Symmetry in Complex Network Systems Connecting Equivariant Bifurcation Theory with Engineering Applications
Upload
others
View
1
Download
0
Embed Size (px)
344 x 292
429 x 357
514 x 422
599 x 487
Citation preview
Visarath In Antonio Palacios
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 ability 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
traffic, 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, synergetics, turbulence,
dynamical systems, catastrophes, instabilities, stochastic
processes, chaos, graphs and networks, cellular automata, adaptive
systems, genetic algorithms and computational intelligence.
The three major book publication platforms of the Springer
Complexity program are the monograph series “Understanding Complex
Systems” focusing on the various applications of complexity, the
“Springer Series in Synergetics”, which is devoted to the
quantitative theoretical and methodological foundations, and the
“Springer Briefs in Complexity” which are concise and topical
working reports, case studies, surveys, essays and lecture notes of
relevance to the field. In addition to the books in these two core
series, the program also incorporates individual titles ranging
from textbooks to major reference works.
Editorial and Programme Advisory Board
Henry Abarbanel, Institute for Nonlinear Science, University of
California, San Diego, USA Dan Braha, New England Complex Systems,
Institute and University of Massachusetts, Dartmouth, USA Péter
Érdi, Center for Complex Systems Studies, Kalamazoo College,
Kalamazoo, 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 Viktor Jirsa, Centre
National de la Recherche Scientifique (CNRS), Université de la
Méditerranée, Marseille, France Janusz Kacprzyk, System Research,
Polish Academy of Sciences, Warsaw, Poland Kunihiko Kaneko,
Research Center for Complex Systems Biology, The University of
Tokyo, Tokyo, Japan Scott Kelso, Center for Complex Systems and
Brain Sciences, Florida Atlantic University, Boca Raton, USA Markus
Kirkilionis, Mathematics Institute and Centre for Complex Systems,
University of Warwick, Coventry, UK Jürgen Kurths, Potsdam
Institute for Climate Impact Research (PIK), Potsdam, Germany
Ronaldo Menezes, Department of Computer Science, Florida Institute
of Technology, Melbourne, FL, USA Andrzej Nowak, Department of
Psychology, Warsaw University, Warsaw, Poland Hassan Qudrat-Ullah,
School of Administrative Studies, York University, Toronto, ON,
Canada 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 Zürich, Zürich, Switzerland Didier
Sornette, Entrepreneurial Risk, ETH Zürich, Zürich, Switzerland
Stefan Thurner, Section for Science of Complex Systems, Medical
University of Vienna, Vienna, Austria
Understanding Complex Systems
Founding Editor: S. Kelso
Future scientific and technological developments in many fields
will necessarily depend upon coming to grips with complex systems.
Such systems are complex in both their composition—typically many
different kinds of components interacting simultaneously and
nonlinearly with each other and their environments on multiple
levels—and in the rich diversity of behavior of which they are
capable.
The Springer Series in Understanding Complex Systems series (UCS)
promotes new strategies and paradigms for understanding and
realizing applications of complex systems research in a wide
variety of fields and endeavors. UCS is explicitly
transdisciplinary. It has three main goals: First, to elaborate the
concepts, methods and tools of complex systems at all levels of
description and in all scientific fields, especially newly emerging
areas within the life, social, behavioral, economic, neuro- and
cognitive sciences (and derivatives thereof); second, to encourage
novel applications of these ideas in various fields of engineering
and computation such as robotics, nano-technology and informatics;
third, to provide a single forum within which commonalities and
differences in the workings of complex systems may be discerned,
hence leading to deeper insight and understanding.
UCS will publish monographs, lecture notes and selected edited
contributions aimed at communicating new findings to a large
multidisciplinary audience.
More information about this series at
http://www.springer.com/series/5394
Visarath In • Antonio Palacios
123
Visarath In Space and Naval Warfare Systems Center San Diego, CA
USA
Antonio Palacios Department of Mathematics, Nonlinear Dynamical
Systems Group
San Diego State University San Diego, CA USA
ISSN 1860-0832 ISSN 1860-0840 (electronic) Understanding Complex
Systems ISBN 978-3-662-55543-9 ISBN 978-3-662-55545-3 (eBook) DOI
10.1007/978-3-662-55545-3
Library of Congress Control Number: 2017947684
© Springer-Verlag GmbH Germany 2018 This work is subject to
copyright. All rights are reserved by the Publisher, whether the
whole or part of the material is concerned, specifically the rights
of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical
way, and transmission or information storage and retrieval,
electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed. The use of
general descriptive names, registered names, trademarks, service
marks, 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. The publisher, the authors and the editors are safe to
assume that the advice and information in this book are believed to
be true and accurate at the date of publication. Neither the
publisher nor the authors or the editors give a warranty, express
or implied, with respect to the material contained herein or for
any errors or omissions that may have been made. The publisher
remains neutral with regard to jurisdictional claims in published
maps and institutional affiliations.
Printed on acid-free paper
This Springer imprint is published by Springer Nature The
registered company is Springer-Verlag GmbH Germany The registered
company address is: Heidelberger Platz 3, 14197 Berlin,
Germany
To our wives, Christine and Irene, and our sons, Beredei, Reynard
and Daniel, and to our beloved parents, for their love and support
throughout these years.
Preface
The seminal work by Lorenz in 1963 [264], and later by May in 1976
[273–275], has led scientists and engineers to recognize that
nonlinear systems can exhibit a rich variety of dynamic behavior.
From simple systems, such as the evolution of single species [314],
an electronic or biological oscillator [423, 424], to more complex
systems, such as chemical reactions [33], climate patterns [153],
bursting behavior by a single neuron cell [208], and flocking of
birds [333, 393], Dynamical Systems theory provides quantitative
and qualitative (geometrical) techniques to study these and many
other complex systems that evolve in space and/or time. Regardless
of the origins of a system, i.e., Biology, Chemistry, Engineering,
Physics, or even the Social Sciences, dynamical systems theory
seeks to explain the most intriguing and fundamental features of
spatio-temporal phenomena.
In recent years, systems made up of individual units coupled
together, either weakly or tightly, have gained considerable
attention. For instance, the dynamics of arrays of Josephson
junctions [18, 99, 100, 160], central pattern generators in
biological systems [85, 226, 227], coupled laser systems [328,
419], synchroniza- tion of chaotic oscillators [315, 426],
collective behavior of bubbles in fluidization [163], the flocking
of birds [393], and synchronization among interconnected biological
and electronic nonlinear oscillators. These are only a few
representative examples of a new class of complex dynamical systems
or complex networks. The complexity arises from the fact that
individual units cannot exhibit on their own the collective
behavior of the entire network. In other words, the collective
behavior is the exclusive result of the mutual interaction that
takes place when multiple units are interconnected in some
fashion.
In most cases, three factors are normally considered when studying
the collective behavior of a complex system. Mainly, the internal
dynamics of each individual unit or cell, the topology of cell
connections, i.e., which cells communicate with each other, and the
type of coupling. More recently, a fourth factor has gained further
attention—symmetry. It is well-known that symmetry alone can
restrict the type of solutions of systems of ordinary– and partial
differential equations, which often serve as models of complex
systems. So it is reasonable to expect that certain aspects of the
collective behavior of a complex system can be inferred from
the
vii
presence of symmetry alone. In fact, the work by Golubitsky [145,
146, 149] lays down the theoretical foundations for a
model-independent analysis to understand, and predict, the behavior
of a dynamical system using, mainly, the underlying symmetries of
the system while separating the fine details of the model. While
this approach has been widely successful in explaining computer
simulations and experimental observations of many different
spatio-temporal phenomena, it has found limited use in the
conceptualization and development of nonlinear devices even though
many of those systems are, inherently, symmetric. And while many
works have been dedicated to study the symmetry-preserving
phenomenon of synchronization [25, 315, 317, 382, 426],
significantly less is known about how one can exploit the rich
variety of collective patterns that can emerge via
symmetry-breaking bifurcations, such as heteroclinic cycles
[63].
Over the past 17 years, we and other colleagues and students have
been attempting to bridge the current gap between the theory of
symmetry-based dynamics, equivariant bifurcation theory, and its
application to developing non- linear devices. At the beginning,
around the year 2000, we were interested in developing new methods
to manipulate frequency in arrays of nonlinear oscillators for
antenna devices. Collaborators from the U.S. Navy had already shown
[166] that small frequency perturbations applied to the end points
of a chain of nonlinear oscillators can lead to a change in the
direction of the radiation pattern. That is, they demonstrated that
beam steering was possible without mechanically rotating an
antenna. The next puzzle that we had to solve was to manipulate the
collective frequency of the array over a broad band without
changing the internal frequency of each individual oscillator. But
just when we were about to solve this problem, we were steered, no
pun intended, into developing a new class of highly sensitive,
low-power and low-cost, magnetic- and electric field sensors.
Theoretical work for this new class of sensors started around 2002.
