small
D = 4 mV, there is only a slight change in the dynamics of the individual
elements, however, the collective behavior changes dramatically (Fig. 13.27).
Fig. 13.26. Dynamics of an array of 64 electrodes at ε = 0.014, without noise,

D = 0 mV. (a) Representative time series of a current of individual element (inset:
the corresponding power spectrum). (b) Mean field time series and its power spec-
trum (inset) (c) Space–time plot of individual currents. (d) Frequency distribution.
(e) Histogram of the phases at t = 50 s
Fig. 13.27. Dynamics of 64 electrodes as in Fig. 13.26, but with optimal noise
intensity, D = 4 mV
The space–time plot shows synchronization clusters and the frequency of 50

oscillators becomes equal, and many oscillators have a similar phase. Now the
mean field h(t) displays coherent oscillations and its power spectrum exhibits a
strong peak at the dominant frequency. Thus, this small noise intensity induces
a coherent behavior. With increasing the noise intensity further (D = 10 mV,
Fig. 13.28), the coherent motion of the mean field breaks down, the frequency
distribution becomes larger, and the phases become more scattered.
Such a noise-enhanced collective synchronization can be characterized by
various measures: the variance of the mean field var(h), the standard devi-
ation of the frequencies of the elements, σω /ωmean . One can also measure
the degree of PS based on the temporal phase locking between all pairs of
oscillators in the ensemble. The distribution of the cyclic relative phase dif-
ference (Δφ mod 2π) between an oscillator and a reference oscillator is con-
structed with a histogram of M bins. Peaks in the distribution exhibit that
there are preferred phase differences between the oscillators. The sharpness
of the distribution
M characterizes the degree of PS, and is quantified by the
entropy H = − k=1 p(k) ln p(k). In the ensemble, the entropy H(i) of the ith
Fig. 13.28. Dynamics of 64 electrodes as in Fig. 13.26, but with excessive noise,
D = 10 mV
oscillator is computed with respect to a randomly chosen reference oscillator.

By averaging over the ensemble and normalizing with the entropy of the uni-
form distribution Hm = ln(M ), one gets the PS index [474]
ρ = (Hm − H )/Hm . (13.35)
Note that ρ shows no sensitive dependence on the reference oscillator. The

PS degree is higher for larger ρ. It is important to emphasize that, in coupled
chaotic oscillators, the phase difference fluctuates around a constant value
even in the perfect phase locking region, yielding ρ < 1.
The coherence of collective oscillation of the ensemble can be efficiently
measured by the signal-to-noise ratio β of the mean field h(t), namely
Ωp
β = Ap , (13.36)
ΔΩ
where Ωp is the frequency of the main peak in the spectrum of h(t), Ap is
the peak height mainly depending on the amplitude of h(t), and ΔΩ is the
half width of the peak reflecting temporal randomness of h(t). Defined in this
way, β measures the sharpness of the main peak in the spectra. A large value
of β indicates that the oscillations of h(t) are rather periodic, thus are highly
coherent.
The results of these measures are shown as a function of the noise inten-
sity in Fig. 13.29 for ε = 0 and ε = 0.014. The mean field increases in a
monotone way for ε = 0, showing that noise can induce a degree of PS even
for uncoupled oscillators, as is consistent with the prediction in Sect. 13.2.2.
With the weak coupling ε = 0.014, the mean field shows a maximum, indicat-
ing a resonant-like behavior, which can also be observed in ρ and β – there is
a small variation for ε = 0 and a resonance curve for ε = 0.014. However, the
frequency distribution (Fig. 13.29c) does not clearly exhibit the trend above:
although for ε = 0.014 there is a minimum at the optimal noise intensity, for
ε = 0 there are larger variations.
The experiments on ensembles of chaotic oscillators confirm the noise-
induced effects predicted by numerical calculations. At an optimal common
a b
−2
0.14
−2.5 0.12
log10 var(h)
0.1
−3
ρ
0.08
0.06
−3.5
0.04
−4 0.02
0 2 4 6 8 10 0 2 4 6 8 10
x 10−3 c d
7000
12 ε=0
6000 ε=0.014
10 5000
σω /ωmean
4000
β
8
3000
6 2000
1000
4
0 2 4 6 8 10 0 2 4 6 8 10
Noise intensity /mV
Fig. 13.29. Different measures of the collective behavior as a function of noise
intensity for ε = 0 (squares) and ε = 0.014 (circles). (a) Variance of mean field
var(h). (b) Temporal PS index ρ. (c) Relative standard deviation of frequencies
σω /ωmean . (d) Coherence factor β
13.4 Noise-Enhanced Synchronization-Like Phenomena 305
noise intensity, increased measure of PS and coherent oscillations are observed;

the effects are stronger with weak inherent coupling.
13.4 Noise-Enhanced Synchronization-Like Phenomena

in Arrays of Coupled Excitable Cells
In most of the previous chapters, we have discussed synchronization in arrays

of coupled self-sustained oscillators. However, many biological and chemical
systems are rather characterized by excitability, as discussed in Sect. 2.7. There
is a threshold below which an external signal cannot generate a significant
response of the system, and the system remains in a quiescent state. A signal
that exceeds the threshold, on the other hand, can evoke a strong, mostly
spike-like response with a refractory period Tr before the system returns
to its quiescent state. Some synchronization-like phenomena (synchrony) in
excitable systems due to a strong enough external forcing have been discussed
in Chap. 12.
In recent years, a great deal of attention has been paid to study the
response of excitable systems in the presence of noise (cf. the review [482]). In
isolated excitable cells, noise alone can generate a sequence of spikes, separated
by fluctuating interspike intervals T . Importantly, there is an optimal inten-
sity of noise where the fluctuation of T is reduced and the spiking sequences
become quite regular (periodic) with T ∼ Tr , a phenomenon known as coher-
ence resonance [483].
Here we will discuss the synchrony behavior when the noise-sustained oscil-
lations of the excitable elements are coupled, and study how arrays of coupled
excitable cells respond to an external forcing.
13.4.1 Phase Synchrony in Chains of Coupled Noisy

Excitable Neurons
We consider a simple but representative model of excitable systems and nerve

pulses [483], the FitzHugh–Nagumo (FHN) neuron model (cf. Sect. 2.7.2).
We analyze the following chain of N diffusively coupled nonidentical FHN
neurons
x3i
ẋi = xi − − yi + d(xi+1 + xi−1 − 2xi ),
3
y˙i = xi − ai + Dξi (t), (13.37)
where ai is a parameter of the ith element, and d is the excitatory coupling

strength. For a single FHN model, if |a| > 1, the system has only one stable
steady state, while |a| < 1 generates a limit-cycle. The system in a steady
state (|a| slightly larger than one) is excitable because it will return to the
steady state only after a large excursion (“near limit-cycle”) when perturbed
away from the steady state. To make the study more general, we suppose
that ai is not identical for the elements, but is uniformly distributed ai ∈
(−1.1, −1). This implementation of the model is important because many
real systems are diffusively coupled, and nonidentity is also more natural in
physics, engineering, or biology. With nonidentical ai , the uncoupled elements
will have different average firing frequencies in the presence of the same level
of noise ξi (t). We assume that the Gaussian noise is uncorrelated in different
elements, i.e., ξi (t)ξj (t ) = δij δ(t − t ). Periodic boundary conditions are
considered. The parameter of the time scale is fixed at = 0.01.
To characterize synchronous behavior in the chain of these nonidentical
excitable systems, we introduce the phase of the elements as in (13.24), where
τk is the time of the kth firing of the element defined by threshold crossing of
xj (t) at x = 1.0. The quantity

φi − φi+1
ρi = sin2 (13.38)
2
measures the PS effect of neighboring elements, as introduced in Chap. 6.

To measure the temporal coherence of the noise-induced motion, we exam-
ine the distribution of the pulse duration Tk = τk+1 − τk . For a single element
subject to noise, this distribution P (T ) has a peak at a certain value of Tk
and an exponential tail at large values [483]. A measure of the sharpness of
this distribution, P (T ), for example,

S = Tk / Var(Tk ), (13.39)
which can be viewed as a signal-to-noise ratio, provides an indication for

the coherence of the firing events. Biologically, this quantity is of importance
because it is related to the timing precision of the information processing in
neural systems. For a single element, it has been shown that S possesses an
optimal value at a certain level of noise [483]. In the coupled ensemble (13.37),
the same measure of coherence (13.39) is computed from the overall distrib-
ution P (T ) constructed by the pulse duration of all the N elements during a
long enough period of time.
Typical behaviors observed in numerical simulations of the system
are [484]:
(1) Without external noise, each element comes to a steady state and no firing
takes place. The steady states are slightly different due to the nonidentity
of the parameters ai .
(2) When strong enough noise is added to the originally quiescent system, the
system becomes excited. The firing of the elements may be in PS if the
coupling is strong enough.
(3) The coherence of the temporal and spatial pattern of the firing process is
strongly enhanced in a certain region of noise and coupling strength.
20
10
S
−0.5
0 −1
0.5
−0.5 −1.5
−1.5
−2.5 −2 lg(D)
lg(d)
Fig. 13.30. Signal-to-noise ratio S in the parameter space (log10 (d), log10 (D)) of
a coupled chain of nonidentical FHN neurons (13.37) with N = 100. The black dots
correspond to the patterns in Figs. 13.31–13.34
The main features of the system are illustrated in Fig. 13.30, where the
results of S in the parameter space (log10 (d), log10 (D)) are shown for a chain
of N = 100 elements. For a fixed coupling strength d, S increases first with
the noise level D, reaches a maximum, then decreases again; i.e., it shows a
typical resonance-like behavior. In general, for a stronger coupling, a higher
level of noise is needed to excite the system. Similarly, for a fixed noise level
D, S increases with increasing d until it reaches an optimal value; after that,
it decreases again.
From Fig. 13.30, one can observe several dynamical regimes in the system.
For very weak coupling (d < 10−2 ), the firing of the elements are essentially
independent, because a noise-induced firing of an element cannot excite its
neighboring lattices. A typical spatiotemporal pattern of xi and ρi for weakly
coupled elements subject to relatively weak noise is shown in Fig. 13.31. Both
the spatial and temporal behaviors are quite irregular. Due to the independent
firing, the frequencies and phases are not synchronized, and the distribution
P (T ) is broad.
With an increase of the coupling strength, the system becomes sensitive to
weak noise because the firing events induced by noise now become the source of
excitation of the neighboring elements and leads to partial synchrony, as seen
in Fig. 13.32. This mutual excitation enhances the coherence of the motion
in the coupled system, as indicated by a narrower distribution P (T ) and a
larger S. The lattice displays clusters of synchrony. The clusters break and
reunite during the evolution, so that each element has a slightly different firing
frequency.
The next regime where S takes the largest values (S ∼ 18) is the most
interesting one, because the system performs a quite regular motion globally,
Fig. 13.31. Synchrony in a chain of coupled noisy excitable FHN neurons (13.37)
with d = 0.005 and D = 0.02. The upper panels show the spatiotemporal structure
of xi and ρi . The time step is 0.2. The lower panels show the average firing frequency
in the chain and the distribution of the pulse duration
as seen in Fig. 13.33. All elements are locked to a relatively large firing fre-
quency, and the distribution of the pulse duration becomes very sharp. After
that, with a stronger coupling, the system keeps the global synchrony regime,
however, the temporal behavior becomes irregular again (Fig. 13.34), corre-
sponding to a decreased S.
The locations of the four representative dynamical regimes are indicated
by the black dots in Fig. 13.30. These regimes are typical for different sizes of
the chain. For a larger chain, the regime of global regular motion (S large) is
wider in the parameter space [484].
Further analysis [485] has shown that high coherence of the phase synchro-
nized oscillations in chains of coupled FHN neurons is supported by spatially
independent noise components, and a correlated or global noise reduces the
temporal coherence of the spiking. In a certain regime of noise intensity and
coupling strength, the disorder in the excitability (inhomogeneous parameter
ai ) can further enhance the coherence.
Fig. 13.32. As in Fig. 13.31, but for d = 0.03 and D = 0.025
Noise-enhanced PS of excitable media can also be characterized by the

phase diffusion constant [486], defined in Sect. 2.3, i.e.,
1 d
Dφ = (Δφ − Δφ )2 , (13.40)
2 dt
where Δφ is the phase difference between an element and the reference ele-
ment, and · denotes average over the elements. In the regime of optimal
noise, Dφ becomes minimal, indicating the best synchrony behavior.
A geometric approach is introduced in [487] for understanding the phe-
nomenon of PS in coupled nonlinear systems in the presence of additive noise.
13.4.2 Noise-Enhanced PS of Coupled Excitable Neurons

by External Forcing
So far, we have shown that noise-sustained oscillations in nonidentical

excitable cells can be synchronized by a suitable coupling strength. The
synchronized oscillation of the chain becomes much more coherent compared
to isolated cells.
Fig. 13.33. As in Fig. 13.31, but for d = 0.25 and D = 0.07. The spatiotemporal
plot of ρ is black, indicating no phase slips
It is interesting to study how the noise-sustained coherent oscillations in

the coupled arrays respond to external period forcing. Consider a uniform
periodic signal forcing the coupled chain [488],
x3i
ẋi = xi − − yi + A cos Ωt + d(xi+1 + xi−1 − 2xi ),
3
y˙i = xi − a + Dξi . (13.41)
Here the parameters are in the excitable regime, = 0.01 and a = −1.05,
with the coupling strength d = 0.05. To demonstrate the significant role of
the coupling, the synchronous behavior of the coupled chain (N = 30) to the
external forcing is compared with a single uncoupled neuron (N = 1).
Now a weak noise D = 10−1.75 is added. Both systems of a single (N = 1)
uncoupled neuron and an array (N = 30) of coupled FHN neurons show
a broad distribution of the spiking intervals T (Fig. 13.35a), although the
coupling has reduced slightly the probability of very long intervals. The two
systems, however, have quite different responses to the same subthreshold
signal with a period Te close to the peak of P (T ) (Fig. 13.35, vertical dashed
lines). An uncoupled neuron may fail to fire a spike at some periods of the
Fig. 13.34. As in Fig. 13.31, but for d = 1.0 and D = 0.04. The spatiotemporal
plot of ρ is black, indicating no phase slips
−1 0
10 (a) N=30 (c)
(b)
−2
10 −100
N=1
P(T)
Δφ
−3 N=1
10 N=1
−4
−200
10 N=30 N=30
10
−5 −300
0 10 20 30 40 0 5 10 15 20 25 0 1000 2000
Fig. 13.35. Different responses of a single (N = 1) and an array (N = 30) of

coupled FHN neurons to a subthreshold periodic signal (13.41) with weak noise
D = 10−1.75 . Distribution of interspike interval at A = 0 (a) and A = 0.03 (b).
Phase difference between the spike train (randomly selected neuron in the array)
and the signal (c). The signal period Te is indicated by the vertical dashed lines in
(a) and (b)
signal, resulting in intervals T ≈ nTe , n = 2, 3, . . .; hence, a few peaks at nTe

show up in P (T ) on the exponentially decaying background (Fig. 13.35b), and
the phases are not locked due to this occasional skipping of spikes (Fig. 13.35c).
In sharp contrast to this, in the chain, P (T ) becomes sharply peaked around
the signal period Te and the phase is locked to the signal.
D=10−1.75 D=10−1.5 D=10−1.3

0.08
0.06
A
0.04
0.02
(a) (b) (c)
0.00
0.08
0.06
A
0.04
0.02
(d) (e) (f)
0.00
0.8 1.2 1.6 2.0 0.8 1.2 1.6 2.0 0.8 1.2 1.6 2.0
Ω Ω Ω
Fig. 13.36. Comparison of the locking behavior of N = 1 (upper panel ) and
N = 30 (lower panel ) neurons (13.41) at various noise intensities D = 10−1.75
(a, d), D = 10−1.5 (b, e), and D = 10−1.3 (c, f ). Filled dots: effective locking region
(|Δω| < 0.002) of the noisy systems; dashed line: the threshold beyond which the
noise-free systems generate sustained spike trains. Above the solid lines is the 1 : 1
superthreshold locking region of the noise-free systems
Figure 13.36 exhibits systematically the locking behavior in the parameter

space (Ω, A) by computing the difference Δω between the spiking frequency
and the signal frequency. In a noise-free excitable system, a sustained and
synchronized response only occurs when the signal exceeds a threshold. We
find that uncoupled neurons and an array of coupled neurons display almost
the same Ω-dependent firing threshold (Fig. 13.36, dashed lines) and the same
1 : 1 superthreshold response region (Fig. 13.36, above the solid lines). A small
noise D = 10−1.75 can induce an occasional skipping of spikes in an uncoupled
neuron, thus the 1 : 1 superthreshold response is no longer perfect. An effective
locking region (|Δω| ≤ 0.002) is found only at a quite large superthreshold
amplitude A (Fig. 13.36a), including a small subthreshold region for small
Ω. At D = 10−1.5 , such an effective locking region shrinks considerably and
it only appears in the superthreshold region (Fig. 13.36b), but it disappears
effectively at an even stronger noise level D = 10−1.3 (Fig. 13.36c), although
the noise-induced spontaneous spike trains are more coherent here. An effec-
tive stochastic phase locking (Δω has a relatively flat region vs. Ω, but with
|Δω| 0.002) can be observed for A below but rather close to the threshold.
A coupled chain, in contrast, displays a much more prominent locking
behavior. At D = 10−1.75 , the superthreshold locking region of the noise-free
0.08
1:1 2:3
1:2
0.06
0.04
A
0.02
2:1 1:1 2:3
1:2
0.00
0.5 1.5 2.5 3.5 4.5
Ω
Fig. 13.37. m : n Arnold tongues for a chain ((13.41), N = 30) with D = 10−1.5
(dots). The dashed line shows the spiking threshold and the solid lines are the border
lines of the m : n superthreshold locking region at D = 0
system remains intact, while a large subthreshold locking region at Ω > ω0 (D)
emerges (Fig. 13.36), where ω0 (D) is the mean firing frequency induced by
noise of intensity D in the chain without external forcing. At D = 10−1.5 , lock-
ing can be achieved with almost vanishing A when Ω ≈ ω0 (D) (Fig. 13.36e).
At D = 10−1.3 , the locking region shrinks a bit (Fig. 13.36f), and it shrinks
further for even larger D.
Higher order m : n locking regimes, where each cell generates m spikes
during every n periods of the signal, are also observed (Fig. 13.37). It is seen
again that an m : n locking can be achieved with almost vanishing A when
mΩ ≈ nω0 (D). The locking regions are no longer confined by the border lines
of the superthreshold locking regions of the noise-free system; in contrast, they
n
become centered around m ω0 (D) which moves with D. We emphasize that an
m : 1 (m > 1) superthreshold locking region does not exist in the noise-free
system, while in the noisy array, a 2 : 1 region is observable.
These results show that the interplay between coupling and noise can have
a significant role in enhancing the resonant response of excitable systems due
to external forcing. Resonances and locking occur due to a matching between
the noise-controlled time scale and that of the signal. Synchrony of the noise-
sustained oscillations in the coupled chain due to external forcing is very
similar to that of self-sustained oscillators in Chap. 3. Such array-enhanced
resonances may be important in neural systems, since coupling and noise
together can establish a much higher sensitivity to both the frequency and
the amplitude of weak external signals.
13.4.3 Resonant Pattern Formation in 2D Arrays
Such noise-sustained synchrony can also generate interesting pattern forma-

tions. To illustrate this, we consider a 2D array of excitable FHN neurons
( = 0.01, a = 1.05) coupled to its four nearest neighbors with a small coupling
strength d = 0.0075. The external periodic signal has a frequency Ω = 3.2.
With this frequency, the threshold for the signal to excite the noise-free FHN
Fig. 13.38. Patterns at t1 (upper ) and t1 + Te (bottom) in the presence of a

subthreshold periodic signal (A = 0.012, Ω = 3.2) at various noise intensities: D =
0.6 × 10−2 (a), D = 1.5 × 10−2 (b), D = 2.5 × 10−2 (c), and D = 4.0 × 10−2 (d). The
moment t1 corresponds to a local maximum of the mean concentration x after a
transient, and Te = 2π/Ω is the signal period. The patterns in (b) and (c) repeat
for every two periods of the signal corresponding to a 2:1 locking
neurons is Ath = 0.02. Here we consider a weak, subthreshold signal amplitude

