and Simulation
Analysis of Chaotic Models,
Attractors and Forms
Christos H. Skiadas
Charilaos Skiadas
Chapman & Hall/CRC
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 334872742
© 2009 by Taylor & Francis Group, LLC
Chapman & Hall/CRC is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Printed in the United States of America on acidfree paper
10 9 8 7 6 5 4 3 2 1
International Standard Book Number13: 9781420079005 (Hardcover)
This book contains information obtained from authentic and highly regarded sources. Reasonable
efforts have been made to publish reliable data and information, but the author and publisher can
not assume responsibility for the validity of all materials or the consequences of their use. The
authors and publishers have attempted to trace the copyright holders of all material reproduced
in this publication and apologize to copyright holders if permission to publish in this form has not
been obtained. If any copyright material has not been acknowledged please write and let us know so
we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information
storage or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copy
right.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222
Rosewood Drive, Danvers, MA 01923, 9787508400. CCC is a notforprofit organization that pro
vides licenses and registration for a variety of users. For organizations that have been granted a
photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and
are used only for identification and explanation without intent to infringe.
Library of Congress CataloginginPublication Data
Skiadas, Christos H.
Chaotic modelling and simulation: analysis of chaotic models, attractors and
forms / Christos H. Skiadas, Charilaos Skiadas.
p. cm.
Includes bibliographical references and index.
ISBN 9781420079005 (hardcover : alk. paper)
1. Chaotic behavior in systems. 2. Mathematical models. I. Skiadas,
Charilaos. II. Title.
Q172.5.C45S53 2009
003’.857dc22 2008038183
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
Preface
“Is there a chaotic world? To what extent? Mathematics, geometry or simply
chaos?”
The above questions arise time and time again when dealing with chaotic phenom
ena and the related attractors and chaotic forms. During the lengthy preparation of
this manuscript, computers were used in the same way that geometers of old used
geometric tools. Simple or complicated ideas are designed, estimated and tested us
ing this new tool. The result is more than 500 graphs and illustrations that ﬁll the
pages in front of you.
Verbal explanations are kept to a minimum, and a lot of eﬀort is devoted to ensur
ing that the presentation of ideas progresses from the most elementary to the most
advanced, in a clear and comprehensible way, while requiring little previous knowl
edge of mathematics. In this way, we hope that a more general audience will beneﬁt
from the material of this book, especially from the large variety of chaotic attractors
included.
A lot of new material is presented alongside the classical forms and attractors
appearing on the literature on chaos. However, most of the illustrations are based on
new simulation methods and techniques. A rather lengthy introduction provides the
reader with an overview of the subsequent chapters, and the interconnection between
them.
Our inspirations came from various disciplines, including geometry from the an
cient Greek and Alexandrian period (a great deal of which is taught in the Greek
school system), mathematics from the developments of the last centuries, astronomy
and astrophysics from recent developments (the Conference on Chaos in Astronomy
organized by George Contopoulos in 2002 was very stimulating) and the amazing
chaotic illustrations of Hubble and Chandra and the vortex movements from ﬂuid
ﬂow theory.
Chaotic theory is developing in a new way that inﬂuences the world around us,
and consequently also inﬂuences our ways of approaching, analyzing and solving
problems. It is not surprising that one of the central models in the chaos literature,
the H´ enonHeiles model, is presented in a paper with the title “The applicability of
the third integral of motion: Some numerical experiments.” Numerical experiments
in 1964 were the basis for many signiﬁcant changes in astronomy in the decades that
followed. In 1963 Edwin Lorenz, in his pioneering work on “Deterministic Nonpe
riodic Flow,” proposed a more prominent title for chaotic modelling, by including
the term “deterministic.” His work spearheaded numerous studies on chaotic phe
nomena. Fifteen years later, Feigenbaum explored the rich chaotic properties of the
simple discrete Logistic model, known since 1845 following the work of Verhulst on
the continuous version of this model, and introduced scientists to the chaotic ﬁeld.
Our intention in this work was to present the main models developed by the pi
oneers of chaos theory, along with new extensions and variations to these models.
The plethora of new models that can arise from the existing ones through analysis
and simulation was surprising.
This book is suitable for a wide range of readers, including systems analysts, math
ematicians, astronomers, engineers and people in any ﬁeld of science and technology
whose work involves modelling of systems. It is our hope that this book will prompt
more people to become involved in the rapidly advancing ﬁeld of chaotic models and
will inspire new developments in the ﬁeld.
Christos H. Skiadas
Charilaos Skiadas
List of Figures
1.1 The (x, y) diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Continuous and discrete R¨ ossler systems . . . . . . . . . . . . . . 4
1.3 Bifurcations of the logistic and the Gaussian . . . . . . . . . . . . 5
1.4 Delay models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 The H´ enon and Holmes attractors . . . . . . . . . . . . . . . . . . 7
1.6 The Lorenz and autocatalytic attractors . . . . . . . . . . . . . . . 8
1.7 A chemical reaction attractor . . . . . . . . . . . . . . . . . . . . 9
1.8 Nonchaotic portraits . . . . . . . . . . . . . . . . . . . . . . . . 9
1.9 A rotation chaotic map . . . . . . . . . . . . . . . . . . . . . . . . 10
1.10 The Ikeda attractor . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.11 Ikeda attractors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.12 Translationreﬂection attractors . . . . . . . . . . . . . . . . . . . 13
1.13 Symmetric forms . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.14 Separation of chaotic forms . . . . . . . . . . . . . . . . . . . . . 15
1.15 Flowers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.16 Orbits in twobody motion . . . . . . . . . . . . . . . . . . . . . . 17
1.17 Motion in the plane . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.18 Motion in space . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.19 Ring systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.20 Spiral galaxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.21 Spiral galaxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.22 Reﬂection forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.23 Spiral Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.24 H´ enonHeiles systems . . . . . . . . . . . . . . . . . . . . . . . . 24
1.25 A twoarmed spiral galaxy . . . . . . . . . . . . . . . . . . . . . . 25
1.26 Chaotic paths in the H´ enonHeiles system (E = 1/6) . . . . . . . . 26
1.27 The Contopoulos system: orbits and rotation forms at resonance
ratio 4/1 (left) and 2/3 (right) . . . . . . . . . . . . . . . . . . . . 27
2.1 The solution to the logistic equation . . . . . . . . . . . . . . . . . 35
2.2 The two Gompertz models . . . . . . . . . . . . . . . . . . . . . . 37
2.3 Chaotic forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.4 A snail’s pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.5 Chaos as randomness . . . . . . . . . . . . . . . . . . . . . . . . 43
3.1 Tracing iterations . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.2 Second order stationary points . . . . . . . . . . . . . . . . . . . . 49
3.3 The logistic and the inverse logistic map for b = 3.5 . . . . . . . . 50
3.4 The third order cycle of the logistic and the inverse logistic map for
b = 1 +
√
8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5 Chaotic region for the logistic and the inverse logistic map at b =
3.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.6 The logistic map . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.7 The logistic map, b ≤ 3 . . . . . . . . . . . . . . . . . . . . . . . 54
3.8 The logistic map, b > 3 . . . . . . . . . . . . . . . . . . . . . . . 54
3.9 Order4 bifurcation of the logistic map; real time and phase space
diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.10 Order8 bifurcation of the logistic map; real time and phase space
diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.11 Order16 bifurcation and real time and phase space diagrams of the
logistic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.12 Order32 bifurcation and real time and phase space diagrams of the
logistic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.13 The order5 bifurcation and real time and phase space diagrams . . 57
3.14 The order6 bifurcation and real time and phase space diagrams . . 57
3.15 The bifurcation diagram of the logistic . . . . . . . . . . . . . . . 59
3.16 Order3 bifurcations . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.17 The order3 bifurcation . . . . . . . . . . . . . . . . . . . . . . . 61
3.18 The May model: an order4 bifurcation (left) and the bifurcation
diagram (right) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.19 Bifurcation diagrams of the Gaussian model . . . . . . . . . . . . 63
3.20 Bifurcation diagram of the Gaussian model (a = 17) . . . . . . . . 63
3.21 Bifurcation diagram of the GL model . . . . . . . . . . . . . . . 63
3.22 Cycles in the GL model . . . . . . . . . . . . . . . . . . . . . . . 64
3.23 Patent applications in the United States . . . . . . . . . . . . . . . 65
3.24 The GRM1 model . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.25 Spain, oxygen steel process (1968–1980) . . . . . . . . . . . . . . 69
3.26 Italy oxygen steel process (1970–1980) . . . . . . . . . . . . . . . 70
3.27 Luxemburg oxygen steel process (1962–1980) . . . . . . . . . . . 71
3.28 Bulgaria oxygen steel process (1968–1978) . . . . . . . . . . . . . 72
3.29 The modiﬁed logistic model (ML) . . . . . . . . . . . . . . . . . 73
3.30 The modiﬁed logistic model (ML) for b = 1.3 and e = 0.076 . . . . 73
3.31 A chaotic stage of the modiﬁed logistic model (ML) for b = 1.3
and e = 0.073 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.32 A third order cycle of the GRM1 model for b = 1.3 and σ = 0.025 74
3.33 A third order cycle of the GRM1 model for b = 1.2 and σ =
0.01175 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.34 A third order cycle of the modiﬁed logistic model for b = 1.3 and
e = 0.0719 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.35 The ﬁrst bifurcation cycle of the GRM1 model for b = 1.3 and
σ = 0.032 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.1 A simple delay oscillation scheme (a = −1) . . . . . . . . . . . . 80
4.2 Variation of the simple delay model . . . . . . . . . . . . . . . . . 80
4.3 The delay logistic model . . . . . . . . . . . . . . . . . . . . . . . 81
4.4 Folds in the delay logistic model . . . . . . . . . . . . . . . . . . 82
4.5 A simple delay model (n = 15) . . . . . . . . . . . . . . . . . . . 83
4.6 Delay model (b = 2); a square pattern . . . . . . . . . . . . . . . . 85
4.7 Delay logistic model . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.8 Oscillations and limit cycles in delay logistic models . . . . . . . . 87
4.9 Chaotic attractors of a delay logistic model . . . . . . . . . . . . . 88
4.10 Chaotic attractors of the logistic delay model (m = 40) . . . . . . . 89
4.11 Comparing the Glass (A) and the logistic (B) delay models (m = 40) 90
4.12 The delay exponential model . . . . . . . . . . . . . . . . . . . . 91
4.13 Delay logistic and delay cosine H´ enon variants . . . . . . . . . . . 92
4.14 Chaotic patterns of the delay cosine model . . . . . . . . . . . . . 93
4.15 Carpetlike forms . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.16 Carpetlike forms . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.17 Squarelike forms . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.18 Irregular patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.19 Islands in the chaotic sea . . . . . . . . . . . . . . . . . . . . . . 96
4.20 Triangular islands . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.1 The H´ enon map . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.2 Orbits in the cosine H´ enon model . . . . . . . . . . . . . . . . . . 102
5.3 Bifurcation forms in the cosine H´ enon model . . . . . . . . . . . . 103
5.4 Higherorder bifurcations in the cosine H´ enon model . . . . . . . . 104
5.5 A thirdorder delay model . . . . . . . . . . . . . . . . . . . . . . 105
5.6 A secondorder delay model . . . . . . . . . . . . . . . . . . . . . 106
5.7 The order2 delay model . . . . . . . . . . . . . . . . . . . . . . . 106
5.8 A complicated delay model . . . . . . . . . . . . . . . . . . . . . 106
5.9 A H´ enon model with a logistic delay term . . . . . . . . . . . . . 107
5.10 A variation of the H´ enon model . . . . . . . . . . . . . . . . . . . 108
5.11 An exponential variant of the H´ enon model . . . . . . . . . . . . . 108
5.12 A 2diﬀerence equations system . . . . . . . . . . . . . . . . . . . 109
5.13 A threedimensional model . . . . . . . . . . . . . . . . . . . . . 110
5.14 The Holmes model (a = 2.765, b = −0.2) . . . . . . . . . . . . . . 110
5.15 The Holmes model (a = 2.4, b = 0.3) . . . . . . . . . . . . . . . . 111
5.16 The sine delay map (b = 0.3) . . . . . . . . . . . . . . . . . . . . 112
5.17 The sine delay model . . . . . . . . . . . . . . . . . . . . . . . . 112
5.18 The sine delay model for large b . . . . . . . . . . . . . . . . . . 113
5.19 The sine delay model for b = 1 . . . . . . . . . . . . . . . . . . . 114
5.20 The sine delay model (a = 0.1) . . . . . . . . . . . . . . . . . . . 114
6.1 The Arneodo model . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.2 A modiﬁed Arneodo model . . . . . . . . . . . . . . . . . . . . . 120
6.3 An autocatalytic attractor . . . . . . . . . . . . . . . . . . . . . . 121
6.4 Threedimensional views of the autocatalytic model . . . . . . . . 122
6.5 A fourdimensional autocatalytic model . . . . . . . . . . . . . . 123
6.6 The R¨ ossler attractor . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.7 Limit cycles in a variant of the R¨ ossler model . . . . . . . . . . . 125
6.8 Spiral chaos in the variant of the R¨ ossler model . . . . . . . . . . 126
6.9 Plane views of the total spiral chaos in the variant of the R¨ ossler
model (a = 0.60) . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.10 View of a simple cycle of the variant of the R¨ ossler model (a = 0.4,
b = 0.3 and c = 4.8) . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.11 Modiﬁed R¨ ossler models . . . . . . . . . . . . . . . . . . . . . . 128
6.12 Another view of the modiﬁed R¨ ossler model . . . . . . . . . . . . 129
6.13 The Lorenz attractor . . . . . . . . . . . . . . . . . . . . . . . . . 131
6.14 The modiﬁed Lorenz model — left to right: (x, y), (y, z) and (x, z)
views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
6.15 The modiﬁed Lorenz chaotic attractor . . . . . . . . . . . . . . . . 132
7.1 The simplest conservative system . . . . . . . . . . . . . . . . . . 137
7.2 Qualitative classiﬁcation of equilibrium points . . . . . . . . . . . 143
7.3 Eggshaped forms . . . . . . . . . . . . . . . . . . . . . . . . . . 146
7.4 A double eggshaped form with an envelope . . . . . . . . . . . . 147
7.5 Symmetric forms . . . . . . . . . . . . . . . . . . . . . . . . . . 149
7.6 A simple nonsymmetric form . . . . . . . . . . . . . . . . . . . . 149
7.7 A symmetric threefold form . . . . . . . . . . . . . . . . . . . . 150
7.8 More complex forms . . . . . . . . . . . . . . . . . . . . . . . . . 152
7.9 Higherorder forms . . . . . . . . . . . . . . . . . . . . . . . . . 153
8.1 The limiting trajectory . . . . . . . . . . . . . . . . . . . . . . . . 159
8.2 Rotating paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
8.3 Rotationtranslation, and the chaotic bulge . . . . . . . . . . . . . 165
8.4 The chaotic bulge . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.5 The chaotic bulge . . . . . . . . . . . . . . . . . . . . . . . . . . 167
8.6 The chaotic bulge . . . . . . . . . . . . . . . . . . . . . . . . . . 167
8.7 Equilibrium points . . . . . . . . . . . . . . . . . . . . . . . . . . 168
8.8 The chaotic bulge . . . . . . . . . . . . . . . . . . . . . . . . . . 169
8.9 Elliptic forms and the chaotic bulge . . . . . . . . . . . . . . . . . 170
8.10 Rotating particles . . . . . . . . . . . . . . . . . . . . . . . . . . 172
8.11 Development of rotating particles over time (a = 0.6). . . . . . . . 172
8.12 Development of rotating particles according to the magnitude of the
translation parameter. The elapsed time is t = 10. . . . . . . . . . 173
8.13 Rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
8.14 The characteristic trajectory . . . . . . . . . . . . . . . . . . . . . 175
8.15 Comparison of the continuous and the discrete models . . . . . . . 177
9.1 Symmetry/reﬂection along a line . . . . . . . . . . . . . . . . . . 180
9.2 Planar isometries are determined by their eﬀect on three points . . 181
9.3 Glide reﬂection seen in two diﬀerent ways . . . . . . . . . . . . . 182
9.4 The composition of two reﬂections . . . . . . . . . . . . . . . . . 183
9.5 A symmetric graph . . . . . . . . . . . . . . . . . . . . . . . . . 185
9.6 The Ikeda attractor . . . . . . . . . . . . . . . . . . . . . . . . . . 187
9.7 The reﬂection Ikeda attractor . . . . . . . . . . . . . . . . . . . . 188
9.8 Comparing rotationreﬂection cases . . . . . . . . . . . . . . . . . 189
9.9 Comparing rotationreﬂection case . . . . . . . . . . . . . . . . . 189
9.10 A reﬂection chaotic attractor resulting from a quadratic rotation an
gle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
9.11 Two other reﬂection attractors from a quadratic rotation angle . . . 190
9.12 Space contraction . . . . . . . . . . . . . . . . . . . . . . . . . . 191
9.13 The eﬀect of the reﬂection parameter on chaotic images . . . . . . 192
9.14 The eﬀect of space contraction on chaotic images . . . . . . . . . 193
9.15 The eﬀect of a complicated rotation angle on chaotic attractors . . 193
9.16 The eﬀect of a complicated rotation angle on chaotic attractors . . 194
9.17 Twopiece chaotic attractors . . . . . . . . . . . . . . . . . . . . . 195
9.18 Reﬂection: a twopiece chaotic attractor . . . . . . . . . . . . . . 195
9.19 A simple area preserving rotationtranslation model . . . . . . . . 196
9.20 Bifurcation diagrams . . . . . . . . . . . . . . . . . . . . . . . . 198
9.21 Circular chaotic forms . . . . . . . . . . . . . . . . . . . . . . . . 199
9.22 Geometric analogue of rotationtranslation (θ = 2) . . . . . . . . . 200
9.23 Geometric analogues of rotationtranslation . . . . . . . . . . . . 201
10.1 Spiral forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
10.2 Chaotic attractors in a noncentral sink . . . . . . . . . . . . . . . 208
10.3 The two symmetric sinks model . . . . . . . . . . . . . . . . . . . 210
10.4 Chaotic attractor in the twosink problem . . . . . . . . . . . . . . 211
10.5 Two distinct vortex forms (d = 1) . . . . . . . . . . . . . . . . . . 212
10.6 The chaotic attractor with strong vortex strength parameter d = 2π 212
10.7 Vortex forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
10.8 Chaotic forms without space contraction . . . . . . . . . . . . . . 214
10.9 Higher order vortices . . . . . . . . . . . . . . . . . . . . . . . . 215
10.10 Spiralling towards the sink . . . . . . . . . . . . . . . . . . . . . 216
10.11 A sinusoidal rotation angle chaotic attractor (c = 0.4, d = 6) . . . . 217
10.12 Sinusoidal rotation angle chaotic attractors . . . . . . . . . . . . . 217
10.13 Sinusoidal rotation angle chaotic attractors . . . . . . . . . . . . . 218
10.14 Chaotic images of rotationtranslation maps . . . . . . . . . . . . 220
10.15 Chaotic images with area contraction rotation angles . . . . . . . . 221
11.1 Elliptic orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
11.2 Attracting bodies in space and relative movement . . . . . . . . . 226
11.3 Motions in space . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
11.4 Ring systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
11.5 Chaotic forms of a simple rotationtranslation model . . . . . . . . 231
11.6 Chaotic forms of a simple rotationtranslation model . . . . . . . . 232
11.7 Chaotic forms of a simple rotationtranslation model . . . . . . . . 233
11.8 Attractors and attracting points in rotationtranslation model . . . . 234
11.9 A rotation image following a ContopoulosBozis paper (approach
ing galaxies) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
11.10 Comparing chaotic forms: rotationreﬂection . . . . . . . . . . . . 236
11.11 A rotationtranslation chaotic image . . . . . . . . . . . . . . . . 237
11.12 Chaotic forms for low relativistic speeds . . . . . . . . . . . . . . 239
11.13 Relativistic rotationtranslation forms for medium and high rela
tivistic speeds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
11.14 Relativistic rotationtranslation forms for high relativistic speeds . 241
11.15 Nonrelativistic chaotic images in the early period of a rotation
translation process with b = 1 . . . . . . . . . . . . . . . . . . . . 242
11.16 Other relativistic chaotic images . . . . . . . . . . . . . . . . . . . 243
11.17 Various relativistic chaotic forms . . . . . . . . . . . . . . . . . . 243
11.18 Relativistic and initial chaotic images of a rotationtranslation pro
cess . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
11.19 Two galaxylike chaotic images . . . . . . . . . . . . . . . . . . . 245
11.20 Other relativistic chaotic images . . . . . . . . . . . . . . . . . . . 246
11.21 A relativistic chaotic image when b = 0.9 . . . . . . . . . . . . . . 246
11.22 A relativistic chaotic image when b = 0.9 . . . . . . . . . . . . . . 247
11.23 A galacticlike cluster . . . . . . . . . . . . . . . . . . . . . . . . 247
11.24 Relativistic reﬂectiontranslation . . . . . . . . . . . . . . . . . . 249
11.25 Rotating disks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
11.26 Twoarmed galaxies . . . . . . . . . . . . . . . . . . . . . . . . . 252
11.27 Twoarmed galaxies . . . . . . . . . . . . . . . . . . . . . . . . . 252
11.28 Rotating particles under the inﬂuence of an attracting mass . . . . 255
11.29 Boxlike paths of a twoarmed spiral galaxy . . . . . . . . . . . . 256
11.30 An asymptotically stable chaotic attractor . . . . . . . . . . . . . . 256
11.31 Rotation speed vs. radius r . . . . . . . . . . . . . . . . . . . . . 257
11.32 Two attracting masses and the disk of rotating particles at time t = 0 259
11.33 Rotating disks with two attracting masses . . . . . . . . . . . . . . 259
11.34 Rotating disks with two attracting masses over a long time . . . . . 260
11.35 Chaotic attractors in the case of two unequal symmetric attracting
masses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
11.36 Rotating disc at early times . . . . . . . . . . . . . . . . . . . . . 262
11.37 Multiarm images . . . . . . . . . . . . . . . . . . . . . . . . . . 263
11.38 The eﬀect of two perpendicularly attracting masses . . . . . . . . 264
12.1 The original H´ enonHeiles (y, ˙ y) diagram . . . . . . . . . . . . . . 268
12.2 (y, ˙ y) diagrams for the H´ enonHeiles system . . . . . . . . . . . . 269
12.3 A discrete analogue to the H´ enonHeiles system (E = 1/12) . . . . 270
12.4 The discrete H´ enonHeiles model . . . . . . . . . . . . . . . . . . 271
12.5 Discrete analogues to the H´ enonHeiles system . . . . . . . . . . . 271
12.6 Chaotic paths in the H´ enonHeiles system . . . . . . . . . . . . . 272
12.7 The H´ enonHeiles system . . . . . . . . . . . . . . . . . . . . . . 273
12.8 Boxlike orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
12.9 A simple Hamiltonian: orbits and rotation forms at various times . 276
12.10 Orbits and rotation forms at time t = 1 . . . . . . . . . . . . . . . 278
12.11 Logarithmic potential: orbits and rotation forms at time t = 1 . . . 280
12.12 The Contopoulos system: orbits at resonance ratio 4/1 (left) and
2/3 (right) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
12.13 Another simple Hamiltonian system . . . . . . . . . . . . . . . . 282
13.1 Adding a sinusoidal forcing term to a twodimensional system . . . 286
13.2 The eﬀect of noise on threedimensional models . . . . . . . . . . 287
13.3 The LotkaVolterra system: limit cycle (upper image) and (t, x)
(heavy line) and the (t, y) (light line) graphs . . . . . . . . . . . . 289
13.4 The LotkaVolterra system . . . . . . . . . . . . . . . . . . . . . 290
13.5 The pendulum and the (t, x) oscillations . . . . . . . . . . . . . . . 291
13.6 The (y, ˙ y) curves of a special second order diﬀerential equation sys
tem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
13.7 The dolphin attractor . . . . . . . . . . . . . . . . . . . . . . . . . 293
13.8 The ship and signature attractors . . . . . . . . . . . . . . . . . . 294
13.9 The Ushiki and Tinkerbell attractors . . . . . . . . . . . . . . . . 294
Contents
1 Introduction 1
1.1 Chaos in Diﬀerential Equations Systems . . . . . . . . . . . . . . 1
1.2 Chaos in Diﬀerence Equation Systems . . . . . . . . . . . . . . . 4
1.2.1 The logistic map . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Delay models . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.3 The H´ enon model . . . . . . . . . . . . . . . . . . . . . . 6
1.3 More Complex Structures . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 Threedimensional and higherdimensional models . . . . 8
1.3.2 Conservative systems . . . . . . . . . . . . . . . . . . . . 8
1.3.3 Rotations . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.4 Shape and form . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Chaos and the Universe . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.1 Chaos in the solar system . . . . . . . . . . . . . . . . . . 14
1.4.2 Chaos in galaxies . . . . . . . . . . . . . . . . . . . . . . 19
1.4.3 Galactictype potentials and the H´ enonHeiles system . . . 23
1.4.4 The Contopoulos system . . . . . . . . . . . . . . . . . . 25
1.5 Odds and Ends, and Milestones . . . . . . . . . . . . . . . . . . . 27
2 Models and Modelling 29
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2 Model Construction . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.1 Growth/decay models . . . . . . . . . . . . . . . . . . . . 30
2.3 Modelling Techniques . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3.1 Series approximation . . . . . . . . . . . . . . . . . . . . 33
2.3.2 Empirical model formulation . . . . . . . . . . . . . . . . 36
2.3.3 The calculus of variations approach . . . . . . . . . . . . 38
2.3.4 The probabilisticstochastic approach . . . . . . . . . . . 39
2.3.5 Delay growth functions . . . . . . . . . . . . . . . . . . . 40
2.4 Chaotic Analysis and Simulation . . . . . . . . . . . . . . . . . . 41
2.5 Deterministic, Stochastic and Chaotic Models . . . . . . . . . . . 42
3 The Logistic Model 47
3.1 The Logistic Map . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.1.1 Geometric analysis of the logistic . . . . . . . . . . . . . 47
3.1.2 Algebraic analysis of the logistic . . . . . . . . . . . . . . 51
3.2 The Bifurcation Diagram . . . . . . . . . . . . . . . . . . . . . . 58
3.3 Other Models with Similar Behaviour . . . . . . . . . . . . . . . 61
3.4 Models with Diﬀerent Chaotic Behaviour . . . . . . . . . . . . . 62
3.5 The GRM1 Chaotic Model . . . . . . . . . . . . . . . . . . . . . 64
3.5.1 GRM1 and innovation diﬀusion modelling . . . . . . . . . 64
3.5.2 The generalized rational model . . . . . . . . . . . . . . . 66
3.5.3 Parameter estimation for the GRM1 model . . . . . . . . . 68
3.5.4 Illustrations . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.6 Further Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 71
4 The Delay Logistic Model 79
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2 Delay Diﬀerence Models . . . . . . . . . . . . . . . . . . . . . . 79
4.2.1 Simple delay oscillation scheme . . . . . . . . . . . . . . 79
4.2.2 The delay logistic model . . . . . . . . . . . . . . . . . . 81