The fundamental principles were twofold: to exploit
coupling-induced oscillations in symmetric networks to generate
self-induced oscillations, thus reducing power consumption; and to
exploit symmetry-breaking effects of heteroclinic cycles to enhance
sensitivity. As a starting point, we chose fluxgate magnetometers
as individual units, because their behavior is governed by a
one-dimensional autonomous differential equation. Consequently,
based on the fundamental theory of ODEs, it follows that in the
absence of any forcing term the one-dimensional dynamics of the
individual units cannot produce oscillations. But when the
fluxgates are coupled then the network can, under certain
conditions that depend on the coupling strength, oscillate. This
configuration could demonstrate to skeptics that self-induced
oscillations can indeed be engineered. In practice, the network
would still need, of course, a min- imum amount of energy to kick
it off of its trivial equilibrium state and get the oscillations
going. Overall, we were able to show that, under certain
conditions, the sensitivity response of an array of weakly coupled
fluxgate sensors can increase by four orders of magnitude while
their cost could be simultaneously reduced to a fraction of that of
an individual fluxgate sensor. This technology matured around 2005
with design, fabrication and deployment.
viii Preface
In 2006, we extended the work on magnetic fields to electric field
sensors. These sensors are also governed by one-dimensional,
overdamped, bistable systems of equations. We conducted a complete
bifurcation analysis that mirrors that of the fluxgate magnetometer
and, eventually, translated the research work into a micro- circuit
implementation. This microcircuit was intended to be used for
measuring minute voltage or current changes that may be injected
into the system. The con- ceptualization of these sensors employs
the model-independent approach of Golubitsky’s theory for the study
of dynamical systems with symmetry, while the development of
laboratory prototypes takes into account the model-specific
features of each device which, undoubtedly, may impose additional
restrictions when we attempt to translate the theory into an actual
experiment. For instance, a sensor device that measures magnetic
flux, as oppose to electric field signals, may limit the type of
coupling functions that can be realized in hardware. In other
words, not every ide- alization of a network-based structure can be
readily implemented in the laboratory. This and other similar
restrictions need to be kept in mind while reading this book.
Around that same 2006 year, we started, in tandem, to the work on
electric field sensors, theoretical studies of networks of
Superconducting Quantum Interference Devices (SQUIDs). The work was
suspended until 2009 when we returned to explore in greater detail
the response of networks of non-uniform SQUID loops. The technology
matured by 2012 with applications to antennas and communication
systems. Around that same period, 2007–2009, we went back to the
study of multi-frequency oscillations in arrays of nonlinear
oscillators. In fact, we were able to develop a systematic way to
manipulate collective frequency through cascade networks. The work
matured in 2012 with the modeling, analysis, design and fabrication
of the nonlinear channelizer. This is an integrated circuit made up
of large parallel arrays of analog nonlinear oscillators, which,
collectively, serve as a broad-spectrum analyzer with the ability
to receive complex signals containing multiple frequencies and
instantaneously lock-on or respond to a received signal in a few
oscillation cycles. Again, the conceptualization of the nonlinear
channelizer was based on the generation of internal oscillations in
coupled nonlinear systems that do not normally oscillate in the
absence of coupling. Between 2007–2011, we investigated various
configurations of networks of coupled vibratory gyroscopes. The
investigations showed that networks of vibratory gyroscopes can
mitigate the negative effects of noise on phase drift. But the
results were, mainly, computational and applicable only to small
arrays. Finally, between 2012–2015, we developed the necessary
mathematical approach to study networks of arbitrary size. This
work showed the nature of the bifurcations that lead arrays of
gyroscopes, connected bidirectionally, in and out of
synchronization. The results were applicable to net- works of
arbitrary size.
In the past few years, previous works have led us into new topics.
Networks of energy harvesters, which, interestingly, are governed
by ODEs that resemble those of vibratory gyroscopes. This feature
highlights again the model-independent nature of the analysis of
differential equations with symmetry. In the year 2011, in par-
ticular, we started a new project to study the collective behavior
of spin-torque nano-oscillators. The motivation for this work is a
conjecture by the 2007 Nobel
Preface ix
Laureate, Prof. Albert Fert, about the possibility that
synchronization of nano-oscillators could produce substantial
amounts of microwave power for prac- tical applications.
Determining the regions of parameter space of stable synchro- nized
solutions was a very challenging problem due to the nature
(non-polynomial form) of the governing equations. Finally, this
year we overcame the major diffi- culties by exploiting, again,
equivariant bifurcation theory. And the most recent project that we
started in 2016 is about networks of coupled oscillators for
improving precision timing with inexpensive oscillators, as oppose
to atomic clocks.
Along the way, several patents were approved by the U.S. Patent
Office for the works related to these projects, including:
2007 U.S. Patent # 7196590. Multi-Frequency Synthesis Using
Symmetry Methods in Arrays of Coupled
Nonlinear Oscillators. 2008 U.S. Patent # 7420366. Coupled
Nonlinear Sensor System. 2009 U.S. Patent # 7528606. Coupled
Nonlinear Sensor System for Sensing a Time-Dependent Target
Signal
and Method of Assembling the System. 2011 U.S. Patent # 7898250.
Coupled Fluxgate Magnetometers for DC and Time-Dependent (AC)
Target
Magnetic Field Detection. 2011 U.S. Patent # 8049486. Coupled
Electric Field Sensors for DC Target Electric Field Detection. 2012
U.S. Patent # 8049570. Coupled Bistable Microcircuit for
Ultra-Sensitive Electric and Magnetic Field
Sensing. 2012 U.S. Patent # 8212569. Coupled Bistable Circuit for
Ultra-Sensitive Electric Field Sensing Utilizing
Differential Transistors Pairs. 2015 U.S. Patent # 8994461. Sensor
Signal Processing Using Cascade Coupled Oscillators. 2015 U.S.
Patent # 9097751. Linear Voltage Response of Non-Uniform Arrays of
Bi-SQUIDS. 2016 Under review. Navy Case: 101427. Enhanced
Performance in Coupled Gyroscopes and Elimination of Biasing
Signal in a Drive-free Gyroscope. 2016 Under review. Navy Case:
101950. Arrays of Superconducting Quantum Interference Devices with
Self Adjusting
Transfer to Convert Electromagnetic Radiation into a Proportionate
Electrical Signal to Avoid Saturation.
2016 Under review. Navy Case: 102297. 2D Arrays of Diamond Shaped
Cells Having Multiple Josephson Junctions. 2016 Under review. Navy
Case: 103829. Network of Coupled Crystal Oscillators for Precision
Timing.
x Preface
None of these projects would have been possible without the active
participation of students, joint work with collaborators, and the
financial support from various sources. We would like to thank each
of the students first: John Aven [21, 22], Jeremmy Banning [26],
Katherine Beauvais [30], Susan Berggren [36, 159], Bernard Chan,
Nathan Davies [91], Scott Gassner [132, 133], Mayra Hernandez
[168], Habib Juarez, Tyler Levasseur, Patrick Longhini [261, 262],
Daniel Lyons [266, 267], Antonio Matus [272], Derek Moore, Loni
Olender, Steven Reeves [331] Norbert Renz [332], Richard Shaffer
[359], Brian Sturgis-Jensen, James Turtle [398, 399], Huy Vu [407],
Sarah Wang, Bing Zhu [433]. Special acknowledgement and thanks to
Patrick Longhini, he was the first student that got involved in the
work through his Master and, later on, Ph.D. thesis. He continues
to be an extremely valuable asset to multiple ongoing projects.
Collaborators include: Bruno Ando (Univ. of Catania), Marcio De
Andrade (SPAWAR), Salvatore Baglio (Univ of Catania, Italy), Peter
Blomgren (SDSU), Donald Bowling (NAWC), Pietro-Luciano Buono (Univ.
of Ontario Institute of Technology, Canada), Adi Bulsara (SPAWAR),
Lowell Burnett (QUASAR), Juan Carlos Chaves (HPTi), Ricardo
Carretero (SDSU), Anna Leese de Escobar (SPAWAR), Jocirei Dias
Ferreira (Federal Univ. of Mato Grosso, Brazil), Hugo
Gonzalez-Hernadez (Instituto Tecnologico de Monterrey), Frank
Gordon (SPAWAR), Takachi Hikihara (Kyoto Univ., Japan), Calvin
Johnson (SDSU), Andy Kho (SPAWAR), Daniel Leung (SPAWAR), John F.
Lindner (College of Wooster), Norman Liu (SPAWAR), Joseph M.
Mahaffy (SDSU), LT Jerome McConnon (SPAWAR), Brian K. Meadows
(SPAWAR), Oleg Mukhanov (HYPRES), Joseph Neff (SPAWAR), Suketu Naik
(Weber State Univ.), Martin Nisenoff (M. Nisenoff Associates),
Georgy Prokopenko (HYPRES), Wouter-Jan Rappel (UCSD), LT Sarah Rice
(SPAWAR), Robert Romanofsky (NASA), Vincenzo Sacco (Univ. Catania),
Benjamin Taylor (SPAWAR), Edmond Wong (SPAWAR), Yongming Zhang
(QUASAR).
Many thanks to the Chaos Group at the Oak Ridge National
Laboratory, Stuart Daw, Charles Finney, and Sreekanth Pannala, for
very stimulating discussions. We also wish to acknowledge very
fruitful interactions with John Angus and Ali Nadim who served as
committee members for many of the Ph.D. theses that derived from
related projects. Special thanks to Pietro-Luciano Buono with whom
we have collaborated extensively in recent years to apply advance
methods from equivariant bifurcation theory. One of the authors,
Antonio Palacios, wishes to thank Marty Golubitsky, in particular,
for his mentorship and guidance to learn from him (during a
postdoctoral appointment) the principles and methods for studying
dynamical systems that posses symmetry.
We also wish to acknowledge the financial support provided by
several agencies to conduct the necessary research work that serves
as the foundation of some of the technologies discussed in this
book, including: Army Research Office, Department of Defense,
Department of Energy, the National Science Foundation, the National
Security Agency, the Office of Naval Research, the San Diego
Foundation, and the Space and Naval Warfare Center, San Diego. We
wish to acknowledge the con- tinuous support of Dr. Michael
Shlesinger from ONR.