A = 0.012.
When the media is subjected to noise, various pattern formations can be
observed. Especially, there is a range of noise intensities D, where the spiking
frequency of the neurons is locked to the signal. Interesting resonant patterns
occur in this regime. At a rather weak intensity, e.g., D = 0.6 × 10−2 , noise
alone can hardly initiate target waves. Together with the weak external signal,
noise now excites simultaneous many target waves which, however, have not
been locked by the signal, as seen in Fig. 13.38a. At a larger noise intensity
D = 1.5 × 10−2 , the mean firing frequency ω of all the neurons is locked to
the signal with a ratio 2:1, ω = Ω/2. As seen in Fig. 13.38b, the medium
reorganizes its patterns into two almost uniform domains, each locked to the
signal but with phases differing by π. Such locked patterns having almost
stationary boundaries between the two domains is very similar to standing-
wave patterns observed in self-sustained oscillatory media subject to resonant
forcing [489, 490] or global feedback [491, 492]. At the larger intensity D =
2.5×10−2 , the media remain 2:1 locked with the signal, but the spatial patterns
become almost uniform (Fig. 13.38c). At even larger intensities, e.g., D =
4×10−2 , the locking is lost and the patterns become dense, randomly flushing
clusters (Fig. 13.38d). A more detailed analysis of the noise-sustained resonant
pattern formation can be found in [493] for diffusive coupling of excitable FHN
neurons or in [494] when the media are subjected to stirring of a chaotic flow.
It is important to emphasize that these resonant wave patterns are sustained
by noise. The waves die out and the media relax to the homogeneous steady
states when the noise is ceased in the simulations.
13.5 Conclusions
In this chapter, constructive effects of noise on synchronization and
synchronization-like phenomena of nonlinear systems have been discussed.
– In nonlinear systems with significant contraction regions, such as the

Lorenz system, a common noisy forcing with a suitable intensity can syn-
chronize two uncoupled, identical oscillators, as has been also illustrated
in chaotic neuron models and experimentally observed in a homoclinic
chaotic laser.
– A common noise can also induce a rather high degree of PS in nonidentical
oscillators.
– In weakly coupled oscillators, where the coupling is not strong enough to
generate a high degree of PS, a (correlated or common) noise can enhance
synchronization significantly. The enhancement is found to be maximal
at some optimal noise levels. Noise-enhanced PS has been confirmed in
experiments, e.g., in ensembles of coupled electrochemical oscillators.
– In excitable systems, the noise-sustained spiking can be synchronized by a
suitable coupling. The coherence of the noise-sustained oscillations can be
significantly enhanced due to the coupling. As a result, in a coupled array
of noisy excitable neurons, synchrony to an external period signal can be
achieved with a much weaker signal amplitude. Such a noise-enhanced syn-
chronization can create interesting resonant pattern formations in spatially
extended media.
14
Networks with Complex Topology
14.1 Introduction
In the previous chapters we have considered synchronization of oscillators
which have regular arrangement (1D arrays or 2D lattices) with the coupling
extended to the nearest neighbors (local coupling), see Fig. 1.2, or global cou-
pling among all the osicllators (e.g., Sects. 5.5, 10.9, 13.3.3 and 13.3.4). Such
simple coupling topology of the oscillators is relevant to many experimental
and natural situations.
The locally or globally coupled networks can be considered as regular net-
works. However, there are many real-world systems which are neither locally
nor globally coupled, but often display a much more complicated coupling
topology. Examples include Internet and world-wide-web in communication
systems, neural system or genetic regulation in biology, epidemic spreading
and synchronization in social and ecological systems, etc. [495].
These systems can be characterized by complex networks. We distinguish
three main classes of complex networks [495]:
1. Random networks. About 40 years ago Erdös and Rényi (ER) [496] studied
random networks (they called them random graphs) where a pair of nodes
i and j are connected with a probability p (Fig. 14.1a). In such ER random
networks, the connection is fully random, which has the advantage that
the distance of the shortest path (path length) between any pair of nodes
is very small even for very large networks.
2. Small-world networks (SWNs). While many realistic network systems are
not regular as lattices with only local connections, they are also not fully
random as the ER networks. Especially, many of them display high clus-
tering, i.e., two connected nodes are also connected to a common third
node (thus forming a triangle), which form some local communities. In
their seminal work, Watts and Strogatz [497] proposed the small-world
network (SWN) model. They started with a regular ring of nodes, each
connected to its k nearest neighbors; then with a probability p, each link
318 14 Networks with Complex Topology
(a) (b) (c)
Fig. 14.1. The three basic classes of complex networks. (a) ER random networks,
(b) small-world networks (SWNs), (c) scale-free networks (SFNs)
is rewired, namely, it is cut and reconnected to another node randomly.

When p is small, the resulting network is dominated by the local coupling,
while there are a very few long-range random shortcuts (Fig. 14.1b). Strik-
ingly, the average distance of the shortest path between any pair of the
nodes has been significantly reduced due to this small number of short-
cuts in this small-world regime. Many empirical examples of SWNs have
been documented in various fields ranging from cell biology to scientific
collaboration [499–502]. When p is approaching 1.0, any pair of nodes are
connected in a fully random manner and we move into the regime of ER
random networks. Due to the small-world property, dynamical systems
coupled in this way would display enhanced signal propagation, synchro-
nizability, and computational power, as compared with the original regular
chain of the same size.
3. Scale-free networks (SFNs). In both the SWNs and ER random networks,
the nodes have a characteristic number of connections kj (degree of
node j). The degree kj fluctuates around the mean value K = kj and
follows a Poisson distribution in large SWNs and ER random networks.
The analysis [498] of many large-scale real-world complex systems, how-
ever, reveals that the degrees of nodes are very heterogeneous. In such
networks, for example, in Internet and the worldwide airport network,
most of nodes have small degrees, while a few nodes have a large number
of connections; these nodes are called hubs. The probability P (k) to find
a node with degree k in the network follows a power-law distribution
P (k) ∼ k −γ , (14.1)
so that there does not exist a characteristic degree as in the SWNs or

ER random networks. These networks are thus called scale-free networks
(SFNs). SFNs are characterized by a statistical abundance of hubs with
a very large number of connections k compared with the average degree
value K = k (Fig. 14.1c).
These seminal findings of complex networks have stimulated a great deal of
research interest in structure analysis of real-world complex systems from the
14.1 Introduction 319
viewpoint of network topology. It has been shown that both the small-world
and the scale-free properties are universal in many real-world complex systems
(cf. [495, 503–506] and references therein).
A very important issue in the study of complex systems is the interplay
between structure and dynamics. The topology of the networks can have a
systematic influence on their physical and dynamical properties, such as error
and attack tolerance [507], percolation transition [508, 509], epidemic spread-
ing [510–512], cascading failures [513], etc. For a recent review on dynamical
process of complex networks, refer to [514].
A special aspect of this interplay is synchronization. Synchronization of
oscillators acting on the nodes is one of the widely studied dynamical behav-
ior on complex networks. A basic question is: for which network structure
(topology) we can obtain an optimal synchronization? It has been shown that
the SWNs provide a better synchrony of coupled excitable neurons in the
presence of external stimuli [515], compared with regular arrays with local con-
nections and ER random networks. In pulse-coupled oscillators, synchroniza-
tion becomes optimal in the small-world regime [516], and it is degraded when
the degree becomes more heterogeneous with increased randomness [517,518].
Investigation of phase oscillators [519] or circle maps [520] on SWNs has shown
that when more and more shortcuts are created at larger rewiring probability
p, the transition to the synchronization regime becomes easier [519]. These
observations have revealed that the ability of a network to synchronize is
generally enhanced in SWNs compared with regular chains. This enhanced
synchronization in SWNs has also been analyzed in the context of com-
plete synchronization of identical chaotic systems [521–525]. Physically, this
enhanced synchronizability was attributed to the decreasing of the average
network distance due to the shortcuts.
More recently, it has been shown that the ability of complete synchroniza-
tion of identical chaotic oscillators also depends critically on the heterogeneity
of the degree distribution [526]. In particular, random networks with strong
heterogeneity in the degree distribution, such as SFNs, are more difficult to
synchronize than random homogeneous networks [526], despite the fact that
heterogeneity reduces the average distance between the nodes [527, 528]. The
synchronizability in [521–526] is based on the linear stability of the complete
synchronization state using the spectral analysis of the network coupling matrix.
In this chapter, we discuss synchronization of nonlinear oscillators coupled
in complex networks. Our emphasis is to demonstrate how the network topol-
ogy and the connection weights influence the synchronization behavior of the
oscillators. In Sect. 14.2, we present the general dynamical equations and the
linear stability analysis for CS state when the oscillators are identical. Then
we study phase synchronization (PS) of nonidentical oscillators in SWNs
(Sect. 14.3) and SFNs (Sect. 14.4). Section 14.5 presents a qualitative analy-
sis of the hierarchical synchronization observed in Sect. 14.4. Section 14.6
is devoted to the effects of weighted coupling on synchronizability of the
networks.
14.2 Dynamical Equations and Stability Analysis

The dynamics of a general network of N coupled oscillators is described by

N
ẋj = τj F(xj ) + d Aji Wji [H(xi ) − H(xj )] (14.2)
i=1

N
ẋj = τj F(xj ) − d Gji H(xi ), j = 1, . . . , N, (14.3)
i=1
where xj is the state of oscillator j and F = F(x) governs the dynamics

of each individual oscillator. The parameter τj controls the timescale of the
oscillators. H = H(x) is the output function and d is the overall coupling
strength. A = (Aji ) is the adjacency matrix of the underlying network of
couplings, where Aji = 1 if there is a link from node i to node j and 0
otherwise. Here, we assume that the coupling is bidirectional so that Aij =
Aji , i.e., A is symmetric. Wji is the weight of the incoming strength for the
link from node i to node j. Note that the incoming and output weights can be
in general asymmetric, Wji = Wij . G = (Gji ), defined as Gji = −Aji Wji +
N
δji i=1 Aji Wji , is the coupling matrix which includes both the information
of network topology and connection weights. By definition, the rows of matrix
G have zero sum.
For identical oscillators, i.e., τ1 = τ2 = · · · = τN = 1, the completely
synchronized state {xj (t) = s(t), ∀j | ṡ = F(s)} is a solution of (14.3). The
linear stability of this solution can be assessed in the framework of the master
stability function [523, 529]. Small perturbations of the synchronization state
s are governed by the linear variational equations

N
δ ẋj = JF(s)δxj − dJH(s) Gji δxi , j = 1, · · · , N, (14.4)
i=1
where JF(s) and JH(s) are the Jacobians on s. Equation (14.4) can be diag-
onalized into N decoupled blocks of the form
η̇j = [JF(s) − dΛj JH(s)] ηj , j = 1, · · · , N, (14.5)
where ηj is the eigenmode associated to the eigenvalue Λj of the coupling
matrix G. Here Λ1 = 0 corresponds to the eigenmode parallel to the synchro-
nization manifold, and the other N −1 eigenvalues Λj represent the eigenmodes
transverse to the synchronization manifold. We assume that the networks are
connected (without unconnected subnetworks), so that only one eigenvalue is
zero. The largest Lyapunov exponent (LE) λ(dΛj ) of (14.5) determines the
linear stability of the corresponding eigenmode ηj . The mode is damped if
λ(dΛj ) < 0. The synchronized state s is stable if λ(dΛj ) < 0 for j = 2, · · · , N ,
i.e., all the transverse modes are damped. Note that the same analysis can be
carried out for regular networks, as already discussed in Sect. 7.1. The above
stability analysis will be used to study CS of weighted networks in Sect. 14.6.
14.3 Phase Synchronization in Small-World Networks of Oscillators 321
14.3 Phase Synchronization in Small-World Networks

of Oscillators
In general, the oscillators are not identical, and the stability analysis from
Sect. 14.2 is not applicable. In this section and in Sect. 14.4, we study the PS
of nonidentical oscillators (τj randomly distributed) first in SWNs and then
in SFNs with numerical simulations.
First, we have to generate such complex network structures. The rewiring
procedure for SWNs originally proposed in [497] may result in networks with
some isolated nodes if the degree k of the nodes in the initial regular ring is
small. To avoid this problem, we present the following modified SWN model
proposed in [530]. We start with a regular ring of N nodes, each connected to
its nearest neighbors, i.e., k = 2. Shortcuts are then added between randomly
selected pairs of nodes, with probability p per link of the basic regular ring,
such that typically there are pN shortcuts in the resulting networks. This way,
the total number of connections also increases with p.
The Rössler chaotic oscillators are used as active node elements in the
simulations of the SWNs. The dynamics of the network then reads:
d
N
ẋj = τj F(xj ) + Aji (xi − xj ), j = 1, . . . , N, (14.6)
kj i=1
where x = (x, y, z) represents the Rössler chaotic oscillator: (ẋ = −0.97y − z,

ẏ = 0.97x + 0.15y, ż = x(z − 8.5) + 0.4). We regard here the simple output
function H(x) = x, so that the coupling is added to all variables x, y, and z.
Note that the coupling strength d is normalized by the degree kj of each node,
so that we can scale out effects of the increasing average degree K = kj when
more and more shortcuts are added in the network at larger probability p.
We assume that the time scale parameter τj of the oscillators are uniformly
distributed in the interval [1−Δτ, 1+Δτ ] as in Chaps. 4, 8 and we fix Δτ = 0.4
in the simulations.
We now discuss the synchronization behavior of (14.6) for networks with
different shortcut probabilities p. The degree of synchronization is quantified
N
by the variance Var(X) of the mean field oscillation X = (1/N ) j=1 xj as a
function of the coupling strength d (Fig. 14.2a). We also examine the change of
oscillation amplitudes in individual oscillators with respect to d by measuring
the average value of the variance of xj (t) over all j = 1, · · · , N , Var(xj )
(Fig. 14.2b). We then observe the following types of synchronous behavior:
1. When the shortcut probability p is rather small (p = 0.01), the network
is still dominated by local coupling and it does not display evident syn-
chronization effects over a broad range of d, as indicated by an almost
vanishing mean field X. However, the oscillation amplitude of individual
oscillators changes with d.
50 50
(a) (b)
40 40
<Var(xj)>
30 p=0.01 30
Var(X)
p=0.1
p=0.3
20 20
p=0.5
10 10
0 0
0.0 1.0 2.0 3.0 4.0 5.0 0.0 1.0 2.0 3.0 4.0 5.0
d d
Fig. 14.2. Transition to synchronization regime and oscillation death of various

SWNs of chaotic Rössler oscillators (14.6). (a) The variance Var(X) of the mean
field X as a function of the coupling strength d. (b) The average value of the variance
of all individual oscillators. The network size is N = 1, 024
2. With a larger number of shortcuts at p = 0.1, the network starts to

synchronize and generates a coherent collective oscillation at a strong
coupling strength.
3. At even larger values of p, e.g., p = 0.3 and p = 0.5, the networks dis-
play three regimes: (a) When the coupling strength is increased from very
small values, the trajectory of each oscillator approaches closer and closer
to the unstable steady state x̄ (F(x̄) = 0), as seen by a rapid decrease
of the variance of individual oscillators (Fig. 14.2b). The oscillation fre-
quencies Ωj , which are defined as in Chap. 6, are still distributed in this
regime (Fig. 14.3a, b). (b) When a critical value d1 is reached, all oscilla-
tors become stable at the same steady state x̄, so that Var(X) = 0 and
Var(xj ) = 0, and we observe oscillation death (OD) in SWNs. (c) When
d is further increased to exceed another critical value d2 , the steady state
x̄ becomes unstable again, and the oscillation, first periodic, is restored.
Importantly, the whole network is now in a global synchronization regime,
the frequencies and phases of all oscillators are locked (Fig. 14.3c). Com-
paring the two critical values d1 and d2 of the coherent regimes for p = 0.3
and p = 0.5, one can see that networks with more shortcuts achieve coher-
ent regimes for smaller coupling strengths.
The transition from a nonsynchronous regime to OD and then to a coher-
ent collective oscillation is associated with a series of bifurcations in the sys-
tem (Fig. 14.4). Both the mean field and the individual oscillators display
similar period-doubling bifurcations, as indicated by plotting the local max-
ima of X(t) and x1 (t) for each value of coupling strength d. The bifurcation
diagram is almost the same for any other individual oscillators. In the non-
synchronous regime, x1 (t) and X(t) show chaotic oscillations with large and
small amplitudes, respectively. In a broad range of coupling strength, SWNs
of strongly disordered chaotic oscillators create a simple steady state (OD)
14.3 Phase Synchronization in Small-World Networks of Oscillators 323
1.5
(a)
Ωj
1.0
0.5
1.5
(b)
Ωj
1.0
0.5
1.5
(c)
Ωj
1.0
0.5
0 512 1024
Node index j
Fig. 14.3. Oscillation frequency Ωj of the oscillators in SWNs (14.6) with the
shortcut probability p = 0.5 for different coupling strength. (a) d = 0, (b) d = 0.1,
and (c) d = 1.5
15 15
(a) (b)
10 10
max(x1)
max(X)
5 Global synchronization 5 Global synchronization
OD OD
0 0
0.0 2.0 4.0 6.0 8.0 0.0 2.0 4.0 6.0 8.0
d d
Fig. 14.4. Bifurcation diagram of the mean field (a) and of an individual oscillators
(b) against the coupling strength d in SWNs (14.6) with the shortcut probability
p = 0.5
or periodic dynamics (global synchronization regime). Only at very large

coupling strength the globally synchronized oscillations become chaotic again.
This result offers an efficient tool of controlling chaos in a network of locally
coupled chaotic oscillators by introducing long-range connections.
For a fixed value of the coupling strength d, the two regimes of OD and
global synchronization can also be obtained by adding a sufficient number of
shortcuts (Fig. 14.5). The system behavior is not sensitive to increasing p when
p < 0.02. With a further increase of p, the amplitudes of the oscillators are
reduced and finally the regime of OD is reached, which is stable for networks
in a certain range of p, and afterwards a coherent collective oscillation is
50
40 Var(X)
<Var(xj)>
30
20
10
0
−3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0
log10p
Fig. 14.5. Synchronization behavior vs. the shortcut probability p in SWNs (14.6)
for a fixed coupling strength (here d = 1.5)
observed due to global synchronization, which becomes more pronounced as

more shortcuts are in network.
We have shown that the coupling topology in the SWNs has significant
effects on the synchronization of strongly nonidentical nonlinear oscillators.
Compared to regular networks with local coupling (p ≈ 0), SWNs with many
shortcuts display enhanced synchronization as expressed by the regimes of
OD and global synchronization similar to globally coupled networks. OD has
also been demonstrated in Chaps. 6 and 7 on locally coupled networks, where
it can occur locally in clusters of neighboring elements.
14.4 Synchronization in Scale-Free Networks

of Oscillators
Now we turn to SFNs and show that the heterogeneity in the connection degree
also has significant effects on the synchronization behavior of the oscillators.
We generate random SFNs in the following way [531]: Each node is assigned
to have a number kj ≥ kmin of “half-links” according to the probability
distribution P (k) ∼ k −γ , where γ is the scaling exponent and kmin is a
constant integer. The network is generated by randomly connecting these half-
links to form links, prohibiting self- and repeated links. In the limit γ = ∞,
all nodes have the same degree k = kmin , so that the degree is homoge-
neous in the network, called homogeneous network in the following. Note that
the mean degree K = kj is a function of γ and can be approximated as
K = (γ−1)
(γ−2) kmin for large network size N . In Fig. 14.6, we show the sequence of
degrees kj in descending order. The power-low distribution of kj is shown in
the inset. In what follows, we always label the nodes via decreasing degrees,
k1 ≥ k2 ≥ · · · ≥ kN = kmin .
Here we treat nonidentical oscillators on unweighted SFNs, i.e., Wji = 1
for all connections. The dynamical equations then read
14.4 Synchronization in Scale-Free Networks of Oscillators 325
100
2
10
10−1
P(k)
10−2
kj 10−3
101 102
101 k
0 200 400 600 800 1000