4.3 Time Delay Diﬀerential Equations . . . . . . . . . . . . . . . . . 82
4.4 A More Complicated Delay Model . . . . . . . . . . . . . . . . . 84
4.5 A Delay Diﬀerential Logistic Analogue . . . . . . . . . . . . . . 86
4.6 Other Delay Logistic Models . . . . . . . . . . . . . . . . . . . . 86
4.7 Model Behaviour for Large Delays . . . . . . . . . . . . . . . . . 89
4.8 Another Delay Logistic Model . . . . . . . . . . . . . . . . . . . 91
5 The H´ enon Model 99
5.1 Global Period Doubling Bifurcations in the H´ enon Map . . . . . . 99
5.1.1 Period doubling bifurcations when b = −1 . . . . . . . . . 100
5.1.2 Period doubling bifurcations when b = 1 . . . . . . . . . . 101
5.2 The CosineH´ enon Model . . . . . . . . . . . . . . . . . . . . . . 101
5.3 An Example of Bifurcation and Period Doubling . . . . . . . . . . 103
5.4 A Diﬀerential Equation Analogue . . . . . . . . . . . . . . . . . 103
5.5 Variants of the H´ enon Delay Diﬀerence Equation . . . . . . . . . 104
5.5.1 The thirdorder delay model . . . . . . . . . . . . . . . . 104
5.5.2 Secondorder delay models . . . . . . . . . . . . . . . . . 105
5.5.3 Firstorder delay variants . . . . . . . . . . . . . . . . . . 107
5.5.4 Exponential variants . . . . . . . . . . . . . . . . . . . . 108
5.6 Variants of the H´ enon System Equations . . . . . . . . . . . . . . 109
5.7 The Holmes and Sine Delay Models . . . . . . . . . . . . . . . . 110
5.7.1 The Holmes model . . . . . . . . . . . . . . . . . . . . . 110
5.7.2 The sine delay model . . . . . . . . . . . . . . . . . . . . 111
6 ThreeDimensional and HigherDimensional Models 117
6.1 Equilibrium Points and Characteristic Matrices . . . . . . . . . . 117
6.2 The LotkaVolterra Model . . . . . . . . . . . . . . . . . . . . . . 118
6.3 The Arneodo Model . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.4 An Autocatalytic Attractor . . . . . . . . . . . . . . . . . . . . . 121
6.5 A FourDimensional Autocatalytic Attractor . . . . . . . . . . . . 122
6.6 The R¨ ossler Model . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.6.1 A variant of the R¨ ossler model . . . . . . . . . . . . . . . 124
6.6.2 Introducing rotation into the R¨ ossler model . . . . . . . . 127
6.7 The Lorenz Model . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.7.1 The modiﬁed Lorenz model . . . . . . . . . . . . . . . . 131
7 NonChaotic Systems 135
7.1 Conservative Systems . . . . . . . . . . . . . . . . . . . . . . . . 135
7.1.1 The simplest conservative system . . . . . . . . . . . . . 137
7.1.2 Equilibrium points in Hamiltonian systems . . . . . . . . 138
7.2 Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.2.1 Transformations on linear systems . . . . . . . . . . . . . 140
7.2.2 Qualitative behaviour at equilibrium points . . . . . . . . 142
7.3 EggShaped Forms . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.3.1 A simple eggshaped form . . . . . . . . . . . . . . . . . 144
7.3.2 A double eggshaped form . . . . . . . . . . . . . . . . . 145
7.3.3 A double eggshaped form with an envelope . . . . . . . . 146
7.4 Symmetric Forms . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7.5 More Complex Forms . . . . . . . . . . . . . . . . . . . . . . . . 150
7.6 HigherOrder Forms . . . . . . . . . . . . . . . . . . . . . . . . . 153
8 Rotations 157
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
8.2 A Simple RotationTranslation System of Diﬀerential Equations . 158
8.3 A Discrete RotationTranslation Model . . . . . . . . . . . . . . . 163
8.4 A General RotationTranslation Model . . . . . . . . . . . . . . . 169
8.5 Rotating Particles inside the EggShaped Form . . . . . . . . . . . 171
8.6 Rotations Following an Inverse Square Law . . . . . . . . . . . . 173
9 Shape and Form 179
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.1.1 Symmetry and plane isometries . . . . . . . . . . . . . . . 180
9.2 Isometries in Modelling . . . . . . . . . . . . . . . . . . . . . . . 184
9.2.1 Twodimensional rotation . . . . . . . . . . . . . . . . . . 184
9.3 Reﬂection and Translation . . . . . . . . . . . . . . . . . . . . . 186
9.3.1 Space contraction . . . . . . . . . . . . . . . . . . . . . . 186
9.4 Application in the Ikeda Attractor . . . . . . . . . . . . . . . . . . 187
9.5 Chaotic Attractors and RotationReﬂection . . . . . . . . . . . . . 188
9.6 Experimenting with Rotation and Reﬂection . . . . . . . . . . . . 191
9.6.1 The eﬀect of space contraction on rotationtranslation . . . 191
9.6.2 The eﬀect of space contraction and change of reﬂection
angle on translationreﬂection . . . . . . . . . . . . . . . 192
9.6.3 Complicated rotation angle forms . . . . . . . . . . . . . 193
9.6.4 Comparing rotationreﬂection . . . . . . . . . . . . . . . 194
9.6.5 A simple rotationtranslation model . . . . . . . . . . . . 196
9.7 Chaotic Circular Forms . . . . . . . . . . . . . . . . . . . . . . . 196
9.8 Further Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
9.8.1 The space contraction rotationtranslation case . . . . . . 202
10 Chaotic Advection 205
10.1 The Sink Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.1.1 Central sink . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.1.2 The contraction process . . . . . . . . . . . . . . . . . . . 207
10.2 NonCentral Sink . . . . . . . . . . . . . . . . . . . . . . . . . . 207
10.3 Two Symmetric Sinks . . . . . . . . . . . . . . . . . . . . . . . . 208
10.3.1 Aref’s blinking vortex system . . . . . . . . . . . . . . . 208
10.4 Chaotic Forms without Space Contraction . . . . . . . . . . . . . 213
10.5 Other Chaotic Forms . . . . . . . . . . . . . . . . . . . . . . . . 213
10.6 Complex Sinusoidal Rotation Angle . . . . . . . . . . . . . . . . 216
10.7 A Special RotationTranslation Model . . . . . . . . . . . . . . . 219
10.8 Other RotationTranslation Models . . . . . . . . . . . . . . . . . 219
10.8.1 Elliptic rotationtranslation . . . . . . . . . . . . . . . . . 219
10.8.2 Rotationtranslation with special rotation angle . . . . . . 220
11 Chaos in Galaxies and Related Simulations 223
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
11.2 Chaos in the Solar System . . . . . . . . . . . . . . . . . . . . . 225
11.3 Galaxy Models and Modelling . . . . . . . . . . . . . . . . . . . 228
11.3.1 A special rotationtranslation image . . . . . . . . . . . . 234
11.4 RotationReﬂection . . . . . . . . . . . . . . . . . . . . . . . . . 235
11.5 Relativity in RotationTranslation Systems . . . . . . . . . . . . . 236
11.6 Other Relativistic Forms . . . . . . . . . . . . . . . . . . . . . . 241
11.7 Galactic Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . 247
11.8 Relativistic ReﬂectionTranslation . . . . . . . . . . . . . . . . . 248
11.9 Rotating Disks of Particles . . . . . . . . . . . . . . . . . . . . . 248
11.9.1 A circular rotating disk . . . . . . . . . . . . . . . . . . . 248
11.9.2 The rotating ellipsoid . . . . . . . . . . . . . . . . . . . . 250
11.10 Rotating Particles under Distant Attracting Masses . . . . . . . . . 253
11.10.1 One attracting mass . . . . . . . . . . . . . . . . . . . . . 253
11.10.2 The area of the chaotic region in galaxy simulations . . . . 255
11.10.3 The speed of particles . . . . . . . . . . . . . . . . . . . . 257
11.11 Two Equal Attracting Masses in Opposite Directions . . . . . . . 258
11.11.1 Symmetric unequal attracting masses . . . . . . . . . . . 260
11.12 Two Attracting Equal Nonsymmetric Masses . . . . . . . . . . . . 263
12 GalacticType Potentials and the H´ enonHeiles System 267
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
12.2 The H´ enonHeiles System . . . . . . . . . . . . . . . . . . . . . 268
12.3 Discrete Analogues to the H´ enonHeiles System . . . . . . . . . . 270
12.4 Paths of Particles in the H´ enonHeiles System . . . . . . . . . . . 272
12.5 Other Forms for the Hamiltonian . . . . . . . . . . . . . . . . . . 273
12.6 The Simplest Form for the Hamiltonian . . . . . . . . . . . . . . 275
12.7 Gravitational Attraction . . . . . . . . . . . . . . . . . . . . . . . 275
12.8 A Logarithmic Potential . . . . . . . . . . . . . . . . . . . . . . . 279
12.9 Hamiltonians with a Galactic Type Potential: The Contopoulos
System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
12.10 Another Simple Hamiltonian System . . . . . . . . . . . . . . . . 281
13 Odds and Ends 285
13.1 Forced Nonlinear Oscillators . . . . . . . . . . . . . . . . . . . . 285
13.2 The Eﬀect of Noise in ThreeDimensional Models . . . . . . . . . 285
13.3 The LotkaVolterra Theory for the Growth of Two Conﬂicting
Populations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
13.4 The Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
13.5 A Special SecondOrder Diﬀerential Equation . . . . . . . . . . . 292
13.6 Other Patterns and Chaotic Forms . . . . . . . . . . . . . . . . . 292
14 Milestones 297
References 303
Chapter 1
Introduction
1.1 Chaos in Diﬀerential Equations Systems
Four centuries ago, Newton and other scientists introduced the idea of determin
ism in the mathematical representation of the real world. As a result, future events
in nature could be explained using knowledge of the past. The movements of stel
lar bodies were assumed to follow predetermined paths according to laws that were
known or had to be discovered by the scientiﬁc community. The theory of deter
minism also had theoretical support from philosophy and other scientiﬁc ﬁelds. It
would be diﬃcult to say that determinism, as a general idea, can solve all problems
in nature. The scientists of the last few centuries followed this theory in order to
ﬁnd rational routes in the construction of models. It was essential to simplify our
description of the world ﬁrst, and only then to solve fundamental problems, instead
of entering into obscure paths of uncertainty and later on Chaos.
How then does one introduce a theory that goes against the perfect world of New
tonian dynamics? Our solar system has been rotating for millions of years, and the
experience of the last centuries was in favor of determinism. The sun continues to
rise every day, the moon follows a fourweek period, the stars of Ursa Major are
found every year in predetermined places. How and why could this perfect deter
ministic world lose this certainty and lead a chaotic existence?
It was not easy to convince the scientiﬁc community to introduce uncertainty into
this deterministic world and accept chaotic theories. However, advances in applied
mathematics made this transition easier. At the end of the nineteenth century, the
RungeKutta method for numerical solutions to diﬀerential equations dramatically
improved our ability to get more accurate solutions for numerous models. It was a
real milestone in applied mathematics, especially for the study of highly nonlinear
problems, for which the earlier method by Euler was not very accurate.
Following the introduction of the RungeKutta method, the paths traversed by a
stellar body could now be estimated quite eﬃciently. Although this method was
still quite laborious at those early days, it gave new directions in the way scientists
addressed various problems. Theory and practice could now equally beneﬁt each
other.
It is no surprise that this change inﬂuenced even the French school, traditionally
the most theoretical of all, especially in mathematics and physics. Henri Poincar´ e,
one of the most prominent scientists in 19th century France, tried to resolve the
1
2 Chaotic Modelling and Simulation
famous threebody problem in a contest proposed by the King of Norway (Poincar´ e,
1890b, 1892). Although he succeeded in solving a special case of this problem,
namely the movement of a small mass around two larger masses revolving in two
dimensions, he noticed that the orbits of three bodies moving under a central force
due to gravity are quite complicated and can change drastically under small changes
of the initial conditions.
Nowadays we know that to solve this problem we need a set of more than three
coupled diﬀerential equations, and the paths of the solutions have a chaotic charac
ter. Poincar´ e tried to determine solutions to the restricted threebody problem, when
one mass is very small relative to the other two. He understood that, even in this
simple case, the motion had still a very complicated form. He tried to explain the
phenomenon more generally and to establish a theory describing the chaotic paths of
the solutions to a system of diﬀerential equations: The threedimensional solutions
to a system of nonlinear diﬀerential equations could be chaotic. In stark contrast,
the twodimensional paths do not exhibit a chaotic behaviour, as follows from the
famous Poincar´ eBendixson theorem: Chaos cannot occur in a twodimensional au
tonomous system.
To elaborate a bit, consider the twodimensional autonomous system (1.1):
˙ x = f (x, y)
˙ y = g(x, y)
(1.1)
where ˙ x and ˙ y are the ﬁrst time derivatives. The system (1.1) can take the form of a
twodimensional equation for x and y:
dy
dx
=
g(x, y)
f (x, y)
(1.2)
and we can ﬁnd phase paths in the plane describing the motion, directly from 1.2.
Assuming that we have determined the equilibrium points at which both f (x, y) = 0
and g(x, y) = 0, the Poincar´ eBendixson theorem says that a phase path will either:
a) terminate at an equilibrium point,
b) return to the original point, resulting in a closed path, or
c) approach a limit cycle.
Hence, there are no chaotic solutions. The system of diﬀerential equations (1.3)
illustrates all three cases:
˙ x = −y(y − a)(y − b)
˙ y = x(x − a)(x − b)
(1.3)
The resulting diﬀerential equation for x, y is:
dy
dx
= −
x(x − a)(x − b)
y(y − a)(y − b)
Introduction 3
This equation can be easily integrated, resulting in the solution:
x
4
+ y
4
4
− (a + b)
x
3
+ y
3
3
+ ab
x
2
+ y
2
2
= h
where h is the constant of integration. The paths, with parameters equal to a = 0.8
and b = 0.3, are illustrated in Figure 1.1. There are nine equilibrium points, four
at centres four corners, four in the sides, and one in the middle of the ﬁgure. The
more complex paths pass from the four unstable equilibrium points at the sides. Such
systems, not chaotic in themselves but nonetheless not without interest, are further
discussed in Chapter 7.
FIGURE 1.1: The (x, y) diagram
On the other hand, the simplest threedimensional nonlinear system of diﬀerential
equations, the R¨ ossler model (R¨ ossler, 1976d)
˙ x = −y − z
˙ y = x + ey
˙ z = f + xz − mz
(1.4)
easily produces chaotic paths for appropriate values of the parameters.
The R¨ ossler system has only one nonlinear term, in the last equation. However,
when the parameters have values e = f = 0.2 and m = 5.7, chaotic paths appear with
the form presented in Figure 1.2(a). The R¨ ossler system, along with other interesting
threedimensional systems, is examined in greater detail in Chapter 6.
4 Chaotic Modelling and Simulation
(a) The continuous case (x, y) (b) A discrete analogue
FIGURE 1.2: Continuous and discrete R¨ ossler systems
1.2 Chaos in Diﬀerence Equation Systems
Radical changes in the behaviour appear when a system of diﬀerential equations is
replaced by a similar system of diﬀerence equations, or by a single diﬀerence equa
tion with nonlinear terms. For example, suppose we replace the R¨ ossler system (1.4)
by a corresponding system of diﬀerence equations:
x
t+1
= x
t
−
1
6
(y
t
+ z
t
)
y
t+1
= y
t
+
1
6
(x
t+1
+ ey
t
)
z
t+1
= z
t
+ 0.1 + ex
t+1
z
t
(1.5)
Then, for parameter value e = 0.2, and iterating under the condition x
t+1
= x
t
− 0.4,
an interesting chaotic form like a seashell results (Figure 1.2(b)).
For centuries the mathematical development of methods and tools of analysis con
centrated mainly on diﬀerential equations, and not on diﬀerence equations. It is
therefore not surprising that the logistic model, the most popular chaotic diﬀerence
equation model today, was “invented” only in the last three decades. The logistic
model is described by a very simple equation:
x
t+1
= bx
t
(1 − x
t
) (1.6)
This very simple model already exhibits the key elements of chaotic behaviour. As
the parameter b increases, period doubling bifurcations occur, giving rise to stable
orbits of ever increasing size, culminating in completely chaotic behaviour. On the
other hand, the diﬀerential equation analogue of this model was very popular:
˙ x = bx(1 − x) (1.7)
Introduction 5
Equation (1.7), as expected, gives simple ﬁnite solutions without complications
and chaotic behaviour. This continuous model, called the logistic model, was applied
in demography by Verhulst already in 1838.
Chapter 2, further discusses models and the modelling process in general. Deter
ministic, stochastic and chaotic models are introduced, along with chaotic analysis
and simulation techniques. The subsequent chapters deal with particular classes of
models, and their interesting behaviour.
1.2.1 The logistic map
Chapter 3 is devoted to a detailed analysis and simulation of the chaotic behaviour
of nonlinear diﬀerence equations, starting with the logistic model. The logistic map
and its bifurcation diagram are presented. The bifurcation diagram exhibits the stable
orbit of the x values for each value of the chaotic parameter b. In the case of the
logistic model, the stable orbits become more and more complex as b increases,
culminating in the complete chaos observed for b close to 4.
Other models that are examined, and exhibit similar behaviour, are: the sinusoidal
model, the Gompertz model and a model proposed by May. Also, a class of models
with behaviour diﬀerent than that of the quadratic maps are studied, the Gaussian
type models. As can be seen clearly from the bifurcation diagrams of the logistic
(Figure 1.3(a)) and the Gaussian model (Figure 1.3(b)), there are clear diﬀerences
between the two classes of models, as well as some similarities in the chaotic win
dows.
(a) Logistic (b) Gaussian
FIGURE 1.3: Bifurcations of the logistic and the Gaussian
6 Chaotic Modelling and Simulation
1.2.2 Delay models
Delay models are models where future values depend on values far in the past. For
instance, a simple delay equation is the delay logistic:
x
t+1
= bx
t
(1 − x
t−m
)
which is analysed further in Chapter 4, along with some variations and more com
plicated models.
Delay models can be found in several real life cases when the future development
of a phenomenon is inﬂuenced from the past. The complexity of the delay system
is higher when the delay is large and the equations used are highly nonlinear. Such
chaotic behaviour is expressed by a model proposed by Glass
x
t+1
= cx
t
+ a
x
t−m
1 + x
10
t−m
When the parameters are a = 0.2, c = 0.9 and the delay m = 30, the Glass model
gives a very complicated graph in the phase space and chaotic time oscillations of a
very complex character (Figure 1.4(a)). Figure 1.4(b) presents the map of the delay
logistic, as well as the related oscillations, when the delay is m = 10. The chaotic
parameter is b = 1.2865.
(a) The Glass model (b) The delay logistic
FIGURE 1.4: Delay models
1.2.3 The H´enon model
In some cases a system of two diﬀerence equations is replaced by a single dif
ference delay equation with the same properties. This is the case of the H´ enon
Introduction 7
model (H´ enon, 1976), which can be thought of either as a system of two diﬀerence
equations, or as a single delay equation. The model proposed by H´ enon, which is
now considered one of the standard models in the chaos literature, has the form:
x
t+1
= 1 − ax
2
t
+ y
t
y
t+1
= bx
t
(1.8)
The equivalent delay diﬀerence equation of the H´ enon system is:
x
t+1
= bx
t−1
+ 1 − ax
2
t
(1.9)
For parameter values a = 1.4 and b = 0.3 the famous H´ enon attractor is formed
(Figure 1.5(a)).
(a) The H´ enon attractor (b) The Holmes attractor
FIGURE 1.5: The H´ enon and Holmes attractors
Another model, with very interesting properties, was proposed by Holmes (Holmes,
1979a):
x
t+1
= bx
t−1
+ ax
t
− x
3
t
(1.10)
Fig 1.5(b) presents the Holmes model when the parameters take the values a = 3, b =
0.8.
The H´ enon model, along with other related models, extensions and variations,
such as the Holmes model, are presented in Chapter 5.
8 Chaotic Modelling and Simulation
1.3 More Complex Structures
1.3.1 Threedimensional and higherdimensional models
A number of interesting threedimensional models are studied in Chapter 6. These
include the Lorenz (Lorenz, 1963) and R¨ ossler models, the autocatalytic and Ar
neodo (Arneodo et al., 1980, 1981b) models, and their extensions and variations.
3dimensional views of the Lorenz and autocatalytic attractors are illustrated in Fig
ures 1.6(a) and 1.6(b) respectively.
(a) The Lorenz attractor (3d) (b) The autocatalytic attractor
FIGURE 1.6: The Lorenz and autocatalytic attractors
We also examine a fourparameter model of chemical reactions, which gives rise
to a rich variety of chaotic patterns. A twodimensional view of this chaotic at
tractor, along with the time development of onedimensional oscillations, appears in
Figure 1.7.
1.3.2 Conservative systems
In Chapter 7, we consider conservative systems, that is, systems that have a non
trivial ﬁrst integral of motion, and in particular Hamiltonian systems. The existence
of a ﬁrst integral of motion provides enough information to draw the phase portrait of
the systemin cartesian coordinates. The resulting patterns exhibit a varying degree of
symmetry, depending on the equations of the system. Figure 1.8 shows an example
of two of the forms analysed in this chapter.
Introduction 9
FIGURE 1.7: A chemical reaction attractor
−0.3 −0.2 −0.1 0.0 0.1 0.2 0.3
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
FIGURE 1.8: Nonchaotic portraits
10 Chaotic Modelling and Simulation
1.3.3 Rotations
Chapter 8 is devoted to rotations, reﬂections and translations, and the chaotic
forms and attractors that these give rise to. The models in this category are char
acterized by the presence of rotations and translations, as well as nonlinear terms,
in the system equations. For appropriate values of the parameters, these models give
rise to chaotic patterns and attractors. The most popular of these is probably the
Ikeda model (Ikeda, 1979) (equations (1.14) and (1.13)).
Rotation models express a large variety of systems in various disciplines. Asimple
rotation in the plane, followed by a translation, does not in itself lead to chaotic
forms. The rotation equation, in its usual form, is based on linear functions such
as (1.11):
x
t+1
= x
t
cos(θ
t
) − y
t
sin(θ
t
)
y
t+1
= x
t
sin(θ
t
) + y
t
cos(θ
t
)
(1.11)
However, the introduction of nonlinear terms in system (1.11) gives chaotic so
lutions. First of all, let us consider the eﬀect of adding a quadratic term in both
equations. The system then becomes nonlinear:
x
t+1
= x
t
cos(θ) − y
t
sin(θ) + x
2
t
sin(θ)
y
t+1
= x
t
sin(θ) + y
t
cos(θ) − x
2
t
cos(θ)
(1.12)
When the rotation angle is stable, θ = 1.3, the system gives an interesting two
dimensional map with chaotic regions and islands of stability inside these regions
(Figure 1.9).
FIGURE 1.9: A rotation chaotic map
Another way to add nonlinearity is to assume that the rotation angle θ is a non
linear function, as is the case of the famous Ikeda attractor. The model is based on
a system of two rotation equations with a translation parameter a added to the ﬁrst
Introduction 11
equation, a space parameter b, and a nonlinear rotation angle of the form:
θ
t
= c −
d
1 + x
2
t
+ y
2
t
(1.13)
The rotationtranslation system in this case takes the following form:
x
t+1
= a + b
_
x
t
cos(θ
t
) − y
t
sin(θ
t
)
_
y
t+1
= b
_
x
t
sin(θ
t
) + y
t
cos(θ
t
)
_ (1.14)
This system was ﬁrst proposed by Ikeda (1979) to explain the propagation of light
into a ring cavity. The parameter values in that case where a = 1, b = 0.83, c = 0.4
and d = 6, and they give rise to a very interesting attractor illustrated in Figure 1.10.
FIGURE 1.10: The Ikeda attractor
This map has Jacobian J = b
2
. When b < 1 several chaotic and nonchaotic
patterns can be seen in the various simulations, some of which are illustrated in
Chapter 8.
The properties of the chaotic maps in this category are so rich and often unex
pected. The shape of the Ikeda attractor in the previous standard form depends on
the selection of the values of the parameters. By varying these parameters, very inter
esting new forms appear. A wide variety of these forms are presented and analysed
in Chapter 8. Two of these attractors are illustrated in Figures 1.11(a) and 1.11(b). In
Figure 1.11(a) an attractor pair appears. The parameters are a = 6, b = 0.9, c = 3.1
and d = 6. In Figure 1.11(b), only the parameter c is changed (c = 2.22). Now three
images of the attractors are present.
12 Chaotic Modelling and Simulation
(a) A double Ikeda attractor (b) A triple Ikeda attractor
FIGURE 1.11: Ikeda attractors
1.3.4 Shape and form
In Chapter 9 we study how chaotic analysis can be used as a tool to design forms
that appear in physical systems. Such an analysis is useful in the scientiﬁc ﬁeld re
lated to shape and form. Even the most strange chaotic attractors and forms can have
real signiﬁcance, especially in biological systems but also in ﬂuid ﬂow or in crys
tal formation. The analysis of these systems has moved from the classical analysis
concerning simple and regular forms to chaotic analysis. The aim here is to go be
yond the classical stability analysis. We try to ﬁnd mechanisms to design simple or
complicated and chaotic forms, and to pass from a qualitative to a quantitative point
of view. The seashell formation illustrated earlier in this chapter is one case of an
attractor with a special shape. The equations used are 3dimensional and the original
inspiration comes from the R¨ ossler model.
The main tools for shape formation are rotation and translation, as in the Ikeda
case above, but also reﬂection with translation, based on the following set of equa
tions:
x
t+1
= a + b
_
x
t
cos(2θ
t
) + y
t
sin(2θ
t
)
_
y
t+1
= b
_
x
t
sin(2θ
t
) − y
t
cos(2θ
t
)
_ (1.15)
If the same function as in the Ikeda case is used for the angle θ
t
θ
t
= c −
d
1 + x
2
t
+ y
2
t
then we obtain an attractor (Figure 1.12(a)) which is divided in two new separate
forms (Figure 1.12(b)). The parameters in the ﬁrst case are a = 2.5, b = 0.9, c = 1.6
and d = 9. In the second case, a = 6 and the other parameters remain unchanged.
Introduction 13
The centres of the two attractors are located at:
(x, y) = (a, 0) and
(x, y) =
_
a + ab cos(θ), ab sin(θ)
_
where θ = c −
d
1+a
2
. The reﬂection attractor is on the left side of Figure 1.12(b).
These cases simulate the split of a vortex in two separated forms. The parameter a is
quite higher than in the previous case.
(a) Reﬂection with 1 image (b) Reﬂection with 2 images
FIGURE 1.12: Translationreﬂection attractors
The rotation equation (1.14), applied to the Ikeda case, but now with b = 1, is a
spacepreserving transformation. Interesting forms arise assuming that the rotation
angle is a function of the distance r
t
=
_
x
2
t
+ y
2
t
. For instance, when
θ
t
= c +
d
r
2
t
we obtain the form illustrated in Figure 1.13(a). There is a symmetry axis at x =
a
2
and one equilibrium point in the central part of the ﬁgure, located at (x, y
t+1
) =
_
a
2
, y
t
_
. The ﬁve equilibriumpoints in the periphery are of the ﬁfth order: (x
t+5
, y
t+5
) =
(x
t
, y
t
). One of these points is also located on the x =
a
2
axis, whereas the other four
are located in symmetric places about this axis. The parameters are a = 3, c = 1 and
d = 4.
Furthermore, the Ikeda model with b = 1 provides interesting forms with the same
axis of symmetry, x =
a
2
. Figure 1.13(b) illustrates this case with parameter values
a = 1, c = 5 and d = 5. In the ﬁgure there are two pairs of secondorder equilibrium
points, ﬁve ﬁfthorder and eight eightorder equilibrium points.
14 Chaotic Modelling and Simulation
(a) Fifthorder symmetry (b) Multipleorder symmetries
FIGURE 1.13: Symmetric forms
In Chapter 9, several symmetric and nonsymmetric forms are presented and anal
ysed. The following ﬁgures illustrate a symmetric chaotic form before the separation
step (Figures 1.14(a) and 1.14(b)), in the intermediate stage (Figure 1.14(c)), and af
ter the separation of the chaotic form into two symmetric forms (Figure 1.14(d)).
The parameters are b = 1, c = 2.85 and d = 16, and the parameter a takes the values
a = 18, 35, 45 and 55 respectively. The rotation iterative procedure is used, with
rotation angle θ
t
= c +
d
r
t
.
Another category of patterns includes forms which we call “ﬂowers.” By using
the rotation formula with rotation angle
θ
t
= c +
d
r
δ
t
and with parameter values b = 0.9, a = c = d =
2π
k
, k =
17
12
and δ = 2, the graph in
Figure 1.15(a) appears. When a =
2π
3k
, c = d =
2π
k
and δ = 3, Figure 1.15(b) results.