Preface xi
The book is intended for a broad audience. For engineers who might
be inter- ested in applying ideas and methods from dynamical
systems with symmetry and equivariant bifurcation theory to design
and fabricate novel devices. For mathe- maticians and physicists
who might be interested in translational research work to
extrapolate fundamental research theorems into practical
applications. And for scientists from many disciplines, viz.
Biology, Chemistry, Computer Science, Geology, etc., who might be
interested in the interplay between theory and real-life
applications from the general field of nonlinear science.
The book is organized as follows. In Chap. 1 we present fundamental
ideas of complex networks and bistability, which is a common
feature of many sensor devices; and then we dedicate a few sections
to introduce basic ideas, methods and examples in the analysis of
differential equations (ODEs and PDEs) with symmetry. One
particular class of solutions that rarely appears in generic
versions of systems of differential equation are heteroclinic
cycles. These types of solutions are, how- ever, generic features
of systems with symmetry. We exploit these cycles to enhance
sensitivity and, thus, we dedicate a section to explain what they
are and how they can be found. The book is then organized in two
parts. Part I, Chap. 2 through Chap. 6 is dedicated to
translational research work that already led to mature
technologies. These technologies include networks of fluxgate
magne- tometers; arrays of micro-electronic electric field sensors;
networks of SQUIDs; cascade arrays of nonlinear oscillators for
multi-frequency generators; and a special chapter in honor of the
theoretical work by Pietro-Luciano and Marty Golubitsky: a device
realization of a Central Pattern Generator network of the animal
gaits studied by them. Part II, Chap. 7 through Chap. 10 include,
mainly, theoretical works that have not yet mature into actual
device realizations. The technologies that may derive from these
works are part of ongoing efforts.
San Diego, USA Visarath In 2017 Antonio Palacios
xii Preface
Contents
1 A Unifying Theme . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . 1 1.1 Complex Networks . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2
Bistability . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . 5 1.3 Self-oscillating Networks . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.4 The Role
of Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . 10 1.5 Symmetry-Breaking Bifurcations . . . . . . . . . . .
. . . . . . . . . . . . . 13 1.6 Coupled Cell Systems. . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . 18 1.7
Heteroclinic Connections . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 21 1.8 Representative Projects . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 30
2 Coupled-Core Fluxgate Magnetometer . . . . . . . . . . . . . . .
. . . . . . . . 37 2.1 Fluxgate Technology . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 37 2.2 Modeling
Single-Core Dynamics . . . . . . . . . . . . . . . . . . . . . . .
. 41 2.3 Coupled Single-Domain System . . . . . . . . . . . . . . .
. . . . . . . . . . 43 2.4 Frequency Response . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 52 2.5 Sensitivity
Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . 54 2.6 Alternating Configuration . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . 61 2.7 AC Field Detection. . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.8 Target Signal Contamination . . . . . . . . . . . . . . . . . .
. . . . . . . . . 71 2.9 Effects of Nonhomogeinities . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 74 2.10 Effects of Delay
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . 81 2.11 Laboratory Implementation . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . 88
3 Microelectric Field Sensor . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . 105 3.1 Overview . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.2 Circuit Equations . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 106 3.3 Analysis and Bifurcation
Diagrams. . . . . . . . . . . . . . . . . . . . . . . 110 3.4
Numerical and Experimental Results . . . . . . . . . . . . . . . .
. . . . . 111 3.5 Period and Residence Times Response . . . . . . .
. . . . . . . . . . . . . 118 3.6 SPICE Simulations . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . 120
xiii
4 Superconductive Quantum Interference Devices (SQUID) . . . . . .
. . 127 4.1 History of Superconductivity . . . . . . . . . . . . .
. . . . . . . . . . . . . . 128 4.2 The Josephson Effect and SQUID
Technology . . . . . . . . . . . . . . 131 4.3 Phase-Space Dynamics
of DC SQUID . . . . . . . . . . . . . . . . . . . . 132 4.4 Chimera
States in Non-locally Coupled Arrays. . . . . . . . . . . . . . 138
4.5 The DC Bi-SQUID. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . 141 4.6 Serial Bi-SQUID Array. . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 147 4.7 Parallel
Bi-SQUID Array . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . 151 4.8 Design, Fabrication, and Evaluation . . . . . . . .
. . . . . . . . . . . . . . 154
5 Frequency Conversion . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . 165 5.1 Frequency Up-Conversion . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . 165 5.2
Experiments on Frequency Up-Conversion . . . . . . . . . . . . . .
. . . 176 5.3 Frequency Down-Conversion . . . . . . . . . . . . . .
. . . . . . . . . . . . . 182 5.4 Experiments on Frequency
Down-Conversion . . . . . . . . . . . . . . 191 5.5 Large Frequency
Downconversion Ratios. . . . . . . . . . . . . . . . . . 199 5.6
Nonlinear Channelizer . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . 204 5.7 Experimental Setup of Nonlinear
Channelizer . . . . . . . . . . . . . . 218
6 ANIBOT: Biologically-Inspired Animal Robot . . . . . . . . . . .
. . . . . . 229 6.1 Central Pattern Generators . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . 229 6.2 CPG Network Topology
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.3 Analog Fitzhugh–Nagumo Neuron Circuit . . . . . . . . . . . . .
. . . . 232 6.4 Patterns and Locomotion . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . 235 6.5 Leg Motion. . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
236
7 Gyroscope Systems . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . 237 7.1 Motivation . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
7.2 History of Navigation Systems . . . . . . . . . . . . . . . . .
. . . . . . . . . 238 7.3 Evolution of Gyroscopes . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 240 7.4 Vibratory
Gyroscopes . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . 244 7.5 Bi-Directionally Coupled Ring . . . . . . . . . .
. . . . . . . . . . . . . . . . 249 7.6 Unidirectionally Coupled
Ring . . . . . . . . . . . . . . . . . . . . . . . . . . 264 7.7
Drive-Free Gyroscope System . . . . . . . . . . . . . . . . . . . .
. . . . . . 277 7.8 Hamiltonian Approach . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . 285
8 Energy Harvesting. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . 295 8.1 State of the Art . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
295 8.2 Single Magnetostrictive Energy Harvesting Model . . . . . .
. . . . . 299 8.3 Coupled Energy Harvester System . . . . . . . . .
. . . . . . . . . . . . . . 302 8.4 Computational Bifurcation
Results . . . . . . . . . . . . . . . . . . . . . . . 305 8.5
Hamiltonian Analysis . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . 306 8.6 Experimental Validation . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 314
xiv Contents
9 Spin Torque Nano Oscillators . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . 317 9.1 The Giant Magnetoresistance (GMR)
Effect . . . . . . . . . . . . . . . . 317 9.2 Spin Torque Nano
Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . .
320 9.3 Landau-Lifshitz-Gilbert-Slonczewski (LLGS) Model . . . . .
. . . . 321 9.4 The Synchronization Challenge . . . . . . . . . . .
. . . . . . . . . . . . . . 323 9.5 Series Array . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
9.6 Complex Stereographic Projection . . . . . . . . . . . . . . .
. . . . . . . . 328 9.7 Hopf Bifurcation Curves. . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . 330 9.8 Nonlinear
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . 333 9.9 Locking into Synchronization . . . . . . . . .
. . . . . . . . . . . . . . . . . . 336
10 Precision Timing . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . 339 10.1 History of Precision
Timing Devices . . . . . . . . . . . . . . . . . . . . . 339 10.2
Crystal Oscillators. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . 341 10.3 Two-Mode Oscillator Model. . . . .
. . . . . . . . . . . . . . . . . . . . . . . 343 10.4 Coupled
Crystal Oscillator System . . . . . . . . . . . . . . . . . . . . .
. . 348 10.5 Averaging and Symmetries . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . 352 10.6 Linearization and Spectrum at
the Origin . . . . . . . . . . . . . . . . . . 357 10.7 Stability
and Bifurcation Results. . . . . . . . . . . . . . . . . . . . . .
. . . 360 10.8 Numerical Continuation . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . 366 10.9 Phase Error . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
375 10.10 Experiments . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . 379
References . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . 383
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . 401
Contents xv
Chapter 1 A Unifying Theme
Nonlinear systems can behave very robustly, yet they can also
exhibit high sensitivity to small perturbations [205, 206, 277,
404, 420]. This dichotomy is a consequence of the parameters that
govern the behavior of a given system. In regionswhere changes in
parameters do not yield radically different behavior the system is
expected to behave robustly. However, in regions that contain a
bifurcation point, high-sensitivity can ensue. Close to the onset
of oscillations, in particular, a small perturbation can drasti-
cally alter the behavior of a nonlinear oscillator, it can either
destroy the oscillations completely or it can change their
characteristics such as frequency and amplitude. For instance,
temperature variations can reset a circadian oscillator, see Fig.
1.1 while noise [277] can induce an exogenous oscillator to
oscillate in regions where only steady-state behavior would
dominate the system dynamics absent the noise. Similar sensitivity
features can be used for signal amplification in electrical,
mechanical and optical systems [213]. Exploring those features is
the subject of this chapter and of the entire book.
1.1 Complex Networks
Over the past seventeen years, we and other colleagues and students
have been developing, through theory and experiments, a new
paradigm that combines ideas and methods from the theory of
Nonlinear Dynamical Systems with Symmetry in Mathematics, Physics,
and Engineering, for performance enhancement of nonlinear devices.