Node index j
Fig. 14.6. Degree sequence kj of a random SFNs (14.7) with N = 1, 000 nodes,
kmin = 5 and the scaling exponent γ = 3. The inset shows the power-law distribution
P (k) ∼ k−γ , where the flat tail results from finite size effects
2.0
1.5
Var(X)
1.0
0.5
0.0
0.0 0.2 0.4 0.6 0.8 1.0
d
Fig. 14.7. The variance of the mean field as a function of the coupling strength d
in SFNs with various scaling exponents. The symbols are: circles (γ = 3); squares
(γ = 4), and triangles (γ = ∞). The solid line is the analytically obtained results
(14.11) for globally coupled oscillators. The networks have the same mean degree
K = 10 and size N = 1, 000
d
N
ẋj = τj F(xj ) + Aji (xi − xj ), j = 1, . . . , N, (14.7)
K i=1
where we have chosen the van der Pol oscillators for F here: (ẋ = y, ẏ =
0.3(1 − x2 )y − x) and the coupling is added to both variables x and y (cf.
Sect. 2.2 for the properties of the van der Pol oscillator). Again the timescale
parameters τj follow a uniform distribution in the interval [1 − Δτ, 1 + Δτ ].
In our simulations, we set Δτ = 0.2.
Let us first examine the collective oscillations in the network. Figure 14.7
shows the variance of the mean field X as a function of the coupling strength
d for SFNs with the same mean degree K = 10, but various scaling exponents
γ = 3, 4, and ∞. It is seen that all the networks generate a coherent collective
oscillation when the coupling strength is larger than a critical value dcr ≈ 0.2.
However, networks with more heterogeneous degrees, i.e., smaller γ, generate a
1.2 1.2
Ωj 1.0 1.0
(a) (b)
0.8 0.8
0 500 1000 0 500 1000
Node index j Node index j
Fig. 14.8. The oscillation frequencies Ωj of the oscillators in a SFN (14.7) with
a weak coupling strength d = 0.25. (a) Cluster synchronization regime in a hetero-
geneous network (γ = 3) and (b) global synchronization regime in a homogeneous
one (γ = ∞). The networks have the mean degree K = 10 and size N = 1, 000
weaker degree of collective synchronization as indicated by a smaller variance

of the mean field.
Next we study in more detail the organization of the transition to the
synchronization behavior in the networks. For this purpose, the behavior of
a typical heterogeneous network (γ = 3) is compared with that of a homo-
geneous network (γ = ∞). We start with the weak coupling regime with
d = 0.25. In Fig. 14.8, we show the oscillation frequency Ωj of the oscillators.
In the heterogeneous network, about 80% of nodes are locked to a common
frequency Ω = 1.02, forming a synchronization cluster. Note that most of the
nodes with large degrees kj are synchronized, whereas many nodes with small
degrees are not locked yet. In the homogeneous network, on the contrary, the
frequencies of all nodes are locked so that the network is globally synchronized
and displays a stronger collective oscillation (Fig. 14.7).
The dependence of the synchronization behavior of an oscillator on its
degree kj can be understood by a mean field approximation. Let x̄j =
N
(1/kj ) i=1 Aji xi be the local mean field of all the neighbors connected to
the oscillator j; Equation (14.7) then can be rewritten as
dkj
ẋj = τj F(xj ) + (x̄j − xj ). (14.8)
K
In random networks, the local mean field x̄j can be approximated by the
global mean field X of the network, x̄j ≈ X, for nodes with degrees kj 1.
With this the system is approximated as
dkj
ẋj = τj F(xj ) + (X − xj ), j = 1, 2, · · · , N. (14.9)
K
This approximation means that the oscillators are forced by a common signal
X, with the forcing strength being proportional to their degree kj . As a result,
on average nodes with a larger degree k synchronize better to the mean field.
For the homogeneous network, there is kj = K, and (14.9) is equivalent to a
globally coupled network, as studied in Chaps. 5 and 10,
ẋj = τj F(xj ) + d(X − xj ), j = 1, 2, · · · , N. (14.10)
To study the transition to a global synchronization regime in globally cou-

pled networks, it is assumed that in the regime of global synchronization, the
state of each oscillator has a small deviation from the mean field, xi = X+δxi .
Expanding the system (14.10) with respect to X and neglecting the higher
order terms O(|δxi |2 ), we get the following low-dimensional macroscopic equa-
tions [532]:
Ẋ = F(X) + JF(X)W,
Ẇ = στ2 F(X) + JF(X)W − dW, (14.11)
where JF denotes the Jacobian matrix of F. Here W = (τj − 1)(xj − X) . στ2
is the variance of the distribution of τj . Simulating (14.11) numerically and
computing the variance of X yields that these macroscopic equations give
a good approximation for the homogeneous network (Fig. 14.7). This net-
work displays a weaker collective synchronization compared with the globally
coupled one, and the difference is induced by the rather sparse connections
(K = 10 only). The difference is reduced at larger K, but this change is not
sensitive to K when K 1. This observation suggests that sparse random
networks with a rather uniform degree already display an efficiency of syn-
chronization similar to highly connected or globally coupled networks. This is
of importance in real-world complex networks, such as neural networks, since
sparse connections save a great deal of energy without degrading too much
the function of the network. This could also be of importance for engineering
applications, such as brain–machine interaction.
The mean field approximation in (14.9) indicates that the synchronizability
of an oscillator with respect to X depends on its degree kj . Now we apply a
few measures to quantify synchronization of the nodes with respect to X. We
measure PS by the time-averaged order parameter
2
sj = sin(Δφj ) + cos(Δφj ) 2 , (14.12)
where Δφj is the difference of the phases of an individual oscillator j and the
mean field X = (X, Y ). Here the phases are defined as φj = − arctan(yj /xj )
and φX = − arctan(Y /X) for an individual oscillator j and the mean field,
respectively. The order parameter has the same property as defined in Chaps. 5
and 10, i.e., sj ≈ 0 when there is no phase locking and sj ≈ 1 when the phase is
locked with an almost constant phase difference. As consistent with Fig. 14.8,
we find that sj = 1 in the homogeneous network; while sj < 1 for many nodes
with small degrees in the heterogeneous network (Fig. 14.9a). To get a clear
dependence of s on the degree k, we calculate the average value s(k) among
all nodes with the same degree k, i.e.,
1
s(k) = sj , (14.13)
Nk
kj =k
where Nk is the number of nodes with degree kj = k in the network. Now

there is a more pronounced dependence between s(k) and k (Fig. 14.9b), which
shows that nodes with larger k synchronize better.
0.8
1.0
0.6
0.8
ΔXj
sj
0.4
0.6 0.2
(a) (c)
0.4 0.0
0 500 1000 0 500 1000
100
1.0 (b) (d)
0.8
ΔX(k)
s(k)
0.6 10−1
0.4 1 1 2
10 10 10 101 10
1
10
2
k k
Fig. 14.9. (a) Phase synchronization order parameter sj of node j (14.12) with
respect to the mean field X. Solid line: heterogeneous network (γ = 3); Dotted line:
homogeneous network (γ = ∞). (b) Average value s(k) of nodes with degree k as
a function of k in the heterogeneous network. (c),(d) as (a) and (b), but for the
distance ΔXj and its average value ΔX(k), respectively. The solid line in (d) with
slope α = 0.82 is plotted for reference. The results are averaged over 50 realizations of
random distribution of the time scale parameter τj . The coupling strength d = 0.25,
and N = 1, 000, and K = 10
We have also measured the distance between the state xj and the mean
field X, ΔXj = |X−xj |. In the homogeneous network, on average this measure
is also uniform for all the nodes, whereas in the heterogeneous network, it
again depends on the degree kj (Fig. 14.9c): the larger kj , the smaller ΔXj on
average. The average value ΔX(k) of nodes with degree k can be calculated
similar to (14.13) and shows a power-law dependence on k, ΔX(k) ∼ k −α ,
with α ≈ 0.82 (Fig. 14.9d). We find that the exponent α is largely independent
of γ in the degree distribution P (k) ∼ k −γ .
For the strong coupling regime (d = 1.0), the frequencies of all the oscillators
are locked mutually as well as locked to the mean field; as a result, the PS order
parameter is sj = 1 for all oscillators in both heterogeneous and homogeneous
networks, i.e., the network is globally phase synchronized. Moreover, in the
homogeneous networks, the oscillators are almost completely synchronized
in the sense that ΔXj is small and uniform on average (Fig. 14.10c). This
difference is due to a small phase difference between the oscillator and the
mean field X, Δφj = |φj − φX | (averaged over time) (Fig. 14.10a). In the
heterogeneous network, the synchronization difference ΔXj is still heterogeneous
(Fig. 14.10c), showing a dependence ΔX(k) ∼ k −α with α ≈ 0.87 (Fig. 14.10d),
0.3 0.4
(a) (c)
0.2
Δφj
ΔXj
0.2
0.1
0.0 0.0
0 500 1000 0 500 1000
(b) (d)
10−1 10−1
ΔX(k)
Δφ(k)
10−2 10−2
1 10 100 1 10 100
k k
Fig. 14.10. (a) Average phase difference Δφj between a node j and the mean field
X. Solid line: heterogeneous network (γ = 3); Dotted line: homogeneous network
(γ = ∞). (b) Average value Δφ(k) of nodes with degree k as a function of k in the
heterogeneous network. (c),(d) as (a) and (b), but for the absolute difference ΔXj
and its average value ΔX(k), respectively. The solid lines in (b) and (d) with slope
α = 0.87 are plotted for reference. The results are averaged over 50 realizations of
the initial frequency distribution. The coupling strength d = 1.0, N = 1, 000 and
K = 10
similar to the case of weak coupling in Fig. 14.9d. Again, the synchronization
difference is induced by a heterogeneous phase difference Δφj (Fig. 14.10a).
The average phase difference displays the same power-law dependence on the
degree, Δφ(k) ∼ k −α , with the same α ≈ 0.87 (Fig. 14.10b).
In networks with completely synchronized identical oscillators, desynchro-
nization can be induced due to noise, and the degree of desynchronization
also depends on the connection degree. This is illustrated for oscillators in a
network with N = 100, γ = 3, and kmin = 5. Since the van der Pol oscillators
are nonchaotic, they become completely synchronized for a nonvanishing cou-
pling strength d > 0. Now an independent Gaussian noise from the normal
distribution N (0, 1), with amplitude D = 0.5, is added to each variable of each
oscillator in the network. This noise induces a desynchronization among the
oscillators. The oscillators still have the same average oscillation frequency,
but phase slips can occur. The de-synchronization property is not uniform in
the network, but depends on the degree k of nodes. For the coupling strength
d = 0.25, the time series x1 (t) of the oscillator with a large degree k = 44
is much cleaner compared with xN (t) of the oscillator with a small degree
k = 5, since the former is coupled stronger to the mean field X(t) accord-
ing to (14.9) and becomes more resistant to noise perturbations (Fig. 14.11).
(a) 4.0
2.0
X(t) 0.0
−2.0
−4.0
(b) 4.0
2.0
x1(t)
0.0
−2.0
−4.0
4.0
(c)
2.0
xN(t)
0.0
−2.0
−4.0
0.0 100.0 200.0 300.0 400.0
time
Fig. 14.11. Time series of the mean field X(t) (a), of the node with the maximal
degree k = 44, x1 (t) (b), and of a node with the minimal degree k = 5, xN (t) (c).
The oscillators are identical with τj = 1 in (14.7). The coupling strength d = 0.25,
and N = 100 and kmin = 5
1.0 (b) 100

(a)
80
0.9
60
s(k)
ST
j
0.8 40
20
0.7
4 10 20 40 20 40 60 80 100
k i
Fig. 14.12. (a) Average phase synchronization parameter s(k) vs. k. (b) An effec-
tive cluster; a point is plotted when sij ≥ ST . The network is the same as in
Fig. 14.11
Figure 14.12a shows s(k) as a function of k for this network. These results
indicate that nodes with large degrees form an effective cluster when they all
synchronize closer to the common mean field. Such an effective cluster is shown
in Fig. 14.12b where the bright point corresponds to sji ≥ ST = 0.82. Here
sji is the phase synchronization order parameter between a pair of oscillators
14.5 Mean-Field Analysis of Hierarchical Synchronization 331
j and i. As shown in Fig. 14.12a, the threshold value ST = 0.82 corresponds

to the degree k = 7. We find that all the nodes in this effective cluster have
degrees kj ≥ 7.
Cluster synchronization has been discussed in previous chapters on regu-
lar networks. This phenomenon also appears in complex networks [533–536].
There are two types of cluster formation in very sparse networks displaying
tree-like structures [533–535]. One is self-organized cluster formation where
the nodes within the cluster display internal connections. The other one is
driven cluster formation, where the nodes of one cluster are driven by those
of the others, but do not connect to other nodes of the same cluster. Cluster
formation in this case is related to some symmetry in the networks [537] (e.g.,
a branch in the tree can form a self-organized cluster, and two branches con-
nected to a common node can form driven clusters). Such symmetry vanishes
with increasing connectivity in random networks, and a clear identification
of these types of clusters becomes difficult. We emphasize that the cluster in
Fig. 14.12 is effective in the sense that there is still a synchronization differ-
ence among the oscillators even though they all synchronize to the mean field
strongly.
The results in the above for the van der Pol oscillators can also be observed
in chaotic oscillators, e.g., the Rössler oscillator discussed in Sect. 14.3.
14.5 Mean-Field Analysis of Hierarchical

Synchronization
The results in Sect. 14.4 shows that the synchronization behavior in the SFNs
of oscillators with nonidentity or noise displays a hierarchical structure due
to the heterogeneity in the degrees kj . Nodes with larger degrees synchronize
better and contribute more to the collective oscillations of the network. For a
given mean degree K, the homogeneous network topology seems to be optimal
for synchronization, since all the nodes have equally significant contributions.
In the following we present a qualitative analysis of the hierarchical syn-
chronization in SFNs, based on the mean field approximation in (14.9) and
the linear analysis in Sect. 14.2. The mean field approximation means that the
oscillators are forced by a common signal X, with the forcing strength being
proportional to their degree kj .
Now let us suppose that the oscillators are identical (τj = 1), so that we
can perform a similar linear analysis as in Sect. 14.2. The linear variational
equations (14.9) then read

d
η̇j = JF(X) − kj I ηj , kj 1, (14.14)
K
which have the same form as (14.5), except that Λj is replaced by kj /K and
H(s) by X. Here I is the identity matrix. The largest Lyapunov exponent
λ(kj ) of (14.14) is a function of kj ,
λ(kj ) = −dkj /K. (14.15)
In the presence of parameter mismatches or noise, the dynamics of the

averaged synchronization difference ΔX(k) over large timescales can be
expressed as
d
ΔX(k) = λ(k)ΔX(k) + c, (14.16)
dt
where c > 0 is a constant denoting the level of perturbation, which depends on
the noise level D, the parameter distribution Δτ and the coupling strength d.
From this we get the asymptotic result ΔX(k) = c/|λ(k)|, leading to
ΔX(k) ∼ k −1 , (14.17)
which explains qualitatively the numerically observed scaling in Figs. 14.9d and
14.10d. The deviation of the scaling exponents α ≈ 0.85 from the linear result
α = 1 may result from the mean field approximation and significant nonlinearity,
so that the linear analysis in (14.16) is only a first-order approximation.
The hierarchical synchronization and effective clusters are also oberved
in network of chaotic oscillators where the coupling strength is below the
threshold of CS [538].
14.6 Synchronization Properties of Weighted Networks

In Sect. 14.4 and 14.5, we have treated a uniform coupling strength, i.e.,
Wji = 1, and showed that in such unweighted SFNs, the synchronization
of the oscillators depends strongly on its degree. However, many complex net-
works, where synchronization is relevant, are actually weighted and display
a highly heterogeneous distribution of both degrees and weights [539–542].
Examples include brain networks [543, 544], and airport networks [539] and
the synchronization of epidemic outbreaks in different cities [555, 556]. It
has been observed that heterogeneity in the coupling strength can lead to
desynchronization and localized instability in locally coupled regular net-
works of periodic oscillators [557] or in random networks of pulse-coupled
oscillators [558].
Here, we study the ability of CS (synchronizability) in random networks
of identical oscillators (τj = 1 in (14.2) and (14.3)) with weighted coupling
schemes motivated from real networks. The analysis of real networks [539],
including scientific collaboration networks and airport networks, yields the
following main properties:
1. The weight Wji of a connection between the nodes j and i is strongly
correlated with the product of the corresponding degrees as Wji ∼ (kj ki )θ .
14.6 Synchronization Properties of Weighted Networks 333
2. The average intensity I(k) of nodes with degree k increases as I(k) ∼ k β .

Here the intensity Ij of a node j is defined as the total input weight of the node:

N
Ij = Aji Wji . (14.18)
i=1
In random networks where the degree kj of a node is independent of the degree

of its connected neighbors, so that the degree correlations can be neglected,
the exponents in (1) and (2) are related as β = 1 + θ [539]. In particular,
θ ≈ 0.5 (β ≈ 1.5) for the worldwide airport network, so that the intensities
of the nodes grow faster than their degrees, whereas θ ≈ 0 (β ≈ 1) for the
cond-mat collaboration network [539]. The case θ < 0 (β < 1) corresponds to
a saturation in the capacity of nodes with large degrees and is expected to be
relevant for other networks, such as neuronal networks [543, 544].
Motivated by these observations from real networks, we treat two models
of weighted couplings:
Model I : Wji = (kj ki )θ , (14.19)

Model II : Wji = kjθ , (14.20)
where the weights are defined for the connections of a given network topology
and θ is a tunable parameter. Model I reproduces property (1) and has the
same weighted structure as many real networks [539]. In degree-uncorrelated
networks, Model I also reproduces the expected scaling I(k) ∼ k 1+θ . Model
II incorporates other realistic features and always reproduces property (2).
These models also include many previously studied systems as special cases.
For θ = 0, both models correspond to the widely studied case of unweighed
networks [523, 526, 559], where the weights are uniform (Wji = 1). For Model
II with θ = −1, the intensities are fully uniform (Ij = 1), as in a number of
previous studies about synchronization of coupled maps [522, 533–535].
Note that the weight matrix W is in general asymmetric for Model II.
The corresponding coupling matrix G in (14.2) is also asymmetric and can be
written as G = Dθ L, where D = diag{k1 , k2 , . . . kN } is the diagonal matrix
of the degrees and L = DI − A is the (symmetric) Laplacian matrix. Since
det(Dθ L − ΛI) = det(Dθ/2 LDθ/2 − ΛI) (14.21)
is valid for any Λ, we have that the spectrum of eigenvalues of the matrix G
is equal to the spectrum of a symmetric matrix defined as H = Dθ/2 LDθ/2 .
As a result, all the eigenvalues of G are real and can be ordered as [560, 561]
0 = Λ1 ≤ Λ2 · · · ≤ ΛN . (14.22)
The analysis of synchronizability is based on the master stability approach

presented in Sect. 14.2. For many oscillatory dynamical systems [523, 529],
each transverse eigenmode ηj in (14.5) is stable in a single, finite interval
0.3
ε1 dΛ2 dΛj dΛN ε2
0.0
Λ
−0.3
−0.6
0.0 1.0 2.0 3.0 4.0
ε
Fig. 14.13. A schematic illustration of master stability. The jth eigenmode in
(14.5) is stable when 1 < dΛj < 2 where the largest Lyapunov exponent Λ of
(14.5) is negative
1 < dΛj < 2 , where the thresholds 1 and 2 are determined only by F,
H, and s. The network is thus synchronizable for some d values when all the
transverse eigenmodes are damped (Fig. 14.13), with the following condition:
ε1 < dΛ2 ≤ dΛ3 · · · ≤ dΛN < ε2 . (14.23)
This is equivalent to the condition

ΛN ε2
R≡ < , (14.24)
Λ2 ε1
where the eigenratio R depends only on the network structure, as defined
by the coupling matrix G, and ε2 /ε1 depends only on the dynamics. From
this, it follows that the smaller the eigenratio R the more synchronizable the
network and vice versa [529]. We can hence characterize the synchronizability
of networks using only the eigenratio R.
For numerical simulations we generate the adjacency matrix A with a
model of growing SFNs with aging of nodes [562], which extends the Barabási–
Albert (BA) model [498]. We start with m fully connected nodes and add
a new node with m links at each time step. The minimum degree is then
kmin = m and the mean degree is K = 2m. A node j is randomly selected to
be connected to the new node with a probability Πj which depends on the
degree kj and age τj of the corresponding node, i.e.,
Πj ∼ kj τj−α , (14.25)
where α is the aging exponent controlling the aging effects. α = 0 corresponds

to the usual BA model [498], which generates SFNs with a power-law degree
distribution at γ = 3. For the aging exponent −∞ < α ≤ 0, the growing rule
(14.25) generates SFNs with a power-law tail P (k) ∼ k −γ and the scaling
exponent in the interval 2 < γ ≤ 3 [562], as in most real SFNs.
The weighted coupling schemes of Models I and II in (14.19) and (14.20)
have a significant impact on the synchronizability of the networks. In both
Wij=(kikj)θ Wij=kiθ
103
α = −3 102
2
R 10
101
101 (a) α=0 (b)
0 0
10 10
−3.0 −2.0 −1.0 0.0 1.0 −3.0 −2.0 −1.0 0.0 1.0
θ θ
Fig. 14.14. Eigenratio R (14.24) as functions of θ for Model I (a) and Model II (b)
in growing SFNs with the aging exponent α = 0 (◦) and α = −3 (•). Each symbol
is an average over 50 realizations of the networks, for K = 20 and N = 210 . The
solid lines are the approximations of R by (14.36) with AR = 0.47
102
Wij = 1
R Wij = (kikj)1
101
Wij = ki1
100
−3.0 −2.0 −1.0 0.0
α
Fig. 14.15. Eigenratio R (14.24) vs.α for growing SFNs with θ = 0 (circles) and
θ = −1 in Model I (triangles) and in Model II (dots). Dotted line: RH (K) in (14.35).
Solid lines: approximation by (14.36) with AR = 0.47. The other parameters are
the same as in Fig. 14.14
models, as θ is reduced from zero, the eigenratio R decreases and reaches a

minimum around θ = −1 (Fig. 14.14). The eigenratio R is also shown as a func-
tion of the aging exponent α (Fig. 14.15). For unweighted networks (θ = 0),
the synchronizability decreases when α is reduced. When the networks are
weighted with θ = −1, the synchronizability is clearly enhanced in both
Model I and Model II, and, in particular, appears to become independent
of α in Model II.
The synchronizability of the networks is closely correlated to the hetero-
geneity of the intensity Ij of nodes as defined in (14.18). In Model II, we get
Ij = kj1+θ , (14.26)
so that Ij becomes more homogeneous when θ → −1, and it is fully uniform
at θ = −1, i.e., Ij = 1, for all j, independent of α. In Model I, we have
Ij = kj1+θ kiθ j , (14.27)

where kiθ j = (1/kj ) kiθ is not constant and depends on α, so that the
intensities are not fully uniform at θ = −1. Accordingly, at θ = −1, the eigen-
ratio R for Model I is larger than for Model II and still exhibits a dependence
on α (Fig. 14.15). These observations show that the synchronizability of the
networks is positively correlated with the homogeneity of Ij .
Now we present a more physical and quantitative understanding of the
dependence of synchronizability on the intensity Ij using a mean field approx-
imation of the dynamical system in (14.2). Let us define