1.4 Chaos and the Universe
1.4.1 Chaos in the solar system
Complicated and even chaotic orbits and paths in the solar system are studied
and simulated in Chapter 11. Objects in the solar system mainly follow circular or
elliptic orbits around a centre of mass. In this case the simulation can follow a purely
geometric approach. To simplify the problem the centre of mass is assumed to be
located in the centre of the large mass. The attracting force is, of course, gravity, and
Introduction 15
(a) One symmetric form (b) Starting separation
(c) First separation stage (d) Final separation stage
FIGURE 1.14: Separation of chaotic forms
16 Chaotic Modelling and Simulation
(a) δ = 2 (b) δ = 3
FIGURE 1.15: Flowers
is presented by an inverse square law of the distance. The nondimensional equations
of motion for a body with small mass m rotating around a large body with mass M,
which we consider to stay in equilibrium position, are, in Cartesian coordinates:
¨ x = −
x
_
x
2
+ y
2
_
3/2
¨ y = −
y
_
x
2
+ y
2
_
3/2
(1.16)
The initial conditions are x = 1, y = 0, ˙ x = 0 and ˙ y = ν. The escape velocity is
ν =
√
2, and the circular orbit is obtained when ν = 1. The circular orbit is illustrated
in Figure 1.16(a). In the same ﬁgure, an elliptic orbit is presented, where the initial
velocity is 1 < ν = 1.2 <
√
2 . A hyperbolic orbit is also presented. The initial
velocity is higher than the escape limit
_
ν = 1.45 >
√
2
_
and the particle ﬂies oﬀ to
inﬁnity. Figure 1.16(b) illustrates a large number of elliptic paths for the same value
of the initial velocity as before. The paths remain in the same plane, but deviate from
the original position, forming the shape presented in the ﬁgure.
The motions of two attracting point masses are relatively simple. The equations
of motion for the point mass m in Cartesian coordinates are:
¨ x
1
= −
(x
1
− x
2
)m
2
_
(x
1
− x
2
)
2
+ (y
1
− y
2
)
2
_
3/2
¨ y
1
= −
(y
1
− y
2
)m
2
_
(x
1
− x
2
)
2
+ (y
1
− y
2
)
2
_
3/2
(1.17)
Analogous is the case for the point mass m
2
. The gravity constant is G = 1 and
the total mass of the rotating system is M = m
1
+ m
2
. If we give the two masses
initial positions (x
1
, y
1
) and (x
2
, y
2
), and initial velocities ˙ x
1
, ˙ y
1
, ˙ x
2
and ˙ y
2
, then the
Introduction 17
(a) Circular, Elliptic and Hyperbolic orbits (b) Many elliptic orbits
FIGURE 1.16: Orbits in twobody motion
masses revolve around each other while drifting together, as a pair, through space.
Their centre of mass drifts at a constant velocity. Relative to an observer moving
with the centre of mass, their orbits are ellipses and, in special cases, circles. The
motion of two attracting points in space is presented in Figure 1.17(a). The heavy line
represents the movement of the centre of mass. The masses revolve around the centre
and drift together. The points have masses m
1
= 0.4, m
2
= 0.6 and initial velocities
˙ x
1
= −0.15, ˙ x
2
= 0, ˙ y
1
= −0.45 and ˙ y
2
= 0.25. The same motion, viewed relative
to the centre of mass, is illustrated in Figure 1.17(b). Each pointmass follows an
elliptic path.
(a) Attracting bodies in space (b) Relative movement
FIGURE 1.17: Motion in the plane
18 Chaotic Modelling and Simulation
The threebody problem, reduced to the plane, still gives very complicated and
chaotic orbits. To simplify the problem we assume that the third body, a satellite, is
of negligible mass relative to the other two bodies. The other two, the planets, travel
around the centre of the combined mass in circular orbits in the same plane. The
satellite rotates in the same plane. The total mass of the planets is m
1
+ m
2
= 1. The
equations of motion for the satellite then are:
¨ x = −
m
1
(x − x
1
)
r
3/2
1
−
m
2
(x − x
2
)
r
3/2
2
¨ y = −
m
1
(y − y
1
)
r
3/2
1
−
m
2
(y − y
2
)
r
3/2
2
(1.18)
where (x, y) is the position of the satellite, r
1
, r
2
are the distances of the satellite from
the two planets, m
1
, m
2
are the masses of the two planets, and (x
1
, y
1
) and (x
2
, y
2
) are
the positions of the planets at the same time t.
In the following example the planets have masses m
1
= 0.9, , m
2
= 0.1, and the
position and initial velocity of the satellite are (−1.033, 0) and (0, 0.35) respectively.
The complicated and chaotic paths are illustrated in the following ﬁgures. In Fig
ure 1.18(a), the movement is viewed from the outside of the system, from space. It
shows that the satellite can move between the two revolving planets. Figure 1.18(b)
illustrates the orbit of the satellite viewed in a coordinate system that rotates with the
revolution of the planets (rotating frame), with the Xaxis always passing through the
planets according to the rotation formulas:
X = x cos t + y sin t
Y = −x sin t + y cos t
(1.19)
(a) Motion in space (b) Motion in a rotating frame
FIGURE 1.18: Motion in space
Introduction 19
By using diﬀerence equations to model solar systems, it is possible to simulate
chaotic patterns like the rings of Saturn and other planets. The simplest approach is
to use an equation like the logistic, and to assume that the system rotates following
elliptic orbits. Two realizations appear in the next ﬁgures. The parameter of the
logistic model is b = 3.6 for Figure 1.19(a) and b = 3.68 for Figure 1.19(b). The
rotation formula is:
x
t
= 2r
t
cos t
y
t
= 2r
t
sin t
(1.20)
where r
t+1
= br
t
(1 − r
t
).
(a) b = 3.6 (b) b = 3.68
FIGURE 1.19: Ring systems
1.4.2 Chaos in galaxies
The underlying literature on galactic simulations deals mainly with the Nbody
problem and related simulations. The approach to the Nbody problem follows some
simpliﬁcations. The majority of the models we examined are based on numerical
integration of the equations of motion for N mutually interacting particles with equal
masses, which follow a steady rotational motion.
1
The acceleration of the ith parti
cle is:
N
ji
r
j
− r
i
_
r
2
i j
+ r
2
cut
_
3/2
1
See Sellwood (1983, 1989); Sellwood and Wilkinson (1993).
20 Chaotic Modelling and Simulation
The cutoﬀ radius r
cut
greatly simpliﬁes the numerical computation by eliminating
the infrequent, but very large, accelerations at close encounters. The gravity constant
and the mass of each particle are set to 1. When N is large, the main computational
problem is the large number (∝ N(N − 1)) of operations required to determine the
accelerations.
When a small perturbation is applied to the rotating system, familiar galaxy forms
appear after some iterations. The details are presented in Chapter 11. Two character
istic examples of these forms are the twoarmed spirals illustrated in Figures 1.20(a)
and 1.20(b). The simulations start from an elliptic disk with diﬀerent density dis
tribution. In Figure 1.20(b), the high central density leads to the formulation of a
barred spiral form.
(a) Twoarmed spiral
_
ρ =
c
2
r
4
, t = 100
_
(b) Barred spiral
_
ρ =
c
1
r
8
, t = 200
_
FIGURE 1.20: Spiral galaxies
Another approach to producing a spiral pattern formation is based on the elliptical
epicycle that a star follows around its guiding centre. According to the theory of
kinematic spirals, the oval orbits are nested to form a twoarmed spiral pattern of
a galaxy.
2
The simulation results, illustrated in Figure 1.21(a), are based on the
following equation in polar coordinates (r, θ):
r =
R
g
1 + a cos
_
n(5 − 5R
g
+ φ)
_
The values of the guiding centre are 0.2 < R
g
< 1. The parameters are a =
0.08, n = 2 for the twoarmed spiral and a = 0.15, n = 1 for the onearmed spiral
(Figure 1.21(b)).
2
See also Sparke and Gallagher (2000).
Introduction 21
(a) Twoarmed spiral (b) Onearmed spiral
FIGURE 1.21: Spiral galaxies
Recent discoveries of new galaxies, intergalactic forms, planetary nebulae, black
holes and other objects give inspiration for new simulations of the large variety of
observed forms and shapes. The questions about black matter and how it inﬂuences
the movement and shape of galaxies and other objects change our approach to model
building and the related simulations of galactic forms. In many instances, the galactic
forms arise by using very simple models and simulation tools. The forms presented
below arise from the reﬂection iterative formula (1.20) with parameters a = 1, b = 1,
and rotation angle:
θ
t
= c −
d
r
2
t
The reﬂection procedure leads to mirror image symmetric shapes. In some in
stances a central chaotic bulge is created. Figure 1.22(a) illustrates the beginning of
the mirror image formation, whereas in Figure 1.22(b) the central bulge is already
present. The outer part has the form of an electromagnetic ﬁeld. In both cases the
parameter c is set to 1.6, while the parameter d is 35 for Figure 1.22(a) and 200 for
Figure 1.22(b).
When d = 800, the reﬂection image takes the form presented in Figure 1.22(c).
There is a centralbulge connected with the outer periphery by two symmetric routes.
Figure 1.22(d) also illustrates a case where d = 800, but now a = 0.6. In this case,
only the central bulge remains, while the form in the outer region is just beginning
to take shape.
Rotations, reﬂections and translations with appropriate selection of the equation
of the rotation angle give rise to a plethora of galaxy formations. Figure 1.23(a)
illustrates a simple spiral pattern with one companion, whereas in Figure 1.23(b)
two spiral galaxies with two companions appear. In both cases b = 0.9, c = 2, d = 6,
and a is 0.3 for Figure 1.23(a) and 5 for Figure 1.23(b). The rotation angle in both
22 Chaotic Modelling and Simulation
(a) d = 35 (b) d = 200
(c) d = 800 (d) d = 800, a = 0.6
FIGURE 1.22: Reﬂection forms
Introduction 23
case is given by:
θ
t
= c +
d
r
t
(a) a = 0.3 (b) a = 5
FIGURE 1.23: Spiral Patterns
1.4.3 Galactictype potentials and the H´enonHeiles system
Contopoulos (1958, 1960, 1965) explored the boxlike paths of a body in a galactic
potential, and found that the orbits in the meridian plane of an axisymmetric galaxy
are like Lissajous ﬁgures. In his book (Contopoulos, 2002), Contopoulos explains
that in 1958 he expected the orbits to be ergodic and ﬁll all the space inside the
energy surface. Instead, he found that the orbits did not ﬁll all the available space,
but instead ﬁlled curvilinear parallelograms, as in the case of deformed Lissajous
ﬁgures. Later he could prove (Contopoulos, 1960) that such orbits can be explained
both qualitatively and quantitatively by a formal third integral of motion.
The work of H´ enon and Heiles (1964) on galactic motion is the ﬁrst complete work
that establishes the existence of chaos in galaxy formations. It provided considerable
evidence of the existence of chaotic motions in galaxies. Their work started as a
computer experiment in order to explore the existence and applicability of the third
integral of galactic motion. Chapter 12 is devoted to this work of H´ enonHeiles and
Contopoulos.
The works of Contopoulos and H´ enonHeiles are examples of what we will call
computer experiments. This new type of experiments gave new directions in various
scientiﬁc ﬁelds, and especially in astronomy. Sometimes the computer results were
surprising and contradicted the existing theories of that time.
24 Chaotic Modelling and Simulation
The Hamiltonian applied by H´ enon and Heiles (1964) has the form
H =
1
2
( ˙ x
2
+ ˙ y
2
) +
1
2
_
x
2
+ y
2
+ 2x
2
y −
2y
3
3
_
= h (1.21)
So the potential U(x, y) that is used has the form:
U(x, y) =
1
2
_
x
2
+ y
2
+ 2x
2
y −
2y
3
3
_
or in polar coordinates:
V(r, θ) =
1
2
r
2
+
1
3
r
3
sin(3θ)
(a) The original H´ enonHeiles (y, ˙ y) dia
gram
(b) A discrete alternative (E = 1/12)
FIGURE 1.24: H´ enonHeiles systems
We have reproduced here, via computer simulation, the original ﬁgure Figure 1.24(a)
presented in the H´ enonHeiles paper. H´ enon and Heiles estimated the paths of a par
ticle according to the Hamiltonian (1.21) at the energy level E = h = 1/12, and then
they provided the (y, y
′
) diagram when x = 0. In this low energy level mainly regular
orbits appear. At a higher energy level chaotic regions are formed.
A discrete alternative of the H´ enonHeiles model results in the image presented
in Figure 1.24(b) for the energy level E = 1/12. In this case, c = 0.1. The discrete
model is based on the following set of equations:
x
n+1
= cx
n
+ u
n
y
n+1
= cy
n
+ v
n
u
n+1
= −x
n
− 2x
n+1
y
n+1
v
n+1
= −y
n
− x
2
n+1
+ y
2
n+1
(1.22)
Introduction 25
The Jacobian of the system (1.22) is:
J = 1 + 2(c − 1)y
n+1
+ 4(c − 1)
_
x
2
n+1
+ y
2
n+1
_
It is obvious that for c = 1, J is 1, and therefore the spacepreserving property
is satisﬁed. However, when c takes values less than 1, very interesting ﬁgures arise
(c = 0.1 in Figure 1.24(b)). The similarities with the H´ enonHeiles system (Fig
ure 1.24(a)) are obvious.
FIGURE 1.25: A twoarmed spiral galaxy
Twoarmed spiral galaxy forms can be obtained from simulations of very sim
ple models. The twoarmed spiral galaxy in Figure 1.25 is produced by using the
following rotation model:
x
n+1
= b(x
n
cos a − y
n
sin a) + x
2
n
sin a
y
n+1
= b(x
n
sin a + y
n
cos a) − x
2
n
cos a
(1.23)
In this case, a = 0.2 and b = 0.9. Since the Jacobian is J = b
2
, b is an area
contracting parameter.
The H´ enonHeiles system gives also interesting paths in the (x, y) plane. The paths
are characterized by the level of potential. The form of the potential is triangular.
Totally chaotic trajectories is illustrated in Figures 1.26(a) and 1.26(b). The energy
level in both cases is E = h = 1/6, which is the escape limit. For this limit the
equipotential triangle is drawn. All the (x, y) paths of the particle will be restricted
inside the triangle. The initial values are (x, y) = (0, 0), and v = 0.1. The value of u,
0.5668627, is estimated so that the value of the Hamiltonian is h = 1/6.
1.4.4 The Contopoulos system
A Hamiltonian with a “galactictype” potential was ﬁrst introduced by Contopou
los (1958, 1960) in his pioneering work on galaxies. The potential is based on the
addition of two harmonic oscillators, along with higher order terms. The resulting
26 Chaotic Modelling and Simulation
(a) A chaotic path over a long time period (b) A chaotic path over a short time pe
riod
FIGURE 1.26: Chaotic paths in the H´ enonHeiles system (E = 1/6)
form is
U(x, y) =
1
2
_
w
2
1
x
2
+ w
2
2
y
2
_
− exy
2
with corresponding Hamiltonian:
H =
1
2
_
u
2
+ v
2
_
+ U(x, y) = h
Without loss of generality this Hamiltonian can be simpliﬁed to give the form:
H =
1
2
_
u
2
+ v
2
_
+
1
2
_
k
2
x
2
+ y
2
_
− exy
2
=
1
2
where u, v, x, y, e have been rescaled, and k = w
1
/w
2
is the important resonance ratio.
The equations of motion are given by:
˙ u = −
∂U
∂x
= −k
2
x − 2ey
2
˙ v = −
∂U
∂y
= y − 2exy
(1.24)
Resonant orbits, characterized as 4/1 (left image) and 2/3 (right image), are illus
trated in Figure 1.27. For the 4/1 case the parameters are e = 0.1 and k = w
1
/w
2
=
4/1, and the initial conditions are x = 0.2, y = 0, u = 0 and v = 0.6. For the 2/3
case the parameters are e = 0.1 and k = w
1
/w
2
= 2/3, and the initial conditions are
x = 0.1, y = 0, u = 0 and v = 0.9977753.
Introduction 27
FIGURE 1.27: The Contopoulos system: orbits and rotation forms at
resonance ratio 4/1 (left) and 2/3 (right)
1.5 Odds and Ends, and Milestones
The book ends with Chapter 13, which is a collection of interesting systems of or
variations on systems, including the eﬀect of introducing noise into models, and an
extensive discussion of the very interesting LotkaVolterra system. The interesting
images of Section 13.6 could certainly be considered as works of art. We hope they
will act as an inspiration for future researchers.
The history of chaotic modelling is quite complex. A “milestones” table in Chap
ter 14 should provide a useful reference for future reading.
Chapter 2
Models and Modelling
2.1 Introduction
For many years scientists from all ﬁelds have been trying to describe the real
world by constructing logical forms of varying complexity, called models. The pro
cess of selecting the main characteristics of a real life phenomenon and ﬁnding the
basic mechanisms that describe the development of the phenomenon is called mod
elling, or also model building. Since the beginning of time, humans have been using
models to understand the real world. The development of language enabled man to
form important verbal constructs. These verbal constructs became the ﬁrst models,
expressing not just real objects, but also abstract ideas and innovations. Ancient
Greeks established the ﬁrst schools and academies, in order to encourage learning
and to enhance the dissemination of information and exchange of ideas on scientiﬁc
model building; a process known as dialectics, or more commonly philosophy.
Mathematical model building was also studied in Ancient Greece, mainly as a
subset of geometry. Calculus and analysis came many centuries later. Nevertheless,
their foundations can be found in the Pythagorean theory of fractions and in the
Archimedean theory of polygonal approximations. Indeed Newton’s “Method of
Fluxions,” where his treatment of calculus is presented, also includes a section on the
“Applications on the Geometry of CurveLines.” His explanation of Kepler’s second
lawof planetary motion, as described in his famous letter to Halley,
1
is reminiscent of
Archimedes’s computation of the area of the circle by inscribed polygons. The paper,
titled “De motu corporum in gyrum,” is based on a purely geometric formulation of
a model describing planetary motion.
In this book we analyse models that are expressed in mathematical terms. All
the models we will examine fall under two categories, diﬀerential equation models
and diﬀerence equation models, also called maps. During the modelling process,
the standard methods of linear and nonlinear analysis are extensively used. We pay
particular attention to the methods of nonlinear analysis based on singular points,
equilibrium points and characteristic trajectories, and eigenvalues of the Jacobian,
the characteristic matrix of the system.
1
See the introduction in Acheson (1997).
29
30 Chaotic Modelling and Simulation
2.2 Model Construction
Models are approximations of reality. In some cases, they describe the real sit
uations quite well. However, the exact form of a model is not always possible to
determine. The modelling process thus proceeds in two steps. First, a general form
for the model is provided, typically in the form of a modelling function. This func
tion will depend on some unknown parameters, initial conditions, and possibly other
special characteristics, and thus can describe a number of diﬀerent situations. The
second step in the process is the estimation of these parameters so that the model
approximates the real situation to the best of its potential a process typically known
as “model ﬁtting.” In this section we will see numerous methods commonly used for
constructing models.
2.2.1 Growth/decay models
An important general class of modelling functions are the so called Growth Func
tions. These are functions that express the growth of a system or of some special
characteristics of a system, and they have, in general, a positive ﬁrst time derivative.
Similarly, functions with a negative ﬁrst derivative would express the decline of a
system, and would in general describe a decay process. The mathematical treatment
in both cases is very similar.
It is important here to emphasise that one needs to strike a balance between the the
oretical constructions of models, and the practical application of these models in real
situations. Practical applications that lack a theoretical background do not contribute
much to the development of any scientiﬁc ﬁeld. A purely theoretical approach, on the
other hand, becomes relevant only when supported by related applications (Skiadas,
1994; Skiadas et al., 1994).
Growth functions can be described in a number of diﬀerent ways. We discuss
some of these ways in the sections that follow.
2.2.1.1 Diﬀerential equation models
Models based on diﬀerential equations have been extensively used ever since the
invention of calculus. In such models, the system under consideration is described
by one or more functions that satisfy a system of diﬀerential equations. Thus, one
doesn’t always have a closed formula for the function, but a number of simulation
and approximation techniques are available.
During the modelling process, one usually agrees upon the general formof the sys
tem, based on general considerations of the form of the solutions, and the expected
behaviour of the system. One then proceeds to estimate the parameters involved, and
to study the eﬀect that these parameters have on the behaviour of the solution.
The most widely used growth/decay model is the exponential model, where the
rate change of the system is proportional to the system’s current state. This can be
Models and Modelling 31
described by the simplest ﬁrst order diﬀerential equation:
˙ x = bx (2.1)
Here, as in the rest of this book, we will use Newton’s notation, ˙ x, for the time
derivative of x:
˙ x
DFN
=
∂x
∂t
In equation (2.1), b is the model parameter that needs to be speciﬁed. Positive
values of b will give rise to growth functions, while negative values will describe
decay functions.
Sometimes systems such as (2.1) can be explicitly solved. In this particular case,
as is typically shown in any calculus class, the solution is the familiar exponential
function:
x(t) = x(0)e
bt
Often such an explicit description is not possible, and one has to rely on other quan
titative and qualitative techniques for describing the solutions.
2.2.1.2 Diﬀerence equation models
Diﬀerence equations are the discrete analog of diﬀerential equations. The time
parameter t is here assumed to be taking only discrete, values: t = 0, 1, 2, . . ., corre
sponding to consecutive stages of the system, referred to as steps. The time derivative
is then replaced by a ﬁnite time diﬀerence:
˙ x =
dx
dt
≈
∆x
∆t
=
x
t+1
− x
t
(t + 1) − t
= x
t+1
− x
t
(2.2)
The “rate of change” thus becomes simply the change in the system from the one
stage to the next.
A diﬀerence equation is then nothing more than an algebraic relation between x
t
and its successors, x
t+1
, x
t+2
, . . ., or if you prefer of x
t+1
and its predecessors. These
diﬀerence equations are referred to as maps in the chaotic literature. For instance,
equation (2.1) has the following discrete analog:
x
t+1
− x
t
= bx
t
or more simply:
x
t+1
= bx
t
(2.3)
In diﬀerence equation systems, closed form solutions can sometimes be obtained
by recursion and induction. For equation (2.3), this can be readily achieved and the
solution is
x
t
= x
0
b
t
(2.4)
32 Chaotic Modelling and Simulation
which, as you will notice, greatly resembles the diﬀerential equation solution.
2
2.2.1.3 Stochastic diﬀerential equations
The growth or decline of a system expressed by a variable x
t
over time can often
be formulated by two components:
1. The growth part or inﬁnitesimal growth, µ(x, t) dt
2. A measure expressing the inﬁnitesimal ﬂuctuations, σ(x, t) dw
t
A combination of these two components provides a stochastic diﬀerential equation
of the general form:
dx(t) = µ(x, t) dt + σ(x, t) dw
t
(2.5)
where the growth function x = x(t) can, without loss of generality, be considered to
be bounded (0 ≤ x(t) ≤ 1),
3
the functions µ(x, t), σ(x, t) are to be speciﬁed, and w
t
is
the standard Wiener process (Wiener, 1930, 1938, 1949, 1958).
Every stochastic diﬀerential equation has a deterministic analogue, where the ﬂuc
tuations are assumed to be zero:
dx(t) = µ(x, t)dt
In many applications, the growth rate ˙ x =
dx
dt
is a function only of the magnitude x of
the system, and not of time. Hence the deterministic analogue takes a much simpler
form:
dx(t) = µ(x)dt
2.3 Modelling Techniques
We will now discuss several general techniques for model construction. Series
approximation (section 2.3.1) can help us reduce a complex systemto a much simpler
one that closely resembles it. It is a technique that we will use many times throughout
the book.
The most common source of models is the close investigation of real situations,
and we discuss this process in section 2.3.2. In section 2.3.3, the Calculus of Varia
tions approach is discussed, with its source in the Lagrangian reformulation of clas
2
In fact, if we keep in mind that the b in (2.3) here corresponds to the 1+b of (2.1), we can rewrite (2.4)
as
x
t
= x
0
(1 + b)
t
= x
0
e
t ln(1+b)
≈ x
0
e
bt
for small values of b.
3
Since the system can be rescaled, so that x(t) measures the percentage of the current state of the
system over its “maximum state.”
Models and Modelling 33
sical mechanics. Diﬀusion equations are discussed in section 2.3.4. Finally, sec
tion 2.3.5 discusses delay diﬀerential equations, that have the ability to express the
delay in propagation of changes inherent in many natural and social systems.
2.3.1 Series approximation
When the function x
t
describing a system takes values close to a particular point,
then one can approximate a complex model by expanding the functions in the sys
tem in a Taylor series around that point. This method often gives suﬃcient results
when the original model is simple, smooth and continuous, and the derivatives of the
general equation of the model exist. As an example, consider a diﬀerential system of
the form:
˙ x = f (x) (2.6)
where f (x) is a reasonably behaving function of x. If we are interested in values of x
near 0, then we can replace f with the ﬁrst couple of terms of its Taylor expansion:
f (x) = a
0
+ a
1
x + a
2
x
2
+ a
3
x
3
+ a
4
x
4
+ · · · (2.7)
Here the coeﬃcients of the expansion can be readily interpreted as the derivatives
of f at 0:
a
n
=
f
(n)
(0)
n!
As an example of this approximation process, if we replace f in (2.6) by its linear
approximation, and possibly after replacing x by x + c, we obtain the exponential
model:
˙ x = a
1
x
In the majority of cases, using only the ﬁrst (direction), and possibly second (cur
vature), derivative terms provides an adequate approximation to f , and hence to the
model. Thus, it is not surprising that systems where f is a quadratic polynomial
approximate very well other more complicated systems. Of course, if necessary, the
higher terms of the Taylor expansion may also be used.
Often even this quadratic form can be simpliﬁed further. Symmetry of the system,
quite common in physical systems, often leads to important simpliﬁcations of the
approximation, and can lead directly to an expression of the model equation. For
instance, if the xspace is isotropic,
4
we are led to the functional equation:
f (x) = f (−x)
Since the odd derivatives of an even function vanish at 0, the expansion (2.7), can
be simpliﬁed to:
f (x) = a
0
+ a
2
x
2
+ a
4
x
4
+ · · · (2.8)
4
i.e. f is symmetric around the origin.
34 Chaotic Modelling and Simulation
Probably the most wellknown example of this type is the equation for the kinetic
energy of a mass. If f (v) represents the energy of the system, then a good approxi
mation for small v (close to 0) would be:
f (v) = a
0
+ a
2
v
2
The two terms can be identiﬁed physically as the energy at rest
f (0) = a
0
and the kinetic energy
f
kin
= a
2
v
2
=
1
2
mv
2
On the other hand, economic, social and biological phenomena do not have this
isotropic property. The same lack of symmetry holds for the time space, and the
relevant Taylor approximations tend to include a linear term as well. The constant
term in such a system can often be assumed to be zero.
5
Hence, a simple approx
imation will involve linear and quadratic terms. This approximation gives rise to a
diﬀerential equation known as “The Logistic,” ﬁrst proposed by Verhulst (1845) to
model the population growth in France. Later on, Pearl and Reed (1920) applied this
model to express the population growth in the United States.
The logistic diﬀerential equation has the simple form
˙ x = bx(1 − x) (2.9)
where x is the population at the present time, over the maximum level that the popu
lation could reach in the future. In other words, x is the saturation level.
To see how equation (2.9) follows naturally from (2.6) through the process of a
Taylor series approximation, consider that x is restricted in the interval [0, 1]. We
can consider the value x =
1
2
where the population is at half its potential as the
centre, and assume that the system will exhibit symmetry around x =
1
2
. In that case,
the Taylor approximation to the second order would be:
f (x) = a
0
+ a
2
_
x −
1
2
_
2
=
_
a
0
+
1
4
_
− a
2
_
x − x
2
_
= a + bx(1 − x)
Here a is the rate of growth of the population when the population is at x = 0 or
x = 1, so we can reasonably assume that a = 0. The other parameter, b, is related to
the growth rate at x =
1
2
. We are thus led to equation (2.9).