A unifying theme and goal of this paradigm is to demonstrate,
theoretically and experimentally, that collective behavior, which
is uniquely produced by inter- connected nonlinear devices, can be
exploited to develop novel complex network based systems that can
outperform the function of their individual counterparts. After
all, many sensory systems [225] in animals are controlled by
clusters of neu- ron cells, located in the central nervous system
and physically interconnected to
© Springer-Verlag GmbH Germany 2018 V. In and A. Palacios, Symmetry
in Complex Network Systems, Understanding Complex Systems, DOI
10.1007/978-3-662-55545-3_1
1
Fig. 1.1 Endogenous circadian oscillators of Drosophila and
Mammals. Source Journal of Cell Science 119, 4793–4795 (2006)
produce, somehow, an optimal response, which in many cases
outperforms those of humans. A dog’s sense of smell is said to be a
thousand times more sensitive than that of humans, thanks to the
collective behavior of more than 220 olfactory receptor cells in
its nose, see Fig. 1.2(left), while the temporal resolution of the
com- pound eye (Drosophila melanogaster) of a fruitfly eye is
approximately ten times better than that of humans due to the
collective output of 6400 photoreceptor cells, tightly bound to one
another, see Fig. 1.2(right). These biologically-inspired ideas are
model-independent, so we hope they can stimulate further
development of other sensory systems.
The collaborative work has lead to a broad class of nonlinear
devices been fabri- cated, including: highly-sensitive, low power,
miniaturized magnetic- and electric- field sensors; networks of
SQUID (Superconducting Quantum Interference Devices) for the
development of a sensitive, low noise, significantly lower Size,
Weight and Power (SWaP) antenna integrated with Low-Noise Amplifier
(LNA); arrays of gyro- scopeswith reduced phase drift for improved
navigation; frequency up/down convert- ers; a broad-spectrum
analyzer with the ability to receive complex signals containing
multiple frequencies and instantaneously lock-on or respond to a
received signal in a few oscillation cycles; networks of energy
harvesters with increased power output; an inexpensive precision
timing device through coupled nonlinear oscillators circuits; and
networks of spin-torque nano-oscillators for generating microwave
signals at the nano scale. Figure1.3 shows a composite snapshot of
the multiple projects outlined above. To provide readerswith a
description of the translational work, involving, con-
1.1 Complex Networks 3
Fig. 1.2 Biological sensors made up of coupled cell systems. (Left)
A dog’s smell is thousand times more sensitive than that of humans
thanks to the collective behavior of more than 220 olfac- tory
receptor cells. (Picture courtesy of
http://www.faithfool.wordpress.com). (Right) The temporal
resolution of a firefly is ten times better than in humans due to
the collective output of approxi- mately 6400 photoreceptor cells
tightly coupled to one another. Picture courtesy of “Research Penn
State” [306]
Fig. 1.3 Representative projectswith a unifying theme: to exploit
the collective behavior of complex network systems to model,
analyze and design advanced engineering systems. From top-right, in
counter-clockwise direction, coupled nonlinear oscillators for
antennas; highly-sensitive coupled- core fluxgate magnetometer;
networks of vibratory gyroscopes with optimal phase drift;
thousands of superconductive loops for communication systems;
arrays of spin-torque nano oscillators for microwave signal
generation; and networks of energy harvesting systems
ceptualization, derivation of mathematical models, analysis, design
and fabrication of these technologies, is the main goal of this
book.
The technologies that are derived from the proposed network-based
configuration can be effectively described as hybrid,
discrete-continuous, nonlinear dynamical system: a discrete number
of individual components, in which each unit is governed by a
continuousmodel of nonlinear differential equations, physically
coupled to form the network structure. These networks can also be
described as complex networks in the sense that their response
cannot be achieved by individual units, i.e., only by the mutual
interaction of the individual components.
Although a direct relation to biological systems has yet to be
demonstrated inmore detail, we show proof of concept that a
network-based approach can indeed lead to significant performance
enhancements for certain classes of sensors. In particular:
fluxgatemagnetometers, electric-field sensors, and SQUID sensors.
But the emphasis of the proposed approach onmagnetic and
electric-field sensors is not exhaustive. The basic principles of
cooperative behavior can be applied to enhance the performance of a
wide range of nonlinear systems, such as antennas and radar
systems; gyroscopes, energy harvesters; precision timing devices
and microwave signal generators. By cooperative we mean patterns of
collective behavior. For instance, we show that complete
synchronization, in which all units oscillate in-phase with the
same wave form and amplitude, can lead to significant reduction in
the phase drift that affects the performance of gyroscopes. On the
other hand, heteroclinic cycles are found to be ideal patterns to
increase sensitivity in networks of fluxgate magnetometers.
Similarly, traveling wave patterns can be exploited to improve the
performance of precision timing devices. Which pattern of behavior
is better suited for a particular application is a model dependent
issue that has to be studied on a case-by-case basis. However, the
existence of these, and many other related patterns of collective
behavior are governed, mainly, by a common feature, symmetry.
In all the systems that are described in this book, our objectives
include to inves- tigate the following model-independent
issues:
(i) To investigate, analytically, computationally, and
experimentally, the interplay between the discrete and continuous
characteristics of the network dynamics.
(ii) To determine the fundamental limit of performance enhancement,
e.g., reduc- tion in phase drift or increase in sensitivity, that
can be achieved with a network- based device.
(iii) To develop design rules for nonlinear devices.
To accomplish these goals, we draw on methods from dynamical
systems and bifurcation theory in systemswith symmetry to address
someof the following,model- dependent, fundamental issues:
• What is the optimal network architecture that can produce the
highest-level of performance enhancements at the lowest cost and
with the lowest power con- sumption?
• How can a network be built in hardware? What are the advantages
and disad- vantages of different hardware implementations, and how
do they compare to simulations in software?
1.1 Complex Networks 5
• How can a network be programmed and controlled to operate in
realistic (noisy) environments?
Since the majority of the technologies that are discussed in this
book employ nonlinear oscillator components, we present next a
brief introduction of basic ideas, concepts and principles.
1.2 Bistability
Many complex systems, natural and artificial ones, exhibit
oscillatory behavior, i.e., cyclic behavior that repeats at regular
intervals. Examples include: the rhythmic light pulses of fireflies
[53, 111], see Fig. 1.4, the electrical activity of neuron cells
that make up central pattern generators in biological systems [63,
83, 85, 148, 226, 227, 414], the patterns of lights produced by
arrays of coupled lasers [328, 419], voltage variations in modern
communication systems [315, 426], the growth and decay of
population sizes between competing species [273–275], bubble
formation and evolution in fluidization and mixing processes [163],
and variations in phase and current in arrays of Josephson
junctions [18, 99, 100, 160] in quantum physics.
In the absence of noise, the underlying cyclic fluctuations in a
given system can arise from individual units that oscillate on
their own, also known as endogenous or self-excited oscillators, or
from exogenous units that oscillate only when they are externally
driven or coupled together. Circadian rhythms, which regulate the
daily cycle of many living organisms, plants, and animals, for
instance, are endogenously
Fig. 1.4 Complex interactions among fireflies can lead them to
coordinate the rhythmic flashing lights produced by each individual
firefly. Collectively, the swarm can then achieve synchronization
and oscillate in unison. Source National Geographic
6 1 A Unifying Theme
generated. In fact, the first endogenous circadian oscillation to
be observed was the movement of the leaves of Mimosa pudica, a
plant studied by the French scientist Jean-Jacques dÒetous de
Mairan.1
In addition, bistability—the property that allows a system to rest
in either of two states—underlies the basic oscillatory behavior of
many other natural and artificial systems. Statesmay include
typical invariant sets, such as equilibriumpoints, periodic and
quasi-periodic solutions, and chaotic attractors. In the absence of
an external stimulus, the state variable x(t) of a bistable system
will relax to one of the invariant sets, and it will remain in that
state unless it is switched or forced to another state. It is in
this sense that the system exhibits “memory.” Which invariant set
the system will relax to depends typically on the set of initial
conditions. All bistable systems employ some form of energy source
as the underlying principle that allows them to switch between
states. The source of energy is due typically through external
forcing or through the coupling mechanism.
For instance, dynamic sensors [150, 210, 252, 326, 334], operate as
exoge- nous oscillators with nonlinear input-output
characteristics, often corresponding to a bistable potential energy
function of the form
dx
dt = −∇U (x), (1.1)
where x(t) is the state variable of the natural system or
artificial device, e.g., mag- netization state, and U is the
bistable potential function. Examples include: fluxgate
magnetometers [45, 139], ferroelectric sensors [27], and mechanical
sensors, e.g., acoustic transducers made with piezoelectric
materials. Fig. 1.5(top) illustrates the case of a double-well
potential function U (x) = −ax2 + bx4, whose minima are located at
±xm and the height of the potential barrier between the two minima
is labeled by U0.
Without an external excitation (periodic forcing or noise), the
state point x(t) of the exogenous oscillator described by Eq. (1.1)
will rapidly relax to one of two stable attractors, which
correspond to the minima of the potential energy function U (x). In
the presence of an external periodic forcing term f (t), with
frequency ω, the state variable in U (x + f (t)) can be induced to
oscillate periodically (with a well- defined waveform) between its
two stable attractors−xm and+xm , as is illustrated in Fig.
1.5(bottom). The forcing term is also known as biasing signal in
the engineering literature.
Standard Detection Mechanism. To detect a small target signal (dc
or low- frequency), typically the standard, spectral-based [323,
335, 336, 338], readout mechanism is employed. Assume ε to be the
target signal. When ε = 0, the power spectral density contains only
the odd harmonics of the bias frequency ω. But when ε > 0, the
potential energy function U (x + f (t) + ε) is skewed, resulting in
the appearance of even harmonics; the response at the second
harmonic 2ω is then used
1Source: Wikipedia https://en.wikipedia.org.