N
H̄W
j = (1/Ij ) Wji Aji H(xi ) = (kj /Ij ) Wji H(xi ) j (14.28)
i=1
as the weighted local mean field of all the neighbors connected to the oscil-
lator j, where j denotes average over the kj neighbors of the node j.
Equation (14.2) then can be rewritten as
j − H(xj )].
ẋj = F(xj ) + dIj [H̄W (14.29)
Since the state xi of an oscillator i is not affected directly by the individual

output weights Wji , we may assume that Wji and H(xi ) are statistically
uncorrelated. Consequently for large kj , (14.28) can be approximated as
j ≈ (kj /Ij ) Wji

H̄W j H(xi ) j = H̄j , (14.30)
where

N
H̄j = (1/kj ) Aji H(xi ) (14.31)
i=1
is the unweighted local mean field. Note that in Model II, H̄W j ≡ H̄j since
Wji = Ij /kj . If the network is sufficiently random with a large enough mini-
mum degree kmin , the local mean field H̄j in (14.31) can be approximated by
the global mean field of the network, H̄j ≈ H̄. Moreover, for small perturba-
tions close to the synchronized state s, we may assume H̄j ≈ H(s), and the
system is approximated as
ẋj = F(xj ) + dIj [H(s) − H(xj )], (14.32)
indicating that the oscillators are decoupled and forced by a common oscillator
ṡ = F(s), with the forcing strength being proportional to the intensity Ij . If
there exists some d satisfying ε1 < dIj < ε2 for all j, then all the oscillators
are synchronizable by the common driving H(s), corresponding to CS of the
whole network.
This finding suggests that the eigenratio R can be approximated as
R ≈ Imax /Imin , (14.33)
where Imax and Imin are the maximal and minimal values of the intensities,
respectively. For fully uniform intensities in Model II at θ = −1, R = 1 by this
approximation. But the spectrum of the coupling matrix G actually depends

on the mean degree K. As discussed already in this section, G = D−1 L,
in this case and it has the same spectrum of eigenvalues as a symmetric
matrix defined as H = D−1/2 LD−1/2 . Here H is the normalized Laplacian
matrix [563]. The spectrum of H tends to the semicircle law for large random
networks with arbitrary √expected degrees [564], provided
√ that the minimum
expected degree kmin K. In particular, if kmin K ln3 N , we have
2
max{1 − λ2 , λN − 1} = [1 + o(1)] √ . (14.34)
K
From these, it follows that
√
1 + 2/ K
R ≈ RH (K) ≡ √ . (14.35)
1 − 2/ K
As shown in Fig. 14.15 by the dotted line, (14.35) provides a good approxi-
mation under the weaker condition kmin 1, when the intensity is uniform
Ij = 1 (Model II, θ = −1), regardless of the degree distributions at different
α values. Physically, the dependence on the mean degree K can be under-
stood as due to deviations or fluctuations of the local mean field H̄j from
the global mean field H̄. This effect is similar when the intensities are not
uniform. Thus, we may assume that the contribution due to the number of
connections is statistically independent of the contribution due to the strength
of the connections. Accordingly, for general weighted random networks with
arbitrary distributions of Ij and kj , we get from (14.33) and (14.35)
Imax
R = AR RH (K), (14.36)
Imin
where AR is a constant of the order of 1.
With a single parameter AR = 0.47, (14.36) approximates the eigenratio R
very closely for different networks and weighted coupling schemes, including
the unweighed networks (θ = 0) (Figs. 14.14 and 14.15). The fitting parameter
underestimates R slightly only when the intensities become rather homoge-
neous (Imax /Imin < 3).
These results demonstrate that the synchronizability in random complex
networks is determined by two leading parameters, the mean degree K and
the heterogeneity of the intensity Ij as measured by the ratio Imax /Imin . This
dependence is universal for networks with different degree distribution. More
analysis also shows that it is also universal for networks displaying nontrivial
clustering and degree correlations [565].
The above analysis on identical oscillators can serve as a good approxi-
mation even when the oscillators are not fully identical. Now we consider PS
in weighted complex networks of nonidentical van der Pol oscillators as in
Sect. 14.4:
2.0
θ=0
1.0 θ = −1
0.0
X
−1.0
(a) (b)
−2.0
0 20 40 60 80 100 0 20 40 60 80 100
time time
Fig. 14.16. Time series of the mean field X in random SFNs of nonidentical van der
Pol oscillators with γ = 3. The coupling strengths are (a) d = 0.1 and (b) d = 0.2
d
N
ẋj = τj F(xj ) + Aji Wji (xi − xj ), j = 1, . . . , N, (14.37)
I ∗ i=1
N
where I ∗ = (1/N ) j Ij is the mean value of the intensity. Note that for
the unweighted case, Wji = 1, and I ∗ is just the mean degree, i.e., I ∗ = K,
so that (14.37) becomes the same as (14.7). Here again we assume that the
parameters τj follow a uniform distribution in the interval [1 − Δτ, 1 + Δτ ].
In our simulations we set Δτ = 0.2. The SFNs are generated as in Sect. 14.4
with γ = 3, kmin = 10 (K = 20), and N = 1, 024. In Fig. 14.16, we show time
series of the mean field X for unweighted (θ = 0) and weighted random SFNs
(Model II, θ = −1). For small coupling strength d, neither of the networks
display significant collective behavior (Fig. 14.16a). As the coupling strength
is increased, coherent collective oscillations emerge for both weighted and
unweighted networks, but the oscillations are much more pronounced for the
networks with θ = −1 (Fig. 14.16b).
In Fig. 14.17, we show the variance of the mean field X as a function of d.
The variance is approximately zero for small coupling strength, increases
sharply as d is increased beyond a certain critical value, and saturates
for large d (Fig. 14.17). The overall behavior is similar for weighted and
unweighted networks, but, again, the variance is significantly larger for net-
works with θ = −1 at the same value of d. Moreover, the variance of random
SFNs for θ = −1 is well approximated by the variance of random homoge-
neous networks with the same mean degree (Fig. 14.17), which are networks
that exhibit good phase synchronization properties as shown in Sect. 14.4. In
both the cases, the intensities Ij are fully uniform. These results confirm that
PS in random networks is also universally determined by the mean degree
K and the distribution of the intensity Ij . Synchronization is enhanced in
networks with more homogeneous intensities.
2.0
1.5
Var(X) 1.0
0.5
0.0
0.0 0.2 0.4 0.6 0.8 1.0
d
Fig. 14.17. Variance of the mean field X as a function of d for random SFNs
of van der Pol oscillators with θ = 0 (circles) and θ = −1 (dots). The diamonds
correspond to random homogeneous networks with the same mean degree (K = 20)
of the random SFNs. The results are averaged over 20 realizations
14.7 Conclusions
In this chapter synchronization of nonlinear oscillators on complex networks
has been discussed. We have shown that the topology as well as the weight
of the connections has significant influences on the synchronization of the
networks. The main results are:
– In SWNs, the synchronizability of the networks is enhanced significantly
by adding some shortcuts to the underlying locally coupled networks.
Oscillation death and global synchronization of SWNs have been demon-
strated.
– In SFNs displaying heterogeneous degrees, the synchronization behavior
of a node depends on its degree in unweighted networks where the coupling
strength is uniform for all the connections in the network. Hubs (nodes
with large degrees) mainly contribute to the mean field of the network
and are dominant in forming an effective synchronization cluster.
– Weighted coupling influences significantly the synchronizability of com-
plex networks. In random networks, the synchronizability is universally
controlled by the mean degree and the heterogeneity of the intensities
of nodes. Networks with more homogeneous intensities display enhanced
synchronization behavior.
Synchronization of complex networks is relevant in many real-world sys-
tems. For example, brain networks [543, 544] display a hierarchy of oscillation
and synchronization on various spatial and temporal scales [545], and syn-
chronization of distributed brain activity has been proposed as an important
mechanism for neural information processing [546–548]. The analysis of the
anatomical connectivity of the mammalian cortex [549] and the functional con-
nectivity of the human brain [550–552] have shown that both share typical
features of many complex networks. However, the relationship between the

anatomical and functional connectivities remains one of the major challenges
in neuroscience [546–548]. A meaningful understanding of this relationship
could be obtained by studing the organization of the synchronization clus-
ters in a realistic network of corticocortical connectivity of mammalian cortex
(cat) [553, 554].
Heterogeneous weights in the coupling strength are natural, for example,
in city networks of coupled populations in the synchronization of epidemic
outbreaks [555, 556], due to the heterogeneous populations of the cities. In
communication and other technological networks, the functioning of the sys-
tem relies on the synchronization of interacting units [566]. The study of syn-
chronization in weighted networks [560, 561, 565, 567, 568] with heterogeneity
both in the connection topology and in the connection strength represents an
interesting and important research direction in the future and has important
potential applications in real-world systems.
Real dynamical networks are often growing and changing in their connec-
tion topology and connection weights. Conditions and criteria for global syn-
chronization in such time-varying networks have been discussed in [569–571].
It is very interesting to investigate how effective synchronization patterns
evolve in time due to the evolution of the structures in time-varying networks.
So far, it is assumed that the structural changes in time-varying networks are
independent of the oscillatory dynamics [569–571]. So one of the most interest-
ing research topics is the self-organization of structures and synchronization
dynamics due to the interplay between them [572, 573], as occurs in more
realisitc systems such as the neural and ecological systems. Understanding
the self-organization of large-scale complex structures and dynamics such as
those in the brain is still a big challenge.
Glossary
ALLS - average length of laminar stage

CM - circle map
CTS - characteristic time scale
ER random network - Erdòs and Rényi random network
FHN model - FitzHugh-Nagumo model
GLE - Ginzburg-Landau equation
HH model - Hodgkin-Huxley model
PLL - phase-locked loop
PS - phase synchronization
CPS - chaotic phase synchronization
CS - complete synchronization
GS - generalized synchronization
LE - Lyapunov exponent
OD - Oscillation death
STI - spatiotemporal intermittency
SFN - scale-free network
SWN - small-word network
UPO - unstable periodic orbit
Acknowledgments
We are grateful to our colleagues and collaborators H. Abarbanel, K. Aihara,

V. Anischenko, F.T. Arecchi, I. Belykh, V.N. Belykh, B. Blasius, S. Boccaletti,
H. Braun, U. Freudel, H. Fujisaka, C. Grebogi, P. Hanggi, M. Hasler,
C. Hilgetag, B. Hu, J. Hudson, M. Ivanchenko, I. Kiss, N. Kuchlander,
E. Montbriò, E. Mosekilde, F. Moss, A. Motter, E. Ott, U. Parlitz, L. Pecora,
A. Pikovsky, M. Romano, M. Rosenblum, R. Roy, N.F. Rulkov, V. Shalfeev,
L. Schimansky-Geier, K. Showalter, M.M. Sushik, M. Thiel, G. Zamora,
M. Zaks, L. Zemanova for many fruitful discussions and collaborations on
the subjects covered by this book. We acknowledge financial support from
the Russian Foundation for Basic Research, from SFB 555 (DFG), Humboldt
Foundation, VW Foundation and GoFORSYS (BMBF) from Germany, and
E2C2 and BIOSIM from European Union.
References
1. A.S. Pikovsky, M.G. Rosenblum, and J. Kurths, Synchronization - A universal

concept in Nonlinear Sciences, Cambridge University Press, Cambridge, 2001.
2. I.I. Blekhman, Synchronization of Dynamical Systems, Nauka, Moscow, 1971.
(In Russian).
3. I.I. Blekhman, Synchronization in Science and Technology, Nauka, Moscow,
1981. (In Russian); English Translation: ASME Press, New York, 1988.
4. V.S. Afraimovich, V.I. Nekorkin, G.V. Osipov, and V.D. Shalfeev, Stability,
Structures and Chaos in Nonlinear Synchronization Networks, World Scien-
tific, Singapore, 1994.
5. E. Mosekilde, Yu. Maistrenko, and D. Postnov, Chaotic Synchroniza-
tion. Applications to Living Systems, World Scientific, Singapore, 2002.
6. S.C. Manrubia, A.S. Mikhailov, and D.H. Zanette, Emergence of Dynami-
cal Order: Synchronization Phenomena in Complex Systems, World Scientific,
Singapore, 2004.
7. A.T. Winfree, The Geometry of Biological Time, Springer-Verlag, New York,
1980.
8. Y. Kuramoto, Chemical Oscillations, Waves and Turbulence, Springer Verlag,
Berlin/Düsseldorf, 1984.
9. P.S. Landa, Self-Oscillations in Systems with Finite Number of Degree of Free-
dom, Nauka, Moscow, 1980. (In Russian).
10. P.S. Landa, Nonlinear Oscillations and Waves in Dynamical Systems, Kluwer
Academic Publishers, Dordrecht-Boston-London, 1996.
11. V.S. Anishchenko, V.V. Astakhov, A.B. Neiman, N.T. Vadivasova, and
L. Shimansky-Geier Nonlinear Dynamics of Chaotic and Stochastic Systems.
Tutorial and Modern Development, Springer, Berlin, Heidelberg, 2002.
12. J. Buck, Quart. Rev. Biol. 63, 265 (1988).
13. J. Buck and E. Buck, Scientific Am. 234, 74 (1976).
14. C.S. Peskin, Mathematical Aspects of Heart Physiology, Courant Institute of
Mathematical Science Publication, New York, 1975, pp. 268–278.
15. D.C. Michaels, E.P. Matyas, and J. Jalife, Circulation Research 61, 704 (1987).
16. V. Torre, J. Theor. Biol. 61, 55 (1976).
17. A. Sherman, J. Rinzel, and J. Keizer, Biophys. J. 54, 411 (1988).
18. T.J. Walker, Science 166, 891 (1969).
19. N.E. Diamant and A. Bortoff, Am. J. Physiol. 216(2), 301 (1969).
346 References
20. F.C. Hoppensteadt and E.M. Izhikevich, Weakly Connected Neural Networks,
Springer-Verlag, New York, 1997.
21. C.V. Gray, J.Computat.Neuroscience 1, 11 (1995).
22. W. Singer and C.M. Gray, Ann. Rev. Neorosci. 18, 555 (1995).
23. C.J. DeLuca, A.M. Roy, and Z. Erim, J. Neurophysiol. 70, 2010 (1993).
24. W. Singer, Ann. Rev. Physiol. 55, 349 (1993).
25. P.A. Tass Phase resetting in Medicine and Biology Springer, Berlin, 1999.
26. R.K. Wang, R.D. Traub, and R. Miles, Adv. Neurol. 44, 583 (1986).
27. J.M. Gonzalez-Miranda Synchronization and Control of Chaos. An Introduc-
tion for Scientists and Engineers World Scientific, Singapore, 2004.
28. C.W. Wu Synchronization in coupled chaotic circuits and systems World
Scientific, Singapore, 2002.
29. M.E. Yalcin, J.A.K. Suykens, and J.P.L. Vandewalle, Cellular Neural Net-
works, Multi-Scroll Chaos and Synchronization World Scientific, Singapore,
2005.
30. M.K. McClintock, Nature 229, 244 (1971).
31. B. Blasius, A. Huppert, and L. Stone, Nature 399, 354–359 (1999).
32. D.E. Postnov, A.G. Balanov, and E. Mosekilde, Advances in Complex Systems
1, 181 (1998).
33. R.A. York and R.C. Compton IEEE Trans. Microwave Theory Tech. 39, 1000
(1991).
34. K. Wiesenfeldt, P. Colet, and S.H. Strogatz, Phys. Rev. Lett. 76, 404 (1996).
35. K. Wiesenfeldt, P. Colet, and S.H. Strogatz, Phys. Rev. E 57, 1563 (1998).
36. A. Rodriguez-Angeles and H. Nijmeijer, Int. J. Control 74, 1311 (2001).
37. Y.-H. Liu, Y. Xu, and M. Bergerman, IEEE Trans. Robot. Autom. 15, 258
(1999).
38. Z. Jiang and M. McCall, J. Opt. Soc. Am. 10, 155 (1993).
39. S.Yu. Kourchatov, V.V. Likhanskii, A.P. Napartovich, F.T. Arecchi, and
A. Lapucci, Phys. Rev. A 52, 4089 (1995).
40. A.F. Glova, S.Yu. Kurchatov, V.V. Likhnaskii, A.Yu. Lysikov, and
A.P. Napartovich, Quant. Electron. 26(6), 500 (1996).
41. Y. Tanguy, J. Houlihan, G. Huyet, E.A. Viktorov, and P. Mandel, Phys. Rev.
Lett. 96, 053902 (2006).
42. S. Peles, J.L. Rogers, and K. Wiesenfeld, Phys. Rev. E 73, 026212 (2006).
43. D. Tsygankov and K. Wiesenfeld, Phys. Rev. E 73, 026222 (2006).
44. S. Yanchuk, A. Stefanski, T. Kapitaniak, and J. Wojewoda, Phys. Rev. E 73,
016209 (2006).
45. L. Pecora, Editor. A focus issue on synchronization in chaotic systems. Chaos
7 (1997).
46. J. Kurths, Editor. A focus issue on phase synchronization in chaotic systems.
Int. J. Bifurc. and Chaos, 10, (2000).
47. A.A. Andronov, A.A. Vitt, and S.E. Khaykin, Theory of Oscillations,
Gostekhizdat, Moscow, 1937. (In Russian); English Translation: Pergamon
Press, Oxford - NY - Toronto, 1966.
48. B. van der Pol, Phil. Mag. 3, 64 (1927).
49. N. Minorsky Nonlinear Oscillations Van Nostrand, Princeton, NJ, 1962.
50. M.I. Rabinovich and D.I. Trubetskov, Oscillations and Waves in Linear and
Nonlinear Systems, Kluwer Academic Publishers, Dordrecht-Boston-London,
1989.
References 347
51. V.V. Matrosov, Tech. Phys. Lett., 22(23), 4 (1996).