Equation (2.9) is a separable diﬀerential equation, and so can be easily solved
explicitly, by rewriting it as
1
x(1 − x)
dx = bdt
5
For instance in birth processes or innovation diﬀusion cases.
Models and Modelling 35
The solution then would satisfy:
−ln(1 − x) + ln(x) = bt + C
This last equation can be rewritten in the ﬁnal form:
x(t) =
x
0
e
bt
1 + x
0
_
e
bt
− 1
_ (2.10)
The graph of (2.10) exhibits a very characteristic smooth sigmoid form (Fig
ure 2.1).
−6 −4 −2 0 2 4 6
t
0
x
0
1
x((t)) ==
x
0
e
bt
1 ++ x
0
((e
bt
−− 1))
FIGURE 2.1: The solution to the logistic equation
A diﬀerence equation whose form closely parallels that of equation 2.9, but whose
behaviour is a lot more interesting, also goes under the name “logistic”:
x
t+1
= bx
t
(1 − x
t
) (2.11)
The behaviour of the solutions to (2.11) depends on the value of the parameter b.
When b < 3, they follow a discrete sigmoid form. However, when b > 3 chaotic
oscillations appear. The oscillations become progressively more chaotic as b ap
proaches 4. The behaviour of the solutions of the discrete logistic will be explored
further in Chapter 3.
The correct diﬀerence equation analogue to (2.11), using the approximation of the
ﬁrst derivative by a ﬁnite diﬀerence as in (2.2), is:
x
t+1
= x
t
+ bx
t
(1 − x
t
)
Various second, third and higher order forms of diﬀerential and diﬀerence equa
tions models can be constructed with similar methods as above. These equations
may in some cases exhibit chaotic behaviour. These models will be examined more
closely in subsequent chapters.
36 Chaotic Modelling and Simulation
2.3.2 Empirical model formulation
In many cases in Physics, Biology, Economics, Marketing, Forecasting, etc, the
construction of an appropriate general model follows by the empirical investigation
of realistic situations. For instance, when the growth rate ˙ x is assumed to be propor
tional to the magnitude x of the system, then we end up with the exponential model
discussed on page 31. Another example is based on the relative growth rate
˙ x
x
=
d
dt
ln x = (ln x)
′
= ˙ y
where y, the relative magnitude of the system, is deﬁned as
y =
_
dx
x
= ln x
and can be approximated by
y ≈
∆x
x
That is, the relative magnitude y is a measure of all the relative changes of the system
from its original formation until its current status.
In this case, the standard measure of growth is the relative magnitude of the sys
tem, which is assumed as the metric scale of the system. This assumption leads to
a convenient model construction especially when systems measured in completely
diﬀerent scales are to be compared. Examples of using the relative magnitude of
the system as the standard measure are the Gompertz, Gaussian and Gamma models,
whose descriptions follow.
2.3.2.1 The Gompertz model
The simplest assumption about the relative growth rate is that it is constant. This
leads us to the exponential growth model
˙ x
x
= b discussed already. However, as it is
reasonable to expect that a system has limits to its growth, it would be more realistic
to assume a declining relative growth rate. Assuming that this decline is proportional
to the relative growth y of the system, we are led to a simple diﬀerential equation in
y:
˙ y = −by (2.12)
or
(ln x)
′
= −b ln x (2.13)
which can ﬁnally be rewritten as:
˙ x = −bx ln x (2.14)
This model was ﬁrst proposed in Gompertz (1825). Direct integration gives as
solution the Gompertz function:
x = e
ln(x
0
)e
−bt
Models and Modelling 37
Using the Taylor approximation methods of section 2.3.1, we can examine the
Gompertz system near x = 1: The logarithm ln(x) can be approximated by:
ln(x) ≈ x − 1
Using this approximation in (2.14) would result exactly in the logistic model. So for
x near 1, the Gompertz model can be approximated by the logistic model.
Another interesting variant occurs if we replace x by 1 − x in the Gompertz diﬀer
ential equation, resulting in:
˙ x = −b(1 − x) ln(1 − x)
This model arises by considering the relative decay of the system, instead of the
relative growth. This model has skewness opposite to that of the Gompertz model,
as now the maximum growth rate is achieved when
x = 1 −
1
e
instead of when x =
1
e
. A comparison of the two models is given in Figure 2.2.
Both models are referred to in the literature as “the Gompertz model,” with diﬀerent
disciplines preferring one model over the other. The second variant is favoured in
the actuarial sciences, as it is more intimately related to mortality.
x
0
b
e
1 e 1 −− 1 e
1
x
⋅⋅
FIGURE 2.2: The two Gompertz models
2.3.2.2 The Gaussian and Gamma models
The well familiar Gaussian and Gamma distributions arise naturally from the
relativegrowthrate setting, after some rather reasonable assumptions. For instance,
if may assume that the relative growth rate is a linear function of time, which leads
38 Chaotic Modelling and Simulation
us to the following equation:
˙ y = (ln x)
′
= a − bt (2.15)
where b > 0.
The solution to (2.15) is a Gaussian function, namely:
x = ce
−
(t−µ)
2
2σ
2
where µ =
a
b
, σ
2
= b, and the integration constant c is determined by normaliza
tion.
If, on the other hand, one assumes a slightly diﬀerent decreasing pattern for the
relative growth rate:
˙ y = (ln x)
′
= a(ln t)
′
− λ
then the solution is related to the Gamma function:
x = c(λt)
a
e
−λt
2.3.3 The calculus of variations approach
Calculus of variations is concerned with optimising a functional:
I =
_
L( ˙ x, x, t) dt (2.16)
where the functional L, called the Lagrangian, summarises the dynamics of the sys
tem in question, in terms of ˙ x, x and t. The ﬁrst step in the modelling process in this
case is the construction of an appropriate Lagrangian function. The solution to the
optimisation problem is then given by the EulerLagrange equation:
d
dt
_
∂L
∂˙ x
_
−
∂L
∂x
= 0 (2.17)
As a speciﬁc example, suppose the Lagrangian consists of two basic components:
L( ˙ x, x, t) = G( ˙ x) + V(x)
where G( ˙ x) expresses the forces of growth, and V(x) expresses the limits of the
system. Then the EulerLagrange equation becomes:
d
dt
_
∂G( ˙ x)
∂˙ x
_
−
∂V(x)
∂x
= 0 (2.18)
The form of the functions G( ˙ x) and V(x) is not always known. However, as the
integral (2.16) involves a summation of the inﬁnitesimal parts of a path, it is in many
cases valid to approximate the functions G( ˙ x) and V(x) using Taylor series, especially
when the growth system is continuous and smooth.
Models and Modelling 39
Note further an interesting property of equation (2.18): The ﬁrst order term of
a Taylor series expansion of G( ˙ x) does not contribute to the equation. Hence, the
simplest expression for G( ˙ x) is ˙ x
2
, and this is suﬃcient in many systems. This as
sumption leads to a Lagrangian of the form
L =
1
2
˙ x
2
+ V(x)
Equation (2.17) yields, in this case, the following diﬀerential equation of growth:
¨ x =
∂V
∂x
(2.19)
The simplest interesting expression for V is V(x) =
1
2
bx
2
. Equation (2.19) then
takes a very familiar form:
¨ x = bx
When b > 0, this model expresses a growth process, while when b < 0, it expresses
an oscillating process.
If we consider the Lagrangian as a function of the relative growth rate of a system,
y = ln x, then the exact same analysis will yield the equation:
¨ y =
∂V
∂y
Assuming V is constant, V(y) = c, we are led to the exponential model:
¨ y = 0 ⇒ ˙ y = b ⇒ (ln x)
′
= b ⇒ ˙ x = bx
As we’ve seen before, when b < 0 this equation expresses a decay process.
2.3.4 The probabilisticstochastic approach
A lot of models have their roots in probabilisticstochastic considerations, which
give rise to certain partial diﬀerential equations. These are usually diﬀusion equa
tions. The simple diﬀusion equation (D is the diﬀusion coeﬃcient)
∂u
∂t
= D
∂
2
u
∂x
2
and the FokkerPlanck equation
∂u
∂t
= −
∂(au)
∂x
+ D
∂
2
u
∂x
2
are the basis for many models.
A reactiondiﬀusion equation has the general form
∂u
∂t
= f (u) + D∇
2
u
40 Chaotic Modelling and Simulation
where, as usual, the Laplacian ∇
2
is given by:
∇
2
≡
∂
2
∂x
2
+
∂
2
∂y
2
+
∂
2
∂z
2
A classical model of this type is the socalled FitzHughNagumo model (FitzHugh,
1961, 1969; Nagumo et al., 1962). The onevariable form of this model is given by
˙ u = u(1 − u
2
) + K + D∇
2
u
where K is a parameter.
Simulation of models of this type gives rise to important chaotic cases of pattern
formation. These models explain the development of chaotic waves in ﬂuids and
chemical kinetics.
2.3.5 Delay growth functions
Delay growth models assume that the rate of growth, ˙ x =
dx
dt
, is a function not only
of (x, t), but also of some earlier time (t −T), where T is the delay. The general form
for a continuous ﬁrstorder delay growth model is:
˙ x = f
_
x, (t − T), t
_
A very simple delay growth model is the following multiplicative process:
˙ x = bx
t−T
g(x
t
) (2.20)
Equation (2.20) is not easy to handle. However, an approximation of the delay
term is possible if we retain the ﬁrst two terms of a Taylor series expansion at t:
x
t−T
≈ x
t
− T ˙ x (2.21)
Substituting (2.21) in (2.20) and rearranging terms, we obtain the system:
˙ x =
bxg(x)
1 + bTg(x)
Depending on the the function g(x), many diﬀerent models arise. If, for instance,
g is set to be the decreasing function g(x) = 1 − x, then we obtain the GRM1 model
proposed in Skiadas (1985) to express an innovation diﬀusion process:
˙ x =
bx(1 − x)
σ + (1 − σ)x
(2.22)
where σ = 1+bT. When no delay is present, σ = 1 and (2.22) reduces to the logistic
diﬀerential equation (2.9).
The discrete alternatives of delay growth models, expressed by diﬀerence equa
tions, give chaotic solutions for certain parameter values. These are examined in
some detail in Chapter 4.
Models and Modelling 41
2.4 Chaotic Analysis and Simulation
In the past decades, extensive studies in various ﬁelds led to the creation of a
new ﬁeld of analysis, called “Chaotic Analysis.” The innovation of this new ﬁeld
lies in its multidisciplinary approach to the problems studied, and its use of vari
ous methods and tools not applied before. Most notable among these is the use of
computers for the related analysis and simulation, that has become an integral part
of Chaotic Analysis. Determining the number of equilibrium islands and their loca
tions inside a chaotic sea is done ﬁrst by using chaotic simulations, and only later
by exact calculations, when possible. If we exclude computers and simulations from
chaotic analysis, an essential part of the chaotic phenomena, “the essence of chaos”
according to Lorenz (Lorenz, 1993), will be lost.
As an example of the advantages of simulation, consider the following formation
of two variants of a twodimensional sea urchin, by using the logistic diﬀerence
equation and a simple rotation scheme. The equations used are:
f
t+1
= bf
t
(1 − f
t
)
x
t
= (1 − f
t
) cos(θ
t
)
y
t
= (1 − f
t
) sin(θ
t
)
(2.23)
The rotation angle θ
t
changes by h = 0.01 in each time step, and the chaotic para
meter b is set to b = 1 +
√
8 in Figure 2.3(a) and b = 1.1 +
√
8 in Figure 2.3(b).
(a) (b)
FIGURE 2.3: Chaotic forms
The shape of a snail is formed from the same logistic diﬀerence equation as above,
but now the chaotic band (t, f ), which appears on the left side of Figure 2.4, rotates
42 Chaotic Modelling and Simulation
and simultaneously sinks, due to the addition of an exponential decay factor, in order
to form the twodimensional shape of a snail (right). The system of equations used
is:
f
t+1
= b f
t
(1 − f
t
)
x
t
= (1 − f
t
) cos(θ
t
)e
−at
y
t
= (1 − f
t
) sin(θ
t
)e
−at
(2.24)
The chaotic parameter used here is 1.03 +
√
8, and the time step is h = 0.005.
FIGURE 2.4: A snail’s pattern
2.5 Deterministic, Stochastic and Chaotic Models
Deterministic models are expressed by speciﬁc, deterministic equations, which
give the same results for the same initial conditions. Stochastic models on the other
hand are given by stochastic equations, which may give diﬀerent paths, and dif
ferent ﬁnal results, for the same initial conditions. The chaotic models can act as
deterministic models for some parameter values, but for some other values of the
chaotic parameters they will give radically diﬀerent ﬁnal values for inﬁnitesimally
small changes in the initial values. Chaotic models are in a sense deterministic, in
that the ﬁnal values are determined completely from the precise initial values. From
a practical point of view however, given the sensitivity of chaotic models to the ini
tial conditions, the unpredictability of the ﬁnal results and the nonuniqueness of the
paths followed by the system in chaotic systems are in some ways similar to the
unpredictability and nonuniqueness present in stochastic models.
As an illustration of this idea, one could attempt to use the logistic model as a
random number generator. To achieve that, consider the onedimensional stochastic
Models and Modelling 43
process generated by the following set of diﬀerence equations:
x
t+1
= 1 − 2x
2
t
y
t+1
= y
t
+ ex
t
(2.25)
Here, e is a small parameter (e = 0.17). Several paths of this process are illustrated in
Figure 2.5(a). The resemblance of these paths to those of a Wiener process is strik
ing. Another example, reminiscent of Brownian motion, appears in Figure 2.5(b).
This ﬁgure shows a twodimensional stochastic process based on a fourdimensional
mapping.
(a) Onedimensional diﬀusion (b) Twodimensional diﬀusion
FIGURE 2.5: Chaos as randomness
44 Chaotic Modelling and Simulation
Questions and Exercises
1. Construct the (t, x) graph for the map: x
t+1
= bx
t
. Compare this graph to the
graph of the solution to the diﬀerential equation
dx
dt
= bx. Consider various
values for b, including both negative and positive values, as well as values less
than 1 and greater than 1.
2. Solve the diﬀerential equation
dx
dt
= bx
2
, and graph the solution. Compare
your results to the exponential (Malthus) model
dx
dt
= bx.
3. Solve the more general Exponential diﬀerential equation:
dx
dt
= bx
a
, where the
constant a is both positive and negative. Discuss the special cases resulting
and draw graphs.
4. Provide examples and reallife applications modelled by the above two diﬀer
ential equations, including cases with positive as well as negative values for
the parameters a, b.
5. For each of the following diﬀerential equations, construct both the “correct”
and the “naive” diﬀerence equation analogue, and graph the (t, x) map. Com
pare this map with the solution to the diﬀerential equation.
(a)
dx
dt
= bx
(b)
dx
dt
= bx
2
(c)
dx
dt
= bx
2
(1 − x)
6. Using the Taylor expansion techniques, show that the diﬀerential equation
model
dx
dt
= sin(bx) may be approximated by the exponential model when
x is small. Use a simulation to compare the two models, and verify that the
approximation is indeed good near x = 0.
7. Using the Taylor expansion techniques, what simple model do you ﬁnd that
approximates the model
dx
dt
= cos(bx) for small x?
8. Use the calculus of variations methods described in section 2.3.3 to explain
why the very simple models described there,
dx
dt
= bx and
d
2
x
dt
2
+bx = 0, end up
explaining very frequently reallife situations.
9. The second order diﬀerential equation x
′′
+ bx = 0 provides oscillating so
lutions for positive values of the parameter b. Find the diﬀerence equation
alternative of this diﬀerential equation and check if the oscillating character is
retained in the map form. What is the result when b = 1?
10. What happens in the diﬀerence equation from the previous problem, but for
negative values of b? What is the diﬀerence between the two cases?
Models and Modelling 45
11. Conﬁrm that the maximum growth rates for the two Gompertz models are
indeed at the points
1
e
and 1 −
1
e
respectively. How much is the maximum
growth rate?
Chapter 3
The Logistic Model
3.1 The Logistic Map
The logistic map is given by the logistic diﬀerence equation
x
t+1
= bx
t
(1 − x
t
) (3.1)
Equation (3.1) has been extensively analysed in the last decades and has become
the basis for numerous papers and studies. The analysis of this relatively simple
iterative equation revealed a rich variety of inherent chaotic principles and rules. In
particular, the study of the period doubling bifurcations that appear when the chaotic
parameter b changes from one characteristic value to the next led to the formulation
of a universal law in order to explain the phenomenon, at least for all quadratic
maps (Feigenbaum, 1978, 1979, 1980a,d, 1983).
3.1.1 Geometric analysis of the logistic
We will present in this section the basic properties of the logistic map based on the
classical analysis proposed by Feigenbaum and others. However, we will approach
the subject from a more geometric perspective. In order to understand the evolution
of (3.1), we consider the set of points
_
(x
t
, y
t
)t = 1, 2, . . .
_
where:
y
t
= x
t+1
= f (x
t
) = bx
t
(1 − x
t
)
Thinking of the system (3.1) via this set of points allows us to examine the logistic
system in a geometric way, by considering ﬁrst of all the graphs of the functions
y = f (x) = bx(1 − x) and y = x. Given a value x
t
, we locate x
t+1
= y
t
= f (x
t
) by
locating the point on the graph of y = f (x) with x = x
t
. We then use the line y = x to
project this y coordinate back to the x axis (Figure 3.1(a)).
If we continue this process, we can trace the evolution of the system, as in Fig
ure 3.1(b). A point of particular interest is the point where the graph of f meets the
line y = x, with coordinates x = y = 1 −
1
b
. If x
t
is the coordinate of that point, then
x
t+1
= f (x
t
) = x
t
, and therefore the system is stationary.
47
48 Chaotic Modelling and Simulation
0 1
0
1
x
t
x
t++1
y
t
(a) Tracing an iteration
0 1
0
1
(b) More iterations
FIGURE 3.1: Tracing iterations
Things get more interesting, and useful, when we consider other curves in place
of y = x. Of particular importance is the inverse function to the logistic map, with
equation:
x = f (y) = by(1 − y)
Suppose we follow steps similar to what we did above, except now we use this
inverse logistic instead of the line y = x. Namely, starting from a value x
t
, locate
y
t
= f (x
t
), and then use the inverse logistic to locate x = f (y
t
). The resulting point
is f ( f (x
t
)); in other words it is x
t+2
. This way, we can perform two iterations.
Suppose now that (x, y) = (x
t
, y
t
) is a point on the intersection of the graphs of
f and its inverse. Then we see that x
t+2
= x
t
and y
t+2
= y
t
, so the system enters a
twopoint orbit. Algebraically, we would have the two equations:
y = f (x) = bx(1 − x)
x = f (y) = by(1 − y)
(3.2)
This system can be solved by observing that:
f ( f (x)) = f (y) = x
or in our case:
x = b
2
x(1 − x)
_
1 − bx(1 − x)
_
(3.3)
Equation (3.3) has four roots, corresponding to the four points where the curves
of the logistic map and the inverse logistic map meet. Two of these will be the points
where these maps meet the line y = x, namely the origin and the stationary point we
discussed already. Let us consider for a moment the geometric signiﬁcance of the
other two points. If (x
t
, y
t
) is one of these points, then (x
t+1
, y
t+1
) will be the other
The Logistic Model 49
one, and (x
t+2
, y
t+2
) will be the same as (x
t
, y
t
) and so on. In other words, the system
will alternate between these two states. We will call these points stationary of second
order. These are the ﬁrst bifurcation points of the logistic map (Figure 3.2).
0 1
0
1
FIGURE 3.2: Second order stationary points
Similarly, higher order bifurcation points can be found by considering the points
of intersection of the iterates
1
of f , i.e. systems of the form:
y = f
k
(x)
x = f
m
(y)
(3.4)
Geometrically, tracing the ﬁrst map followed by the second map results in tran
sitioning from x
t
to x
t+k+m
. Algebraically, we would have to solve a polynomial of
degree 2(k + m).
An illustrative example of the bifurcation of the logistic map is presented in Fig
ure 3.3 when b = 3.5. The starting value is x
0
= 0.1. The ﬁnal equilibrium is at one
of the points where the logistic curve joins the inverse logistic curve.
2
1
The nth iterate of f , denoted by f
k
, is the function obtained by performing the function k times in
succession. For instance, f
2
(x) = f ( f (x)), and in general f
n+1
(x) = f ( f
n
(x)).
2
In this example, the polygonal segments are related to f and its inverse, hence each subsequent x
point is the result of two applications of f .
50 Chaotic Modelling and Simulation
FIGURE 3.3: The logistic and the inverse logistic map for b = 3.5
Figure 3.4 illustrates a third order bifurcation cycle resulting when b = 1 +
√
8.
The three points in the stationary orbit are the points of intersection of the quartic
y = f
2
(x) with the inverse x = f (y).
3
FIGURE 3.4: The third order cycle of the logistic and the inverse logistic map
for b = 1 +
√
8
The value b = 1 +
√
8 ≈ 3.828 is the smallest value of b for which this third order
orbit is present. Higher values of b quickly lead to chaotic behaviour. Figure 3.5 for
instance presents a chaotic region for the logistic and the inverse logistic maps, when
b = 3.9. A large number of repetitions is drawn.
3
Those two curves have 5 points of intersection, but two of these points are the ﬁrst order stationary
points.
The Logistic Model 51
FIGURE 3.5: Chaotic region for the logistic and the inverse logistic map at
b = 3.9
3.1.2 Algebraic analysis of the logistic
We will now proceed to analyse the behaviour of the logistic map from a more
algebraic/analytic point of view. For a given value of b, the graph of the logistic is a
parabola passing through the origin (0, 0) and the point (0, 1). The maximum of the
parabola is obtained when x =
1
2
, and y = f (x) =
b
4
, as can be seen by calculating
the ﬁrst derivative of the logistic function:
f
′
(x) = b(1 − 2x)
and equating it to zero. Keeping in mind that f (x) will be used as the x for the next
iteration step, we see that we can only hope to constrain the system when 0 ≤ b ≤ 4.
The map y = f (x) has two ﬁxed points, (0, 0) and
_
x
0
= 1 −
1
b
, 1 −
1
b
_
. The deriva
tives of f at the critical points are:
f
′
(0) = b and f
′
(x
0
) = 2 − b
We would like to know in which cases these points are attracting.
4
When x is
close to 0, we have
f (x) = bx(1 − x) = bx − bx
2
≈ bx
so if x
t
is close to 0, x
t+1
is about equal to bx
t
. Hence, if b < 1. x
t+1
is even closer to
0, and consequently 0 will be an attracting point. The same is true when b = 1, since
then z
t+1
= x
t
− x
2
t
< x
t
. However, this is not the case any more when b > 1; x
t+1
would tend to move away from 0.
Incidentally, b > 1 is exactly the range of values of b for which the second ﬁxed
point, x
0
= 1−
1
b
, enters the picture, since it is now a point in the square [0, 1] ×[0, 1].
4
A set S of points is called attracting, if starting from any point suﬃciently close to S the map stays
close to S .
52 Chaotic Modelling and Simulation
We will now carry a similar analysis at this point. A Taylor expansion of f at x
0
will
provide us with some insight into the behaviour of the system near the point:
5
f (x) = f (x
0
) + f
′
(x
0
)(x − x
0
) +
f
′′
(x
0
)
2
(x − x
0
)
2
Since we want to determine if the system will move towards x
0
or not, we consider
the diﬀerence from x
0
:
f (x) − x
0
= ( f (x
0
) − x
0
) + f
′
(x
0
)(x − x
0
) +
f
′′
(x
0
)
2
(x − x
0
)
2
(3.5)
We are interested in determining for which values of b, a small x − x
0
will result
in an even smaller f (x) − x
0
. When x − x
0
is small, the ﬁrst term in (3.5), f (x
0
) − x
0
,
will dominate the other two, if it is nonzero. so, in order for x
0
to be attracting, it is
necessary that f (x
0
) − x
0
= 0, i.e that x
0
is a ﬁxed point. In that case, equation (3.5)
becomes
f (x) − x
0
= f
′
(x
0
)(x − x
0
) +
f
′′
(x
0
)
2
(x − x
0
)
2
(3.6)
For x − x
0
close to 0, the second order term will be negligible, and so:
f (x) − x
0
≡ f
′
(x
0
)(x − x
0
)
Therefore, if  f
′
(x
0
) < 1, then x
0
is an attracting point, while if  f
′
(x
0
) > 1, then
x
0
is not attracting. When f
′
(x
0
) = 1, equation (3.5) becomes
f (x) − x
0
= (x − x
0
) +
f
′′
(x
0
)
2
(x − x
0
)
2
and the sign of f
′′
(x
0
) determines whether x
0
is attracting or not.
In our case, the ﬁxed point is x
0
= 1 −
1
b
, and the derivative, as computed earlier,
is:
f
′
(x
0
) = 2 − b
More generally, the Taylor expansion of f around x
0
in this case is:
f (x) − x
0
= (2 − b)(x − x
0
) − b(x − x
0
)
2
Consequently, x
0
is an attracting point for 1 ≤ b < 3. An example of the attraction
process is illustrated in Figure 3.6 (b = 2.9, x
0
= 0.07).
When b > 3, the ﬁxed point is no longer an attracting point, and a twopoint,
second order, attracting orbit appears. In order to obtain this orbit geometrically, the
second order diagram (x
t
, x
t+2
), i.e. the graph of f ( f (x)), is used, in addition to the
(x
t
, x
t+1
) diagram, i.e. the graph of y = f (x), and the (x
t+1
, x
t
) diagram, i.e. the graph
of x = f (y).
With a parameter of b = 2.7, we obtain Figure 3.7(a). It can be seen from the
graph, that the only ﬁxed points of f ( f (x)) are the two ﬁxed points of x. When
5
Note that, since f is a polynomial of degree 2, f actually equals its second order Taylor expansion.
The Logistic Model 53
FIGURE 3.6: The logistic map
b = 3, all the curves pass through the point
_
2
3
,
2
3
_
(Figure 3.7(b)). The slopes of the
tangent lines to the curves at this point are −1 for y = f (x) and x = f (y), and 1 for
y = f ( f (x)).
When b > 3, we can compute the second order orbit by solving the equation
f ( f (x)) = x
that is:
b
2
x(1 − x) (1 − bx(1 − x)) = x (3.7)
Equation (3.7) has four roots: The two ﬁxed points of f (x
1
= 0 and x
2
= 1 −
1
b
),
as well as two new roots, given by:
x
3,4
=
(b + 1) ±
√
(b + 1)(b − 3)
2b
It is clear now why these points appear only when b ≥ 3. This twopoint orbit is
attracting for b > 3, and remains attracting as long as b is such that the slope of the
order2 graph at the attracting points does not exceed −1. This leads to the equation:
b
2
− 2b − 5 = 0
which provides the upper value of b, b = 1 +
√
6 ≈ 3.4495, for which this two
point orbit is attracting. Figures 3.8(a) and 3.8(b) illustrate this order2 bifurcation.
Figure 3.8(a) shows the case where b = 3.43 and x
0
= 0.01. The order2 bifurcation
appears in the form of a square with corners located at the 4 characteristic points
of the diagram. The tangent line to the order2 graph at the larger of these points
appears, with slope very close to −1. One can also clearly see that the ﬁxed point of
y = f (x) is no longer an attracting point.
Another point to consider is the value of x at which the inverse of the logistic curve
cuts the logistic curve at its maximum. This leads to the equation
b
3
− 4b
2
+ 8 = 0
54 Chaotic Modelling and Simulation
(a) b = 2.7 (b) b = 3
FIGURE 3.7: The logistic map, b ≤ 3
(a) b = 3.43 (b) b = 1 +
√
5
FIGURE 3.8: The logistic map, b > 3
The Logistic Model 55
which has three real roots. Two of them are acceptable, b = 2 and b = 1+
√
5. These
are the same points where the order2 curve meets the main diagonal.
6
Figure 3.8(b)
illustrates the case where b = 1 +
√
5, with x
0
= 0.25.