1.2 Bistability 7
Fig. 1.5 (Top) Bistable Potential U (x) = −ax2 + bx4. (Bottom)
Switching between wells of a potential function can be achieved by
a sufficiently large biasing signals (or noise) greater than the
potential barrier
x
U0
to detect and quantify the target signal, as is shown in Fig. 1.6.
The standard readout mechanisms has some drawbacks, however. Chief
among them is the requirement of a large onboard power to provide a
high-amplitude, high-frequency, bias signal. The feedback
electronics can also introduce their own noise floor into the
measurement process, and finally, a high-amplitude, high-frequency,
bias signals often increase the noise floor of the system.
In biological systems, bistability is a key feature for
understanding and engineer- ing cellular functions such as: storing
and processing information by the human brain during the
decision-making process [6]; regulation of the cell cycle [368,
400]; sporu- lation, which controls the timing and dynamics of
dramatic responses to stress [406]; design and construction of
synthetic toggle switches [131]; and in gene regulatory networks
responsible for embryonic stem cell fate decisions [78]. In
chemical sys-
8 1 A Unifying Theme
0.2 0.4 0.6 0.8 1 10−8
10−6
10−4
10−2
10−6
10−4
10−2
P S
D (f
) Fig. 1.6 Power Spectrum Decomposition of the oscillations in a
bistable overdamped system subject to a periodic forcing. (Left) In
the absence of an external signal the PSD shows only even
harmonics. (Right) In the presence of an external signal, however,
the PSD exhibits odd as well as even harmonics. Typically, the
first even harmonic is used as a detection mechanism
tems bistability is central to the analysis of relaxation kinetics
[422]. In mechanical systems, bistablemechanisms are commonly
employed in the design and fabrications
ofMicro-Electro-Mechanical-Systems (MEMS) versions of relays,
valves, clips, and threshold switches [175, 325]. In electronics,
hysteresis and bistability are combined to design and fabricate
Schmitt trigger circuits, which convert analog input signals to
digital output signals [353]. In neuroscience, bistability is at
the core of Hopf models [86, 178, 180], which describe the
input/output response of neurons through differential equations of
the form
τi dVi
dt = −Vi + g(Vi ), (1.2)
where τi is a suitable time constant that controls how quickly
unit(neuron) i responds to a stimulus, Vi is the output (typically
voltage) of unit i , and g is the activation function, which
normally represents a saturation nonlinearity property of
neurons.
Alternatively, the phenomenon of Stochastic Resonance [127, 128,
171, 204, 206, 277] can also serve as a mechanism for generating
oscillations, see Fig. 1.7. Briefly speaking, stochastic resonance
refers to noise-induced transitions which, due to the nonlinear
nature of a system dynamics, become synchronized with the period of
an external force. A distinctive feature of stochastic resonance is
a rapid increase in a system’s output Signal-to-Noise Ratio (SNR)
under weak coupling followed by a slower decrease in SNR for
stronger coupling. At intermediate noise intensities, the system
exhibits maximum SNR. This feature is inherently due to the
nonlinear nature of a system’s dynamics and cannot be reproduced by
linear systems.
1.3 Self-oscillating Networks 9
Fig. 1.7 (Left) Motion of a particle on a double-well potential
function. In the absence of noise, the system dynamics quickly
settles into an equilibrium point. Which equilibrium point is
selected depends on the initial conditions. (Right) In the presence
of noise, the system dynamics now lingers intermittently between
the two equilibrium states of the deterministic system,
independently of initial conditions
1.3 Self-oscillating Networks
A fundamental idea in the new paradigm that we have developed for
performance enhancement in nonlinear systems is simple: make
exogenous oscillators behave as endogenous ones. That is, we seek
to minimize power consumption, and simulta- neously enhance output
response, by replacing force-driven exogenous oscillators with
self-oscillating networks. The paradox can be resolved through
symmetry. This point deserves a little more explanation.
Many systems are known to oscillate only when they are driven by an
external force. However, when they are connected in some fashion,
the symmetry of the resulting topology of connections, i.e., which
units are coupled with each other, and the nonlinear
characteristics of each individual unit, can be exploited to induce
the interconnected network to generate a collective pattern of
oscillation via an appropriate coupling function, see Fig.
1.8.
From a mathematical standpoint, the choice of coupling function can
be any type of function, leading to a wide range of network
solutions. From an engineering standpoint, the coupling function is
restricted, however, by the type of system or technology being
used. This is, in other words, a model-dependent feature of the
system. For instance, fluxgate magnetometers coupled through
magnetic flux are restricted to unidirectional coupling since
directing magnetic flux in both directions can be extremely
complicated to achieve. Mechanical gyroscopes can be coupled,
however, bidirectionally through a series of mass-spring
systems.
An immediate advantage of the self-oscillating network approach is
the cost reduc- tion that can be achieved by eliminating
power-driven components which are typi- cally very expensive. A
more subtle advantage is to exploit the collective pattern of
oscillation to enhance output response. Consider for instance a
sensor device (mag- netic or electric) whose behavior (as is
described earlier) is governed by a potential
10 1 A Unifying Theme
2
1
5
−0.5
0
0.5
1
Time
0 0.2 0.4 0.6 0.8 1 10−5
100
105
Frequency
D Fig. 1.8 (Left) Representative example of self-oscillating
networks. Individually, each unit cannot oscillate. (Right) Under
certain coupling topologies, and the internal dynamics of each
unit, the network can, collectively, oscillate
function of the form (1.1). A self-oscillating network would
eliminate the need of the large onboard power that is required for
a biasing signal to overcome the energy- barrier of the system and
thus induce the required oscillations. Furthermore, the network
oscillations can yield (under certain conditions) wave forms whose
char- acteristics (frequency, amplitude, and phase) can exhibit
higher sensitivity to very small target signals.
The symmetry-based approach to study nonlinear systems, while
nonstandard, is not entirely new among the mathematics community.
However, there is much less familiarity with the techniques of
symmetry-breaking bifurcation, and applications of equivariant
bifurcation theory, developed by Golubitsky and Stewart [146, 147,
149], as they apply to the engineering, design and fabrication, of
complex systems. For this reason, we dedicate the next few sections
to introduce, first, the mathemati- cal formalism to describe
symmetry-breaking bifurcations; networks architecture of
interconnected devices, through the coupled cell formalism and,
then, a more com- prehensive discussion of the existence of
heteroclitic cycles in nonlinear systems.
1.4 The Role of Symmetry
A general approach for the analysis of complex network systems has
been to derive a detailed model of its individual parts, connect
the parts and note that the system contains some sort of symmetry,
then attempt to exploit this symmetry in order to simplify
numerical computations. This approach can result in very
complicated models that are difficult to analyze even
numerically.
Our approach in this book is, however, aimed at promoting a
unifying theme for the use of symmetry in a systematic way. First,
determine the conditions for
1.4 The Role of Symmetry 11
the existence and stability of collective patterns of behavior of a
complex system. Then perform translational work to transfer
theorems and principles into prototypes devices that conform, as
close as possible, to a particular model. For instance, the
phenomenon of coupling-induced oscillations that regulates the
dynamics of the newly postulated coupling-based sensor devices is
dictated, mainly, by the groupZN , of cyclic permutations of N
objects, i.e., units connected with a preferred direction or
directionally. It is in this sense that the ideas and methods of
our approach are device-independent so that similar principles can
be readily applied to understand, and hopefully, to enhance the
performance of a wide variety of sensor devices so long as the
symmetry conditions are satisfied.
A few more details of how symmetry can appear in governing
equations, such as in systems of differential equations, are now in
order.
Definition 1 Symmetry is a geometrical concept that describes the
set of transfor- mations that leave an object unchanged.
In complex systems with continuous nonlinear behavior, the objects
are the gov- erning equations,which typically consist of systemsof
ordinary differential equations (ODEs) or partial differential
equations, and the transformations are the changes in the
underlying variables that leave the equations unchanged.More
formally, consider the following system of ODEs
dx
dt = f (x,λ), (1.3)
where x ∈ Rn , λ ∈ Rp is a vector of parameters and f : Rn × Rp → R
is a smooth function. Let γ be a particular time-independent
transformation inRn . Direct substi- tution of the transformed
variable γx into (1.3) yields γ x = f (γx,λ). Consequently, for the
system of ODEs (1.3) to remain unchanged, f must commute with γ. In
other words, f (γx) = γ f (x). In practice, the set of all
transformations that commute with f forms a group Γ . We then
arrive to the following formal definition.
Definition 2 A system of ODEs such as (1.3) is said to have Γ
-symmetry if
f (γx,λ) = γ f (x,λ), (1.4)
for all x ∈ Rn and for all γ ∈ Γ .
Equation (1.4) also implies that f is Γ -equivariant. But more
importantly, it implies that if x is a solution of (1.3) then so is
γx(t) for all γ ∈ Γ . In fact, the collection of points γx(t), for
all γ ∈ Γ forms a set called the group orbit of Γ :
Γ x = {γ x : γ ∈ Γ }.
Furthermore, the concept of group orbit applies to any point x(t),
not just equi- librium solutions.