52. See, for example, P. Hadley and M.R. Beasley, Appl. Phys. Lett. 50,
621 (1987).
53. S.Y. Kung VSLI array processors, New Jersey, Prentice-Hall, 1988.
54. B. Scholz-Reiter, J.T. Tervo, and M. Freitag, Physica A 363, 32 (2006).
55. R. Kapral and K. Showalter, Editors, Chemical Waves and Patterns, Kluwer,
Dordrecht, 1995.
56. V. Petrov, Q. Quyang, and H.L. Swinney, Nature 388, 655 (1997).
57. Phase-Locked Loops and Their Applications, edited by W.C. Lindsey and
M.K. Simon (IEEE Press, New York, 1978).
58. O.E. Rössler, Phys. Lett. A 57, 397 (1976).
59. G.V. Osipov, B. Hu, Ch. Zhou, M. Ivanchenko, and J. Kurths, Phy. Rev. Lett.
91, 24101 (2003).
60. G. Fisher, Plane algebraic curves, American Mathematical Society, Provi-
dence, Rhode Island, 2001.
61. C. Sparrow, The Lorenz Equations: Bifurcations, Chaos, and Strange Attrac-
tors, Springer-Verlag, Berlin,1982.
62. R.N. Madan, Chua circuit: A paradigm for chaos, World Scientific, Singapore,
1993.
63. M.P. Kennedy, IEEE Trans. Circ. Syst. 41, 771 (1994).
64. O. Papy and H. Kawakami, Trans. IEICE E79-A, 1568 (1996).
65. V.S. Anishchenko, Dynamical Chaos - Models and Experiments, World
Scientific, Singapore, 1995.
66. W. Lauterborn, T. Kurz, and M. Wiesenfeldt, Coherent Optics. Fundamentals
and Applications, Springer-Verlag, Berlin, Heidelberg, New York, 1993.
67. I.Z. Kiss, Q. Lv, J.L. Hudson, Phys. Rev. E, 71, 035201(R) (2005).
68. D. Maraun and J. Kurths, Geophys. Res. Lett., 32, L15709 (2005).
69. D. Gabor, J. IEE London, 93, 429 (1946).
70. P. Panter, Modulation, Noise, and Spectral Analysis, McGraw-Hill, New York,
1965.
71. Y. Pomeau and P. Manneville Phys. Lett., 75A, 1 (1979).
72. Y. Pomeau and P. Manneville Phys. Lett., Commun. Math. Phys., 74, 189
(1980).
73. J.E. Hirsch, B.A. Huberman, D.J. Scalapino, Phys. Rev. A, 25, 519 (1982).
74. I.P. Keener, Trans. AMS., 26(2), 589, (1980).
75. R. Klages and J.R. Dorfman Phys. Rev. Lett., 74, 387, (1995).
76. A. Katok and B. Hasselblatt, Introduction to the modern theory of dynamical
systems, Cambridge University Press, Cambridge, 1995.
77. H.G. Schuster, Deterministic chaos, An introduction, VCH-Verlag, Weinheim,
1988.
78. K.T. Alligood, T.D. Sauer, and J.A. Yorke, Chaos: An introduction to dynam-
ical systems, New York, 1992.
79. E. Ott, Chaos in dynamical systems, Cambridge University Press, Cambridge,
1992.
80. M. de Sousa Vieira, A.J. Lichtenberg, and M.A. Liberman, Phys. Rev. A, 46,
R7359, (1992).
81. G. Goldztein and S.H. Strogatz, Int. J. of Bifurcations and Chaos, 5, 983,
(1995).
82. W.C. Lindsey, Synchronization systems in communication and control,
Prentice-Hall, Englewood Cliffs, New Jersey, 1972.
348 References
83. M. Steirade, D.A. McCormick, and T.J. Sejnowski, Science 262, 679 (1993).
84. F. Amzica and M. Steirade, Electroencephalogr. Clin. Neurophysiol. 2, 69
(1998).
85. W. Schultz, J. Neurophysiol. 80, 1 (1998).
86. H.D. Abarbanel et al., Phys. Usp. 39, 337 (1996) [Usp. Fiz. Nauk 166, 363
(1996).].
87. E.M. Izhikevich, IEEE Trans. on Neural Networks, (2004).
88. H.D. Abarbanel et al., Neural Comput. 8, 1567 (1996).
89. R.D. Pinto et al., Phys. Rev. E 62, 2644 (2000).
90. R.C. Elson et al., Phys. Rev. Lett. 81, 5692 (1998).
91. N.F. Rulkov, Phys. Rev. Lett. 86, 183 (2001).
92. R. FitzHugh, Biophys. J. 1, 445 (1961).
93. J. Nagumo, S. Arimoto, and S. Yoshizawa, Proc. IRL 50, 2061 (1960).
94. C.H. Luo and Y. Rudy, Circ. Res. 68, 1501 (1991).
95. I.S. Aranson, A.V. Gaponov-Grekhov, M.I. Rabinovich, A.V. Rogalsky, and
R.Z. Sagdeev, “Lattice models of nonlinear dynamics of nonequilibrium
media” (Preprint: Inst. of Appl. Phys., Gorky.) 163 (1987).
96. S.H. Strogatz Sync. The Emerging Science of Spontaneous Order. Theia
(Hyperion), New York, 2003.
97. L. Glass and M.C. Mackey, From Clocks to Chaos: The Rhythms of Life.
Princeton University Press, Princeton, NJ, 1988.
98. R.L. Stratonovich, Topics in the Theory of Eandom Noise. Gorddon and
Breach, New York, 1963.
99. A.N. Malakhov, Fluctuations in Self-Oscillatory Systems., Nauka, Moscow,
1968 (in Russian).
100. S. Boccaletti, J. Kurths, G. Osipov, D.L. Valladares, and C.S. Zhou, Physics
Reports 366, 1 (2002).
101. V.S. Anishchenko and T.E. Vadivasova, Synchronziation of Self-Oscillations
and Noise-Induced Oscillations. J. of Communications Technology and Elec-
tronics 47, 117 (2002).
102. E.V. Appleton, Proc. Cambridge Phil. Soc. (Math and Phys. Sci.) 21, 231
(1922).
103. A.A. Andronov and A.A. Vitt, Zhurnal prikladnoj Fiziki (J. Apll. Phys.) 7, 3
(1930) (in Russian).
104. A.A. Andronov and A.A. Witt, Archiv für Elektrotechnik 24, 99 (1930).
105. L.I. Mandelshtam and N.D. Papaleksi, Collected works by L. I. Mandelshtam,
2 (Izd. Akademii Nauk, Moscow), 13 (1947) (in Russian).
106. M.L. Cartwright and J.E. Littlewood, J. London Math. Soc., 20, 180 (1945).
107. M.L.Cartwright, J. Inst. Elec. Eng., 95, 88 (1948).
108. N.N. Bogolyubov and Yu.A. Mitropolsky, Asymptotic Methods in the Theory
of Nonlinear Oscillations, Gordon and Breach, New York, 1961.
109. E.F. Stone, Phys. Lett. A 163, 367 (1992).
110. A.S. Pikovsky, M.G. Rosenblum, G.V. Osipov, and J. Kurths, Physica D, 104,
219 (1997).
111. K. Josić and D.J. Mar, Phys. Rev. E 64, 066234 (2001).
112. M.A. Zaks, E.-H. Park, M.G. Rosenblum, and J. Kurths, Phys. Rev. Lett. 82,
4228 (1999).
113. E.H. Park, M. Zaks, and J. Kurths, Phys. Rev. E 60, 6627 (1999).
114. M.A. Zaks, E.-H. Park, and J. Kurths, Int. J. Bifurcation and Chaos, 10, 2649
(2000).
References 349
115. C. Schafer, M.G. Rosenblum, J. Kurths, and H.H. Abel, Nature 392, 239
(1998).
116. C. Schäfer, M.G. Rosenblum, H.H. Abel, and J. Kurths, Phys. Rev. E 60, 857
(1999).
117. A. Pikovsky, G. Osipov, M. Rosenblum, M. Zaks, and J. Kurths, Phys. Rev.
Lett., 79, 47, (1997).
118. E.R. Rosa, E. Ott, and M.H. Hess, Phys. Rev. Lett. 80, 1642 (1998).
119. A.S. Pikovsky, M. Zaks, M. Rosenblum, G. Osipov, and J. Kurths, Chaos, 7,
680 (1997).
120. A.S. Pikovsky and P. Grassberger, J. Phys. A 24, 4587 (1991).
121. D.Y. Tang, M.Y. Li, and C.O. Weiss, Phys. Rev. A 46, 676 (1992).
122. M. Arjona, J. Pujol, and R. Corbalan, Phys. Rev. A 50, 871 (1994).
123. A.M. Kulminskii and R. Vilasecar, J. Mod. Optic 42, 2295 (1995).
124. N.W. Mureithi, M.P. Paidoussis, and S.J. Price, J. Fluid Struct. 8, 853(1994).
125. H. Okamoto, N. Tanaka, and M. Naito, J. Phys. Chem. 102, 7353 (1998).
126. R. Richter, J. Peinke, and A. Kittel, Europhys. Lett. 36, 675 (1996).
127. D. Gillet, Astron. Astrophys. 259, 215 (1992).
128. D.L. Feng et al., Phys. Rev. E 54, 2839 (1996).
129. J.H. Cho et al., Phys Rev. E 65, 036222 (2002).
130. J.K. Bhattacharjee and K. Banerjee, Phys. Rev. A, 29 2301 (1984).
131. S.Y. Kim, Phys. Rev. E, 59, 2887 (1999).
132. L.D. Landau and E.M. Lifshitz Mechanics, Pergamon Press Ltd., 1976.
133. Similar equation appears in studies of the thermal convection between hori-
zontal plates when periodic driving force is applied. Behavior of the system
near the convictive threshold was studied in G. Ahlers, P.C. Hohenberg and
M. Lücke, Phys. Rev. A, 32 3493 (1985).
134. A.R. Yehia, D. Jeandupeux, F. Alonso, and M.R. Guevara, Chaos, 9, 916,
1999.
135. V.S. Afraimovich and L.P. Shilnikov In Methods of Qualitatively Theory of
Differential Equations, Gorky, 3, 1983. (In Russian); English translation Amer.
Moth. Soc. Transl. (2) 149, 201 (1991).
136. G.I. Dykman, P.S. Lnada, and Yu.I. Neimark Chaos, Solitons and Fractals,
1(4) 339 (1991).
137. C. Hugenii, Horoloquim Oscilatorium, Parisiis, France, 1673.
138. A. Mayer, Proc. Gorky State University, 2, 3 (1935) (in Russian).
139. V.I. Gaponov, Zh. Tech. Fiz. (J. Tech. Phys.) 6, 5 (1936) (in Russian).
140. A.S. Bremsen and I.S. Feinberg, Zh. Tech. Fiz. (J. Tech. Phys.) 11 959 (1941)
(in Russian).
141. K.F. Teodorchik, Dokl. Akad. Nauk SSSR (Sov. Phys. Dokl.), 40, 63 (1943)
(in Russian).
142. D.W. Storti and R.H. Rand, Int. J. Non-Linear Mech., 17, 3, 143 (1982).
143. R.H. Rand and P.J. Holmes, Int. J. Non-Linear Mech., 15, 387 (1980).
144. G.B. Ermentrout, Physica D, 41, 219 (1990).
145. D.G Aroson, G.B. Ermentrout, and N. Kopell, Physics D, 41, 403 (1990.)
146. M. Poliashenko, S.R. McKay, and C.W. Smith, Phys. Rev. A, 44, 3452 (1991).
147. H.G. Schuster and P. Wagner, Prog. Theor. Phys., 81, 939 (1989).
148. D.V.R. Reddy, A. Sen, and G.L. Jhonston, Phys. Rev. Lett., 80, 5109 (1998).
149. S.H. Strogatz, Nature (London), 394, 316 (1998).
150. A. Takamatsu, T. Fujii, and I. Endo, Phys. Rev. Lett., 85, 2026 (2000).
350 References
151. M. Dhamala, V.K. Jirsa, and M. Ding, Phys. Rev. Lett., 92, 074104 (2004).
152. R.H. Rand, A.H. Cohenm, and P.J. Holmes, in Neural Control and Rhythmic
Movements in Vertebrates, A.H. Cohen, S. Grillner, and S. Rossignol, eds.,
MIT Press, Cambridge, MA, 1989.
153. E.M. Izhikevich, SIAM J. Appl. Math. 60, 1789 (2000).
154. M.V. Ivanchenko, G.V. Osipov, V.D. Shalfeev, and J. Kurths, Physica D, 189,
8 (2004).
155. T. Chacraborty and R.H. Rand, Int. J. Non-Linear Mech., 23, (1988), 369–376.
156. H. Fujisaka and T. Yamada, Prog. Theor. Phys. 69, 32 (1983).
157. V.S. Afraimovich, N.N. Verichev, and M.I. Rabinovich, Izvestia-Vysshikh-
Uchebnykh-Zavedenii-Radiofizika 29 (9), 1050 (1986).
158. L.M. Pecora and T.L. Carroll, Phys. Rev. Lett. 64, 821 (1990).
159. K. Cuomo and A.V. Oppenheim, Phys. Rev. Lett. 71, 65 (1993).
160. J.Y. Chen, K.W. Wong, L.M. Cheng, and J.W. Shuai, Chaos, 13, 508 (2003).
161. A.S. Dmitriev and A.I. Panas, Dynamical chaos: new information carrier in
communication systems, Fizmatlit, Moscow, 2002 (In Russian).
162. A.S. Pikovsky, Sov. J. Commun. Technol. Electron. 30, 85 (1985).
163. M.G. Rosenblum, A.S. Pikovsky, and J. Kurths, Phys. Rev. Lett. 76, 1804
(1996).
164. U. Parlitz et al., Phys. Rev. E 54, 2115 (1996).
165. G.V. Osipov, A.S. Pikovsky, M.G. Rosenblum, and J. Kurths, Phys. Rev. E
55, 2353 (1997).
166. M.G. Rosenblum, A.S. Pikovsky, and J. Kurths, Phys. Rev. Lett. 78, 4193
(1997).
167. N.F. Rulkov, M.M. Sushchik, L.S. Tsimring, and H.D.I. Abarbanel, Phys. Rev.
E 51, 980 (1995).
168. L. Kocarev and U. Parlitz, Phys. Rev. Lett. 76, 1816 (1996).
169. J.Y. Chen et al., Phys. Rev. E 64, 016212 (2001).
170. K.J. Lee, Y. Kwak, and T.K. Limm, Phys. Rev. Lett. 81, 321 (1998).
171. W.-H. Kye and C.-M. Kim, Phys. Rev. E 62, 6304 (2000).
172. B. Hu, G.V. Osipov, H.-L. Yang, and J. Kurths, Phys. Rev. E, 67 066216
(2003).
173. N.F. Rulkov, Chaos, 6, 262 (1996).
174. P. Ashwin, J. Buescu, and I. Stewart, Phys. Lett. A, 193, 126 (1994).
175. T. Endo, J. Franklin Institute, B 331, 1859 (1994)
176. A. Sato and T. Endo, Proc. NDES’94, 117 (1994).
177. A.K. Kozlov and V.D. Shalfeev, Proc. NDES’95, 233 (1995).
178. N. Smith, C. Crowley, and M. Kennedy, Proc. NDES’96.
179. M.V. Korzinova, V.V. Matrosov, and V.D. Shalfeev, Int. J. Bifurcation and
Chaos, 9, 963 (1999).
180. V.D. Shalfeev and G.V. Osipov, Proc. NDES’97, 139, (1997).
181. H.L. Yang, Phys. Rev. E 63, 026213 (2001).
182. V.V. Matrosov and V.D. Shalfeev, Radiophys. Quantum Electron. 41, 1033
(1998).
183. G.V. Osipov and J. Kurths, Phys. Rev. E65, 016216 (2002).
184. G.V. Osipov, A.S. Pikovsky, and J. Kurths, Phys. Rev. Lett. (2002).
185. F.C. Hoppensteadt and E.M. Izhikevich, Weakly Connected Neural Networks,
Springer-Verlag, New York, 1997.
186. N. Kopell and G.B. Ermentrout, Common. Pure Appl. Math. 39, 623 (1986).
References 351
187. G.B. Ermentrout and N. Kopell, J. Math. Biol. 29, 195 (1991).
188. H. Takana, A. Lichtenberg, and S. Oishi, Phys. Rev. Lett 78, 2104 (1997).
189. K. Tsang and K. Nagi, Phys. Rev. E 54, R3067 (1996).
190. J. Rogers and L. Wille, Phys. Rev. E 54, R2193 (1996).
191. S. Strogatz, R. Mirollo, and P. Matthews, Phys. Rev. Lett 68, 2730 (1992).
192. H. Daido, Phys. Rev. Lett 77, 1406 (1996).
193. H. Daido, Phys. Rev. Lett 78, 1683 (1997).
194. S. Watanabe, H. Zant, S. Strogatz, and T. Orlando, Physica D 97, 429 (1996).
195. H. van der Zant, T. Orlando, S. Watanabe, and S. Strogatz, Phys. Rev. Lett
74, 174 (1995).
196. A. Ustinov, M. Cirillo, and B. Malomed, Phys. Rev. B 47, 8357 (1993).
197. A. Ustinov, M. Cirillo, B. Larsen, V.A. Gbozunov, P. Carelli, and G. Rotoli
B. Malomed, Phys. Rev. B 51, 3081 (1995).
198. Z. Zheng, B. Hu, and G. Hu, Phys. Rev. E 57, 1139 (1998).
199. M. Antoni and S. Ruffo, Phys. Rev. E 52, 2361 (1995).
200. R. Fishman and D. Stroud, Phys. Rev. B 38, 290 (1988).
201. C. Yokoi, L. Thang, and W. Cou, Phys. Rev. B 37, 2173 (1988).
202. P. Liao and R.A. York, IEEE Trans. on Microwave Theory and Techniques
41, 1810 (1993).
203. T. Heath, K. Wiesenfeldt, and R.A. York, Int. J. Bifurc. and Chaos 10, 2619
(2000).
204. M. Gabbay, M.L. Larsen, and L.S. Tsimring, Phys. Rev. E 70, 066212 (2004).
205. D. Topaj and A. Pikovsky, Physica D, 170, 118 (2002).
206. G.B. Ermentrout and N. Kopell, I, SIAM J. Math. Anal. 15, 215 (1984).
207. Z. Zheng, B. Hu, and G. Hu, Phys. Rev. Let. 81, 5318 (1998).
208. Z. Zheng, B. Hu, and G. Hu, Phys. Rev. E 62, 402 (2000).
209. H.F. El-Nashar, Y. Zhang, H.A. Cerdeira, and F. Ibiyinka, Chaos 13, 1216
(2003).
210. A.V. Gaponov-Grekhov, M.I. Rabinovich, and I.M. Starobinetz, JETP Lett.,
39, 561 (1984) [in Russian].
211. G.V. Osipov, Radiophysics 31, 624 (1988) [in Russian].
212. I.S. Aranson, A.V. Gaponov-Grekhov, M.I. Rabinovich, and I.M. Starobinetz,
JETP 90, 1707 (1986) [in Russian].
213. V.S. Anishchenko, I.S. Aranson, D.E. Postnov, and M.I. Rabinovich, Acad. of
Sci. USSR Papers 286, 1120 (1986) [in Russian].
214. V.S. Anishchenko, D.E. Postnov, and M.A. Safonova, JETP Lett., 11, 1505
(1985) [in Russian].
215. S.P. Kuznetsov and A.S. Pikovsky, Radiophysics 28, 308 (1985) [in Russian].
216. V.S. Afraimovich, M.I. Rabinovich, and V.I. Sbitnev, JETP Lett., 11, 338
(1985) [in Russian].
217. I.S. Aranson, A.V. Gaponov-Grekhov, and M.I. Rabinovich, Physica D, 33, 1
(1988).
218. D. Ruelle and F. Takens, Comm. Math. Phys., 20, 167 (1979).
219. S. Newhouse, D. Ruelle, and F. Takens, Comm. Math. Phys., 64, 35 (1979).
220. A.J. Lichtenberg and M.A. Liberman, Regular and stochastic motion,
(Springer-Verlag, New York - Heidelberg - Berlin), 1983.
221. A.T. Winfree and J. Theoret. Biol. 16, 15 (1967).
222. Y. Kuramoto, in Proc. of Int. Symp. on Mathematical Problems in Theoretical
Physics, H. Araki, ed., Lecture Notes in Physics 39 (Springer, New York,
1975).
352 References
223. S.H. Strogatz, D.M. Abrams, A. McRobie et al., Nature 438, 7064 (2005).
224. K. Lonngren and A. Scott (eds.) Solitons in Action. (Acad. Press, New York,
San Francisco, London) (1978).
225. J.S. Eilbeck, P.S. Lomdahl, O.H. Olsen, and M.R. Samuelsen, J. of Appl.
Phys., 57, 861 (1985).
226. O.H. Olsen, P.S. Lomdahl, A.R. Bishop, and I.S. Eilbeck, J. Phys. C.: Solid
State Phys. 18, 54 (1985)
227. A.R. Bishop, M.G. Forest, D.W. McLaughlin, and E.A. Overmann, Physica
D, 23, 293 (1986).
228. A.R. Bishop and P.S. Lomdahl, Physica D, 18, 54 (1986).
229. P.M. Marcus and Y. Imry Solid. State Comm. 33, 345 (1980).
230. L.Sh. Tsimring, Phys. Rev. E 48, 3446 (1993).
231. L.N. Korzinov and M.I. Rabinovich, Izv. VUZov, Applied Nonlin. Dynam. 2,
59 (1994) [in Russian].
232. M. Bazhenov, M. Rabinovich, and L. Rubchinsky, J. Stat. Phys. 83, 1165
(1996).
233. Y. Braiman, J.F. Lindner, and W.L. Ditto, Nature 378, 465 (1995).
234. Y. Braiman, W.L. Ditto, K. Wiesenfeld, and M.L. Spano, Phys. Lett. A 206,
54 (1995).
235. N.E. Diamant, P.K. Rose, and E.J. Davidson, Amer. J. Physiol. 219, 1684
(1970).
236. A.B. Zobnin, M.I. Rabinovich, and M.M. Sushchik, Izv. Akad. Nauk SSSR,
ser. Fizika Atmosfery i Okeana 26, 1289 (1990) [in Russian].
237. S.K. Sarna, E.E. Daniel, and Y.I. Kinoma, Amer. J. Physiol. 221, 166 (1971).
238. D. Robertson-Dunn and D.A. Linkens, Med. Biol. Engng. 12, 750 (1974).
239. B.H. Brown, H.L. Duthie, A.R. Horn, and R.H. Smallwood, Amer. J. Physiol.
229, 384 (1975).
240. R.J. Patton and D.A. Linkens, Med. Biol. Engng. Computing 16, 195 (1978).
241. S.D. Drendel, N.P. Hors, and V.A. Vasiliev, in Dynamics of Cell Popula-
tions (Nizhny Novgorod University Press, Nizhny Novgorod, 1984), p. 108
[in Russian].
242. V.A. Vasiliev, Yu.M. Romanovsky, and V.G. Yakhno, Autowave Processes
(Nauka, Moscow, 1987) [in Russian].
243. Y. Yamaguchi and H. Shimizu, Physica D 11, 212 (1984).
244. K. Bar-Eli, Physica D 14, 242 (1985).
245. G.B. Ermentrout, in Nonlinear Oscillations in Biology and Chemistry. Lecture
Notes in Biomathematics, Vol. 66, ed. H. Othmer (Springer, Berlin, 1986).
246. G.B. Ermentrout and W.C. Troy, SIAM J. Appl. Math. 39, 623 (1986).
247. I.S. Aronson and L. Kramer, Rev. Mod. Phys. 74, 99 (2002).
248. D. Somers and N. Kopell, Physica D 89, 169 (1995).
249. T.E. Vadivasova, G.I. Strelkova, and V.S. Anishchenko, Phys. Rev. E 63,
036225 (2001).
250. D. Hansel and H. Sompolinsky, Phys. Rev. Lett. 68, 718 (1992).
251. D. Hansel, Int. J. Neural Sys. 7, 403 (1996).
252. F. Pasemann, Physica D 128, 236 (1999).
253. P. Varona, J.J. Tores, H.D.I. Abarbanel, M.I. Rabinovich, and R.C. Elson,
Biol. Cybern. 84, 91 (2001).
254. G. Balazsi, A. Cornell-Bell, A.B. Neiman, and F. Moss, Phys. Rev. E 64,
041912 (2001).
References 353
255. Y. Horio, T. Taniguchi, and K. Aihara, Neurocomputing 64, 447 (2005).