The period4 bifurcation appears in Figure 3.9. For better presentation, and in or
der to locate the characteristic points, we introduce the inverse (x
t+3
, x
t
) and (x
t
, x
t+4
)
diagrams. The 4point orbit can be thought of either as the points of intersection of
the (x
t
, x
t+4
) diagram (purple line) with the line y = x, or as the points of intersection
of the x
t+3
, x
t
diagram (blue line) with the curve y = f (x). The parameter b is con
strained in the range 1 +
√
6 < b < 3.5440903 · · ·. There are 4 characteristic values
for x.
FIGURE 3.9: Order4 bifurcation of the logistic map; real time and phase
space diagrams
The right of Figure 3.9 is the real time diagram (x
t
, t), whereas the left is the phase
space diagram (x
t
, x
t+4
). In the real time diagram the order4 oscillations appear. The
values of x
t
, after enough time has elapsed, oscillate between theses four limiting
levels.
When 3.5440903 · · · < b < 3.5644073 · · · the period8 bifurcation appears. Fig
ure 3.10 illustrates this case. The chaotic parameter is b = 3.5644, and the inverse
function is given by the (x
t+7
, x
t
) diagram. The order8 logistic bifurcation is ex
pressed by the (x
t
, x
t+8
) diagram. A real time diagram of the chaotic oscillations
appears on the right of Figure 3.10.
The period16 bifurcation is illustrated in Figure 3.11. The range covered by the
parameter b is 3.5644073 · · · < b < 3.5687594 · · ·. For the application we set b =
3.5687594 · · ·. Figure 3.12 presents the period32 bifurcation (b = 3.56969). The
range for b is 3.5687594 · · · < b < 3.5696916 · · ·.
6
Since if y = f (x), then asking for f (y) = x is the same as asking for f ( f (x)) = x.
56 Chaotic Modelling and Simulation
FIGURE 3.10: Order8 bifurcation of the logistic map; real time and phase
space diagrams
FIGURE 3.11: Order16 bifurcation and real time and phase space diagrams
of the logistic
FIGURE 3.12: Order32 bifurcation and real time and phase space diagrams
of the logistic
The Logistic Model 57
The order5 bifurcation appears in Figure 3.13. The inverse function is given by
the (x
t+4
, x
t
) diagram. The order5 logistic bifurcation is expressed by the (x
t
, x
t+5
)
diagram. The chaotic oscillations appear in the real time diagram (right).
FIGURE 3.13: The order5 bifurcation and real time and phase space diagrams
The order6 bifurcation appears in Figure 3.14 (b = 3.6267). The inverse function
is here given with the (x
t+5
, x
t
) diagram. The order6 logistic bifurcation is expressed
by the (x
t
, x
t+6
) diagram. The chaotic oscillations appear in the real time diagram
(right).
FIGURE 3.14: The order6 bifurcation and real time and phase space diagrams
58 Chaotic Modelling and Simulation
3.2 The Bifurcation Diagram
The period doubling and other properties of the logistic map are presented in the
socalled bifurcation diagram. This is a twodimensional graph (b, x
t
) formed as
follows: For each value of the chaotic parameter b, the values of x
t
that form the
stable orbit are plotted. Figure 3.15(a) shows how the bifurcation diagram looks for
b < 3.5. For 0 < b < 1, there is only one ﬁxed point, at x = 0. For 1 < b < 3, the
ﬁxed point is x = 1 −
1
b
. At b = 3, this splits up into the ordertwo orbit, which at
3.449 becomes unstable and splits up into an order4 orbit and so on. The greyed out
lines show the values that these orbits would have had if they were still stable.
The full bifurcation diagram for the logistic map appears in Figure 3.15(b). The
ﬁrst bifurcation point is located at
_
b = 3, x = 1 −
1
b
_
, and the next two at b ≈ 3.4495,
the next four at b ≈ 3.54409 and so on, according to the theory. These points can
be calculated by solving the appropriate equation of the form f
n
(x) = x, though this
becomes exceedingly diﬃcult as n increases. The construction of the bifurcation
diagram typically follows these steps:
1. Break the range of b ([0, 4]) into small intervals.
2. For each interval, choose a b in that interval (say the left endpoint), and per
form repeated applications of the iterative process x
n+1
= f (x
n
) until stability
is achieved. At this point, the desired values x
n
will be determined.
3. Move on to the next value of b. Use the previously found stable x values as
starting points for the iteration process, to achieve stability much quicker.
The characteristic values of the parameter b at each bifurcation stage follow the
Feigenbaum law. According to this law the sequence of period doubling, presented
in the above ﬁgure by a bifurcation tree, is due to a quantitative convergence of the
parameter b, which tends to a universal parameter δ = 4.6692016 · · ·. This parameter
is given by the expression
δ = lim
i→∞
b
i−1
− b
i
b
i
− b
i+1
The behaviour of the system can be understood better by considering a scaled
version. More generally, if we consider a system
x
n+1
= f (x
n
)
then we consider the function:
g(x) = lim
n→∞
1
f
2
n
(0)
f
2
n
_
x f
2
n
(0)
_
Then, under some mild conditions on the map f , g satisﬁes the universal equation:
g(αx) = −αg
_
g (−x)
_
(3.8)
The Logistic Model 59
0
1
b
0 1 3 3.449 4
(a) The ﬁrst two bifurcations
(b) The whole diagram
FIGURE 3.15: The bifurcation diagram of the logistic
60 Chaotic Modelling and Simulation
Here α is the second Feigenbaum constant, related to the ratio of the width between
the prongs in a bifurcation, relative to the width in the previous bifurcation.
The function g can be approximated by a ﬁnite polynomial, and from that it is
possible to estimate α = 2.50290787 · · ·, with the normalisation convention g(0) =
1 (Lanford, 1982a).
Other interesting properties arising from the bifurcation diagram come from ob
serving the three “windows” which appear in a range of characteristic values of the
parameter b around b
6
= 3.627, b
5
= 3.74 and b
3
= 1 +
√
8. These values of b give
sixth, ﬁfth and third order bifurcations respectively, and the related period doubling
bifurcations lead to 12, 10 and 6 and higher order bifurcation forms. The period3
bifurcation can be seen in Figure 3.16(a). In Figure 3.16(b) the order 2 × 3 = 6
bifurcation is illustrated (b = 3.845).
(a) Order3 bifurcation (b) Order 2 × 3 bifurcation
FIGURE 3.16: Order3 bifurcations
A change of variables
z =
1
2
b(1 − 2x)
results in the following expression for the logistic equation:
z
t+1
= z
2
t
+ c
where
c =
b
2
−
b
2
4
From this point of view, the ﬁrst period doubling occurs at c = −
3
4
, which corre
sponds to b = 3. The second period doubling occurs at c = −
5
4
, which corresponds
to b = 1 +
√
6, while the period3 solutions appear at a saddlenode bifurcation at
The Logistic Model 61
c = −
7
4
, corresponding to b = 1 +
√
8. An illustration of the period3 bifurcation
using the zform of the logistic equation appears in Figure 3.17.
FIGURE 3.17: The order3 bifurcation
3.3 Other Models with Similar Behaviour
Functions with varying equation forms show chaotic behaviour similar to the lo
gistic. One of them is the sinusoidal map, with equation
x
t+1
= b sin(x
t
)
where b is restricted to the interval 0 < b ≤ π. A more convenient form is:
x
t+1
= b sin(πx
t
)
Now the interval for the parameter is 0 < b ≤ 1.
Another model is the Gompertz type model (Gompertz, 1825):
x
t+1
= −bx
t
ln(x
t
)
where 0 < b ≤ e = 2.71829 · · ·.
Another model was proposed by May (May, 1972, 1974, 1976) to describe the
population growth of a single species population, which is regulated by an epidemic
disease at high density:
x
t+1
= x
t
e
r(1−x
t
)
The chaotic behaviour of this model is similar to that of the logistic model, as
illustrated by the bifurcation diagram (Figure 3.18, right). The order4 bifurcation is
presented on the left of Figure 3.18 (r = 2.6).
62 Chaotic Modelling and Simulation
FIGURE 3.18: The May model: an order4 bifurcation (left) and the
bifurcation diagram (right)
3.4 Models with Diﬀerent Chaotic Behaviour
Some Gaussian models showspecial chaotic behaviour, mainly due to the quadratic
exponent in the equations of these models. A simple Gaussian equation is:
x
t+1
= b + e
−ax
2
t
The bifurcation diagram for a = 4 and −1 < b < 1 is shown on the left of
Figure 3.19(a). The model has two period doubling bifurcations, and then it returns
to the original state in a reverse order. A further increase in the parameter a of the
quadratic term, a = 7.5, gives a full sequence of period doubling bifurcations and
order3 saddlenode bifurcations (Figure 3.19(b), right).
When the parameter a takes higher values, the chaotic behaviour of the model
changes radically. This is presented in Figure 3.20 (a = 17, −1.2 < b < 1).
A rich bifurcation diagram appears when the Gaussian model is combined with
the logistic to form the GL model:
x
t+1
= b − x
t
(1 − x
t
) + e
−ax
2
t
When a = 7.5, the bifurcation diagram in the range −1.1 < b < 0.7 has the
form illustrated in Figure 3.21. There are two main bifurcation regions, with period
doubling bifurcations and order3 and order5 saddlenode bifurcations. The order5
bifurcation is presented in Figure 3.22(a) (a = 7.5, b = 0.19).
Figure 3.22(b) illustrates a perfect order12 circle (a = 7.5, b = 0.11).
The Logistic Model 63
(a) a = 4 (b) a = 7.5
FIGURE 3.19: Bifurcation diagrams of the Gaussian model
FIGURE 3.20: Bifurcation diagram of the Gaussian model (a = 17)
FIGURE 3.21: Bifurcation diagram of the GL model
64 Chaotic Modelling and Simulation
(a) Order5 bifurcation (b) Order12 cycle
FIGURE 3.22: Cycles in the GL model
3.5 The GRM1 Chaotic Model
In this section we study the chaotic behaviour of the generalized rational (GRM1)
innovation diﬀusion model. The deterministic continuous version of this model was
proposed, analysed and applied in Skiadas (1985). Here, the chaotic behaviour is ex
pressed through the discrete alternative to the continuous GRM1 model. The model
shows symmetric and nonsymmetric behaviour expressed by a parameter σ. When
the diﬀusion parameter b and the σ parameter are in the range b/σ ≥ 2, then the
chaotic aspects of the model appear. A method is proposed for ﬁtting the model to
the data. Time series data expressing the cumulative percentage of steel produced by
the oxygen process in various countries are used. Characteristic graphs of the chaotic
behaviour are given and applications are presented.
3.5.1 GRM1 and innovation diﬀusion modelling
It has become commonplace to call this the information age, but an even more
appropriate name might be the innovation age. The number of patent applications
that the U.S. Patent and Trademark Oﬃce receives every year has been increasing
exponentially since 1985, with the total number of patent applications exceeding
450, 000 in 2006 (Figure 3.23).
While not all patents translate to new products or new production methods, this
growth clearly demonstrates a tendency, and this explosion of innovation activity
presents signiﬁcant challenges. Companies must have an appropriate way to de
scribe the competitive dynamics in a market (Modis, 1997). Several innovation dif
fusion models have been presented, analysed and applied to real life data (Bass, 1969;
Mahajan and Schoeman, 1977; Sharif and Kabir, 1976; Skiadas, 1985, 1986, 1987;
The Logistic Model 65
1970 1980 1990 2000
1
0
0
2
0
0
3
0
0
4
0
0
Year
P
a
t
e
n
t
s
(
t
h
o
u
s
a
n
d
s
)
G G
G
G
G
G
G
G G
G
G G G G G G G
G G
G
G
G
G
G
G
G
G
G G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
FIGURE 3.23: Patent applications in the United States
Modis and Debecker, 1992). A main direction of these applications focused on the
nonsymmetric behaviour of the models, expressed by speciﬁc parameters.
The GRM1 model is based on a family of generalised rational models. It was
proposed as a relatively simple but very ﬂexible model to express asymmetry during
the innovation diﬀusion process (Skiadas, 1985, 1986). This model is expressed by
the following diﬀerential equation
˙
f = b
f (F − f )
F − (1 − σ) f
(3.9)
where f is the number of adopters at time t, F is the total number of potential
adopters, b is the diﬀusion parameter, and σ is a dimensionless parameter. The
GRM1 model has a point of inﬂection, varying from 0 to F as σ decreases from∞to
0. Another interesting property of σ is that it provides a measure of the asymmetry
of the model. Perfect symmetry appears for σ = 1 when, equation (3.9) reduces to
the logistic:
˙
f = b f
_
1 −
f
F
_
(3.10)
In the logistic model, bifurcation and further chaotic behaviour appear when 2 <
b ≤ 3. However, various applications of the logistic model in several disciplines
showed that the parameter b of the logistic model lies in very low limits, lower than
1. Thus, by using the logistic model it is not possible to express chaotic behaviour
in real situations, as the estimated values of b fail to reach the limit at which chaotic
behaviour appears. On the other hand, data provided for various cases shows that
66 Chaotic Modelling and Simulation
oscillations and chaotic behaviour appear quite frequently, and especially when the
diﬀusion process is close to the upper limit F. Moreover, when the logistic model is
applied in the form
x
t+1
= bx
t
(1 − x
t
) (3.11)
where x
t
= f
t
/F, then bifurcation and chaotic behaviour appear at even higher values
of b, when 3 < b ≤ 4.
As we show here, the GRM1 model exhibits chaotic behaviour for values of b that
are quite low and are in accordance with the values estimated in real situations. This
is accomplished with the help of the ﬂexible parameter σ, which gives a measure of
the asymmetry of the model. The chaotic behaviour of the model is analysed and
illustrated by using appropriate graphs, especially (t, f ) diagrams.
3.5.2 The generalized rational model
The discrete version of the continuous model (3.9) is given by:
f
t+1
= f
t
+ b
f
t
(F − f
t
)
F − (1 − σ) f
t
(3.12)
Some interesting properties of this model are illustrated in Figures 3.24(a) to 3.24(e).
In Figure 3.24(a), the proposed model takes the classical sigmoid form, whereas
in Figure 3.24(b), a bifurcation appears as a simple oscillation. In Figure 3.24(c),
a more complicated oscillation with four distinct oscillating levels appears, whereas
in Figures 3.24(d), 3.24(e), and 3.24(f) a totally chaotic form appears. In all cases
presented here the starting value is f
0
= 1, the upper limit F = 100, b = 0.3 and σ
takes various values. The value selected for b is within the range 0.1 to 0.5, which is
valid in real situations. By varying the dimensionless parameter σ, several forms of
the model appear.
It is very important to consider the estimation of the values of the parameters b and
σ for which bifurcation appears. The presence of the ﬁrst oscillations and the onset
of chaos, which follows, is a signiﬁcant factor when studying innovation diﬀusion
systems. According to the theory of chaotic models, bifurcation for the GRM1 model
starts when:
f
′
t+1
= −1
f
t+1
= f
t
(3.13)
Using equation (3.12), the resulting condition on the parameters b and σ is:
b
σ
= 2
When
b
σ
> 2, oscillation and chaotic behaviour appear by gradually augmenting the
fraction
b
σ
. When σ = 1, which is the case for the logistic model, bifurcation appears
for b > 2.
It is also possible to obtain an analytical form for the values of f
t
between the ﬁrst
two bifurcation points. To achieve this we consider the equation
f
t+2
= f
t
The Logistic Model 67
(a) b = 0.3 and σ = 2 (b) b = 0.3 and σ = 0.13
(c) b = 0.3 and σ = 0.12 (d) b = 0.3 and σ = 0.10
(e) b = 0.3 and σ = 0.09 (f) b = 0.3 and σ = 0.08
FIGURE 3.24: The GRM1 model
68 Chaotic Modelling and Simulation
The exact formula of the solutions is:
f
t
= F
(b + 2) ±
_
b(b+2)(b−2σ)
(b−2σ+2)
2(b + σ − 1)
(3.14)
For the logistic model (σ = 1) this reduces to:
f
t
= F
(b + 2) ±
√
(b + 2)(b − 2)
2b
(3.15)
When b > 2σ in (3.14), or b > 2 in (3.15), the system oscillates at the values
of f
t
given by the above formulas respectively. When b is higher than these values,
four distinct oscillating levels appear, and later eight and ﬁnally 2
n
points. For suf
ﬁcient speciﬁcally high values of b, n is very high and the system exhibits chaotic
oscillations.
3.5.3 Parameter estimation for the GRM1 model
The parameters of the discrete GRM1 model are estimated by an iterative non
linear regression analysis algorithm by minimizing the sum of squared errors (S =
S S E):
S =
ǫ
2
t
=
n
t=1
(y
t
− f
t
)
2
where ǫ
t
is the error term of the stochastic equation:
y
t
= f
t
+
n
i=1
∂f
t
∂a
i
∆a
i
+ ǫ
t
where y
t
denotes the provided data and f
t
is calculated for every t from the GRM1
equation, given a set of initial values for the parameters a
i
. The estimation of para
meters is highly sensitive to the presence of oscillations and chaotic oscillations in
the provided data. For a better ﬁt, it was decided to use the nonlinear estimation
method proposed by Nash for the discrete logistic model for only three parameters
of the model, and ﬁxing the dimensionless parameter σ. This parameter is gradually
changed during the iterative procedure, until the sum of squared errors is minimised.
The starting values for the partial derivatives need an estimation of the following
forms given a set of initial values for the model parameters:
∂f
1
∂b
=
f
0
(F − f
0
)
F − (1 − σ) f
0
∂f
1
∂f
0
= 1 + b
F
2
− 2F f
0
+ (1 − σ) f
2
0
(F − (1 − σ) f
0
)
2
∂f
1
∂F
= bσ
_
f
0
F
_
2
(3.16)
The Logistic Model 69
FIGURE 3.25: Spain, oxygen steel process (1968–1980)
Following the estimation of the initial values of the partial derivatives, the iterative
procedure continues the estimation by using the formulae:
∂f
t+1
∂b
=
∂f
t
∂b
(1 + bk
t
) +
f
t
(F − f
t
)
F − (1 − σ) f
t
∂f
t+1
∂f
0
=
∂f
t
∂f
0
(1 + bk
t
)
∂f
t+1
∂F
=
∂f
t
∂F
(1 + bk
t
) +
bσf
2
t
(F − (1 − σ) f
t
)
2
(3.17)
where
k
t
=
_
F
2
− 2F f
t
+
(1 − σ) f
t
F − (1 − σ) f
t
_
2
3.5.4 Illustrations
Time series data expressing the cumulative percentage of steel produced by the
oxygen process
7
in various countries are used from Poznanski (1983). Figure 3.25
illustrates the diﬀusion of oxygen steel technology in Spain from 1968 to 1980, for
a period of 13 years. The actual data includes 18 years, but it is more appropriate to
study the last part of the time series, as this part shows the characteristic oscillations
that are of special interest to us. The small circles indicate the actual data, the dotted
line is the path of the logistic model and the solid line is the path of the GRM1 model.
Parameter estimates and the sum of squared errors are summarised in Table 3.1.
The parameter b for the logistic model ﬁt is relatively high, but is still far from the
7
In this process, pure oxygen is introduced in order to burn the carbon inside melted steel, so as to
improve the quality of the produced steel. Before the introduction of this method, air, consisting of a
mixture of oxygen and nitrogen, was used instead.
70 Chaotic Modelling and Simulation
TABLE 3.1: Parameter estimates and sum of
squared errors (SSE) for logistic and GRM1 models in
Spain from 1968 to 1980
Model b l F σ(b/σ) SSE
Logistic 0.6309 24.373 51.474 — 72.838
GRM1 0.2331 25.779 51.736 0.084 (2.775) 41.748
FIGURE 3.26: Italy oxygen steel process (1970–1980)
value needed for the start of bifurcation (b = 2). The form of the logistic path
presented in the Figure 3.25 has a smooth form. The model fails to express the oscil
lating behaviour of the actual case studied. On the other hand, the GRM1 model
has a value for b lower than that of the logistic model, but the extra parameter
σ accounts for the presence of oscillating and chaotic behaviour, as the fraction
b/σ = 2.775 > 2. The estimated values for l and F are very close for both mod
els. The ability of the GRM1 model to follow the oscillating behaviour of actual data
is illustrated in the Figure 3.25, and is also expressed by the strong improvement of
the sum of squared errors (SSE).
Figure 3.26 illustrates the diﬀusion of oxygen steel technology in Italy from 1970
to 1980. The process ends in an oscillating form. The discrete logistic fails to express
these oscillations, whereas the discrete GRM1 shows a considerable ﬂexibility in
approximating the data. The sum of the squared errors is very low for the GRM1
model, compared to that of the logistic, as is shown in Table 3.2. The fraction b/σ =
3.5292 for the GRM1 model accounts for the chaotic behaviour.
Data for the diﬀusion of the oxygen steel process in Luxemburg are of consider
able interest, as they cover a wide scale, from 1.5% during 1962 to 100% in 1980
(Figure 3.27). The GRM1 model showed a good ﬂexibility, as it covers both the fast
growth at the ﬁrst stages of the diﬀusion process, as well as the sudden turn to the
high platform of 100%. Also, the small ﬂuctuations at the end of the process are
simulated quite well, as the fraction b/σ = 3.609 accounts for the chaotic region of
The Logistic Model 71
TABLE 3.2: Parameter estimates and sum of squared
errors (SSE) for logistic and GRM1 models in Italy from
1970 to 1980
Model b l F σ(b/σ) SSE
Logistic 0.5447 35.957 44.473 — 15.330
GRM1 0.08823 36.0402 44.4614 0.025 (3.5292) 7.431
FIGURE 3.27: Luxemburg oxygen steel process (1962–1980)
the model. Figure 3.27 illustrates the case of Luxemburg for the following estimated
values for the parameters: b = 0.1931, l = 7.968, F = 99.669 and σ = 0.0535. The
mean squared error is MS E = 20.872.
The ﬂexibility and ability of the GRM1 model to simulate growth processes that
show oscillations, as well as chaotic oscillations, at the end of the process, are
demonstrated in the following case of the diﬀusion of oxygen steel technology in
Bulgaria from 1968 to 1978 (Figure 3.28). The estimates for the parameters are
b = 0.04046, l = 49.2425, F = 58.412 and σ = 0.012. The sum of squared errors
is S S E = 21.431 and the fraction b/σ = 3.3718 indicates that the model is in the
region of chaotic behaviour.
3.6 Further Discussion
Several problems arise when applying the logistic map to real world data. Firstly,
there are many cases where nonchaotic or even nonoscillating behaviour is present,
even when the sigmoid logistic shape approaches the highest values, namely 1 =
100%. But, the analysis of the logistic map shows that bifurcation, and thus oscillat
72 Chaotic Modelling and Simulation
FIGURE 3.28: Bulgaria oxygen steel process (1968–1978)
ing behaviour, starts at b = 3, which corresponds to x = 1 −1/b, or x = 2/3 (66.66%
level). If we follow the logistic model, all the processes exceeding this level must
show oscillating or chaotic behaviour when approaching a higher level. However,
this is not the case. Many logisticlike timeseries data approach high levels without
showing oscillating or chaotic behaviour.
The second problem when applying the logistic map to timeseries data is that,
when a high value for the parameter b is introduced, the intermediate stages of the
process are far apart to give reasonable results during simulation. As a simple exam
ple, consider a logistic process with b = 3 starting at x = 0.10. The next time period
of the process must be 0.27, followed by 0.59 and 0.725. There are very few growth
processes that can cover the gap between 10% and 72.5% of the total process in
only three time periods. Especially in diﬀusion and innovation diﬀusion processes,
the development is slower, but eventually the process reaches very high values, and,
then, oscillating or even chaotic behaviour appears.
One approach is to use a logistic model with a varying function for the parameter
b. This is the method employed in the GRM1 model, where the term b/(1−(1−σ)x)
is replacing the parameter b. Another approach is to select a transformed logisticlike
model. This is easily achieved by selecting a modiﬁed logistic model of the form:
x
n+1
= bx
n
(1 − x
n
)
e
where 0 < e < 1. For e = 1, this model reduces to the logistic model, whereas, for
e = 0, it reduces to the exponential model. The ﬁxed point is at x = 1−1/b
(1/e)
. This
is the minimum level where the ﬁrst bifurcation starts.
The modiﬁed logistic model (ML), for b = 1.3 and e = 0.09, is illustrated in
Figure 3.29. The model approaches the equilibrium limit 0.9458. An oscillating
form of the modiﬁed logistic at b = 1.3 and e = 0.076 is presented in Figure 3.30 A
chaotic process at b = 1.3 and e = 0.073 for the modiﬁed logistic model is presented
in Figure 3.31
The Logistic Model 73
FIGURE 3.29: The modiﬁed logistic model (ML)
FIGURE 3.30: The modiﬁed logistic model (ML) for b = 1.3 and e = 0.076
FIGURE 3.31: A chaotic stage of the modiﬁed logistic model (ML) for b = 1.3
and e = 0.073
74 Chaotic Modelling and Simulation
FIGURE 3.32: A third order cycle of the GRM1 model for b = 1.3 and σ =
0.025
Another serious problem when applying the logistic model is that the oscillations
following the bifurcation process are quite large for most real life situations. On the
other hand, in the cases of the GRM1 and the modiﬁed logistic, the amplitude of the
oscillations varies according to the values of the parameters selected. A comparative
example is given by observing the third order cycle for the logistic, the GRM1 and
the modiﬁed logistic models (Figure 3.32). The interval of this third order cycle
is included inside the chaotic region as illustrated in the bifurcation diagram. By
observing this third order cycle, valuable information for the chaotic region can be
obtained. During the third order cycle, the logistic model oscillates between 0.1494
and 0.9594, that is, the oscillations cover about 80% of the total amplitude. On the
contrary, the GRM1 model is more ﬂexible. The oscillations during the third order
cycle can vary according to the values selected for the parameter σ. In Figure 3.32,
σ = 0.025. For this value of σ the GRM1 model shows a third order cycle for
b = 1.3. The maximum value is 0.9693, and the minimum value is 0.7047. The
oscillations therefore cover only 27% of the total amplitude. Accordingly, for lower
values of the parameter σ the amplitude of the oscillations during the third order
cycle is smaller.
When the parameters for the GRM1 model are b = 1.2 and σ = 0.01175, the
higher value of the third order cycle is x
max
= 0.9768 and the lower value is x
min
=
0.7846. The third order oscillations cover 19% of the total accepted amplitude of the
process. The (x, t) oscillation diagram is presented in Figure 3.33.
During the ﬁrst time periods the process follows a growth path. Later on, chaotic
oscillations appear and the process is stabilised, giving third order chaotic oscilla
tions.
The third order cycle for the modiﬁed logistic model is illustrated in Figure 3.34.
The parameters selected are e = 0.0719 and b = 1.3. The minimum value is
x
min
= 0.8069 and the maximum value is x
max
= 0.9987. The resulting third or
der oscillations cover 20% of the total amplitude.
The Logistic Model 75
FIGURE 3.33: A third order cycle of the GRM1 model for b = 1.2 and
σ = 0.01175
FIGURE 3.34: A third order cycle of the modiﬁed logistic model for b = 1.3
and e = 0.0719
FIGURE 3.35: The ﬁrst bifurcation cycle of the GRM1 model for b = 1.3 and
σ = 0.032
76 Chaotic Modelling and Simulation
Figure 3.35 illustrates the GRM1 model at the ﬁrst bifurcation stage. The process
ends at a simple oscillation scheme. The parameters are b = 1.3 and σ = 0.032.