12 1 A Unifying Theme
Fig. 1.9 Circuit realization of a Van der Pol oscillator. The
governing equations exhibit reflectional symmetry with respect to
the state variables
Example 1 As an example, consider the van der Pol circuit depicted
in Fig. 1.9. IL
and IC are the currents across the inductor L and capacitor C ,
respectively. IR is the current across two resistors R1 and R2
located inside the rectangle labeled R in which F(V ) = −V/R1 + V
3/(3R2
2). The dynamics of the circuit shown in Fig. 1.9, after rescaling,
is governed by the
following second order scalar ODE
d2V
where δ = 1/(R2C), p = R2/R1, ω = 1/ √
LC . After a change of variables, we can rewrite the model equation
(1.5) as a first order system of the form
dx
(1.6)
where x(t) = V (t). We can then find two transformations that leave
this system unchanged: the identity transformation γ1 = id, where
γ1(x, y) → (x, y), and a second transformation, which can be
described abstractly as γ2 = −1, so that γ2(x, y) → (−x,−y). The
identity transformation is always a symmetry of any system, while
the second transformation γ2 corresponds to a reflection through
the origin in the phase space R2. Furthermore, it can be shown that
γ1 and γ2 are the only transformations that leave (1.6) unchanged.
Together, γ1 and γ2 form the group Z2 = {γ1, γ2} of symmetries of
the Van der Pol oscillator (1.6).
But individual solutions can also exhibit symmetry. For instance,
consider equi- librium solutions. A solution xe of Eq. (1.3) is an
equilibrium or steady-state if and only if
f (xe) = 0.
The symmetries of equilibrium points of a Γ -equivariant ODE form a
subgroup of Γ , which we define next.
1.4 The Role of Symmetry 13
Definition 3 Let xe represent an equilibrium or steady-state
solution of a Γ - equivariant system of ODEs. The symmetries of xe
form the isotropy subgroup Σ of Γ , which is defined by
Σxe = {γ ∈ Γ : γ · xe = xe} . (1.7)
1.5 Symmetry-Breaking Bifurcations
Equivariant systems of ODEs always posses a trivial solution x0
whose isotropy subgroup is the entire group of symmetries of the
model equations [146, 147, 149]. That is,Σx0 = Γ . But as
parameters are varied the system can exhibit a new solution with
less symmetry. That is, Σx ⊂ Γ . It is then said that the system
has undergone a symmetry-breaking bifurcation. We consider in this
chapter two types of symmetry- breaking bifurcations, steady-state
and Hopf bifurcations. The former case leads to new equilibrium
solutions while the latter to periodic oscillations. We describe
next each of these two cases.
Steady-State Bifurcations. The iconic picture in Fig. 1.10 of the
milk drop coronet by the pioneering work on speed photography by
Harold E. Edgerton illustrates best the phenomenon of steady-state
symmetry-breaking bifurcation. The pool of milk in its unperturbed
or trivial state is symmetric under arbitrary rotations and
reflections on a plane, which form the orthogonal group O(2). The
perturbation by the droplet breaks, however, the O(2) symmetry of
the trivial solution and it induces a crown-like shape with lesser
symmetry. The 24-sided polygon that appears by joining the
individual clumps now has D24-symmetry, where DN is the dihedral
group of symmetries of an N -gon.
Example 2 Consider now Euler’s column buckling experiment,
illustrated in Fig. 1.11. It consists of an elastic beam subjected
by a compressive force. Upon a critical value of the compressive
force the beam deforms into one of two buck- led states, either to
the right or to the left. Which state actually appears depends on
model-dependent features such as material imperfections or thermal
fluctuations. According to Bernoulli–Euler beam theory, a
mathematical model for the angle θ(t) between the undeformed road
and the tangent of the deformed rod is
E Iθ ′′ (x) + P sin θ(x) = 0, (1.8)
where x is the material coordinate, E is the elastic modulus, I is
moment of inertia, P is the compressive force and L is the length
of the beam. Equation (1.8) pos- sesses reflectional symmetry θ →
−θ, which is described by the group Γ = Z2. The unperturbed
unbuckled state is the trivial solution with Γ -symmetry. Assuming
boundary conditions θ(0) = θ(L) = 0, two nontrivial solutions are ±
sin(πx/L). The isotropy subgroup of these solutions is the trivial
group 1. It can be shown that these two solutions emerge through a
pitchfork bifurcation off of the trivial solution,
14 1 A Unifying Theme
Fig. 1.10 This iconic picture of the milk drop illustrates the
phenomenon of symmetry-breaking bifurcations. An unperturbed pool
of milk is invariant under arbitrary rotations and reflections on a
plane, which form the orthogonal group O(2). The crown-like shape
that emerges under the perturbation by the droplet is a 24-sided
polygon whose symmetries are described by the dihedral group D24.
Source Harold E. Edgerton, Milk Drop Coronet, 1957. 2010
Massachusetts Institute of Technology
Fig. 1.11 Euler beam experiment. An elastic beam is subjected to a
compressive force. Upon reaching a certain threshold value of the
compressive force, the trivial solution, unbuckled state, losses
stability and a buckled state, right or left, emerges through a
pitchfork bifurcation. Source Wikipedia
1.5 Symmetry-Breaking Bifurcations 15
i.e., via symmetry-breaking of Z2 symmetry. Now, if the beam were
to be cylindri- cal instead of rectilinear then the group of
symmetries of the experiment becomes Γ = O(2), just as in the milk
drop experiment. But now the symmetry-breaking mechanism could
lead, in principle, to a buckle mode with reflectional
symmetry.
A critical concept in the symmetry-based analysis of differential
equations is that of a representation of a group. Let Γ be a Lie
group and V a vector space. A representation ofΓ is a homomorphism
fromΓ to the group ofmatrices GL(V). That is, ρ : Γ → GL(V ). Thus,
a group element γ ∈ Γ describes the abstract structure of the
group, while ρ(γ) indicates how each group element acts on the
objects, e.g., equations. It is common practice to refer to V as
“the representation of γ”.
Absolute irreducibility is yet another concept that is used
systematically to deter- mine the generic type of bifurcations that
can occur in a symmetric system of ODEs. For instance, absolute
irreducibility excludes the existence of purely imaginary eigen-
values in the linearization of a model. Consequently, periodic
oscillations cannot emerge via Hopf bifurcations for absolutely
irreducible spaces. A brief definition follows but more details can
be found in [149, 186].
Definition 4 A representation of a group Γ on a vector space V is
absolutely irre- ducible if the only linear mappings on V that
commute with Γ are scalar multiples of the identity.
It is also awell-known fact that symmetry forces systemsofODEs to
have invariant linear subspaces. In particular, the fixed-point
subspace of a solution is the invariant subspace where the isotropy
subgroup acts trivially:
Definition 5 Suppose that Σ ⊂ Γ is a subgroup. Then the fixed-point
subspace
Fix(Σ) = { x ∈ Rn : σx = x ∀σ ∈ Σ
}
is a flow invariant subspace [149].
Fixed point subspaces describe the regions of phase space where a
particular solution resides. This suggests a model-independent
strategy to find solutions of symmetric systems of ODEs. Restrict
the equations to Fix(Σ) and then solve for the solutions. Since
Fix(Σ) is, in general, lower dimensional that the entire space then
it might be significantly easier to solve the restricted equations.
A critical observation is the fact that it might be possible, under
certain conditions, to predict the type of solutions of a symmetric
system of ODEs without having to solve for the solutions. Details
are formalized by the Equivariant Branching Lemma.
Theorem 1 (Equivariant Branching Lemma [149]) Let Γ ⊆ O(n) be a
compact Lie group acting absolutely irreducibly on Rn. Let
dx
16 1 A Unifying Theme
be a Γ -equivariant bifurcation problem so that
f (0,λ) = 0 (d f )0,λ = c(λ)I.
Assume c′(0) = 0 and let Σ ⊆ Γ satisfy
dim Fix(Σ) = 1.
Then there exists a unique branch of solutions to f (x,λ) = 0
bifurcating from (0, 0), where the symmetry of the solution is Σ
.
Example 3 Consider again the Euler beam experiment. Recall that the
symmetries of the experiment are described by the group Γ = Z2.
Furthermore, Σ = 1 is an isotropy subgroup in which Fix(Σ) = R, so
that dim Fix(Σ) = 1. Thus, by the Equivariant Branching Lemma, we
can predict the existence of a branch of steady- state solutions to
the idealized model
x = f (x,λ),
with trivial symmetry Σ = 1. Indeed, a Lyapunov–Schmidt reduction
[76] of the model Eq. (1.8) yields
x = λx − x3.
This reduced problem satisfies all conditions of the Equivariant
Branching Lemma. The steady-state solutions with trivial symmetry
are ±√
x . They emerge via a pitch- fork bifurcation.
Hopf Bifurcation. Symmetries of periodic solutions may arise in one
of two forms. As purely spatial symmetries or as a combination of
space and time symmetries. In the former case, a periodic solution
x(t) is fixed at every moment in time by some γ ∈ Γ , so that γ is
a purely spatial symmetry. This is similar to the symmetries of
steady-states discussed above.
Now, in the latter case, the solution trajectory is fixed by a
combination of the spatial action of γ ∈ Γ and a phase-shift θ ∈
S1, where S1 is the circle group of phase shifts acting on 2π
periodic functions. That is,
(γ, θ) · x(t) = γx(t + θ) = x(t), ∀t.
As it was the case of steady-state solutions, both types of
symmetries can be formally described through an extended version of
the isotropy subgroup for periodic oscillations.
Definition 6 Let x(t) represent a periodic solution of a Γ
-equivariant system of ODEs. The symmetries of x(t) form the
isotropy subgroup, which is defined by
1.5 Symmetry-Breaking Bifurcations 17
} . (1.10)
Observe that the case θ = 0 corresponds to purely spatial
symmetries. We will come back to discuss the issue of purely
spatial symmetries in more detail a little later in the next
section; within then context of coupled cell systems.