256. H. Nakano and T. Saito, IEEE Trans. on neural networks 15, 231 (2004).
257. K.S. Fink, G. Johnson, T. Carrol, and L. Pecora, Phys. Rev. E 61, 5080 (2000).
258. A.C.H. Rowe and P. Etchegoin, Phys. Rev. E 64, 031106 (2001).
259. E.R. Hunt, P.M. Gade, and N. Mousseau, Europhys. Lett. 60, 827 (2002).
260. H.G. Winful and L. Rahman, Phys. Rev. Lett. 65, 1575 (1990).
261. E. Mozly, T.C. Newell, P.M. Alsing, V. Kovanis, and A. Gavrielides, Phys.
Rev. Lett. 51, 5371 (1995).
262. R. Roy and K.S. Thornburg, Phys. Rev. Lett. 72, 2009 (1994).
263. F. Rogister, K.S. Thornburg, L. Fabiny et al., Phys. Rev. Lett. 92, 093905
(2000).
264. F. Rogister and R. Roy, Laser Physics, 15, 313 (2005).
265. I.Z. Kiss and J.L. Hudson, Phys. Chem. Chem. Phys. 4, 2638 (2003).
266. I.Z. Kiss, W. Wang, and J.L. Hudson, Phys. Chem. Chem. Phys. 2, 3847
(2000).
267. J.F. Heagy, L.M. Pecora, and T.L. Carroll, Phys. Rev. Lett. 74, 4185 (1994).
268. L.M. Pecora, Phys. Rev. E 58, 347 (1998).
269. L. Pecora, T. Carroll, G. Johnson, D. Mar, and K.S. Fink, Int. J. of Bif. and
Chaos 10, 273 (2000).
270. V.N. Belykh and E. Mosekilde, Phys. Rev. E, 54, 3196, (1996).
271. M. Hasler, Yu. Maistrenko, and O. Popovich, Phys. Rev. E, 58, 6843, (1998).
272. V. Belykh, I. Belykh, and M. Hasler, Phys. Rev. E, 62, 6332, (2000).
273. H.L. Yang, Phys. Rev. E 64, 026206 (2001).
274. G. Hu, Y. Zhang, H.A. Cerdeira, and S. Chen, Phys. Rev. Lett. 85, 3377
(2000).
275. X. Zhang, M. Fu, J. Xiao, and G. Hu, Phys. Rev. E 74, 015202(R) (2006).
276. L. Brunnet, H. Chatè, and P. Mannevile, Physica D 78, 141 (1994).
277. A. Goryachev and R. Kapral, Phys. Rev. Lett. 76, 1619 (1996).
278. K. Wiesenfeldt, P. Colet, and S.H. Strogatz, Phys. Rev. Lett. 76, 404 (1996).
280. M. Thiel, M.C. Romano, and J. Kurths, Phys. Lett. A 330, 343 (2004).
281. B. Blasius and L. Stone, Int. J. Bif. and Chaos, 10, 2361 (2000).
282. M.K. Schwartz, L.S. Mills, K.S. McKelvey, L.F. Ruggiero, L.F., and
F.W. Allendorf, Nature 415, 520 (2002)
283. J. Vandermeer, L. Stone, and B. Blasius, Chaos, Solitons and Fractals 12, 265
(2001).
284. B. Blasius, E. Montbrió, and J. Kurths, Phys. Rev. E 67, 035204 (2003).
285. R.M. May, Nature, 261, 459 (1976).
286. R.F. Costantino, J.M. Cushing, B. Dennis, and R.A. Desharnais, Nature, 375,
227 (1995).
287. J. Wisdom, Astron. J., 87, 577 (1982).
288. M. Henon, Commun. Math. Phys., 50, 69 (1976).
289. J.F. Heagy, Physica D, 57, 436 (1992).
290. K. Ikeda, H. Daido, and O. Akimoto, Phys. Rev. Lett., 45, 709 (1980).
291. N.F. Rulkov, I. Timofeev, and M. Bazhenov, J. Neurophys. 76, 231 (2004).
292. R. Kapral, in Theory and Applications of Coupled Map Lattices, (Ref. [1]),
Chap. 5, p. 135.
293. F. Daviaud, M. Dubois, and P. Berge, Europhys. Lett. 9, 441 (1989).
294. S. Ciliberto and P. Bigazzi, Phys. Rev. Lett. 60, 286 (1988).
354 References
295. S. Bottin, F. Daviaud, O. Dauchot, and P. Manneville, Europhys. Lett. 43,

171 (1998).
296. M.M. Degen, I. Mutabazi, and C.D. Andereck, Phys. Rev. E 53, 3495 (1996).
297. G. Colovas and C.D. Andereck, Phys. Rev. E 55, 2736 (1997).
298. Theory and Applications of Coupled Map Lattices, edited by K. Kaneko (Wiley,
1993), and references therein.
299. H. Chate, Nonlinearity 7, 185 (1994).
300. M.G. Zimmermann, R. Toral, O. Piro, and M. San Miguel, Phys. Rev. Lett.
85, 3612 (2000).
301. Here boundary conditions do not allow synchronization on the maximal indi-
vidual frequency.
302. R.M. Harris-Warrick et al., in Dynamics Biological Networks: The Stomato-
gastric Nervous system, edited by R.M. Harris-Warrick et al. (MIT Press,
Cambrige, MA, 1992).
303. See A. Szücs et al., NeuroReport 11, 563 (2000) and references therein.
304. H. Sakaguchi, Prog. Theor. Phys., 79, 39 (1988).
305. S.H. Strogatz, Physica D, 143, 1 (2000).
306. K. Kaneko, “Theory and application of coupled map lattices”, Wiley,
Chichester, 1993.
307. K. Kaneko, Physica D 75, 55, (1994).
310. B. Hu and Z. Liu, Phys. Rev. E 62, 2114, (2000).
311. F.S. de San Roman, S. Boccaletti, D. Maza, and H. Mancini, Phys. Rev. Lett.
81, 3639, (1998).
312. G.S. Gil and S.C. Gupta, IEEE Trans. Commun. 20, 454, (1972).
313. M. de Sousa Vieira, A.J. Lichtenberg, and M.A. Lieberman, Int. J. of Bif. and
Chaos 1, 691, (1994).
314. N.F. Rulkov and A.R. Volkovskii, Phys. Lett. A 173, 332, (1993).
315. A.R. Volkovskii, IEEE Trans. Circuits Syst. 44, 913, (1997).
316. Y. Kuramoto, Physica D 50, 15, (1991).
317. A. Treves, Network 4, 259, (1993).
318. M. Usher, H.G. Schuster, and E. Niebur, Neural Comp. 5, 570, (1993).
319. M. Tsodyks, I. Mitkov, and H. Sompolinsky, Phys. Rev. Lett. 71, 1280, (1995).
320. T.L. Carroll, Biol. Cybern 73, 553, (1995).
321. C.J. Perez Vicente, A. Arenas, and L.L. Bonilla, J. Phys. A 29, L9, (1996).
322. A. Diaz-Guilera, C.J. Perez, and A. Arenas, Phys. Rev. E 57, 3820, (1998).
323. T.D. Frank, A. Daffertshofer, C.E. Peper, P.J. Beek, and H. Haken Physica D
144, 62, (2000).
324. V. Torre, J. Theor. Biol. 61, 55, (1976).
325. J. Jalife, J. Physiol. 356, 221, (1984).
326. D.C. Michaels, E.P. Matyas, and J. Jalife, Circulation Res. 61, 704, (1987).
327. T.L. Carroll, J. Heagy,and L.M. Pecora, Phys. Lett. A 186, 225, (1994).
328. L. Pecora and T. Carroll, Phys. Rev. Lett., 80, 2109, (1998).
329. W.C. Lindsey, “Synchronization systems in communication and control”,
(Prentice-Hall, Englewood Cliffs, New Jersey), 1972.
330. J. Jost and M.P. Joy, Phys. Rev. E 65, 016201 (2001).
331. G.V. Osipov and M.M. Sushchik, Phys. Lett. A 201, 205, (1995).
332. V.N. Belykh, G.V. Osipov, N. Kuckländer, and J. Kurths, Physica D 200 81
(2005).
References 355
333. L. Fabiny et al., Phys. Rev. A 47, 4287 (1993).

334. R.A. Oliva and S.H. Strogatz, Int. J. Bifurcation Chaos Appl. Sci. Eng. 11,
2359 (2001).
335. Y. Aizawa, Prog. Theor. Phys 56, 703 (1976).
336. D. Christiansen, P. Alstrom, and M.T. Levinsen, Physica D 56, 23 (1992).
337. R. Eckhorn et al., Biol. Cybern. 60, 121 (1987).
338. J. Guckenheimer and P. Holmes Nonlinear Oscillations, Dynamical Systems,
and Bifurcations of Vector Fields (Springer, Berlin, Heidelberg, 1990).
339. J.D. Murray Mathematical Biology (Springer, Berlin, Heidelberg, 1989).
340. H. Haken Synergetics. An Introduction (Springer, Berlin, Heidelberg, 1977).
341. M. Gorman, P.J. Widman, and K.A. Robbins, Phys. Rev. Lett. 52, 2241
(1984).
342. M. Rabinovich et al., Phys. Rev. Lett. 87, 068102 (2001).
343. J. Karbowski and G.B. Ermentrout, Phys. Rev. E 65, 031902 (2002).
344. H. Haken, J.A.S. Kelso, and H. Bunz, Biol. Cybern 51, 347 (1985).
345. A.H. Cohen, P.J. Holmes, and R.H. Rand, J. Math. Biol. 13, 345 (1982).
346. P. Tass and H. Haken, Z. Phys. B 100, 303 (1996).
347. A.W. Hübler and E. Luscher, Naturwissenschaft 76, 67 (1989).
348. E.A. Jackson, Phys. Lett. A 151, 478 (1990)
349. E. Ott, C. Grebogy, and J.A. Jorke, Phys. Rev. Lett. 64, 1196 (1990).
350. S. Boccaletti, C. Grebogy, Y.-C. Lai, H. Mancini, and D. Maza, Physics Report
329, 103 (2000).
351. G.V. Osipov and M.M. Sushchik, Izv. Vuzov. Prikladnaya Nelineinaya
Dinamika 2, 24 (1994) [In Russian].
352. G.V. Osipov and M.M. Sushchik, Bulletin of the Russian Academy of Science.
Physics. Supplement, Physics of Vibrations 61, 41 (1997).
353. L.L. Rubchinsky, M.M. Sushchik, and G.V. Osipov, Mathematics and Com-
puters and Simulation 58, 443 (2002).
354. K. Okuda and Y. Kuramoto, Progr. Theor. Phys. 86, 1159 (1991).
355. M.B. Mineev, G.S. Mkrtchyan, and V.V. Schmidt, J. Low-Temp. Phys. 45,
497 (1981).
356. T. Holst, J.B. Hansen, N. Groonbech–Jensen, and J.A. Blackburn, Phys. Rev.
B. 42, 127 (1990).
357. S. Continnho, P. Pitanga, and P. Lederer, Phys. Rev. B 23, 4567 (1981).
358. O.M. Braun, Y.U.S. Kivshar, and A.N. Kosevich, J. Phys. C: Solid State Phys.
21, 3881 (1988).
359. J. Pouget and G.A. Maugin, Physica D 14, 88 (1984).
360. G.A. Maugin and A. Miled, Phys. Rev. B 33, 4830 (1986).
361. L. Junge and U. Parlitz, Phys. Rev. E 62, 438 (2000).
362. J. Bragard, F.T. Arecci, and S. Boccaletti, Int. J. of Bif. and Chaos, 10, 2381
(2000).
363. J. Bragard and S. Boccaletti, Phys. Rev. E 62, 6346 (2000).
364. H. Chaté, A. Pikovsky, and O. Rudzick, Physica D 131, 17 (1999).
365. M. Zhan, Z.G. Zheng, G. Hu, and X.H. Peng, Phys. Rev. E, 62, 3552, (2000).
366. C.H. Gray, P. König, A.K. Engel, and W. Singer, Nature 338, 334 (1989).
367. B.I. Shraiman, A. Pumir, W. van Saarloos, P.C. Hohenberg, H. Chaté, and
M. Holen, Physica D 57, 241 (1992).
368. H. Chaté, Nonlinearity, 7, 185 (1994).
369. J. Lega, Eur. J. Mech. B: Fluids 10, N 2 – Suppl. (1991) 145.
356 References
370. Y. Braiman, W.L. Ditto, K. Wiesenfeld, and M.L. Spano, Phys. Lett. A 206,
54 (1995).
371. N. Mousseau, Phys. Rev. Lett. 77, 968 (1996).
373. J.F. Lindner, B.S. Prusha, and K.E. Klay, Phys. Lett. A 231, 164 (1997).
374. A. Gavrielides, T. Kottos, V. Kovanis, and G.P. Tsironis, Phys. Rev. E 58,
5529 (1998).
375. J.F. Hirsch, B.A. Huberman, and D.J. Scalpiano, Phys. Rev. A 25, 519 (1982).
376. S.C. Venkataramani, B.R. Hunt, and E. Ott, Phys. Rev. E 54, 1346 (1996).
377. W. Wiesenfeld and F. Moss, Nature 373, 33 (1995).
378. A.R. Bulsara and L. Gammaitoni, Phys. Today 49, 39 (1996).
379. L. Gammaitoni, P. Hanggi, P. Jung, and F. Marchesoni, Rev. Mod. Phys. 70,
223 (1998).
380. V.S. Anishchenko, A.B. Neiman, F. Moss, and L. Schimansky-Geier, Physics–
Uspekhi 42, 7 (1999).
381. J.F. Lindner, B.K. Meadows, W.L. Ditto, M.E. Inchiosa, and A.R. Bulsara,
Phys. Rev. E 53, 2081 (1996).
382. M. Locher, G.A. Johnson, and E.R. Hunt, Phys. Rev. Lett. 77, 4698 (1996).
Phys. Rev. E 58, 2843 (1998).
383. J.F. Lindner, S. Chandramouli, A.R. Bulsara, M. Locher, and W.L. Ditto,
Phys. Rev. Lett. 81, 5048 (1998).
384. M. Locher, D. Cigna, and E.R. Hunt, Phys. Rev. Lett. 80, 5212 (1998).
385. S. Kadar, J. Wang, and K. Showalter, Nature 391, 770 (1998).
386. P. Jung and G. Mayer-Kress, Phys. Rev. Lett. 74, 2130 (1995).
387. H. Hempel, L. Schimansky-Geier, and J. Garcia-Ojalvo, Phys. Rev. Lett. 82,
3713 (1999).
388. A. Zaikin, J. Garcia-Ojalvo, L. Schimansky-Geier, and J. Kurths, Phys. Rev.
Lett. 88, 010601 (2002).
389. E.I. Volkov, E. Ullner, A.A. Zaikin, and J. Kurths, Phys. Rev. E 68, 026214
(2003).
390. H. Zhonghuai, Y. Lingfa, X. Zuo, and X. Houwen, Phys. Rev. Lett. 81, 2854
(1998).
391. M.A. Santos and J.M. Sancho, Phys. Rev. E 59, 98 (1999).
392. A.T. Stamp, G.V. Osipov, and J.J. Collins, Chaos, 12, 931 (2002).
393. Chaos, Focus Issue: Mapping and Control of Complex Cardiac Arrhythmias,
12, Editors: L. Glass and D. Christini, 3 (2002).
394. American Heart Association, Heart Disease and Stroke Statistics-2003 Update
(American Heart Association, Dallas, 2002).
395. D.L. Hayes, M.A. Lloyd, and P.A. Friedman, Cardiac Pacing and Defibrilla-
tion: A Clinical Approach (Futura Pub Co, 2000.)
396. J. Jalife and R.A. Gray, Acta. Physiol. Scand., 157, 123 (1996)
397. J. Jalife, R.A. Gray, G.E. Morley, and J.M. Davidenko, Chaos, 8, 79 (1998).
398. A.V. Panfilov, Chaos, 8, 57 (1998).
399. F.X. Witkowski, L.J. Leon, P.A. Penkoske, W.R. Giles, M.L. Spano, W.L.
Ditto, and A.T. Winfree, Nature, 392, 78 (1998).
400. L. Tung, Proc. IEEE., 84, 366 (1996).
401. O. Steinbock, J. Schütze, and S.C. Müller, Phys. Rev. Lett., 68, 248 (1992).
402. V.N. Biktashev and A.V. Holden, Chaos Solitons Fractals, 5, 575 (1995).
403. R.-M. Mantel and D. Barkley, Phys. Rev. E, 54, 4791 (1996).
References 357
404. S. Grill, V.S. Zykov, and S.C. Müller, Phys. Rev. Lett., 75, 3368 (1995).
405. V.S. Zykov, A.S. Mikhailov, and S.C. Müller, Phys. Rev. Lett., 78, 3398
(1997).
406. A.M. Pertsov, J.M. Davidenko, R. Salomonsz, W.T. Baxter, and J. Jalife,
Circ. Res., 72, 631 (1993).
407. G.V. Osipov, B.V. Shulgin, and J.J. Collins, Phys. Rev. E, 58, 6955 (1998).
408. G.V. Osipov and J.J. Collins, Phys. Rev. E, 60, 54 (1999).
409. M. Allessie, C. Kirchhof, G.J. Scheffer, F. Chorro, and J. Brugada, Circulation,
84, 1689 (1991).
410. A. Capucci, F. Ravelli, G. Nollo, A.S. Montenero, M. Biffi, and G.Q. Villani,
J. Cardiovasc. Electrophysiol., 10 319 (1999).
411. E.G. Daoud, B. Pariseau, M. Niebauer, F. Bogun, R. Goyal, M. Harvey,
K.C. Man KC, S.A. Strickberger, and F. Morady, Circulation, 94, 1036 (1996).
412. J.M. Kalman, J.E. Olgin, M.R. Karch, and M.D. Lesh, J. Cardiovasc. Elec-
trophysiol., 7, 867 (1996).
413. C. Kirchhof, F. Chorro, G.J. Scheffer, J. Brugada, K. Konings, Z. Zetelaki,
and M. Allessie, Circulation, 88, 736 (1993).
414. B.H. KenKnight, P.V. Bayly, R.J. Gerstle, D.L. Rollins, P.D. Wolf, W.M.
Smith, and R.E. Ideker, Circ. Res., 77, 849 (1995).
415. K.J. Lee, Phys. Rev. Lett. 79, 2907 (1997).
416. R. Mantel and D. Barkley, Phys. Rev. E, 54, 4791 (1996).
417. A.T. Winfree, Physica D, 49, 125 (1991).
418. F. Xie, Z. Qu, J.N. Weiss, and A. Garfinkel, Phys. Rev. E, 59, 2203 (1999).
419. G. Gottwald, A. Pumir, and V. Krinsky, Chaos, 11, 487 (2001).
420. A.L. Bassett, S. Chakko, and M. Epstein, J. Hypertens., 15, 915 (1997).
421. T.R. Chay, J. Electrocardiol., 28, 191 (1995).
422. V.I. Krinsky and K.I. Agladze, Physica D, 8, 50 (1983).
423. V.S. Zykov, Ann. N.Y. Acad. Sci., 591, 75 (1990).
424. A.T. Winfree, Ann. N.Y. Acad. Sci., 591, 190 (1990).
425. J.J. Tyson and J.P. Keener, Physica D, 32, 327 (1988).
426. A.N. Zaikin and A.M. Zhabotinsky, Nature, 225, 535 (1970).
427. M.R. Guevara, A. Shrier, and L. Glass, Am. J. Physiol., 254, H1 (1988).
428. T.J. Lewis and M.R. Guevara J. Theor. Biol., 146, 407 (1990).
429. Z. Qu and A. Garfinkel, IEEE Trans. Biomed. Eng., 46, 1166 (1999).
430. S. Rush and H. Larsen, IEEE Trans. Biomed. Eng., 25, 389 (1978).
431. Z. Qu, J.N. Weiss, and A. Garfinkel, Phys. Rev. E 61, 727 (2000).
432. P.Q. Anderson, J.E. Knoben, and W.G. Troutman, in Handbook of Clinical
Drug Data. 9 th ed. Stamford, CT: Appleton & Lange; 345 (1999).
433. S.C. Vlay, in A Practical Approach to Cardiac Arrhythmias. 2nd ed. Boston,
MA: Little, Brown & Company; 309 (1996).
434. N. Marwan, N. Wessel, U. Meyerfeldt, A. Schirdewan, and J. Kurths, Phys.
Rev. E 66, 026702 (2002).
435. L. Gammaitoni, P. Hanggi, P. Jung, and F. Marchesoni, Rev. Mod. Phys. 70,
223 (1998).
436. A. Neiman, A. Silchenko, V. Anishchenko, and L. Schimansky-Geier, Phys.
Rev. E 58, 7118 (1998).
437. Z.F. Mainen and T.J. Sejnowski, Science 268, 1503 (1995).
438. O.N. Bjornstad, R.A. Ims, and X. Lambin, Trends Ecol. Evol. 14, 427 (1999).
439. P.J. Hudson and I.M. Cattadori, Trends Ecol. Evol. 14, 1 (1999).
358 References
440. F. Jeltsch and C. Wissel, Ecol. Modelling, 75/76, 111 (1994).