The maximum and minimum levels of the oscillation are illustrated by the lines with
two and one dots respectively. These levels are obtained when x
n+2
= x
n
= x. The
resulting fourth order equation can be solved explicitly, giving the values for the ﬁrst
bifurcation levels. The fourth order equation for x has, among others, the roots that
satisfy the relation
x
n+1
= x
n
= x
or, after substituting from the GRM1 map
x =
bx(1 − x)
1 − (1 − σ)x
or after some simpliﬁcation:
x
_
x −
b − 1
b − 1 + σ
_
= 0 (3.18)
The roots of this equation are x
1
= 0 and x
2
=
b−1
b−1+σ
. The resulting equation for
x
n+2
= x
n
= x can then be transformed to
x
_
x −
b − 1
b − 1 + σ
_
×
×
_
b(b + 1 − σ)x
2
− (b + 1 − σ)(b + 1)x + (b + 1)
_
= 0 (3.19)
Clearly, this fourth order equation has the two roots from (3.18) as well as two
other roots:
x
3,4
=
(b + 1) ±
_
(b + 1)
2
−
4b(b+1)
b+1−σ
2b
It is clear that the equation expressing every higher bifurcation level includes the
solutions of the equations expressing the lower bifurcation stages. What is diﬀerent
is the stability of solutions.
The Logistic Model 77
Questions and Exercises
1. Show that the ﬁxed point of the map y = bx(1 − x) is y = x = 1 −
1
b
.
2. Compute the derivative of the transformation y = f ( f (x)), where
f (x) = bx(1 − x)
at the 2point orbit given in equation (3.7), and determine from this the range
of values of b for which this orbit is attracting.
3. Use the logistic map as a random number generator: For various high values
of b, start from an arbitrary value of x, and perform 100, 000 repetitions. Com
pare the distribution of the 100, 000 values to that of the uniform distribution,
as given by the random number generator of your computer. If you found
diﬀerences, how do you explain these diﬀerences?
4. The logistic map is deﬁned on the interval: 0 < x
t
≤ 1. Show that this is
indeed the case, i.e. that, starting from an x in the interval [0, 1], the process
will remain in the interval. Find the related interval for the map x
t+1
= 1+ax
2
t
.
For this latter case, draw the bifurcation diagram and ﬁnd the ﬁrst and the
second bifurcation points.
5. Consider the correct diﬀerence equation analogue to the logistic model, as
described in page 31. Find the interval where the map is deﬁned and ﬁnd the
formula for the ﬁrst and second order bifurcation points.
6. Prove that the Gompertz map is deﬁned into the interval 0 < x
t
≤
1
e
.
7. Transform the Gompertz and the Mirror Gompertz model to the related dis
crete maps, and ﬁnd the interval where every map is deﬁned. Draw the bifur
cation diagram for both cases and ﬁnd the ﬁrst bifurcation point.
8. Draw the bifurcation diagram for the map x
t+1
= b + e
−ax
4
t
.
9. Draw the bifurcation diagram for the map x
t+1
= b + e
−ax
3
t
, and compare the
results with the previous case.
Chapter 4
The Delay Logistic Model
4.1 Introduction
Delay models are applied in a large variety of cases in many scientiﬁc ﬁelds. Con
sequently, it is diﬃcult to classify the particular models used and to study them sys
tematically. In this section we start with the study of simple delay models, and then
proceed to study more complicated models. Special attention is given to the develop
ment and presentation of the most frequently used delay models, as well as to related
extensions and new delay models of particular interest. Of major importance are the
applications of delay models in physiological systems (Glass and Pasternack, 1978;
Glass and Mackey, 1988; Glass, 1988, 1991; Glass and Zeng, 1990; de Olivera and
Malta, 1987; Cabrera and de la Rubia, 1996; Celka, 1997; Sharkovsky, 1994; Ueda
et al., 1994; Bunner, 1999).
Delay models can be divided in two broad categories: The delay models expressed
by a diﬀerence equation
x
t+m
= f (x
t+m−1
, x
t+m−2
, . . . , x
t
) (4.1)
where m is the delay, and the delay models expressed by a diﬀerential equation
˙ x = f (x
t
, x
t−T
) (4.2)
where T is the parameter expressing the delay.
4.2 Delay Diﬀerence Models
4.2.1 Simple delay oscillation scheme
Simple oscillation schemes can arise very easily when using delay diﬀerence equa
tions. As an example, consider the simple ﬁrstdegreepolynomial delay form:
x
t+2
= x
t+1
+ ax
t
(4.3)
When a = −1, and for initial values x
1
= 0.2, x
2
= 0.3, the resulting twodimensional
map has the hexagonal form illustrated at the top left of Figure 4.1. The hexagon
79
80 Chaotic Modelling and Simulation
at the top right is formed by considering random initials values (x
1
, x
2
), uniformly
chosen from the interval [0, 1]. This iterative scheme generates a discrete oscillation
over time (bottom of Figure 4.1).
FIGURE 4.1: A simple delay oscillation scheme (a = −1)
A variation of the hexagon in Figure 4.1 can be obtained by introducing a Eu
clidean distance term in equation (4.1):
x
t+1
= x
t
+ a
x
t−1
√
x
t
+ x
t−1
Figure 4.2 shows the corresponding twodimensional map for a negative value of
a, related to the size of the graph.
FIGURE 4.2: Variation of the simple delay model
The Delay Logistic Model 81
4.2.2 The delay logistic model
The delay logistic model is given by the equation:
x
t+1
= bx
t
(1 − x
t−m
) (4.4)
It has several interesting properties, depending of course on the values of b, but,
mainly, on the values of the delay parameter m. Clearly, for m = 0 equation (4.4)
reduces to the logistic equation. The main diﬀerence when the delay is present is the
formation of a limit cycle (Figure 4.3(b)). There is also a marked time period, during
which the system takes zero values. The resulting (x
t
, x
t−m
) diagram can be seen at
the top of Figure 4.3(b), with the (t, x
t
) diagram at the bottom.
The oscillations of the map (4.4) are of a particular form. Each oscillation is
followed by a linear term with value of zero. The distance between two successive
oscillations depends on the delay parameter: The higher the delay, the higher the dis
tance. The simplest case is illustrated in Figures 4.3(a) and 4.3(b). (The parameters
are m = 1, b = 2.271 and m = 10, b = 1.2865 respectively).
(a) m = 1 (b) m = 10
FIGURE 4.3: The delay logistic model
The special case with delay m = 1 has an interesting property, which can be seen
when examining the left corner (corresponding to the origin) of the (x
t−m
, x
t
) diagram
closer. The graph exhibits a small “fold,” as illustrated in Figure 4.4(a). The m = 5
case is even more interesting, with three folds (Figure 4.4(b), b = 1.44413).
Table 4.1 summarises the values of b used in our simulations. These are the max
imum values for b for which the model makes sense.
82 Chaotic Modelling and Simulation
(a) Delay logistic (m = 1) (b) Delay logistic (m = 5)
FIGURE 4.4: Folds in the delay logistic model
TABLE 4.1: Limiting b values for the delay logistic
m 1 2 3 4 5 · · · 10
b 2.271 1.8395 1.6393 1.52195 1.44413 · · · 1.2865
4.3 Time Delay Diﬀerential Equations
The general form of a time delay diﬀerential equation is given by
˙ x = f (x
t
, x
t−T
) (4.5)
where T > 0 is the time delay. The simplest time delay diﬀerential equation form is
the linear equation
˙ x = −ax
t−1
(4.6)
The transformation T = at turns (4.6) into
˙ y = −y
T−a
where y
T
= x
T/a
.
A very interesting observation here is that the parameter a is directly related to the
delay; The delay increases along with a. Theory suggests that for a =
π
2
the delay
equation has periodic solutions, whereas for a <
π
2
solutions converge to zero, and
for a >
π
2
, solutions diverge to inﬁnity (Hoppensteadt, 1993).
Using a linear spline approximation method, we can perform a simulation of the
above simple time delay equation. The key step in this method is to take the delay
The Delay Logistic Model 83
interval T and divide it into n equal parts. The iterative process is given by the
following scheme:
˙ y
0
= −ay
n
˙ y
1
= −n(y
1
− y
0
)
.
.
.
˙ y
i
= −n(y
i
− y
i−1
)
.
.
.
˙ y
n
= −n(y
n
− y
n−1
)
(4.7)
This system was solved by using a fourth order RungeKutta method of numerical
integration, proposed at the end of the nineteenth century by the German mathemati
cians Carl Runge and Wilhelm Kutta (Runge, 1895; Kutta, 1901). The method is
widely used and, when properly applied, gives good results. The selected time incre
ment h must be suﬃciently small, and the number of intervals n large enough. Sev
eral iterations were tried out for various values of n. When n = 15 and h = 0.001, the
accuracy of the method is good. Figure 4.5 illustrates the twodimensional Poincar´ e
map,
1
and the periodic solutions of the system when a approaches
π
2
. The oscillations
are not stable; small changes of the parameter a from the value
π
2
lead to diverging
or converging oscillations.
FIGURE 4.5: A simple delay model (n = 15)
The behaviour of the simple oscillating scheme simulated above shows that oscil
1
A Poincar´ e map is a twodimensional representation of a multidimensional ﬂow of a nonlinear
system. Geometrically, this map results by cutting the curves of a multidimensional ﬂow by a plane. In
our case, a delay equation with delay m is equivalent to an mdimensional system, and several Poincar´ e
maps may be selected, for instance (x
t
, x
t+m
), (x
t+m−1
, x
t+m
), (x
t
, x
t+1
), (x
t
, x
t+2
) and so on. Each of these
maps leads to diﬀerent graphs, showing us diﬀerent aspects of the ﬂow.
84 Chaotic Modelling and Simulation
lations appear very naturally in a system with memory. This is because the model ex
pressing the behaviour of the system integrates into the diﬀerentialdiﬀerence equa
tions the past as one or more delay variables. As a consequence, delay means oscil
lations.
Experience conﬁrms the appearance of oscillations in very many systems where
delays are inherent. Delays in human decisions on social, economic, technolog
ical or political systems may lead to oscillations. Short or longterm economic
cycles, social or political changes, innovation or technological cycles may be ex
pected when time delays are present. Biological cycles, biological clocks and other
oscillating mechanisms of humans and other living organisms are some of the nu
merous cases where delays are present in nature. Chemical oscillations and the
BelousovZabontiski reactions modelled and studied extensively by Prigogine and
his coworkers (Prigogine and Lefever, 1968; Prigogine et al., 1969; Nicolis and
Prigogine, 1981; Prigogine, 1995, 1996, 1997) are also examples of the eﬀect of de
lays in the formation of the intermediate substrates until the formation of the ﬁnal
product.
4.4 A More Complicated Delay Model
Using the same method presented above, we simulate a more complicated delay
model, which has a more stable oscillating behaviour. The general delay equation
for this model is
˙ x = −ax
t−1
_
1 − x
b
t
_
(4.8)
The oscillations of (4.8) are quite stable when a >
π
2
, and they exhibit a very inter
esting behaviour for some values of a. The parameter b is an integer, and two distinct
cases are considered, depending on the parity of b. When b = 1, the model (4.8) re
duces to a continuous delay logistic model.
Of special importance is the case where b = 2. When b = 2 and a = 14, sim
ulation results in the very important form shown in Figure 4.6. The resulting two
dimensional map is an almost perfect square, irrespective of the starting point, as
long as this point is inside the square, i.e. if the starting values are in [0, 1]. This
rectangular form shows for any a >
π
2
, with the sharpness of the corners depending
on the value of a. In the case studied in Figure 4.6, the linear spline approximation
method was used with n = 30, h = 0.001 and initial value x = 0.2, and it gave
very good results. The rectangular oscillations appear to be quite perfect and may be
useful in several applications.
The oscillations are quite diﬀerent when b is odd. The behaviour in this case is
similar to that of the discrete delay logistic model (Figure 4.7, b = 1, a = 3.7,
n = 15). The delay equation in this case is:
˙ x = −ax
t−1
(1 − x
t
) (4.9)
The Delay Logistic Model 85
FIGURE 4.6: Delay model (b = 2); a square pattern
FIGURE 4.7: Delay logistic model
86 Chaotic Modelling and Simulation
4.5 A Delay Diﬀerential Logistic Analogue
The appropriate diﬀerential equation analogue to the delay diﬀerence equation (4.9)
is:
˙ x
m
= bx(1 − x
m
) (4.10)
In order to perform a simulation of (4.10), the method of linear spline approxima
tion presented in page 83 will be used. The intermediate step is s and the iteration
step is d. The delay term is included only in the ﬁrst equation. It is the last term to
the right of the ﬁrst equation. By changing this delay term, or replacing it by another,
many interesting cases arise.
x
1
= x
1
+ bx
1
(1 − x
m
)d
x
2
= x
2
−
m
s
(x
2
− x
1
)d
.
.
.
x
m
= x
m
−
m
s
(x
m
− x
m−1
)d
(4.11)
Figure 4.8(a) illustrates the (x
1
, x
m+1
) and (t, x
1
) diagrams for the case where b =
1.22, d = 0.01, s = 3 and m = 14. There is a clear similarity with the previous case
of the diﬀerence equation above.
The simulation producing Figure 4.8(b) is based on another delay equation:
˙ x
m
= bx
m
(1 − x
2
) (4.12)
The parameters in Figure 4.8(b) are m = 15, s = 6, d = 0.001 and b = 2.6. As
can be seen from the ﬁgure, the (x
1
, x
m+1
) graph is an almost perfect square. Close
inspection, however, will show that the corners are not very sharp. This is due to the
choice of the parameters and of the integration step. Nevertheless, equation (4.12) is
a continuous equation very appropriate for expressing a square. The oscillation turns
out to exhibit a step form.
In both cases presented above, b was selected close to its upper limit. However,
one can have several intermediate cases of particular interest.
4.6 Other Delay Logistic Models
Another form of a delay logistic model is expressed by the following diﬀerence
equation:
x
t+1
= ax
t
+ bx
t−m
(1 − x
t−m
) (4.13)
The Delay Logistic Model 87
(a) Delay Logistic (m = 14) (b) An almost perfect square
FIGURE 4.8: Oscillations and limit cycles in delay logistic models
This is a delay model that integrates the inﬂuence of a logistic like interaction to
the future. As a simple example, consider the impact of a “wordofmouth” com
munication between x adopters and (1 − x) potential adopters of an innovation or an
idea, communicated m time units in the past, and inﬂuencing people at the present
time. This happens very frequently in marketing, economic, social or political sys
tems. The impact of this timedelayed inﬂuence to the evolution of x in the future
will be studied by varying the width of the delay interval m and the values of the
parameters a and b. The ax
t
term represents the current adopters who continue, and
the bx
t−m
(1 − x
t−m
) term represents the new adopters.
In this model, limit cycles, bifurcation and chaotic behaviour appear. A large
variety of oscillations and chaotic oscillations appear, depending on the level of the
delay. But, even in the simplest cases of one or two units of delay, the behaviour of
the system is quite complicated. High delay levels ensure chaotic behaviour.
The simplest form of (4.13) is when m = 1. The more interesting forms appear in
Figure 4.9(a). The six realisations, A through F, illustrate the behaviour of the system
when b is 1.706, 1.717, 1.751, 1.758, 1.77 and 1.795 respectively. Throughout, a was
set to 0.9 and x
0
was set to 0.1. As b increases, limit cycles are followed by chaotic
behaviour.
Figure 4.9(b) examines the case where m = 2. Again, limit cycles, bifurcation
and then chaos appear as b increases. ﬁgures A through F correspond to b values of
1.2, 1.22, 1.241, 1.252, 1.259 and 1.269 respectively (a = 0.9 and x
0
= 0.1).
The m = 3 time delay has an interesting eﬀect on the twodimensional graph (Fig
ure 4.9(c)). The bifurcation is more complicated, and leads to a chaotic picture with
a period doubling. The parameter a and the initial value x
0
are as in the preceding
example, whereas b is 0.96, 0.965, 0.99, 0.991, 0.994 and 1.016 respectively.
When m = 5, a = 0.9 and x
0
= 0.95, the timedelay eﬀect is more apparent. (Fig
ure 4.9(d)). The values of b are 0.74, 0.743, 0.747, 0.754, 0.758 and 0.762 respec
88 Chaotic Modelling and Simulation
(a) m = 1 (b) m = 2
(c) m = 3 (d) m = 5
FIGURE 4.9: Chaotic attractors of a delay logistic model
The Delay Logistic Model 89
tively. The form of the graphs is similar to the m = 3 case. Limit cycles, bifurcation
and chaotic behaviour are again present.
4.7 Model Behaviour for Large Delays
When the time delay m is large or extremely large, the form of the graphs tends
to be stable for changes of the delay parameter and keeps the form of the chaotic
attractors. The model is expressed in a high dimensional space, and the form of
the attractors is inﬂuenced by the delay mechanisms, not merely by the form of
the delay equation. This can be seen by comparing the logistic delay model to the
Glass model (Mackey and Glass, 1977). The two models look completely diﬀerent in
terms of their analytic equations, but the chaotic attractors and the resulting chaotic
oscillations are very similar.
Figure 4.10 illustrates the twodimensional graphs for the logistic delay model
when m = 40, a = 0.9 and x
0
= 0.95. The parameter b here takes the values
0.36, 0.37, 0.38, 0.3813, 0.3818 and 0.41 respectively.
Graph A represents a classical limit cycle, whereas graph B corresponds to a ﬁrst
order bifurcation, and graphs C and D show second and third order bifurcations
respectively. Higher order bifurcations appear in graph E, and chaotic behaviour is
present in graph F.
FIGURE 4.10: Chaotic attractors of the logistic delay model (m = 40)
The chaotic behaviour presented in graph F of Figure 4.10 shows great similarities
to that arising from a discrete delay Glass model of the form
x
t+1
= ax
t
+ b
x
t−m
1 + x
10
t−m
(4.14)
90 Chaotic Modelling and Simulation
The Glass model (graph A) and the logistic model (graph B) are compared in
Figure 4.11. The delay parameter is m = 40 and a = 0.9 for both models, and b is
set to 0.18 for the Glass model and 0.4 for the logistic model. The similarities are
striking, both in the twodimensional attractors as well as in the chaotic oscillations,
presented in the lower part of the ﬁgure.
FIGURE 4.11: Comparing the Glass (A) and the logistic (B) delay models
(m = 40)
Various general models were also tested, with similar results. Such general models
are of the form
x
t+1
= ax
t
+ F(x
t−m
)
Examples of speciﬁc models that exhibited behaviour similar to that of the Glass
and the delay logistic models in the chaotic region are:
LG2 x
t+1
= ax
t
+ bx
t−m
(1 − x
2
t−m
)
LG3 x
t+1
= ax
t
+ bx
t−m
(1 − x
3
t−m
)
GRM1 x
t+1
= ax
t
+ b
x
t−m
(1 − x
t−m
)
1 − (1 − c)x
t−m
Gauss x
t+1
= ax
t
+ be
−c(x
t−m
−k)
2
Gompertz x
t+1
= ax
t
− bx
t−m
ln(x
t−m
)
(4.15)
We refer to the ﬁve models as delay LG2, delay LG3, delay GRM1, delay Gauss
and delay Gompertz respectively. Chaotic behaviour is expressed for the following
values of the parameters:
In all these cases, the attractors have the same shape as that illustrated in Fig
ure 4.11.
A diﬀerent chaotic behaviour is exhibited by the following delay exponential
model:
x
t+1
= x
t
+ bx
t−m
e
−cx
t−m
(4.16)
The Delay Logistic Model 91
TABLE 4.2: Parameter values on which various
delay models exhibit chaotic behaviour
Model LG2 LG3 GRM1 Gauss Gompertz
a 0.9 0.9 0.95 0.9 0.9
b (c, k) 0.3 0.243 0.1 (0.84) 0.17 (4, 0.7) 0.28
Model (4.16) gives a relatively stable limit cycle for various values of the parameters.
Figure 4.12 illustrates this limit cycle and the related oscillations when the para
meters of the model take the values a = 0.9, b = 3.2 and c = 0.8. A shape like that
of a pigeon drawn by a modern painter appears.
FIGURE 4.12: The delay exponential model
4.8 Another Delay Logistic Model
A somewhat diﬀerent delay logistic model is given by the following diﬀerence
equation:
x
t+1
= bx
t−m
+ ax
t
(1 − x
t
) (4.17)
When m = 1, this system reduces to a system equivalent to that of the H´ enon
model:
x
t+1
= bx
t−1
+ ax
t
(1 − x
t
) (4.18)
The analysis of system (4.18) when b = −1 shows that it has a stable ﬁxed point
92 Chaotic Modelling and Simulation
when
x = 1 −
2
a
which turns into the period2 orbit with:
x =
a + 2 ±
√
(a − 6)(a + 2)
2a
The H´ enon model is usually presented in another form, as a twodimensional
model expressed by a system of diﬀerence equations:
x
t+1
= by
t
+ 1 − ax
2
t
y
t+1
= x
t
(4.19)
However, note that the second equation of (4.19) implies that y
t
= x
t−1
. We can
therefore substitute this into the ﬁrst equation, and then the following delay equation
results:
x
t+1
= bx
t−1
+ 1 − ax
2
t
(4.20)
Equation (4.20) is equivalent to (4.18), as can be easily observed in Figure 4.13(b).
The small attractor produces the large attractor by rescaling and translation. The
large attractor is the (x
t
, x
t+1
) image produced by equation (4.20) when a = 1.4 and
b = 0.3. The small attractor is the one produced by equation (4.18) when a = 3.17
and b = 0.3.
(a) Delay cosineH´ enon (b) Delay logisticH´ enon
FIGURE 4.13: Delay logistic and delay cosine H´ enon variants
In Figure 4.13(a), the two attractors look as if they were produced by reﬂection
and translation of one to the other. The attractor on the top right of Figure 4.13(a)
is the H´ enon attractor provided by the system (4.19). The cross near this attractor is
The Delay Logistic Model 93
located at the origin (0, 0), whereas the cross close to the attractor on the left of the
same ﬁgure is located at (−π, −π). This attractor results from the system
x
t+1
= by
t
+ a cos(x
t
)
y
t+1
= x
t
(4.21)
and its corresponding delay equation
x
t+1
= bx
t−1
+ a cos(x
t
) (4.22)
This system looks very diﬀerent from (4.19), algebraically, yet the attractors are very
similar. The attractor in Figure 4.13(a) is formed when b = 0.3 and a = 3.22.
Equation (4.22) retains some of the properties of the H´ enon map, but it also con
tains other interesting properties, presented in Figure 4.14 (a = −0.4, b = −1). The
map (4.21) is an areapreserving map, as its Jacobian is J = b = −1. The inﬂuence
of the cosine function in the delay equation is clear: The various forms are repeated
many times by following 45
◦
and 135
◦
degrees of symmetry.
FIGURE 4.14: Chaotic patterns of the delay cosine model
Figure 4.15(a) shows the bifurcation diagram (a, x) for 0 < a <
π
2
+ 0.1 and
b = −1. The parameter value a =
π
2
gives rise to an interesting carpetlike form
(Figure 4.15(b)). An inﬁnite number of islands are included inside the chaotic sea.
Another carpetlike attractor appears in Figure 4.16(a). Here b = −1 and a =
π
2
− 0.5. As a is less than the characteristic value
π
2
, the islands cover more space
inside the chaotic sea. The chaotic sea almost disappears in Figure 4.16(b). The
islands have now become very ordered geometric structures, and cover almost the
entire plane. Here a =
π
2
− 1.5.
There is a systematic way of drawing these graphs. The center of the squarelike
forms is located at distances proportional to π, whereas the outer dimension of these
squares is also equal to π. The squares are also related to each other in terms of their
creation over time. With the exception of the central square, all the other squares are
generated in groups of four.
94 Chaotic Modelling and Simulation
(a) The bifurcation diagram (a, x) (b) A carpetlike map
FIGURE 4.15: Carpetlike forms
(a) A carpetlike pattern (b) Groups of squareforms
FIGURE 4.16: Carpetlike forms
The Delay Logistic Model 95
Figure 4.17(a) illustrates the same map as in Figure 4.16(b), but in the latter case
only the centers and the corresponding squares are present. When the control para
meter is very small, a = 0.01, almost perfect squares appear. This is illustrated in
Figure 4.17(b). The chaotic sea almost disappears and the nonsymmetric islands
become geometric objects. The original central square is drawn (starting with an
initial value a bit less than
π
2
), along with the ﬁrst group of four surrounding squares
(starting with an initial value a bit more than
π
2
).
(a) Squarelike forms (b) The generating squares
FIGURE 4.17: Squarelike forms
The case of b = 1 is illustrated in Figure 4.18(a). The control parameter here is
a =
π
3
. The building blocks for the construction of the above carpetlike form are two
interrelated objects close to the origin (marked by a cross in Figure 4.18(b), where a
was set to 0.001). The two forms have the shape of almost perfect squares, with their
centers located at (−
π
2
,
π
2
) and at (
π
2
, −
π
2
).
The bifurcation diagram (b, x) appears in Figure 4.19(a). Period doubling and to
tally chaotic regions are clearly present. At the end of the diagram the chaotic region
has a stochastic character. The oscillations seem to cover the entire plane. Interesting
windows and special forms appear when a takes the special values
π
3
,
π
2
,
2π
3
, π and
3π
2
.
The value
3π
2
is the lower limit for the control parameter a. Beyond it, the chaotic
sea covers the entire plane, with no islands present. Figure 4.19(b) shows the case
where a =
π
2
. On the outer part of each pattern we can see 6 small islands.
When a =
2π
3
the ﬁxed point bifurcates into a period2 orbit. This is illustrated in
Figure 4.19(c). The six islands from the preceding case are now separated from the
original form, and are inside the chaotic sea. On the boundary one can now see 4
islands. The system is at the starting point of the bifurcation. In Figure 4.19(d) the
bifurcation has been completed, and two distinct chaotic objects appear. The control
parameter at this point is a = π.
When the control parameter is increased, a triangular formappears (Figure 4.20(a),
96 Chaotic Modelling and Simulation
(a) Irregular patterns (b) The generating squares
FIGURE 4.18: Irregular patterns
(a) Bifurcation diagram (b, x) (b) Islands in the chaotic sea, a =
π
2
(c) Islands in the chaotic sea, a =
2π
3
(d) The order2 bifurcation
FIGURE 4.19: Islands in the chaotic sea
The Delay Logistic Model 97
a =
6π
5
−0.008). When a gets closer to
6π
5
, the triangular form splits into three objects
around the central one (Figure 4.20(b), a =
6π
5
− 0.02).
(a) A triangular island (b) An order3 bifurcation
FIGURE 4.20: Triangular islands
The H´ enon model was introduced here in order to compare it to the other delay
models. However, the H´ enon model is a very interesting model, deserving its own
study. A more detailed analysis of the H´ enon model will be the subject of the next
chapter.
98 Chaotic Modelling and Simulation
Questions and Exercises
1. Determine the sides of the large hexagon appearing on the right of Figure 4.1.
2. Draw the (x
t−m
, x
t+1
) and (t, x
t+1
) graphs for the Gompertz delay model and for
various positive values of the parameters a, b and m.
3. Draw the graph of the same model for m = 40, a = 0.9 and b = 0.28, and
compare the resulting ﬁgure with that of the Glass and the logistic models.
4. Replace the cos(x
t
) term into the delay cosine map (Equation (4.22)) by Taylor
approximation to the cosine
cos(x
t
) = 1 −
x
2
t
2
and draw the bifurcation diagram for b = −1 and compare the resulting ﬁgure
with that provided for the delay cosine model. Find the characteristic points
and compare them to those of the delay cosine.
5. Repeat the analysis and graphs of the previous problem, but now for the delay
map
x
t+1
= −x
t−1
+ a
_
1 −
x
2
t
c
_
where c is a positive parameter. Find similarities and diﬀerences with the
previous case.
6. Find the Jacobian of the map in the previous problem, and generalise your
result to a map of the form
x
t+1
= −x
t−1
+ af (x
t
)
where f (x
t
) is a continuous smooth function.
7. Can you draw carpet like forms by using the map resulting from the approxi
mation of the cosine? Justify your answer.
8. What happens to the number of “folds” in Figure 4.4(b) as the delay parameter
m increases?
9. Examine what happens when the parameter b in (4.8) is allowed to take non
integer values.