Generically, the existence of Hopf bifurcations in a symmetric
system of ODEs is determined by Γ -simple irreducible
representations. Formally,
Definition 7 A representation W of Γ is Γ -simple if either W is
composed of two copies of an absolutely irreducible representation,
so that W = V ⊕ V , or W is non-absolutely irreducible for Γ
.
In either of these two cases, it can be shown [149, 186] that Γ
-simple represen- tations lead to Jacobian matrices with the
following structure
(d f )(0,0) = [ 0 −Im
Im 0
] , (1.11)
where m = n/2. We can now state the equivalent of the Equivariant
Branching Lemma for Hopf
bifurcations with symmetry.
dx
dt = f (x,λ), x ∈ Rn, λ ∈ R (1.12)
be a Γ -equivariant bifurcation problem with Σ ⊂ Γ × S1. Assume the
linearization of (1.12) satisfies Eq. (1.11) with eigenvalues σ(λ)
± w(λ)i , each of multiplicity m and also
σ′(0) = 0.
If the action of Γ is Γ -simple on Rn and Σ satisfies
dim Fix(Σ) = 2
then there exists a unique branch of periodic solutions to f (x,λ)
= 0, with period near 2π, bifurcating from (0, 0), where the
symmetry of the solution is Σ .
Example 4 (Premixed Flame Dynamics) A mixture of either isobutane
and air, or propane and air, are burned on a circular porous plug
burner in a low pressure (0.3– 0.5atm) combustion chamber. The
process allowed for control of the pressure, flow rate, and fuel to
oxidizer ratio to within 0.1%. The simplest cellular pattern
generated by the burner is a large single cell with O(2) symmetry,
as is shown in Fig. 1.12.
Changes in the experimental parameters (type of fuel, pressure,
total flow, and equivalence ratio) resulted in different cellular
patterns in the flame front. In
18 1 A Unifying Theme
Fig. 1.12 Combustion experiments conducted by M. Gorman et al., at
the University of Houston [152] showcase cellular pattern
instability. Simplest pattern that appears is a homogeneous flame
front with the same O(2) symmetry as that of the circular
burner
Fig. 1.13 Four snapshots from the two-cell state of the flame front
that rotates clockwise. States with the opposite geometrical sense
(i.e., related by reflections) rotate counter clockwise
particular, it has been shown [311] that rotating flames emerge via
Hopf bifurcations that break theO(2) symmetry of the trivial or
homogeneous flame front. Figure1.13 illustrates a two-cell state
that rotates clockwise.
1.6 Coupled Cell Systems
Anaturalmathematical framework for the analysis of complex systems,
such as those that consist of arrays of coupled nonlinear
oscillators, is that of coupled cell system, Fig. 1.14 shows a
representative example with N = 4 cells coupled unidirectionally
(with varying coupling strength among different nodes) along a
ring. By a “cell” we mean an individual component or unit that
possesses its own dynamical behavior.
1.6 Coupled Cell Systems 19
Fig. 1.14 Representative example of a coupled cell system. Each
“cell” or unit has its own internal dynamics governed and the cells
are coupled to one another, in this case, unidirectionally with
varying coupling strength λ j i between nodes j and i
f1 f2
λN3
In what follows, we assume N identical cells, and consider the
internal dynamics of each cell to be governed by a k-dimensional
continuous-time system of differential equations of the form
d Xi
dt = f (Xi ,λ), (1.13)
where Xi = (xi1, . . . , xik) ∈ Rk denotes the state variables of
cell i and λ = (λ1, . . . ,λp) is a vector of parameters. Observe
that f is independent of i because the cells are assumed to be
identical. In engineering applications, such as in nonlinear
antenna technology, for instance, it is common for the cell
dynamics to be described by electrical oscillators, typically a Van
der Pol oscillator [115, 166].
Definition 8 A network of N cells is a collection of identical
cells interconnected in some fashion. We model the network by the
following system of coupled differential equations
d Xi
ci j h(Xi , X j ), (1.14)
where h is the coupling function between two cells, the summation
is taken over those cells j that are coupled to cell i , and ci j
is a matrix of coupling strengths.
Additionally, if we let X = (X1, . . . , X N ) denote the state
variable of the network, then we can write (1.14) in the simpler
form
d X
dt = F(X),
where the dependence in the parameters λ has been omitted for
brevity.
Local and Global Symmetries. Following the work of Dionne et al.
[97, 98], we distinguish local symmetries from global symmetries as
follows.
20 1 A Unifying Theme
Definition 9 Let O(N ) be the group of orthogonal transformations
in Rn . Then L ⊂ O(k) is the group of local or internal symmetries
of individual cells if, for all l ∈ L, we have
f (l Xi ,λ) = l f (Xi ,λ).
While local symmetries are dictated by f , global symmetries are
described by the coupling pattern. More precisely,
Definition 10 G ⊂ O(N ) is the group of global symmetries of the
network if, for all σ ∈ G, we have
F(σ X) = σF(X).
Considering again the case of Hopf bifurcation with symmetry, it
turns out that the set of all spatial symmetries also forms a
group:
K = {γ ∈ Γ : γX (t) = X (t) ∀t}.
In addition, there is also a subgroup H that preserves the overall
trajectories of a periodic solution but not necessarily the phases.
This group is defined as
H = {γ ∈ Γ : γ{X (t)} = {X (t)} ∀t}.
It is easier to explain the differences between these two subgroups
within the context of a coupled cell system, as is shown in the
next example.
Example 5 Consider the four-cell network of Fig. 1.15. The state of
the entire net- work is described by the spatio-temporal pattern X
(t) = (x(t), x(t + T/2), x(t), x(t + T/2)), where T is the common
period of oscillations. Observe that exchang- ing cells x1 and x3
or cells x2 and x4 leaves the pattern unchanged. In other words,
the network exhibits purely spatial symmetry generated by
reflections across both diagonals. These generators form the group
K = Z2 of purely spatial symmetries. Now if we were to permute the
cells cyclically, i.e., x1 → x2, x2 → x3, x3 → x4 and x4 → x1, the
overall combined trajectory of X (t) would remain the same, without
paying attention to the phase differences. It follows that H = Z4
is the group of symmetries that preserves the trajectory. In this
sense, H consists of the spatial parts (without phases) of the
spatio-temporal symmetries of X (t). Together, (H, K ) define the
spatio-temporal symmetries of the collective pattern, which can be
identifiedwith the isotropy subgroup ΣX (t) that was defined
earlier on.
While these examples are representative cases of symmetry-breaking
bifurca- tions we should point out that it also possible for the
changes in parameters to lead to symmetry-preserving bifurcations.
This can occur, for instance, within the context of networks of
interconnected nodes xi when all the individual nodes in the
emerging pattern behave exactly the same, i.e., x1 = x2 = · · · =
xN . If the bifurcation is of steady-state type then the emerging
solution is classified as a symmetric equilib- rium. But if the
nodes oscillate then we have symmetry-preserving Hopf
bifurcation
1.6 Coupled Cell Systems 21
Fig. 1.15 Representative example of a four-cell network with
identical nodes or cells coupled unidirectionally
x1
x(t)
x2
λ
λ
yielding synchronized oscillations Xsync = (x1(t) = x2(t) = . . .
xN (t)). The syn- chronized solution is associatedwith the
irreducible representationV = [1, 1, . . . , 1] and its isotropy
subgroup is the original group of symmetries, i.e.,ΣXsync = Γ ,
which explains why the bifurcation is called
symmetry-preserving.
Further example of Hopf symmetry-breaking patterns of oscillations
will be described later on in Chap.10 within the context of coupled
crystal oscillators for precision timing devices.
1.7 Heteroclinic Connections
In addition to the advantages of low power consumption that can be
achieved by coupling-induced oscillations, the presence of symmetry
in coupled nonlinear devices can lead to non-generic behavior,
e.g., heteroclitic cycles, which can be exploited to enhance
further sensitivity performance. In this case, the cycle involves
solution trajectories that connect sequences of equilibrium points.
Near the onset of the cycle, infinite-period oscillations emerge.
Then, due to the large-period of oscil- lation, the presence of a
target signal can create a large asymmetry, thus rendering a
heteroclinicwaveformhighly sensitive to detect target signals via
symmetry-breaking effects. Heteroclinic cycles are said to be
non-generic features of nonlinear systems because they typically do
not exist or it is very difficult to produce them. The presence of
symmetry can lead, however, to invariant subspaces of the phase of
a dynamical system through which cyclic connections are
facilitated. We make extensive use of these type of solutions, so
this section is dedicated to describing heteroclinic cycles and
their existence in greater detail.
Definition 11 In simple terms, a heteroclinic cycle is a collection
of solution tra- jectories that connects sequences of equilibria,
periodic solutions, and/or chaotic sets [63, 122, 238–240, 278]. As
time evolves, a typical nearby trajectory stays for increasingly
longer periods near each solution before it makes a rapid excursion
to the next solution.