441. A.S. Pikovsky, in Nonlinear and Turbulent Processes in Physics, edited by R.Z.
Sagdeev (Harwood Academic Publisher, New York, 1984), pp. 1601–1604.
442. A.S. Pikovsky, Radiophys. Quantum Electron. 27, 576 (1984).
443. R.V. Jensen, Phys. Rev. E 58, R6907 (1998).
444. K. Matsumoto and I. Tsuda, J. Stat. Phys. 31, 87 (1983).
445. H.P. Herzel and W. Ebeling, Phys. Lett. A 111, 1 (1985).
446. H.P. Herzel and B. Pompe, Phys. Lett. A 122, 121 (1987).
447. L. Yu, E. Ott and Q. Chen, Phys. Rev. Lett. 65, 2935 (1990).
448. A.S. Pikovsky, Phys. Lett. A 165, 33 (1992).
449. A. Mritan and J.R. Banavar, Phys. Rev. Lett. 72, 1451 (1994); ibid. 73, 2932
(1994).
450. A.S. Pikovsky, Phys. Rev. Lett. 73, 2931 (1994).
451. L. Longa, E.M.F. Curado, and A.A. Oliveira, Phys. Rev. E 54, R2201 (1996).
452. H. Herzel, and J. Freund, Phys. Rev. E 52, 3238 (1995).
453. P.M. Gade, and C. Basu, Phys. Lett. A 217, 21 (1996).
454. E. Sanchez, M.A. Matias, and V. Perez-Munuzuri, Phys. Rev. E 56, 4068
(1997).
455. C.-H. Lai, and C.S. Zhou, Europhys. Lett. 43, 376 (1998).
456. R. Toral, C.R. Mirasso, E. Hernandez-Garcia, and O. Piro, Chaos 11, 665
(2001).
457. C.S. Zhou and J. Kurths, Phys. Rev. Lett. 88, 230602 (2001)
458. L. Arnold, Random Dynamical Systems, (Springer, Berlin Heidelberg 1998).
459. C.S. Zhou and T.L. Chen, Neurocomputing 30, 293 (2000).
460. C.S. Zhou and C.-H. Lai, Physica D 135, 1 (2000).
461. N. Platt, E.A. Spiegel, and C. Tresser, Phys. Rev. Lett. 70, 279 (1993).
462. A.L. Hodgkin and A.F. Huxley, J. Physiol. (London) 117, 500 (1952).
463. H.A. Braun, H. Bade, and H. Hensel, Pflugers Arch, Ges. Physiol. Menschen
Tiere 386, 1 (1980).
464. H.A. Braun, M.T. Huber, M. Dewald, K. Schäfer, and K. Voigt, Int. J. Bifur-
cation Chaos, 8, 881 (1998)
465. U. Feudel, A. Neiman, X. Pei, W. Wojtenek, H.A. Braun, M. Huber and
F. Moss, Chaos, 10, 231 (2000).
466. L.P. Shilnikov, Math. USSR Sb. 10, 91 (1970).
467. E.M. Izhikevich, Int. J. Bifurcation Chaos, 10, 1171 (2000).
468. C.S. Zhou and J. Kurths, Chaos, 13, 401 (2003).
469. E. Allaria, F.T. Arecchi, A. Di Garbo, and R. Meucci, Phys. Rev. Lett. 86,
791 (2001).
470. A.N. Pisarchik, R. Meucci, and F.T. Arecchi, Eur. Phys. J. D 13, 385 (2001).
471. C.S. Zhou, J. Kurths, E. Allaria, S. Boccaletti, R. Meucci and F.T. Arecchi,
Phys. Rev. E, 67, 066220 (2003).
472. A. Uchida, R. McAllister, and R. Roy, Phys. Rev. Lett. 93, 244102 (2004).
473. R.L. Stratonovich, Topics in the Theory of Random Noise (Gordon and
Breach, New York, 1963)
474. P. Tass , M.G. Rosenblum, J. Weule, J. Kurths, A. Pikovsky, J. Volkmann, A.
Schnitzler, and H.-J. Freund, Phys. Rev. Lett. 81, 3291 (1998).
475. J. Garcia-Ojalvo and J.M. Sancho, Noise in Spatially Extended Systems
(Springer, New York, 1999).
476. C.S. Zhou, J. Kurths, E. Allaria, S. Boccaletti, R. Meucci, and F.T. Arecchi,
Phys. Rev. E, 67, 015205R (2003).
References 359
477. M. Zaks, E.H. Park, M. Rosenblum, and J. Kurths, Phys. Rev. Lett. 82, 4228
(1999).
478. C.S. Zhou, J. Kurths, I.Z. Kiss, and J.L. Hudson, Phys. Rev. Lett. 89, 014101
(2002).
479. E. Ott, Chaos in Dynamical Systems (University Press, Cambridge, 1993).
480. I.Z. Kiss and J.L. Hudson, Phys. Rev. E 64, 046215 (2001).
481. I.Z. Kiss, Y. Zhai, J.L. Hudson, C.S. Zhou and J. Kurths, Chaos, 13, 267
(2003).
482. B. Lindner, J. Garcia-Ojalvo, A. Neiman, and L. Schimansky-Geier 2004 Phys.
Rep. 392 321–424.
483. A.S. Pikovsky and J. Kurths, Phys. Rev. Lett. 78, 775 (1997).
484. B. Hu and C.S. Zhou, Phys. Rev. E 63, 026201 (2001).
485. C.S. Zhou, J. Kurths, and B. Hu, Phys. Rev. Lett. 87, 098101 (2001).
486. A. Neiman, L. Schimansky-Geier, A. Cornell-Bell, and F. Moss, Phys. Rev.
Lett. 83, 4896 (1999).
487. J. Balakrishnan, Phys. Rev. E 73, 036206 (2006).
488. C.S. Zhou, J. Kurths, and B. Hu, Phys. Rev. E 67, 030101(R) (2003).
489. V. Petrov, Q. Ouyang, and H.L. Swinney, Nature (London) 388, 655 (1997).
490. A.L. Lin, M. Bertram, K. Martinez, and H.L. Swinney, A. Ardelea and
G.F. Carey, Phys. Rev. Lett. 84, 4240 (2000).
491. V.K. Vanag, L. Yang, M. Dolnik, A.M. Zhabotinsky, and I.R. Epstein, Nature
(London) 406, 389 (2000).
492. M. Kim, M. Bertram, M. Pollmann, A. von Oertzen, A.S. Mikhailov, H.H.
Rotermund, and G. Ertl, Science 292, 1357 (2001).
493. C.S. Zhou and J. Kurths, Phys. Rev. E 69, 056210 (2004).
494. C.S. Zhou and J. Kurths, New J. Phys. 7, 18 (2005).
495. R. Albert and A.-L. Barabási, Rev. Mod. Phys. 74, 47 (2002).
496. P. Erdös and A. Rényi, Publ. Math. Inst. Hung. Acad. Sci. 5, 17 (1959).
497. D.J. Watts and S.H. Strogatz, Nature (London) 393, 440 (1998).
498. A.-L. Barabási and R. Albert, Science 286, 509 (1999).
499. H. Jeong, B. Tombor, R. Albert, Z.N. Oltavi, and A.-L. Barabási, Nature 407,
651 (2000).
500. M.E.J. Newman, Proc. Natl Acad. Sci. USA 98, 404 (2001).
501. L.A.N. Amaral, A. Scala, M. Barthélémy, and H.E. Stanley, Proc. Natl Acad.
Sci. USA 97, 11149 (2000).
502. O. Sporns, G. Tononi, and G.M. Edelman, Cereb. Cortex 10, 127 (2000).
503. D.J. Watts, Small Worlds (Princeton University Press, Princeton, 1999).
504. S.H. Strogatz, Nature (London) 410, 268 (2001).
505. S.N. Dorogovtsev and J.F.F. Mendes, Adv. Phys. 51, 1079 (2002).
506. M.E.J. Newman, SIAM Rev. 45, 167 (2003).
507. R. Albert, H. Jeong, and A.-L. Barabási, Nature (London) 406, 378 (2000).
508. D.S. Callaway, M.E.J. Newman, S.H. Strogatz, and D.J. Watts, Phys. Rev.
Lett. 85, 5468 (2000).
509. R. Chen, K. Erez, D. ben Avraham, and S. Havlin, Phys. Rev. Lett. 85, 4626
(2000).
510. R. Pastor-Satorras and A. Vespignani, Phys. Rev. Lett. 86, 3200 (2001).
511. R.M. May, and A.L. Lloyd, Phys. Rev. E 64, 066112 (2001).
512. M.E.J. Newman, Phys. Rev. E 64, 016128 (2002).
513. A.E. Motter, Phys. Rev. Lett. 93, 098701 (2004).
360 References
514. S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, and D.-U. Hwang, Phys. Rep.
424, 175 (2006).
515. L.F. Lago-Fernández, R. Huerta, F. Corbacho, and J.A. Sigüenza, Phys. Rev.
Lett. 84, 2758 (2000).
516. N. Masuda and K. Aihara, Biol. Cybern. 90, 302 (2004).
517. X. Guardiola, A. Diaz-Guilera , M. Llas, and C. J. Perez, Phys. Rev. E 62,
5565 (2000).
518. H. Hasegawa, Phys. Rev. E 70, 066107 (2004).
519. H. Hong, M. Y. Choi, and B.J. Kim, Phys. Rev. E 65, 026139 (2002).
520. A.M. Batista, S.E.D. Pinto, R.L. Viana, and S.R. Lopes, Physica A 322, 118
(2003).
521. P.M. Gade and C.K. Hu, Phys. Rev. E 62, 6409 (2000).
522. J. Jost and M.P. Joy, Phys. Rev. E 65, 016201 (2001).
523. M. Barahona and L.M. Pecora, Phys. Rev. Lett. 89, 054101 (2002).
524. S. Sinha, Phys. Rev. E 66, 016209 (2002).
525. I. Belykh, V. Belykh, and M. Hasler, Physica D 195, 159 (2004).
526. T. Nishikawa, A.E. Motter, Y.-C. Lai, and F.C. Hoppensteadt, Phys. Rev.
Lett. 91, 014101 (2003).
527. F. Chung and L. Lu, Proc. Natl. Acad. Sci. U.S.A. 99, 15879 (2002).
528. R. Cohen and S. Havlin, Phys. Rev. Lett. 90, 058701 (2003).
529. L.M. Pecora and T.L. Carroll, Phys. Rev. Lett. 80, 2109 (1998).
530. M.E.J. Newman, C. Moore, and D.J. Watts, Phys. Rev. Lett. 84, 3201 (2000).
531. M.E.J. Newman, S.H. Strogatz, and D.J. Watts, Phys. Rev. E 64, 026118
(2001).
532. S. De Monte, F. d’Ovidio, and E. Mosekilde, Phys. Rev. Lett. 90, 054102
(2003).
533. S. Jalan and R.E. Amritkar, Phys. Rev. Lett. 90, 014101 (2003).
534. S. Jalan and R.E. Amritkar, Physica A 321, 220 (2003).
535. S. Jalan and R.E. Amritkar, Physica A 346, 13 (2005).
536. L.G. Morelli, H.A. Cerdeira, and D.H. Zanette, Eur. Phys. J. B 43, 243 (2005)
537. I. Stewart, Nature (London) 427, 601 (2004).
538. C.S. Zhou and J. Kurths, Chaos 16, 015104 (2006).
539. A. Barrat, M. Barthélemy, R. Pastor-Satorras, and A. Vespignani, Proc. Natl.
Acad. Sci. U.S.A. 101, 3747 (2004).
540. S.H. Yook, H. Jeong, A.-L. Barabási, and Y. Tu, Phys. Rev. Lett. 86, 5835
(2001).
541. M.E.J. Newman, Phys. Rev. E 64, 016132 (2001).
542. L.A. Braunstein, S.V. Buldyrev, R. Cohen, S. Havlin, and H.E. Stanley, Phys.
Rev. Lett. 91, 168701 (2003).
543. D.J. Felleman and D.C. Van Essen, Cerebral Cortex 1, 1 (1991).
544. J.W. Scannell, G.A.P.C. Burns, C.C. Hilgetag, M.A. O’eil, and M.P. Yong,
Cerebral Cortex 9, 277 (1999).
545. C.J. Stam and E.A. de Bruin, Hum. Brain Mapp. 22, 97 (2004).
546. E. Salinas and T.J. Sejnowski, Nature Neurosci. 2, 539 (2001).
547. P. Friés, Trends Cogn. Sci. 9, 474 (2005).
548. A. Schnitzler and J. Gross, Nature Neurosci. 6, 285 (2005).
549. O. Sporns, D.R. Chialvo, M. Kaiser, and C.C. Hilgetag, Trends Cogn. Sci. 8,
418 (2004).
550. C.J. Stam, Neurosci. Lett. 355, 25 (2004).
References 361
551. V.M. Eguı́luz, D.R. Chialvo, G. Cecchi, M. Baliki, and A.V. Apkarian, Phys.
Rev. Lett. 94, 018102 (2005).
552. R. Salvador et al., Cereb. Cortex 15, 1332 (2005).
553. C.S. Zhou, L. Zemanová, G. Zamora, C.C. Hilgetag, and J. Kurths, Phys. Rev.
Lett. 97, 238103 (2006); New J. Phys. (2007) (to appear).
554. L. Zemanová, C.S. Zhou, and J. Kurths, Physica D 224, 202 (2006).
555. B.T. Grenfell, O.N. Bjornstad, and J. Kappey, Nature (London) 414, 716
(2001).
556. N.C. Grassly, C. Fraser, and G.P. Garnett, Nature( London) 433, 417 (2005).
557. J.G. Restrepo, E. Ott, and B.R. Hunt, Phys. Rev. Lett. 93, 114101 (2004).
558. M. Denker, M. Timme, M. Diesmann, F. Wolf, and T. Geisel, Phys. Rev. Lett.
92, 074103 (2004).
559. X.F. Wang, Int. J. Bifurcation Chaos Appl. Sci. Eng. 12, 885 (2002).
560. A.E. Motter, C.S. Zhou, and J. Kurths, Europhys. Lett. 69, 334 (2005).
561. A.E. Motter, C.S. Zhou, and J. Kurths, Phys. Rev. E 71, 016116 (2005).
562. S.N. Dorogovtsev and J.F.F. Mendes, Phys. Rev. E. 62, 1842 (2000).
563. F.R.K. Chung, Spectral Graph Theory (AMS, Providence, 1994).
564. F. Chung, L. Lu, and V. Vu, Proc. Natl. Acad. Sci. U.S.A. 100, 6313 (2003).
565. C.S. Zhou, A.E. Motter, and J. Kurths, Phys. Rev. Lett. 96, 034101 (2006).
566. G. Korniss, M.A. Novotny, H. Guclu, Z. Toroczkai, and P.A. Rikvold, Science
299, 677 (2003).
567. M. Chavez, D.-U. Hwang, A. Amann, H.G.E. Hentschel, and S. Boccaletti,
Phys. Rev. Lett. 94, 218701 (2005).
568. D.-U. Hwang, M. Chavez, A. Amann, and S. Boccaletti, Phys. Rev. Lett. 94,
138701 (2005).
569. J. Lu, X. Yu and G. Chen, Physica A 334, 281 (2004).
570. I.V. Belykh, V.N. Belykh, and M. Hasler, Physica D 195, 188 (2004).
571. R.E. Amritkar and C. Hu, Chaos, 16, 015117 (2006).
572. C.S. Zhou and J. Kurths, Phys. Rev. Lett. 96, 164102 (2006).
573. T. Gross, C.J. Dommar D’Lima, and B. Blasius, Phys. Rev. Lett. 96, 208701
(2006).
Index
A Bistability, 53, 60, 61, 64, 65, 237, 256,

Action potential, 29, 53, 254, 255 260, 262, 266
Adler equation, 21, 37, 57, 71, 91 Brain activity, 339
Amplitude
complex, 38, 59, 131, 132, 137, 138, C
234, 236, 237 Cardiac
instantaneous, 18 arrhythmia, 251, 252
Amplitude equation, 133, 136 cell, 33, 251–255
Analytic signal, 18 model, 251, 253, 255, 260, 261, 263
Antiphase synchronization, 61, 152 pacemaker, 251, 252
pacing, 251–253, 258
Appleton, Edward, 35
tissue, 34, 52, 251–254, 256
Arnold tongue, 39–41, 44, 45, 49, 74,
88, 291, 292, 313 Cardiorespiratory system, 44
Cell
Asymptotic method, 37, 132
excitable, 29, 251, 270, 305, 309
Attractor
pacemaker, 3, 187
chaotic, 14, 17, 18, 21, 35, 40–43, 45,
Cellular automata, 7
46, 48, 50, 68–71, 74, 83, 88, 91,
Chain of oscillators
151, 154, 218, 278, 279, 291, 293
homoclinic, 291
bifurcation, 45, 46, 71, 74, 218, 234
spiral wave, 177
strange, 111
Chemical reaction, 169, 273
Circle map
B
coupled, 55, 90, 98, 150, 188, 197
Beat frequency, 159 chaotic, 155
Belousov Zhadotinsky reaction, 4, 273 identical, 187, 189
Bifurcation nonidentical, 90, 199
Andronov Hopf, 38, 63, 108, 150, 234 Coherence
longwave, 192, 197, 198, 210 noncoherence, 16, 72, 73, 83, 85–87,
period doubling, 322 92, 95, 204, 205, 207, 208
saddlenode, 23, 24, 38, 46, 47, strong, 70
63, 74, 113 Communication, 4, 90, 187, 215, 231,
shortwave, 152, 192, 195, 198, 210 317
364 Index
Complete synchronization, 4, 68, 70, 72, two, 15, 31, 90, 110, 111, 114, 119,
76, 80, 85, 86, 269 152, 179, 188, 249, 251–253, 255,
Convection, 18, 19, 169 257, 261, 262, 265, 270, 278, 279
Correlation Degree, 16, 25, 70, 72, 117, 121, 124,
cross, 72 140, 147, 163
degree, 333, 337 Distribution
increasing, 295 contraction region, 275
noise, 293 coupling
strong, 70, 72, 73, 84 local, 226, 228, 318
Coupled global, 56, 124, 179, 230, 231, 295,
map, 77, 91, 169, 170, 172, 185, 189, 297, 298, 301, 318
190, 192, 195, 197, 211, 333 of degree, 319, 328, 334, 337
unidirectionaly, 90, 103, 128, 219, 220 gaussian, 124, 157, 269, 271
Coupling of phase, 133, 141, 176
asymmetric, 189, 195, 197, 210, 239,
240, 320, 333 E
diffusive, 131, 150, 152, 171, 189, 207,
Embedding, 174, 178
216, 221, 222, 305, 306, 314
Energy, 252, 271, 327
dissipative, 58, 61, 62, 235, 241
Ensemble average, 17
global, 6, 55, 103, 123–125, 213, 225,
Entrainment, 5, 8, 38, 63, 68, 84, 90, 96,
230, 231, 295–299, 301, 317, 324,
123, 171, 180, 200, 230
325, 327
Excitability, 305, 308
mean field, 179, 180–185, 295
Excitable system, 4, 8, 11, 29–34, 36,
nearest neighbor, 6, 125, 131, 225,
52, 269, 312
313, 317, 321
nonlinear, 188, 189, 213
reactive, 234, 241, 242, 244 F
scalar, 152 Fast motion, 12, 31, 172
symmetric, 90, 112, 113, 115, 117, Feedback, 213–216, 218–220, 222, 224,
124, 189, 190, 195–198, 239, 240, 225, 228, 230, 231, 281, 282, 284,
241, 320, 333, 337 314
unidirectional, 106, 110, 213, 214, Filter, 24, 215, 224
218, 234 Firefly, 22
Coupling function, 57, 58, 105, 114 Focus, 5, 7, 9, 38, 48, 59, 105, 170, 178,
Crosscorrelation, 72 198, 200, 235, 278
Fokker–Planck equation, 271
D Force
Data, 34, 90, 137, 140, 166, 259 periodic, 4, 35, 36, 40, 47, 52, 130,
Decoherence, 167 179, 251, 270, 288, 289
Demodulation, 4 random, 269, 270, 273
Detuning, 68 Fourier, 152
Differential equation, 18, 19, 28, 29, 48, Fractal, 178, 185
170, 234, 236 Frequency
Diffusion, 76, 152, 240 instantaneous, 14, 69, 121
constant, 16, 17, 41, 70, 81–84, 87, mean, 16, 18, 19, 28, 35, 36, 40,
253, 309 43, 61, 63, 69, 74, 80–83, 85–88,
Dimension 113, 116, 119, 120, 131, 133, 153,
one, 7, 104, 110, 243, 247, 251–253, 155, 161, 217, 218, 220, 221, 223,
255, 256, 258–262, 264, 274, 278 227–229, 272, 296–298
Index 365
natural, 12, 36, 37, 45, 56, 151, 153, K