Chapter 5
The H´enon Model
The H´ enon model is a twodimensional model introduced and analysed by H´ enon
(1976). This model can also be expressed by a delay diﬀerence equation:
x
t+1
= bx
t−1
+ 1 − ax
2
t
(5.1)
By introducing a new variable, y
t
= x
t+1
, (5.1) is transformed into a system of two
ordinary diﬀerence equations:
x
t+1
= y
t
+ 1 − ax
2
t
y
t+1
= bx
t
(5.2)
The Jacobian determinant of (5.2) is given by:
det J =
¸
¸
¸
¸
¸
¸
¸
∂x
t+1
∂x
t
∂x
t+1
∂y
t
∂y
t+1
∂x
t
∂y
t+1
∂y
t
¸
¸
¸
¸
¸
¸
¸
= −b
When b = 1, the resulting map is area preserving, but the system is unstable. The
system becomes stable for 0 < b < 1. A variety of maps appear when a and b vary
over a range of values, and the resulting maps take the form of chaotic attractors.
We will see a variety of these attractors in this chapter. The most known chaotic
attractor of the H´ enon model appears when a = 1.4 and b = 0.3, and is illustrated in
Figure 5.1.
5.1 Global Period Doubling Bifurcations in the H´enon
Map
Global period doubling bifurcations in the H´ enon map have been studied by many
researchers, and a very detailed analysis can be found in Murakami et al. (2002).
The analysis of the H´ enon map is based on (5.1) and (5.2). The ﬁxed points (x
t+1
=
x
t
, y
t+1
= y
t
) of the map are the solutions of the secondorder polynomial:
ax
2
− (b − 1)x − 1 = 0
namely
x
t
=
(b − 1) ±
_
(b − 1)
2
+ 4a
2a
, y
t
= x
t
99
100 Chaotic Modelling and Simulation
FIGURE 5.1: The H´ enon map
The period2 orbit can be found from the relations x
t+2
= x
t
and y
t+2
= y
t
. The
resulting fourthorder equation for x has the two extra solutions
x
t
=
−(b − 1) ±
_
−3(1 − b)
2
+ 4a
2a
, y
t
= x
t
5.1.1 Period doubling bifurcations when b = −1
When b = −1, the map has the stable ﬁxed point
x
t
=
−1 +
√
1 + a
a
, y
t
= x
t
for −1 < a < 3. This turns into the period2 orbit
x
1
=
1 +
√
a − 3
a
, y
1
=
1 −
√
a − 3
a
x
2
=
1 −
√
a − 3
a
, y
2
=
1 +
√
a − 3
a
(5.3)
at the value a = 3. This orbit remains stable for 3 < a < 4. The second period
doubling takes place at a = 4, giving rise to the period4 orbit with
x
1
=
1
√
a
, y
1
=
1
√
a
_
1 −
2
√
a
_
1/2
x
2
= −y
1
, y
2
= x
1
x
3
= x
1
, y
3
= −y
1
x
4
= y
1
, y
4
= x
1
(5.4)
The third period doubling bifurcation takes place when a = 4.12045, giving rise
to a period8 orbit. This last value of a is the root of:
4a(2
√
a − a) + 1 = 0
The H´ enon Model 101
5.1.2 Period doubling bifurcations when b = 1
When b = 1, the ﬁxed points of (5.2) are at
x = ±
1
√
a
, y = x (5.5)
These are both unstable when a > 0. The period2 orbit consists of the points
x = ±
1
√
a
, y = −x
This orbit becomes unstable when a = 1, and we have a bifurcation into a period4
orbit. The period4 orbit undergoes a period doubling bifurcation at a =
8−
√
12
4
=
1.13397 and leads to a period8 orbit. The period8 orbit bifurcates into the period
16 orbit at a = 1.15135. We leave the details to the reader.
5.2 The CosineH´enon Model
During the simulation process related to the global period doubling bifurcation,
diﬃculties arise on the estimation of the starting parameters, as well as of the local
or global instabilities related to the period doubling. It is, therefore, more convenient
to use an analogue of the H´ enon system, based on trigonometric functions:
x
t+1
= by
t
+ 2a (cos (x
t
) − 1) + 1
y
t+1
= x
t
(5.6)
This model is called the cosine H´ enon model. The relation between the two models
is that the H´ enon model (5.2) is the model obtained from (5.6) when the cosine term
is replaced by its secondorder Taylor approximation:
cos(x
t
) = 1 −
x
2
t
2
Therefore, this new model retains many of the properties of the H´ enon model,
but it also has other distinct properties. This model accepts a wider range of initial
values in the (x, y) plane. Furthermore, the values of the control parameter related
to the period doubling bifurcations of the H´ enon map are close to the corresponding
values of the control parameter for the cosine model (b = −1). The ﬁxed point at
the critical value a = 3 is illustrated in Figure 5.2(a). Also present in this case are 7
islands on the boundary. Figure 5.2(b) illustrates the period2 orbit at a = 3.23. The
system is in subcritical state, just before the completion of the next global period
doubling. We can see 11 islands on the boundary.
Figure 5.2(c) illustrates the period4 orbit of the cosine map with b = −1 and
a = 4.11.
102 Chaotic Modelling and Simulation
(a) The ﬁxed point (a = 3) (b) The period2 orbit
(c) The period4 orbit
FIGURE 5.2: Orbits in the cosine H´ enon model
The H´ enon Model 103
5.3 An Example of Bifurcation and Period Doubling
Figures 5.3(a) and 5.3(b) illustrate some interesting cases regarding the bifurcation
and period doubling when the parameter b of the cosine H´ enon model takes the
value b = 1. The system will undergo period doubling when a = 1. Figures 5.3(a),
5.3(b), 5.4(a) and 5.4(b) show the map when the control parameter is a = 0.8, 0.98,
1.02 and 1.3 respectively. In Figure 5.3(a), a double structure appears. After the
period doubling, we see that, in each of the original structures, the central island has
turned into three separate forms. In Figure 5.4(b), the period4 islands are completely
separated.
(a) a = 0.8 (b) a = 0.98
FIGURE 5.3: Bifurcation forms in the cosine H´ enon model
5.4 A Diﬀerential Equation Analogue
A diﬀerential equation analogue to the H´ enon model is given by the following
system of two coupled diﬀerential equations
˙ x = by + 1 − ax
2
− x
˙ y = x − y
(5.7)
104 Chaotic Modelling and Simulation
(a) a = 1.02 (b) a = 1.3
FIGURE 5.4: Higherorder bifurcations in the cosine H´ enon model
This system has ﬁxed points at
x = y =
b − 1 ±
_
(b − 1)
2
+ 4a
2a
When b = −1, these ﬁxed points become:
x = y =
−1 ±
√
1 + a
a
which is precisely the same as the ﬁxed points for (5.2).
Similarly, when b = 1, the provided value is x = y = ±
1
√
a
which is also the same
value as that provided from the diﬀerence equation case.
5.5 Variants of the H´enon Delay Diﬀerence Equation
A number of variants of the H´ enon model give rise to some interesting attractors.
We discuss a couple of these variants, involving either higherorder delays (sections
5.5.1, 5.5.2), or simply changes in the way the ﬁrstorder terms are introduced (sec
tions 5.5.3, 5.5.4).
5.5.1 The thirdorder delay model
An extension to higher delay dimensions is given by
x
t+1
= bx
t−3
+ ax
t
(1 − x
t
) (5.8)
The H´ enon Model 105
Similar to (5.2), the diﬀerence delay equation (5.8) can be turned into a system of
four ordinary diﬀerence equations. The number of time delays in the delay diﬀerence
equation is equal to the number of new equations of the ordinary diﬀerence equation
system. The model (5.8) exhibits three timedelay periods, and is characterised by a
very interesting attractor when the parameters are a = 3.2 and b = 0.2 (Figure 5.5).
FIGURE 5.5: A thirdorder delay model
5.5.2 Secondorder delay models
The following variant, with a secondorder delay term, is extremely simple, having
only one parameter b:
x
t+1
= −bx
t
+ x
t
x
t−2
(5.9)
The attractors of the model are however very interesting. Figure 5.6 for instance
shows the attractor when b = 1.37.
Another interesting variant with a secondorder delay is the following:
x
t+1
= x
t
+ a(x
t−2
− x
3
t−2
) (5.10)
When a = 0.52, the resulting map and the related oscillations (t, x) are illustrated in
Figure 5.7.
A more complicated model is given by:
x
t+1
= bx
2
t−1
− ax
t
+ x
t
x
t−2
(5.11)
This is a twoparameter delay model with both ﬁrst and secondorder delay terms.
There is a characteristic limit cycle obtained when the parameters have values a =
1.2 and b = 0.75. The repeated oscillations and the limit cycle are presented in
Figure 5.8.
106 Chaotic Modelling and Simulation
FIGURE 5.6: A secondorder delay model
FIGURE 5.7: The order2 delay model
FIGURE 5.8: A complicated delay model
The H´ enon Model 107
5.5.3 Firstorder delay variants
Interesting ﬁrstorder delay variants of the H´ enon model arise when one of the two
terms on the righthand side of (5.1) is altered.
For example, if we replace the quadratic term with a delay logistic term, the re
sulting model has the form:
x
t+1
= bx
t−1
+ ax
t
(1 − x
t−1
) (5.12)
Figure 5.9 illustrates the twodimensional map, and the related oscillations, when
a = 2.35 and b = 0.2.
FIGURE 5.9: A H´ enon model with a logistic delay term
Another variant arises by altering the ﬁrst term:
x
t+1
= bx
t
x
t−1
+ 1 − ax
2
t
(5.13)
For a = 1.72 and b = 0.6, the resulting map and related oscillations are illustrated in
Figure 5.10.
As in the case of the original H´ enon model, it is more convenient, for further
analysis, to transform (5.13) into a system of two diﬀerence equations:
x
t+1
= x
t
y
t
+ 1 − ax
2
t
y
t+1
= bx
t
(5.14)
The Jacobian determinant of (5.14) is
det J = −bx
t
.
This indicates that the model (5.14) has properties diﬀerent from those of the H´ enon
model.
108 Chaotic Modelling and Simulation
FIGURE 5.10: A variation of the H´ enon model
5.5.4 Exponential variants
The introduction of an exponential term into (4.20) gives rise to a variant with a
very complex attractor (Figure 5.11, a = 0.8, b = 0.36), which we will call the Henia
attractor. The equation is
x
t+1
= bx
t−1
+ 1 − ax
2
t
e
−ax
t−1
(5.15)
FIGURE 5.11: An exponential variant of the H´ enon model
The H´ enon Model 109
5.6 Variants of the H´enon System Equations
Other interesting variants arise from altering the system (5.2). An example is the
following:
x
t+1
= bx
t
+ cy
t
− 2x
t
y
t
y
t+1
= a(y
t
− x
t
)
(5.16)
For appropriate values of the parameters, interesting stoneshaped attractors appear
(Figure 5.12 (a = 0.8, b = 0.48, c = 1.1)).
FIGURE 5.12: A 2diﬀerence equations system
Another set of extensions involves the introduction of a third variable:
x
t+1
= bx
t
− y
t
− x
t
z
t
y
t+1
= −cz
t
+ x
t
z
t
z
t+1
= a(z
t
− x
t
)
(5.17)
Depending on the values of the three parameters, this model exhibits limit cycles,
bifurcation and ﬁnally chaotic behaviour. Figure 5.13 illustrates these cases. The
parameters a and c are ﬁxed at 0.95 and 0.4 respectively, while b takes the values
0.62, 0.628, 0.6337, 0.6348, 0.6355 and 0.637 respectively. The graphs represent the
(x, y) map.
110 Chaotic Modelling and Simulation
FIGURE 5.13: A threedimensional model
5.7 The Holmes and Sine Delay Models
5.7.1 The Holmes model
The Holmes model can be thought of as an extension of the H´ enon model (Holmes,
1979a). The general form of the Holmes model is:
x
t+1
= bx
t−1
+ ax
t
− x
3
t
The Holmes model, when b = −0.2 and a = 2.765, provides a characteristic sigmoid
map and quite complicated chaotic oscillations (Figure 5.14). When b > 0, the
FIGURE 5.14: The Holmes model (a = 2.765, b = −0.2)
Holmes model gives a rich sigmoid form (Figure 5.15, b = 0.3 and a = 2.4).
The H´ enon Model 111
FIGURE 5.15: The Holmes model (a = 2.4, b = 0.3)
5.7.2 The sine delay model
The model proposed by Holmes is an approximation
1
of a sinusoidal one, called
the sine delay model, of the form:
x
t+1
= bx
t−1
+ a sin(x
t
) (5.18)
This model, for relatively small values of a, shows similar behaviour to that of the
Holmes model. However, when a takes large values, the map changes into a series
of sigmoid forms. The number of maxima or minima of the resulting curves can be
estimated according to the number of solutions of the equation
(1 − b)x − a sin(x) = 0.
In the example presented in Figure 5.16, the delay parameter is ﬁxed at b = 0.3 in
both cases, and the parameter a is set to a = 3.0772 for the case A and a = 2×3.0772
for the case B. It is clear, by observing the resulting maps, that by doubling the value
of a, the number of maxima or minima is also doubled.
When b is small, there is a close relation between the maxima and minima of this
map, and the maxima and minima of the function
y = f (x) = (1 − b)x − a sin(x).
The relation can be seen in Figure 5.17(a). Note further that the range of x values for
the map ends exactly at the last solution to f (x) (the solutions to f (x) = 0 are exactly
the ﬁxed points of the sine delay map). Figure 5.17(b) shows the case where b = 0.3
and a = 6.5(1 − b)π.
When b takes values closer to 1, say b = 0.88, the situation changes radically, as
the attractor is now separated in two very similar parts (Figure 5.18(a)). There is now
1
The approximation can be seen by replacing sin(x) with its third degree Taylor approximation, x −
1
6
x
3
.
112 Chaotic Modelling and Simulation
FIGURE 5.16: The sine delay map (b = 0.3)
(a) Solutions to the ﬁxed point equation for
the sine delay model
(b) Maxima and minima of the sine de
lay model and the corresponding ﬁxed point
equation
FIGURE 5.17: The sine delay model
The H´ enon Model 113
little relation with the maxima and minima of the ﬁxed point equation. The values
a = 3 and b = 0.8 produce the very interesting attractor of Figure 5.18(b).
(a) b = 0.88 (b) a = 3, b = 0.8
FIGURE 5.18: The sine delay model for large b
Things change quite a bit when b = 1. In this case, the map takes the form
x
t+1
= x
t−1
+ a sin(x
t
).
The resulting simulation is shown in Figure 5.19(a) for a = 1.2. The process follows
a random sequence that gradually occupies the entire (x
t−1
, x
t
) plane. The basic
forms of the two generating shapes can be seen more clearly in Figure 5.19(b), where
a = 0.8.
Both shapes are identical but they are a ninetydegrees rotation of each other. Both
shapes become closer to a square with rounded corners as the parameter a takes very
small positive values (Figure 5.20,a = 0.1).
114 Chaotic Modelling and Simulation
(a) a = 1.2 (b) Close look at the two shapes, a = 0.8
FIGURE 5.19: The sine delay model for b = 1
FIGURE 5.20: The sine delay model (a = 0.1)
The H´ enon Model 115
Questions and Exercises
1. Verify that the ﬁxed points and perioddoubling bifurcations of the H´ enon map
for b = 1 and b = −1 are as described in section 5.1.
2. Examine the behaviour of the H´ enon map for b = 3.
3. For each of the maps (5.9), (5.10) and (5.11):
(a) Determine the ﬁxed points and secondorder ﬁxed orbits, and their sta
bility.
(b) Rewrite them as systems of 3 diﬀerence equations without delay, and
calculate the Jacobian determinants of these systems.
4. Determine the ﬁxed points of the sine delay model given by equation (5.18)
when b = 1, and their stability. Use this information to explain the carpetlike
forms arising when b = 1 (Figure 5.19(a)).
5. Continuing from the previous question, run simulations with diﬀerent para
meter values, and examine whether the carpetlike formations are present, and
how they spread across the plane.
6. Explain why for small b there is a relationship between the maxima and min
ima of the sine delay model (x
t
, x
t+1
) map and the maxima and minima of the
equation for the ﬁxed points of the map.
7. Estimate the number of maxima and minima of the ﬁxed point equation for the
sine delay model:
f (x) = (1 − b)x − a sin(x)
Hint: Find an equation for those maxima and minima from f
′
(x) = 0, and
estimate the n beyond which f (c) > 0 at the minima.
Chapter 6
ThreeDimensional and
HigherDimensional Models
In this chapter, we present a number of higherdimensional models that exhibit
chaotic behaviour. Before we delve into particular examples, though, a general dis
cussion of the behaviour of such systems near their equilibrium points is necessary.
6.1 Equilibrium Points and Characteristic Matrices
Let us consider a general autonomous system of the form
˙ x = ˙ x(x, y, z)
˙ y = ˙ y(x, y, z)
˙ x = ˙ z(x, y, z)
(6.1)
and suppose x
0
= (x
0
, y
0
, z
0
) is an equilibrium point for it, i.e. a point where
˙ x
0
= ( ˙ x
0
, ˙ y
0
, ˙ z
0
) = (0, 0, 0).
Consider the matrix:
A =
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
∂˙ x
∂x
∂˙ x
∂y
∂˙ x
∂z
∂˙ y
∂x
∂˙ y
∂y
∂˙ y
∂z
∂˙ z
∂x
∂˙ z
∂y
∂˙ z
∂z
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
Then, at a point x = x
0
+h suﬃciently close to x
0
, we can assume, using a ﬁrst order
Taylor approximation, that ˙ x is given by:
˙ x = ˙ x
0
+ A · h = A · h
Suppose now that the matrix A has a real eigenvalue, say λ
1
, with eigenvector
h
1
. Then, the orbit of a particle placed at x
0
+ δh
1
will move away from x
0
, in the
direction of the vector h
1
, since its position after a short time t would be:
x(t) ≈ (x
0
+ δh
1
) + ˙ x · t = x
0
+ (δ + λ
1
t)h
1
.
117
118 Chaotic Modelling and Simulation
So any equilibrium point with a positive eigenvalue will be unstable. To determine
these eigenvalues, we need to compute the roots of the characteristic polynomial
A − λI =
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
∂˙ x
∂x
− λ
∂˙ x
∂y
∂˙ x
∂z
∂˙ y
∂x
∂˙ y
∂y
− λ
∂˙ y
∂z
∂˙ z
∂x
∂˙ z
∂y
∂˙ z
∂z
− λ
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
The matrix A − λI is called the characteristic matrix of the system.
6.2 The LotkaVolterra Model
As an illustration of the above theory in a simpler case, let us consider a classical
twodimensional system, the LotkaVolterra model, also known as the predatorprey
model. In its simpliﬁed form, the LotkaVolterra model has the form:
˙ x = x − xy
˙ y = −y + xy
(6.2)
In the terminology of Chapter 7, this is a conservative system with ﬁrst integral of
motion:
f (x, y) = xye
−x−y
It has two equilibrium points, at (x, y) = (0, 0) and at (x, y) = (1, 1). The characteris
tic polynomial is given by
¸
¸
¸
¸
¸
¸
−y + 1 − λ −x
y x − 1 − λ
¸
¸
¸
¸
¸
¸
= λ
2
− (x − y)λ + (x + y − 1)
At (x, y) = (0, 0), the characteristic equation is
λ
2
− 1 = 0
and the system has two real eigenvalues λ
1,2
= ±1. The eigenvectors correspond to
the two directions where there are no predators, and where there are no prey. Since
one of these eigenvalues is positive, the system is unstable. At (x, y) = (1, 1), the
characteristic equation is
λ
2
+ 1 = 0
and the system has two imaginary eigenvalues, λ
1,2
= ±i. The system will cycle
around this point.
Using linearization, we see that near (0, 0), the system can be replaced by the
simpler system
˙ x = x
˙ y = −y
ThreeDimensional and HigherDimensional Models 119
whose orbits are the hyperbolas xy = c. Near (1, 1), if we use coordinates (ξ, η) =
(x − 1, y − 1) centred at 1, 1, the system can be replaced by the simpler system
˙
ξ = −η
˙ η = ξ
whose orbits are counterclockwise circles ξ
2
+ η
2
= (x − 1)
2
+ (y − 1)
2
= c.
6.3 The Arneodo Model
The simplest threedimensional chaotic models are based on a system of three
linear diﬀerential equations, on which we add a nonlinear term. An example of
such a system is the Arneodo model:
˙ x = y
˙ y = z
˙ z = −
_
z + sy − mx + x
2
_
(6.3)
where s and m are parameters.
This model was proposed by Arneodo et al. (1979, 1980, 1981a), in their work on
systems that exhibit multiple periodicity and chaos. The Jacobian determinant of the
model is
J = −m + 2x.
There are two stationary (equilibrium) points, the origin (0, 0, 0), and (m, 0, 0).
In Figure 6.1, the (x, y) diagram of the system is shown (s = 3.8 and m = 7.5, the
initial values are (x
0
= 0.5 and y
0
= z
0
= 1). An integration method was used with
integration step d = 0.001.
A slight variation, replacing the last equation in (6.3) with
˙ z = −
_
z + sy − mx + x
3
_
leads to a model with a more stable behaviour. Figure 6.2 shows the projections of
the threedimensional solution path of this system to the three planes (x, y), (x, z) and
(y, z) respectively. The parameters are as above, except that m = 5. The bottom part
of this ﬁgure shows the chaotic oscillations in the (y, t) space.
The resulting paths of this modiﬁed system have a somewhat more complicated
form than those of (6.3). There are three stationary points:
( 0, 0, 0)
(
√
m, 0, 0)
(−
√
m, 0, 0)
(6.4)
The Jacobian of the modiﬁed system is J = −m + 3x
2
.
120 Chaotic Modelling and Simulation
FIGURE 6.1: The Arneodo model
FIGURE 6.2: A modiﬁed Arneodo model
ThreeDimensional and HigherDimensional Models 121
6.4 An Autocatalytic Attractor
Autocatalytic reactions often give rise to chaotic phenomena. The representation
of chemical reaction models by mathematical nonlinear models is possible and has
been studied extensively (Prigogine and Lefever, 1968; Prigogine et al., 1969; Nico
lis, 1971; Nicolis and Portnow, 1973; Nicolis and Prigogine, 1977, 1981; Nicolis
et al., 1983). It is more diﬃcult to ﬁnd a good representation of the numerous in
termediate stages of a chemical reaction. The number of these stages inﬂuences the
complexity of the system, and, consequently, gives rise to a process with the chaotic
behaviour. Nonlinear modelling theory suggests that at least one nonlinear part
must be present in each reaction cycle in order for chaotic phenomena to appear.
A threedimensional chaotic model is perhaps the simplest case, representing an
autocatalytic reaction with three stages. Such a model, for appropriate parameter val
ues, may demonstrate chaotic behaviour, giving rise to a threedimensional chaotic
attractor. A typical such model is given by the following system:
˙ x = k + λz − xy
2
− x
˙ y =
xy
2
+ x − y
e
˙ z = y − z
(6.5)
When the parameters of system (6.5) take the values k = 0.7567, λ = 0.3 and e =
0.013, Figure 6.3 arises. On the right side we see the twodimensional (x, y) phase
portrait, while on the left side we see the temporal variation of chaotic oscillations in
the y direction (i.e. the (t, y) diagram).
FIGURE 6.3: An autocatalytic attractor
122 Chaotic Modelling and Simulation
A threedimensional view (x, z, y) of the phase portrait of the autocatalytic reac
tion is presented in Figure 6.4(a). The view from the (y, z, x) coordinate system is
presented in Figure 6.4(b). The parameters in both cases are k = 2.5, m = 0.017 and
e = 0.013, and the integration step is d = 0.0001. In this case the system (6.5) has
been modiﬁed slightly, with the ﬁrst equation replaced by:
˙ x =
_
1
1 + k
+ m
_
(k + z) − xy
2
− x
(a) In the (x, z, y) coordinate system (b) In the (y, z, x) coordinate system
FIGURE 6.4: Threedimensional views of the autocatalytic model
6.5 A FourDimensional Autocatalytic Attractor
The following fourdimensional model expresses a sequence of four chemical au
tocatalytic reactions:
˙ x
1
= x
1
− x
1
x
2
− 0.24x
2
1
+ 200x
4
+ a
˙ x
2
= x
1
x
2
− x
2
− 10x
2
x
3
˙ x
3
= 0.01 + 10x
2
x
3
− 20x
3
˙ x
4
= 0.24x
2
1
− 100x
4
(6.6)
The nonlinear character of (6.6) is due to the second order term in x
1
, which
appears in the ﬁrst and fourth equations, and to the couplings x
1
x
2
in the ﬁrst and the
ThreeDimensional and HigherDimensional Models 123
second equations and x
2
x
3
in the second and third equations. The control parameter,
a, has value −0.12, and the integration step is d = 0.003.
Figure 6.5(a) illustrates, from top to bottom, the (x
1
, t), (x
2
, t), (x
3
, t) and (x
4
, t) di
agrams. The chaotic behaviour is similar, yet more complicated than in the previous
cases, whereas the time oscillations are of a purely chaotic nature, with several peaks
and intermediate chaotic stages. This is a very good example of chemical oscillations
and the related chaotic waves. Figure 6.5(b) shows the threedimensional (x
1
, x
2
, x
3
)
diagram representing the chaotic attractor of (6.6).
(a) Temporal chaotic oscillations (b) The (x
1
, x
2
, x
3
) diagram
FIGURE 6.5: A fourdimensional autocatalytic model
6.6 The R¨ossler Model
One of the simplest threedimensional models is the R¨ ossler model (see R¨ ossler,
1976d):
˙ x = −y − z
˙ y = x − ez
˙ z = f + xz − mz
(6.7)
The model contains a single nonlinear term, xz, which enters in the third equation,
and three parameters, e, f and m.
Figure 6.6(a) is a threedimensional view of the R¨ ossler model when the para
meters are: e = 0.2, f = 0.2 and m = 5.7. Figure 6.6(b) shows, from left to right,
124 Chaotic Modelling and Simulation
the (x, y), (x, z) and (y, z) views, along with the temporal variation in the x, y and z
coordinates respectively.
(a) A threedimensional view of the R¨ ossler
attractor
(b) Chaotic oscillations and twodimensional
views
FIGURE 6.6: The R¨ ossler attractor
6.6.1 A variant of the R¨ossler model
An interesting variant of the R¨ ossler model is given by the following system of
diﬀerential equations:
˙ x = −y − z
˙ y = x + ay
˙ z = bx + xz − cz
(6.8)
where a, b and c are the parameters of the model.
In all the cases that follow, the parameters b and c are speciﬁed in terms of a as
follows:
b = a
c = 1 +
1
a
(6.9)
In Figure 6.7(a), a is set to 0.2, and the ﬁrst order limit cycle of system (6.8)
appears.
The order two limit cycle is presented in Figure 6.7(b), where a = 0.26. Fig
ure 6.7(c) illustrates the ﬁrst order cycle when a = 0.35, whereas in Figure 6.7(d) we
see the second order cycle (a = 0.42).
When a = 0.5, total chaos ensues. This is illustrated in Figure 6.8(a). This chaotic
form is usually referred to as spiral chaos, after the characteristic spiral paths of the
resulting chaotic image.
ThreeDimensional and HigherDimensional Models 125
(a) First order cycle, (a = 0.2) (b) Second order cycle, a = 0.26
(c) First order cycle, a = 0.35 (d) Second order cycle, a = 0.42
FIGURE 6.7: Limit cycles in a variant of the R¨ ossler model
126 Chaotic Modelling and Simulation
A more complicated case of spiral chaos appears when a = 0.6. Now the trajec
tories pass close to the origin. A threedimensional view of the total spiral chaos is
illustrated in Figure 6.8(b). Figure 6.9 shows, from left to right, the (x, y), (y, z) and
(x, z) views for the same value of a.