For a more precise description of heteroclinic cycles and their
stability, see Mel- bourne et al. [278], Krupa and Melbourne [240],
the monograph by Field [122], and
22 1 A Unifying Theme
the survey article by Krupa [239]. The existence of structurally
stable heteroclinic cycles is considered a highly degenerate
feature of both types of systems, continuous and discrete. In other
words, typically they do not exist. In continuous systems, where
the governing equations normally consist of systems of differential
equations, it is well-known that the presence of symmetry can,
however, lead to structurally stable, asymptotically stable, cycles
[121, 156]. First, symmetry forces certain subspaces of the
phase-space to be invariant under the governing equations. Then,
cycles are formed through saddle-sink connections between
equilibria and/or periodic solutions that lie on the invariant
subspaces. Since saddle-sink connections are structurally sta- ble
so are the cycles. Homoclinic cycles, on the other hand, are a
specific case of heteroclinic cycles in which the sequence of
connections joins invariant solutions (equilibria, periodic
solutions or chaotic sets) which belong to the same group orbit.
Why are heteroclinic connections important to nonlinear sensor
devices?
From a dynamical systems point of view, it is desirable to
construct symmet- ric networks of sensors that can produce
oscillations via symmetry-breaking global bifurcations of
heteroclinic connections because near a heteroclinic cycle the
emer- gent oscillations tend to have a very large period of
oscillation, which renders their waveform highly sensitive to
symmetry-breaking effects caused by external signals. Indeed, as
one gets closer to the bifurcation point of the oscillations, their
period increases exponentially and so does the symmetry-breaking
response. This effect offers the possibility of significantly
increasing the sensitivity response of a network- based sensor
device by careful-tuning its operation close to the onset of
oscillations. These facts have lead us to consider a new readout
mechanism referred to as Res- idence Times Detection or (RTD)
mechanism, as is described in Chap.2, Sect. 2.5. Next we present a
few more details about the existence and stability of heteroclinic
connections.
Finding Heteroclinic Cycles. For systems whose symmetries are
described by the continuous group O(2), i.e., the group of
rotations and reflections on the plane, Armbruster et al. [16] show
that heteroclinic cycles between steady-states can occur stably,
and Melbourne et al. [278] provide a method for finding cycles that
involve steady-states as well as periodic solutions. Let Γ ⊂ O(N )
be a Lie subgroup (where O(N ) denotes the orthogonal group of
order N ) and let g : RN → RN be Γ -equivariant, that is,
g(γX) = γg(X),
d X
dt = g(X).
Note that N = kn in an n-cell system with k state variables in each
cell. Equivari- ance of g implies that whenever X (t) is a
solution, so is γX (t). Using fixed-point subspaces, Melbourne et
al. [278] suggest a method for constructing heteroclinic cycles
connecting equilibria. Suppose that Σ ⊂ Γ is a subgroup. Then the
fixed- point subspace
1.7 Heteroclinic Connections 23
Fig. 1.16 Pattern inside lattice of subgroups that suggests the
existence of heteroclinic cycles
Fix(Σ) = {X ∈ RN : σX = X ∀σ ∈ Σ}
is a flow invariant subspace. The idea is to find a sequence of
maximal subgroups Σ j ⊂ Γ such that dim Fix(Σ j ) = 1 and
submaximal subgroups Tj ⊂ Σ j ∩ Σ j+1
such that dim Fix(Tj ) = 2, as is shown schematically in Fig. 1.16.
In addition, the equilibrium in Fix(Σ j ) must be a saddle in
Fix(Tj ) whereas the equilibrium in Fix(Σ j+1) must be a sink in
Fix(Tj ).
Such configurations of subgroups have the possibility of leading to
heteroclinic cycles if saddle-sink connections between equilibria
in Fix(Σ j ) and Fix(Σ j+1) exist in Fix(Tj ). It should be
emphasized that more complicated heteroclinic cycles can exist.
Generally, all that is needed to be known is that the equilibria in
Fix(Σ j )
is a saddle and the equilibria in Fix(Σ j+1) is a sink in the
fixed-point subspace Fix(Tj ) (see Krupa andMelbourne [240]) though
the connections can not, in general, be proved. Since saddle-sink
connections are robust in a plane, these heteroclinic cycles are
stable to perturbations of g so long as Γ -equivariance is
preserved by the perturbation. For a detailed discussion of
asymptotic stability and nearly asymptotic stability of
heteroclinic cycles, which are also very important topics, see
Krupa and Melbourne [240].
Near points of Hopf bifurcation, this method for constructing
heteroclinic connec- tions can be generalized to include time
periodic solutions as well as equilibria. Mel- bourne, Chossat, and
Golubitsky [278] do this by augmenting the symmetry group of the
differential equations with S1 — the symmetry group of
Poincare–Birkhoff normal form at points of Hopf bifurcation—and
using phase-amplitude equations in the analysis. In these cases the
heteroclinic cycle exists only in the normal form equations since
some of the invariant fixed-point subspaces disappear when symme-
try is broken. However, when that cycle is asymptotically stable,
then the cycling like behavior remains even when the equations are
not in normal form. This is proved by using asymptotic stability to
construct a flow invariant neighborhood about the cycle and then
invoking normal hyperbolicity to preserve the flow invariant neigh-
borhood when normal symmetry is broken. Indeed, as is shown by
Melbourne et al. [278], normal form symmetry can be used to produce
stable cycling behavior even in systems without any spatial
symmetry. More generally, it also follows that if an asymptotically
stable cycle can be produced in a truncated normal form equation
(say truncated at third or fifth order), then cycling like behavior
persists in equations with higher order terms— even when those
terms break symmetry— and the cycling like behavior is
robust.
24 1 A Unifying Theme
X
Z
Y
(a)
0 100 200 300 400 500 600 700 800 900 1000 0
0.5
1
1.5
x
0 100 200 300 400 500 600 700 800 900 1000 0
0.5
1
1.5
y
0 100 200 300 400 500 600 700 800 900 1000 0
0.5
1
1.5
z
t
(b)
Fig. 1.17 Heteroclinic cycle found between three equilibrium points
of the Guckenheimer and Holmes system. a Saddle-sink connections in
phase-space, b Time series evolution of a typical nearby
trajectory. Parameters are: μ = 1.0, a = 1.0, b = 0.55, c =
1.5
The Guckenheimer–Holmes Cycle. Figure1.17 illustrates a cycle
involving three steady-states of a system of ODE’s proposed by
Guckenheimer and Holmes [156]. Observe that as time evolves a
nearby trajectory stays longer on each equilibrium.
The group Γ in this example has 24 elements and is generated by the
following symmetries
(x, y, z) → (±x,±y,±z) (x, y, z) → (y, z, x)
Note that, in fact, this is a homoclinic cycle since the three
equilibria are on the group orbit given by the cyclic generator of
order 3. The actual system of ODE’s can be written in the following
form
x1 = μx1 − (ax2 1 + bx2
2 + cx2 3 )x1
3 + cx2 1 )x2
1 + cx2 2 )x3.
In related work that describes cycling chaos, Dellnitz et al. [94]
point out that the Guckenheimer–Holmes system can be interpreted as
a coupled cell system (with three cells) in which the internal
dynamics of each cell is governed by a pitchfork bifurcation of the
form
xi = μxi − ax3 i ,
where i = 1, 2, 3 is the cell number. As μ varies from negative to
positive through zero, a bifurcation from the trivial equilibrium
xi = 0 to nontrivial equilibria xi = ±√
μ occurs. Guckenheimer and Holmes [156] show that when the strength
of the remaining terms in the system of ODE’s (which can be
interpreted as cou- pling terms) is large, an asymptotically stable
heteroclinic cycle connecting these bifurcated equilibria exists.
The connection between the equilibria in cell one to the
1.7 Heteroclinic Connections 25
equilibria in cell two occurs through a saddle-sink connection in
the x1x2−plane (which is forced by the internal symmetry of the
cells to be an invariant plane for the dynamics). As Dellnitz et
al. (1995) further indicate, the global permutation sym- metry of
the three-cell system guarantees connections in both the x2x3-plane
and the x3x1-plane, leading to a heteroclinic connection between
three equilibrium solutions.
A Cycle in a System with Circular Symmetry. Melbourne et al. [278]
prove the existence of robust, asymptotically stable heteroclinic
cycles involving time periodic solutions in steady-state/Hopf and
Hopf/Hopf mode interactions in systems with O(2)-symmetry. In these
symmetry breaking bifurcations each critical eigenvalue is doubled
by symmetry—so the centermanifold for a steady-state/Hopfmode
interac- tion is six-dimensional and for a Hopf/Hopf mode
interaction it is eight-dimensional. It is well known that O(2)
symmetry-breaking Hopf bifurcations at invariant equi- libria lead
to two types of periodic solutions: “standing waves” (solutions
invariant under a single reflection for all time) and ”rotating
waves” (solutions whose time evolution is the same as spatial
rotation).
Figure1.18 shows a cycle connecting a steady-state with a
standingwave obtained from a steady-state/Hopf mode interaction by
numerically integrating a general sys- tem of ODE’s with
O(2)-symmetry, which has the form
Fig. 1.18 Heteroclinic cycle connecting a steady-state with a
standing wave in a system with O(2) × S1 symmetry with S1 symmetry
due to normal form. Parameters are: c1 = λ − 1.5ρ − 4N , c2 = 1.3,
c3 = −9, p1 = 1.2λ − 3ρ − N , p2 = 4, p3 = 4, q1 = 0.8λ + 7, and
all other coefficients set to zero
26 1 A Unifying Theme
dz
P(z) = P1
[ z1 z2
] ,
where δ = |z2|2 − |z1|2, C j = c j + iδc j+1, c j are real-valued
O(2) × S1-invariant functions and P j = p j + q j i are
complex-valued O(2) × S1-invariant functions depending on two
parameters λ and μ. The time series in this figure are taken from
three different coordinates: x0 is a coordinate in the steady-state
mode and x1, x2 are coordinates in the Hopf mode. In these
coordinates a sta
LOAD MORE