154, 160–163, 167, 228, 238, 270, Kink, 176
272 Kuramoto model, 124
observed, 87, 114, 119, 121, 122,
155, 157–159, 171, 173, 174, 215, L
218–221, 223, 224, 227, 298 Laser, 3, 16, 19, 47, 169, 215, 278,
Frequency locking, 69n4, 70, 73, 80, 87, 282–285, 290, 292
200, 217, 218, 234, 291, 296 chaotic, 270, 281, 289, 291
Full synchronization, 4, 133, 293 Lattice, 121, 170, 197, 199, 307
Funnel attractor, 14, 16, 18, 35, 69, 73, of oscillators, 6, 7, 119, 152, 160
151, 153, 162–165, 218 twodimensional, 119, 152, 168, 270,
317
G Limit cycle, 11–14, 23, 24, 36–39
Generalized synchronization, 4, 68 Locking, 5, 8, 37, 40, 41, 43–47
Generator, 3, 4, 9, 12, 16, 28, 36, 109, Logistic map, 273, 275
188, 213, 216, 281 lorenz system
Global coupling, 6, 301, 317 periodically driven, 42–44
LotkaVolterra, 166
lyapunov exponent, 50, 117, 152, 156,
H
157, 218, 219, 273, 320, 332, 334
Hard transition, 64, 74, 87, 88, 115, 116,
122, 141, 142, 155, 158, 160, 211,
M
227, 295
Manifold, 64, 320
Heart, 3, 34, 252
stable, 152, 278–283, 289
Hilbert transform, 18
unstable, 278–280, 283
Homoclinic chaos, 278, 282 Map, 11, 20, 24–29, 45, 47–51
Hub, 318 coupled, 77, 91, 169, 170, 172, 185,
Huygens C., 4 189, 190, 192, 195, 197, 211, 333
Hyperchaos, 70, 71, 80, 96–98, 157 Mean field, 124, 179–182, 184, 185,
Hysteresis, 53 295–298, 301, 302
Medium, 177, 179, 253–255, 259,
I 261–265, 314
Identical synchronization, See Complete Mixing, 28
synchronization Modulation, 16, 51, 145, 147, 193, 234,
Inpahse synchronization, 61, 64, 239 289–291
Instability, 48, 112, 195, 197, 198, 276, Multiplier, 198, 214n1, 215, 247
332 Multistability, 114, 118, 130–143, 248
Interaction, 3, 44, 75, 123, 160, Mutual information, 286, 287
161, 166, 167, 172, 176, 215, 233,
239, 327 N
Intermittency, 14, 19, 22, 47, 50, 54, 73, Natural frequency, 12, 36, 37, 45, 153,
74, 78, 79, 123, 169 167
Intestine, 3, 130, 131 Nervous system, 354 (in ref only)
Network
J regular, 317, 320, 324, 331, 332
Josephson junction, 3, 22, 24, 103, complex, 5, 198, 317, 318, 319, 321,
130, 233 327, 331, 332, 337, 339, 340
366 Index
Network (Continued ) Pendulum, 4, 22–24, 103, 125, 130, 216

unweighted, 332, 335, 338, 339 Period, 3, 4, 8, 12, 14, 36, 37, 40, 41, 45
weighted, 332, 335, 337–340 Periodic orbit, 25, 41–46, 88, 273, 275,
Neuron, 3, 11, 22, 28–30, 52, 169, 278, 293, 294
177–180, 182 Perturbation, 31, 34, 52, 124, 145, 146,
Node, 23, 24, 38, 46, 47, 63, 74, 113, 241, 242, 278, 283–285, 288, 291,
210, 252 292, 320, 329, 332, 336
Noise Phase
additive, 32, 247, 274, 309 coherence, 76, 81, 84, 308
uncorrelated, 249, 306 definition, 12–19, 176, 218
Gaussian, 157, 269, 271, 279–281, difference, 37, 39–41, 62–64, 66, 67,
288, 292, 297, 300, 306, 329 69, 71, 80–82, 88, 90, 91, 98,
intrinsic, 282–285, 289 105–109, 112,
Numerical prediction, 299 diffusion, 16, 17, 41, 76, 81–84, 89
entrainment, see Phase, locking
O equation, 56, 105, 131, 134, 222
Observed frequency, 121, 155, 157, 158, initial, 12, 23
173, 218–220, 298 instantaneous, 8, 12, 18, 158, 163, 207
Order parameter, 124, 125, 180, 181, locked loop, 9, 22, 24, 25, 89, 103,
230, 231, 296, 297, 327, 328, 330 106, 187, 189, 213
Oscillation death, 75, 143, 144, 150, locking, 8, 37, 40, 41, 43, 44–46, 71,
160, 161, 168, 233, 248, 322, 339 72, 76, 82–84, 86, 89, 92, 114, 153,
Oscillations 171, 179, 286, 290, 291, 302, 303
chaotic, 4, 8, 18, 40, 42, 109, 293 plane, 23, 32
coherent, 40, 302, 305, 310, 322, 323, point, 18, 45, 273
325 portrait, 13, 14, 83, 88, 97, 141
collective, 130, 177, 233, 248, 296, shift, 8, 37, 57, 106, 107, 116, 237
303, 322, 323, 325, 326, 331, 338 slip, 14, 22, 42–44, 46, 57, 72, 73, 95,
harmonic, 13, 48 98, 163, 176, 269–272, 290, 292,
noiseinduced, 304 293–296, 298–300, 310, 311, 329
non-synchronous, 8 space, 18, 23, 37, 39, 45, 87, 88, 97,
Oscillator 98, 107, 110–112, 134, 141, 273,
chaotic, 4, 11, 14, 16, 18, 20, 36, 39– 274, 276–280, 282, 283, 285, 296
41, 45, 54, 55, 68, 77, 81, 83, 151, synchronization, 5, 18, 35, 36, 39,
155, 160, 165, 167, 189, 213, 217, 41–43, 45–47, 49, 51, 54
223–225, 228, 303, 304, 319, 321 anomalous, 167
coupled, 8, 55, 56, 68, 76, 103–105, automatic, 214, 224, 231
112, 113, 119, 123, 127, 129, chaotic, 4, 9, 35, 49, 80, 89, 90, 93,
130, 150–152, 154, 155, 157, 158, 189, 248
160, 165, 178, 180, 213–215, 217, external, 4, 35, 37–39, 52, 53, 179,
220–224, 228, 300, 305, 320 182
electrochemical, 16, 270, 297, 299, global, 8, 105, 106, 108, 113–116,
315 118, 123, 126–128
linear, 215 noiseenhanced, 270
weakly nonlinear, 37, 38, 56 transition, 81, 87–89, 98
volume, 117
P Phase-coherent attractor, 14, 16, 68, 69,
Pacemaker 71, 73, 74, 82, 155, 163
cell, 3, 187 Poincaré system, 216, 225–228
Index 367
Population, 3, 9, 123, 169, 189, 207, Signal-to-noise ratio, 303, 306, 307
270, 301, 340 Slow motion, 31
Potential, 29–31, 33, 34, 53, 230, 251, Small world, 317–319, 321
254–257, 271, 279, 297, 300, 340 Soft transition, 63, 87, 115–117, 122,
Power spectrum, 12, 68, 114, 301, 302 123, 128, 141, 142, 155, 156, 167,
Pulse, 31, 34, 52, 53, 126, 177, 187, 171, 203, 211, 221, 227
252–259, 261, 265, 305, 306, 308, Spatiotemporal chaos, 177, 178, 187
319, 332 spectrum
Lyapunov, 84, 89, 117, 155, 157
Q eigenvalues, 333, 337
Quasiperiodic, 59, 109–112, 114, 179, Spike
253, 264, 266 train, 29, 30, 278, 280, 281, 311, 312
Spiral wave, 34, 53, 177, 251–258,
R 260–266
Random walk, 242 Splay state, 193
Stability, 7, 59, 80, 91, 113, 152,
Relaxation oscillator, 56, 149
189–191, 193–196, 198, 204,
Repeller, 45, 46
209, 242, 257, 279, 319–321,
Resonance 333, 334
coherent, 305 Stimulus, 33, 52, 53, 251, 252, 254–264,
parametric, 48 269
stochastic, 249, 269 Stroboscopic technique, 40
array-enhanced, 313 Suppression of oscillations, 160
Return time Symmetry, 115, 117, 140, 147, 242, 331
average, 41, 45, 294 Synchronization
Resonance-like behaviour, 307 Cluster, 114, 116, 118, 120, 128, 130,
Rhythm, 3, 5, 7, 8, 22, 28, 53, 179, 180, 135, 145, 146, 149, 150, 181, 204,
182, 252 207, 227, 302, 326, 339, 340
Riddling, 47 external, 4, 35, 37, 39, 41, 47, 52
Rotation number, 25, 26, 45, 90–93, 95, global, 8, 105, 106, 108, 113–116,
99, 173, 200, 203–208, 211 118, 123, 126–129, 133–136,
Rotator, 21, 22, 24, 55, 56, 66, 111, 113, 140, 144
119, 120, 123, 126, 127, 206, 233 mutual, 4, 28, 55, 77, 103, 114, 123,
Roessler system, 166 130, 155, 180, 187, 238
Synchronization region, 40, 42, 45, 49,
50, 61, 64, 67, 72, 74, 75, 84, 88,
S
91, 92, 95, 97, 183–185, 190–192,
Saddle, 23, 24, 38, 42, 46, 47, 63, 74, 196, 197, 204, 210, 271, 272, 290,
110, 113, 278–282, 284, 289–291 291, 293
Saddlenode, 23 Synchronization transition, 7, 41, 73,
Scale free, 318, 319, 324, 325, 327 99, 117, 119, 120, 122, 124, 129,
Self-sustained oscillator, 12, 36, 305, 167, 185, 222, 228
313, 314 Synchrony, 8, 53, 125, 251, 252, 254–261,
Shannon entropy, 287 270
Signal System
chaotic, 45, 89, 90 autonomous, 51, 77
periodic, 35, 39, 178, 179, 218, 291, bistable, 249
292, 310, 311, 313–315 distributed, 127
368 Index
System (Continued ) Transient, 82, 149, 276, 283, 314

excitable, 11, 29, 33, 36, 52, 54, 292, Turbulence, 7, 109–111, 131, 169, 185,
305, 306, 312, 313, 315 239–241
neural, 269, 306, 313, 317 V

nonlinear, 269, 273, 285, 309, 315 van der Pol oscillator, 11–13, 16, 36–38,
55, 58, 75, 76, 92
T
W
Thermodynamic limt, 140 Wave number, 192, 194
Torus, 110, 111 Wavelength, 152, 196
Temporal Disorder Rhythms in Physiological Systems
in Human Oscillatory Systems Editors: H. Haken, H. P. Koepchen
Editors: L. Rensing, U. an der Heiden, Quantum Noise 3rd Edition
M. C. Mackey By C.W. Gardiner, P. Zoller
The Physics of Structure Formation Nonlinear Nonequilibrium
Theory and Simulation Thermodynamics I Linear and Nonlinear
Editors: W. Güttinger, G. Dangelmayr Fluctuation-Dissipation Theorems
Computational Systems – Natural and By R. Stratonovich
Artificial Editor: H. Haken
Self-organization and Clinical
From Chemical to Biological Psychology Empirical Approaches
Organization Editors: M. Markus, to Synergetics in Psychology
S. C. Müller, G. Nicolis Editors: W. Tschacher, G. Schiepek,
Information and Self-Organization E. J. Brunner
A Macroscopic Approach to Complex Nonlinear Nonequilibrium
Systems 2nd Edition By H. Haken Thermodynamics II Advanced Theory
Propagation in Systems Far from By R. Stratonovich
Equilibrium Editors: J. E. Wesfreid, Limits of Predictability
H. R. Brand, P. Manneville, G. Albinet, Editor: Yu. A. Kravtsov
N. Boccara
On Self-Organization
Neural and Synergetic Computers An Interdisciplinary Search
Editor: H. Haken for a Unifying Principle
Cooperative Dynamics in Complex Editors: R. K. Mishra, D. Maaß, E. Zwierlein
Physical Systems Editor: H. Takayama Interdisciplinary Approaches
Optimal Structures in Heterogeneous to Nonlinear Complex Systems
Reaction Systems Editor: P. J. Plath Editors: H. Haken, A. Mikhailov
Synergetics of Cognition Inside Versus Outside
Editors: H. Haken, M. Stadler Endo- and Exo-Concepts of Observation
Theories of Immune Networks and Knowledge in Physics, Philosophy
Editors: H. Atlan, I. R. Cohen and Cognitive Science
Editors: H. Atmanspacher, G. J. Dalenoort
Relative Information Theories
and Applications By G. Jumarie Ambiguity in Mind and Nature
Multistable Cognitive Phenomena
Dissipative Structures in Transport
Editors: P. Kruse, M. Stadler
Processes and Combustion
Editor: D. Meinköhn Modelling the Dynamics
of Biological Systems
Neuronal Cooperativity
Editors: E. Mosekilde, O. G. Mouritsen
Editor: J. Krüger
Synergetic Computers and Cognition Self-Organization in Optical Systems
A Top-Down Approach to Neural Nets and Applications in Information
2nd edition By H. Haken Technology 2nd Edition
Editors: M. A. Vorontsov, W. B. Miller
Foundations of Synergetics I
Distributed Active Systems 2nd Edition Principles of Brain Functioning
By A. S. Mikhailov A Synergetic Approach to Brain Activity,
Behavior and Cognition
Foundations of Synergetics II By H. Haken
Complex Patterns 2nd Edition
By A. S. Mikhailov, A. Yu. Loskutov Synergetics of Measurement, Prediction
and Control By I. Grabec, W. Sachse
Synergetic Economics By W.-B. Zhang
Predictability of Complex Dynamical Systems
Quantum Signatures of Chaos
By Yu.A. Kravtsov, J. B. Kadtke
2nd Edition By F. Haake
Interfacial Wave Theory of Pattern Formation Synergetic Phenomena in Active Lattices
Selection of Dentritic Growth and Viscous Patterns, Waves, Solitons, Chaos
Fingerings in Hele-Shaw Flow By Jian-Jun Xu By V. I. Nekorkin, M. G. Velarde
Asymptotic Approaches in Nonlinear Brain Dynamics
Dynamics New Trends and Applications Synchronization and Activity Patterns
By J. Awrejcewicz, I. V. Andrianov, in Pulse-Coupled Neural Nets with Delays
L. I. Manevitch and Noise By H. Haken
Brain Function and Oscillations From Cells to Societies
Volume I: Brain Oscillations. Models of Complex Coherent Action
Principles and Approaches By A. S. Mikhailov, V. Calenbuhr
Volume II: Integrative Brain Function. Brownian Agents and Active Particles
Neurophysiology and Cognitive Processes Collective Dynamics in the Natural and Social
By E. Başar Sciences By F. Schweitzer
Asymptotic Methods for the Fokker-Planck
Nonlinear Dynamics of the Lithosphere
Equation and the Exit Problem in Applications
and Earthquake Prediction
By J. Grasman, O. A. van Herwaarden
By V. I. Keilis-Borok, A.A. Soloviev (Eds.)
Analysis of Neurophysiological Brain
Nonlinear Fokker-Planck Equations
Functioning Editor: Ch. Uhl
Fundamentals and Applications
Phase Resetting in Medicine and Biology By T.D. Frank
Stochastic Modelling and Data Analysis
By P. A. Tass Patterns and Interfaces
in Dissipative Dynamics
Self-Organization and the City By J. Portugali By L.M. Pismen
Critical Phenomena in Natural Sciences
Synchronization in Oscillatory Networks
Chaos, Fractals, Selforganization and Disorder:
Concepts and Tools 2nd Edition By D. Sornette By G. Osipov, J. Kurths, C. Zhou
Spatial Hysteresis and Optical Patterns

By N. N. Rosanov
Nonlinear Dynamics of Chaotic and Stochastic
Systems Tutorial and Modern Developments
2nd Edition
By V. S. Anishchenko, V. Astakhov,
A. Neiman, T. Vadivasova,
L. Schimansky-Geier

Synchronization in Oscillatory Networks: Grigory V. Osipov Jürgen Kurths Changsong Zhou

Uploaded by

Copyright:

Available Formats

Synchronization in Oscillatory Networks: Grigory V. Osipov Jürgen Kurths Changsong Zhou

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Synchronization in Oscillatory Networks: Grigory V. Osipov Jürgen Kurths Changsong Zhou

Uploaded by

Copyright:

Available Formats

S P R I N G E R S E R I E S Springer ..

Editorial and Programme Advisory Board

With 221 Figures

Professor Dr. Jürgen Kurths

Dr. Changsong Zhou

Library of Congress Control Number: 2007927818

The formation of collective behavior in large ensembles or networks of coupled

amplitude-dependent eﬀects, for instance, oscillations death and amplitude

Part I: Basics on Synchronization and Paradigmatic Models

3 Synchronization Due to External Periodic Forcing . . . . . . . . . 35

4 Synchronization of Two Coupled Systems . . . . . . . . . . . . . . . . . . 55

Part II: Synchronization in Geometrically Regular Ensembles

5 Ensembles of Phase Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

5.3.2 Synchronization Transitions in Chains

6 Chains of Coupled Limit-Cycle Oscillators . . . . . . . . . . . . . . . . . 129

7 Ensembles of Chaotic Oscillators with a Periodic-Doubling

7.6 Anomalous Collective Behavior of Coupled

8 Intermittent-Like Oscillations in Chains

9 Regular and Chaotic Phase Synchronization

10 Controlling Phase Synchronization

10.9 Synchronization of Globally Coupled Chaotic Oscillators . . . . . 230

11 Chains of Limit-Cycle Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . 233

12 Chains and Lattices of Excitable Luo–Rudy Systems . . . . . . . 251

Part III: Synchronization in Complex Networks and Inﬂuence

13 Noise-Induced Synchronization in Ensembles of Oscillatory

14 Networks with Complex Topology . . . . . . . . . . . . . . . . . . . . . . . . . . 317

Basics on Synchronization and Paradigmatic

1.1 Synchronization Phenomena in Nature, Physics,

employed for frequency stabilization of generators, for synthesizing frequen-

synchronization implies that the output of a driving system is associated with

1.2 Goal of the Book

There are some instructive monographs on synchronization but they are

1,1 1,2 1,N

2,1 2,2 2,N

M,1 M,2 M,N

Fig. 1.1. Lattice structure of nearest-neighbor coupled systems

Even the possibility of the existence of a total synchronization regime in a

Ẋj = F (Xj ) + d1 H(Xj+1 − Xj ) + d2 H(Xj−1 − Xj ),

The model structure of (1.1) or (1.2) is analogous to network models of

1.3 Terminological Remarks

We will characterize the rhythm of oscillations by the characteristic time

mΩi = nΩj , (1.3)

|mφi (t) − nφj (t)| = Const, (1.4)

(2) Non-strict phase locking (the phase shift is bounded)

|mφi (t) − nφj (t)| < Const, (1.5)

where φi (t) and φj (t) are somehow-introduced instantaneous phases of

1.4 Bibliographical Remarks

First, we would like to mention the recently published book by A. Pikovsky,

Synchronization phenomena studied in this book are mainly based on the

2.1 Harmonic Oscillator: Amplitude, Frequency

has the solution

where A is the constant amplitude of oscillations,

is the linear increasing instantaneous phase, φ0 is the initial phase, and ω0 is

2.2 Van der Pol Oscillator: Quasi-Harmonic

where ω is the natural frequency of the oscillations and the parameter of

First phase deﬁnition. In a straightforward generalization of the harmonic

of the phase φ(t) looks like intermittency consisting of relatively large

2.3 Rössler Oscillator: From Phase-Coherent to Funnel

(a) (b) (c)