(a) a = 0.50 (b) Total spiral chaos, a = 0.60
FIGURE 6.8: Spiral chaos in the variant of the R¨ ossler model
The characteristic matrix of the R¨ ossler model is
A − λI =
_
¸
¸
¸
¸
¸
¸
¸
¸
_
−λ −1 −1
1 a − λ 0
b + z 0 x − c − λ
_
¸
¸
¸
¸
¸
¸
¸
¸
_
The corresponding characteristic polynomial is
A − λI =
¸
¸
¸
¸
¸
¸
¸
¸
−λ −1 −1
1 a − λ 0
b + z 0 x − c − λ
¸
¸
¸
¸
¸
¸
¸
¸
= −λ
3
+ (x + a − c)λ
2
− (ax − ac + z + b + 1)λ + (x + az + ab − c)
This polynomial determines the stability of the system at its equilibrium points. In
our case, the equilibrium points are the solutions of the system:
y + z = 0
x + ay = 0
bx − cz + xz = 0
(6.10)
The ﬁrst equilibrium point is at the origin, M
1
= (0, 0, 0). The most interesting
paths are in the neighbourhood of this point. The other equilibrium point is located
ThreeDimensional and HigherDimensional Models 127
FIGURE 6.9: Plane views of the total spiral chaos in the variant of the R¨ ossler
model (a = 0.60)
at:
M
2
= (x, y, z) =
_
c − ab, −
c − ab
a
,
c − ab
a
_
(6.11)
The characteristic equation at the origin is
λ
3
+ (c − a)λ
2
+ (1 + b − ac)λ + (c − ab) = 0
The characteristic equation at M
2
is
λ
3
− a(1 − b)λ
2
+
_
1 − a
2
b +
c
a
_
λ − (c − ab) = 0 (6.12)
If a, b, c are such that c − ab > 0, then equation (6.12) has at least one positive root.
This in turn makes M
2
into an unstable point.
The variant of the R¨ ossler model is illustrated in Figure 6.10. The (x, y) graph is
presented, where the parameters are a = 0.4, b = 0.3 and c = 4.8. In the same ﬁgure,
the characteristic polynomial is illustrated. It is a thirddegree polynomial with only
one real and two complex roots (eigenvalues).
6.6.2 Introducing rotation into the R¨ossler model
A simple way to introduce rotation into the R¨ ossler model is to insert a simple
rotation in two of the equations of the model. The resulting equations have the form:
˙ x = −z sin(h) − y cos(h)
˙ y = x cos(h) + ey sin(h)
˙ z = f + xz − mz
(6.13)
The rotation angle h is a function of the distance r from the origin. For the follow
ing simulations, we used
h = 0.27
_
x
2
+ y
2
.
128 Chaotic Modelling and Simulation
FIGURE 6.10: View of a simple cycle of the variant of the R¨ ossler model
(a = 0.4, b = 0.3 and c = 4.8)
(a) (b)
FIGURE 6.11: Modiﬁed R¨ ossler models
ThreeDimensional and HigherDimensional Models 129
The inﬂuence of the rotation appears in Figure 6.11(a). There is a complicated
movement, ﬁrst in the (x, y) plane and then in the positive z direction and back, and
ﬁnally the trajectories continue in the (x, y) plane. The parameters in this case were
e = 0.12, f = 0.4 and m = 5.7. A fourth order RungeKutta procedure was used,
with an integration step d = 0.005.
A diﬀerent set of parameters for the model, e = 0.2, f = 0.2 and m = 4.6, gives
the paths presented in Figure 6.11(b). The movement in the z direction now tends to
higher limits, whereas the paths around the origin follow concentric elliptical orbits.
A better threedimensional illustration appears in Figure 6.12, where we increased
f to 0.3.
FIGURE 6.12: Another view of the modiﬁed R¨ ossler model
6.7 The Lorenz Model
In 1963, Edward Lorenz (Lorenz, 1963) proposed an idealized model of a ﬂuid
like air or water; the warm ﬂuid below rises, while cool ﬂuid above sinks, setting
up a clockwise or counterclockwise current. The Prandtl number σ, the Rayleigh
number r and the Reynolds number b are parameters of the system. The width of
the ﬂow rolls is proportional to b. x is proportional to the circulatory ﬂuid ﬂow
velocity. If x > 0 the ﬂuid rotates clockwise. y is proportional to the temperature
diﬀerence between ascending and descending ﬂuid elements, and z is proportional to
the distortion of the vertical temperature proﬁle from its equilibrium. The equations
130 Chaotic Modelling and Simulation
of this threedimensional model are:
˙ x = −σx + σy
˙ y = −xz + rx − y
˙ z = xy − bz
(6.14)
The origin, (0, 0, 0), which corresponds to a ﬂuid at rest, is an equilibrium point
for all r. For 0 < r < 1, it is stable, while, for r ≥ 1, it is unstable. Two other
equilibrium points exist when r ≥ 1:
c
+
=
_ _
b(r − 1),
_
b(r − 1), r − 1
_
c
−
=
_
−
_
b(r − 1), −
_
b(r − 1), r − 1
_ (6.15)
These represent convective circulation (clockwise and counterclockwise ﬂow). At
the equilibrium points c
±
, the Lorenz model has two purely imaginary eigenvalues:
λ = ±i
_
2σ(σ + 1)
σ − b − 1
when
r = r
h
=
σ(σ + b + 3)
σ − b − 1
.
Using the standard settings
σ = 10
b =
8
3
(6.16)
the Hopf bifurcation point
1
is
r
h
= 24
14
19
≈ 24.73684
For r
h
< r, all three equilibrium points of the Lorenz model are unstable. Lorenz
found, numerically, that the system behaves “chaotically” whenever the Rayleigh
number r exceeds the critical value r
h
: All solutions are sensitive to the initial condi
tions, and almost all of them are neither periodic solutions nor convergent to periodic
solutions or equilibrium points.
A threedimensional view of the Lorenz attractor appears in Figure 6.13(a). Two
small circles correspond to the unstable equilibrium points c
±
. The parameters take
the standard values (6.16). The vertical axis is z and the horizontal axis is y.
In Figure 6.13(b), a twodimensional (x, z) view of the Lorenz attractor is pre
sented. This is the most common illustration of this famous attractor. Perhaps the
famous “butterﬂy eﬀect” associated with the presence of chaos owes its name to this
ﬁgure.
1
See Strogatz (1994); Kuznetsov (2004).
ThreeDimensional and HigherDimensional Models 131
(a) A threedimensional view (b) The butterﬂy attractor
FIGURE 6.13: The Lorenz attractor
6.7.1 The modiﬁed Lorenz model
A modiﬁed Lorenz model is given by the following system of equations:
˙ x = −σx + σy
˙ y = rx − xz
˙ z = xy − bz
(6.17)
The only way (6.17) diﬀers from the original Lorenz model is that the term −y is
excluded from the second equation. Thus, the resulting model is simpler.
The equilibrium points are the solutions of the following system of equations:
−σx + σy = 0
rx − xz = 0
xy − bz = 0
(6.18)
The characteristic matrix of the modiﬁed Lorenz model is
A − λI =
_
¸
¸
¸
¸
¸
¸
¸
¸
_
−σ − λ σ 0
r − z −λ −x
y x −b − λ
_
¸
¸
¸
¸
¸
¸
¸
¸
_
Two equilibrium points appear, located at:
M
+
=
_ √
br,
√
br, r
_
M
−
=
_
−
√
br, −
√
br, r
_ (6.19)
The third equilibrium point, M = (x, y, z) = (0, 0, 0), is easily seen to be unstable.
132 Chaotic Modelling and Simulation
FIGURE 6.14: The modiﬁed Lorenz model — left to right: (x, y), (y, z) and
(x, z) views
The chaotic paths of the modiﬁed Lorenz model, for s = 10, b =
8
3
and r = 27,
are shown in Figure 6.14. Three twodimensional chaotic graphs appear; from left to
right, they are the (x, y), (y, z) and (x, z) views of the attractor.
The image on the left of Figure 6.15(a) illustrates the (x, y) view of the modiﬁed
Lorenz attractor. The image on the right presents the Poincar´ e (x, y) diagram of the
same attractor resulting when the plane Z = r cuts the chaotic paths. This graph
gives an illustrative example of the bifurcation.
(a) The modiﬁed Lorenz model: left, (x, y)
view; right, cut of the chaotic attractor by the
plane Z = r
(b) A 3D (y, z, x) view
FIGURE 6.15: The modiﬁed Lorenz chaotic attractor
Figure 6.15(b) illustrates a threedimensional (y, z, x) view of the modiﬁed Lorenz
chaotic attractor. The two stationary points are visible. There appear to be many
similarities with the Lorenz chaotic paths presented earlier in this chapter.
ThreeDimensional and HigherDimensional Models 133
Questions and Exercises
1. Show that the linearisation scheme described in section 6.1 is not applicable
to the system:
˙ x = y
˙ y = −x
3
Show, furthermore, that the system is stable at the origin.
2. The system
˙ x = −y
˙ y = x
is stable at the origin, while the system
˙ x = y
˙ y = x
is not. Explain why this is so.
3. Linearise the system
˙ x = µx − y − x
_
x
2
+ y
2
_
˙ y = x + µy − y
_
x
2
+ y
2
_
at (0, 0), and determine the stability of the system at the origin.
4. Consider the system:
˙ x = y
˙ y = −x
2
(a) Characterise the stability of the system at the origin.
(b) Solve the system and draw various level curves.
5. Consider the system:
˙ x = −2xy
˙ y = 2xy − y
(a) Solve the system, and check the stability at the ﬁxed point.
(b) Draw the level curves, and discuss the form of these curves for various
values of the integration parameter.
6. (a) Find the ﬁxed points and stability of the system:
˙ x = y
˙ y = −sin(x)
134 Chaotic Modelling and Simulation
(b) Compare the behaviour of the system to that of the system:
˙ x = y
˙ y = −sin(x)
(c) In both cases, replace sin(x) with its ﬁrstorder Taylor approximation,
and examine the resulting systems.
(d) What is the diﬀerences when we use the thirdorder Taylor approxima
tion for sin(x)?
7. Let b > 0. Find the ﬁxed points of the system
˙ x = y
˙ y = −bx(x − 1)
and determine the stability of the system at the points.
8. Compute the eigenvalues of the Lorenz model.
9. Consider the system:
˙ x = a − (b + 1)x + x
2
y
˙ y = bx − x
2
y
(a) Find the equilibrium points of the system.
(b) Determine the system’s stability at these points.
(c) Analyse the special cases when b = 1 and b = −1.
10. Show that when c − ab > 0, equation (6.12) has at least one positive root.
11. Determine if the equilibrium points M
+
, M
−
for the Lorenz model, given by
equation (6.19), are stable.
12. Explain the butterﬂy eﬀect.
Chapter 7
NonChaotic Systems
In this chapter we discuss various aspects of nonchaotic systems, that will be helpful
in analysing other, more complex, systems. In section 7.1 we deﬁne conservative and
Hamiltonian systems, while in section 7.2 we give a brief account of linear systems
of diﬀerential equations, and the qualitative behaviour of their solutions near the
equilibrium point. These are models for the local behaviour of more complicated
systems near their equilibrium points.
In the remaining sections, we look at a number of interesting conservative systems.
7.1 Conservative Systems
In this section we consider twodimensional conservative systems, that is, systems
that have a nontrivial ﬁrst integral of motion. For a system
˙ x = F(x, y, t)
˙ y = G(x, y, t)
(7.1)
A ﬁrst integral of motion is a function f (x, y), of x and y, that is independent of time.
As a consequence:
d
dt
f (x, y) = ˙ x
∂f
∂x
+ ˙ y
∂f
∂y
= 0 (7.2)
There is a direct connection between the level curves of the ﬁrst integral and the
trajectories of the system: If f is a ﬁrst integral of the system (7.1), then f is constant
on every trajectory of the system. Thus, every trajectory is part of some level curve
of f , determined by the value of f at the initial conditions (x
0
, y
0
). Hence, each level
curve of f is a union of trajectories.
The ﬁrst integral derives its name from the usual method of computing it, by direct
integration of the diﬀerential equation:
dy
dx
=
G
F
(7.3)
This follows from (7.2), as follows. We consider a particular level curve of f ,
f (x, y) = c, and consequently we think of f as an implicit equation for y as a function
135
136 Chaotic Modelling and Simulation
of x. In that case, we have
∂f
∂x
+
∂f
∂y
dy
dx
= 0
which gives us:
d
dt
f (x, y) = ˙ x
∂f
∂x
+ ˙ y
∂f
∂y
= −˙ x
dy
dx
∂f
∂y
+ ˙ y
∂f
∂y
=
∂f
∂y
_
−˙ x
dy
dx
+ ˙ y
_
(7.4)
Equating the last term to zero, and using equation (7.1), we end up with equa
tion (7.3). The value of the integration parameter is directly related to the level
of the curve.
Since F, G in general depend on t, this integral may not be independent of t. How
ever, in the case where F, G are independent of the time t, integration of (7.3) will
result in a ﬁrst integral of motion for the system. We will examine a number of
particular such systems in this chapter.
Before we proceed, let us elaborate a bit more on the equations. As we saw
in (7.2), the ﬁrst integral of motion, f (x, y), and consequently the implicit equations
for the trajectories of the system, are determined by the equation:
F
∂f
∂x
+ G
∂f
∂y
= 0 (7.5)
Note that this equation will remain true if F, G are both multiplied by the same
function of x, y. To see this geometrically, notice that the trajectories are determined
completely by the direction of their tangent vector at any point, and this direction is
that of the vector ( ˙ x, ˙ y) = (F, G). Therefore, multiplying both F, G with the same
factor has no eﬀect on this direction.
This means that there is a family of conservative systems sharing the same ﬁrst in
tegral of motion. From all these systems, there is one that stands out. Note that (7.5)
can be rewritten as:
F
∂f
∂y
= −
G
∂f
∂x
Rescaling F, G amounts to selecting a value for this ratio, so a natural choice would
be to set it to 1, in which case F, G, and hence the solutions of the system, are
completely determined from f by:
F =
∂f
∂y
and G = −
∂f
∂x
(7.6)
In this case, we will call f the Hamiltonian of the system, and we will call the
system a Hamiltonian system. All the systems we will consider in this section are
Hamiltonian.
Equations (7.6) provide a necessary condition for the Hamiltonian to exist, since
they imply that:
∂F
∂x
= −
∂G
∂y
=
∂
2
H
∂x∂y
NonChaotic Systems 137
In the case where F, G are independent of t, this is also suﬃcient, and the Hamilto
nian can be computed by simply solving the system (7.6).
Hamiltonian systems play a central role in Hamiltonian Mechanics. The equations
of motion of Hamiltonian systems can be completely described using their Hamilto
nian, from equations (7.5) and (7.6).
7.1.1 The simplest conservative system
The simplest conservative system is given by
˙ x = y
˙ y = −x
(7.7)
The corresponding diﬀerential equation for the integral of motion is
dy
dx
= −
x
y
(7.8)
The solutions to (7.8) are concentric circles
x
2
+ y
2
= h
where h is a positive constant (Figure 7.1). Consequently, the system is conservative.
Its ﬁrst integral of motion is:
H(x, y) =
1
2
_
x
2
+ y
2
_
The level curves are therefore concentric circles, whose radius is determined by the
initial conditions (x
0
, y
0
).
FIGURE 7.1: The simplest conservative system
138 Chaotic Modelling and Simulation
Note that this is indeed a Hamiltonian system, as:
∂H
∂y
= y = ˙ x and −
∂H
∂x
= −x = ˙ y
We could have come to the same conclusions by solving the system:
∂H
∂x
= ˙ x = y
∂H
∂y
= −˙ y = x
Integrating the ﬁrst equation would yield H(x, y) =
1
2
y
2
+ c(x), and substituting into
the second equation would yield c(x) =
1
2
x
2
+ c.
7.1.2 Equilibrium points in Hamiltonian systems
Before we proceed to more complicated examples, it is worthwhile discussing
the behaviour of a system in a neighbourhood of an equilibrium point. Equilibrium
points are found by setting ˙ x and ˙ y to 0. In the case of a Hamiltonian system, this
corresponds to
∂H
∂x
=
∂H
∂y
= 0
So the equilibrium points of a Hamiltonian system are precisely the critical points of
the Hamiltonian H. For instance, in the system 7.7 the origin is the only equilibrium
point.
If (x
0
, y
0
) is an equilibrium point, then the Taylor expansion of H around (x
0
, y
0
)
will be:
H(x, y) = H(x
0
, y
0
) +
1
2
h
xx
(x − x
0
)
2
+h
xy
(x − x
0
)(y −y
0
) +
1
2
h
yy
(y −y
0
)
2
+· · · (7.9)
where
h
xx
=
∂
2
H
∂x
2
_
x
0
, y
0
_
, h
xy
=
∂
2
H
∂x∂y
_
x
0
, y
0
_
, h
yy
=
∂
2
H
∂y
2
_
x
0
, y
0
_
are the secondorder partial derivatives of H at (x
0
, y
0
), and the terms ignored are
of higher order in x − x
0
, y − y
0
. If, for simplicity, we change coordinates so that
(x
0
, y
0
) = (0, 0), and we further denote h
0
= H(x
0
, y
0
) = H(0, 0), then the equa
tion (7.9) becomes:
H(x, y) = h
0
+
1
2
h
xx
x
2
+ h
xy
xy +
1
2
h
yy
y
2
+ · · ·
NonChaotic Systems 139
So, the behaviour of the trajectories of the system, i.e. of the level curves of H, near
the equilibrium point, is determined by the quadratic form:
1
1
2
_
h
xx
x
2
+ 2h
xy
xy + h
yy
y
2
_
(7.10)
Depending on the number of real roots of the characteristic polynomial
2
k(X) =
h
xx
+ 2h
xy
X + h
yy
X
2
, these trajectories may look elliptic, hyperbolic, or parabolic.
3
If k(X) has no roots, then the trajectories near the equilibrium point are elliptical,
and the point is a stable equilibrium point. If k(X) has two distinct roots, then the
trajectories are hyperbolas, and their asymptotes are the lines with slopes the two
roots (this will be elaborated further in the next section).
For instance, the system (7.7) has an equilibrium point at (0, 0), whereupon H is
written, locally, as:
1
2
x
2
+
1
2
y
2
In this case, the polynomial in question is 1 + X
2
, which has no real roots. Hence,
the system will exhibit elliptical orbits near (0, 0).
4
7.2 Linear Systems
Near an equilibrium point (we will assume this point is 0, 0), system (7.1) can be
approximated by a linear system:
˙ x = ax + by
˙ y = cx + dy
(7.11)
where a, b, c and d are real numbers or, in a more general analysis, real functions of
t of a special form.
We will now discuss in greater detail the classical theory of system (7.11). This
system has, as already mentioned, an equilibrium point at (x, y) = (0, 0). We will
1
This quadratic form is the quadratic form associated to the Hessian matrix:
_
¸
¸
¸
¸
¸
¸
_
∂
2
H
∂x
2
∂
2
H
∂x∂y
∂
2
H
∂y∂x
∂
2
H
∂y
2
_
¸
¸
¸
¸
¸
¸
_
2
This number depends of course on the discriminant of the polynomial, which is the determinant of
the Hessian matrix.
3
That is, in the case that the critical point is nondegenerate, i.e. in the case where the characteristic
polynomial is not identically zero.
4
In this case, of course, the orbits will remain elliptical even away from 0, 0.
140 Chaotic Modelling and Simulation
denote by A the matrix of coeﬃcients:
A =
_
a b
c d
_
The eigenvalues of this matrix are called the eigenvalues of system (7.11), and
they are given by the equation
A − λI =
¸
¸
¸
¸
¸
¸
a − λ b
c d − λ
¸
¸
¸
¸
¸
¸
= 0
or, in simpler terms:
λ
2
− (a + d)λ + (ad − bc) = 0
If we note, that a + d and ad − bc are respectively the trace and determinant of A,
we can write the equation for the eigenvalues as:
λ
2
− tr(A)λ + det (A) = 0
The solutions to this are
λ
1,2
=
tr(A) ±
√
∆
2
(7.12)
where, as usual, ∆ denotes the discriminant:
∆ = tr(A)
2
− 4 det(A) (7.13)
Before we proceed to a classiﬁcation of equilibrium points according to the qual
itative behaviour of the solutions near these points, we will discuss the eﬀects of
various transformations on the shape of the matrix A.
7.2.1 Transformations on linear systems
We start by considering a simple coordinate rescaling:
u = kx v = k
′
y (7.14)
The transformed system (7.11) then takes the form
˙ u = au + rbv
˙ v =
1
r
cu + dv
where r =
k
k
′
. So the matrix of the new system, A
′
, is:
A
′
=
_
a rb
1
r
c d
_
NonChaotic Systems 141
The trace and determinant thus remain invariant:
tr(A
′
) = tr(A)
det(A
′
) = det(A)
A more general change of coordinates amounts to the transformation
_
u v
_
= C
_
x y
_
(7.15)
where C is an invertible matrix of constants. The corresponding matrix A
′
is then a
matrix similar to A:
A
′
= CAC
′
Hence, this again keeps the trace and determinant invariant.
Another simple transformation involves time rescaling:
5
u(t) = x(kt) v(t) = y(kt) (7.16)
System (7.11) then becomes:
˙ u = kau + kbv
˙ v = kcu + kdv
Thus the matrix of the system becomes
A
′
= kA
and hence
tr(A
′
) = k tr(A)
det(A
′
) = k
2
det(A)
According to (7.12) and (7.13), the eigenvalues are simply rescaled by k.
Finally, the transformation
u(t) = e
−kt
x(t) v(t) = e
−kt
y(t) (7.17)
results in a system with matrix
A
′
= A − kI
with the corresponding changes this implies to the trace and determinant:
tr(A
′
) = tr(A) − 2k
det(A
′
) = k
2
− k tr(A) + det(A)
Note however, that this transformation does keep the discriminant
∆ = tr(A)
2
− 4 det(A)
invariant, and hence it amounts to a parallel translation of the eigenvalues by −k.
This transformation changes radically the behaviour of solutions near (0, 0), but in a
very controlled way, and we will utilise it in the next section.
5
This does not alter in any essential way the behaviour of solutions near the point (0, 0), provided
k > 0.
142 Chaotic Modelling and Simulation
7.2.2 Qualitative behaviour at equilibrium points
We will now proceed to analyse the qualitative behaviour of solutions near equi
librium points, using the transformations from section 7.2.1. First oﬀ, the sign of
tr A is of paramount importance. Using transformation (7.17), we can always reduce
to the case where tr(A) = 0, i.e. where d = −a, and we will do so for the next couple
of paragraphs. The condition tr(A) = 0 is exactly the condition for system (7.11) to
be conservative. In that case, the system has the ﬁrst integral:
f (x, y) = cx
2
− 2axy − by
2
We have to distinguish a couple of cases here.
1. ∆ = 0, i.e. the ﬁrst integral f is degenerate, and there is only one, double,
eigenvalue, namely 0. Up to a change of coordinates, we may assume f (x, y) =
x
2
, i.e. that the linear system is:
˙ x = 0
˙ y = x
The solutions are then:
x(t) = A
y(t) = At + B
This case is illustrated in Figures 7.2(a) and 7.2(b).
2. ∆ < 0. Then, there are two complex eigenvalues, which, after performing
transformation (7.16), we may assume to be ±i. Then, up to a change of
coordinates, we may assume f (x, y) = x
2
+ y
2
, i.e. that the linear system
is
˙ x = −y
˙ y = x
The solutions are then:
x(t) = Acos(t) + Bsin(t)
y(t) = Bcos(t) − Asin(t)
This case is illustrated in Figures 7.2(c) and 7.2(d).
3. ∆ > 0. Then there are two distinct real eigenvalues, which, after performing
tranformation (7.16), we may assume to be ±1. After a change of coordinates,
we may assume that f (x, y) = xy, i.e. that the system is:
˙ x = x
˙ y = −y
The solutions are then:
x(t) = Ae
t
y(t) = Be
−t
NonChaotic Systems 143
(a) ∆ = 0 , tr(A) = 0 (b) ∆ = 0, tr(A) > 0
(c) Center point, ∆ < 0, tr(A) = 0 (d) Spiral source, ∆ < 0 , tr(A) > 0
(e) Saddle point, ∆ > 0, tr(A) = 0 (f) ∆ > 0 , tr(A) > 0
FIGURE 7.2: Qualitative classiﬁcation of equilibrium points
144 Chaotic Modelling and Simulation
This case is illustrated in Figures 7.2(e) and 7.2(f).
To summarise, the behaviour of solutions near an equilibrium point falls under one
of the following cases:
1. ∆ = 0. This is the degenerate case.
2. ∆ > 0. Then λ
1,2
are both real. The behaviour in this case depends a lot on their
sign. If they are both negative, then all solutions tend toward the equilibrium
point, which in this case is called a stable point, or also a stable node. If they
are both positive, then the solutions tend away from the equilibrium point, and
the point is classiﬁed as unstable.
Otherwise, λ
1,2
are both real, but with opposite signs. In this case, there is a
stable direction and an unstable direction, indicated by the negative and pos
itive eigenvalues respectively. The equilibrium point is then called a saddle
point.
3. ∆ < 0. Then λ
1,2
have nonzero real and imaginary parts. If the real part is
negative, then the solutions spiral toward the equilibrium point, which is called
in this case a stable spiral or a spiral sink. If, on the other hand, the real part
is positive, then the solutions spiral away from the critical point, and the point
is then called an unstable point or a spiral source.
A special case is when λ
1,2
have only an imaginary part (i.e. when the real part
is zero). The critical point is then called a center.
7.3 EggShaped Forms
7.3.1 A simple eggshaped form
We now proceed to examine certain Hamiltonian systems that give rise to very
interesting eggshaped forms. The ﬁrst such system is:
˙ x = y(y − b)
˙ y = x
(7.18)
This is a Hamiltonian system, since
∂
∂x
_
y(y − b)
_
= 0 = −
∂
∂y
(x)
The Hamiltonian is easily seen to be
H(x, y) = −
x
2
2
− b
y
2
2
+
y
3
3
+ h
NonChaotic Systems 145
This system has two equilibrium points, at
(x, y) = (0, 0) and (x, y) = (0, b)
The ﬁrst of these is a stable point, as the quadratic form,
−
1
2
_
x
2
+ by
2
_
has two imaginary roots:
λ
1,2
= ±
i
√
b
The second point, however, is unstable, since its characteristic polynomial,
−2 + 2bX
2
= 2(
√
bX − 1)(
√
bX + 1)
has two real roots:
λ = ±
1
√
b
These are exactly the slopes of the two tangent lines at (0, b).
For convenience, we will choose the integration constant in H to be such, that H
is 0 at this, second, equilibrium point, namely:
h =
b
3
6
The level curve that passes through (0, b) is shown, marked with a heavy line, in
Figure 7.3(a) (b = 0.6). It has equation:
H(x, y) = −
x
2
2
− b
y
2
2
+
y
3
3
+
b
3
6
The sharp (top) corner of the eggshaped form is located at the equilibrium point
(0, b), whereas the bottom of the form is at the point
_
0, −
b
2
_
.
6
At this point, the speed
of the particle is given by ˙ x =
3
4
b
2
. The boundaries of the eggshaped form in the x
direction are when y = 0, namely x = ±
_
b
3
3
. The speed at these points is ˙ y = ±
_
b
3
3
respectively.
7.3.2 A double eggshaped form
A double eggshaped form with interconnected sharp corners is expressed by the
following system:
˙ x = −y(y − a)(y − b)
˙ y = x
(7.19)
6
This is found by solving the cubic polynomial H(0, y) = 0 This will have a double root at y = b, and
one more root.
146 Chaotic Modelling and Simulation
−0.3 −0.2 −0.1 0.0 0.1 0.2 0.3
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
(a) An eggshaped form
−0.3 −0.2 −0.1 0.0 0.1 0.2 0.3
0.0
0.5
1.0
(b) A ﬁgure eight form
FIGURE 7.3: Eggshaped forms
where we will assume that 0 < b < a. The Hamiltonian in this case is
7
H(x, y) = −
y
4
4
+ (a + b)
y
3
3
− ab
y
2
2
−
x
2
2
+
b
3
12
(b − 2a)
Figure 7.3(b) shows the trajectories when a = 1 and b = 0.6.
The ﬁrst eggshaped form at the top of the ﬁgure has a stable equilibrium point
at (a, 0), whereas the other eggshaped form has a stable equilibrium point located
at (0, 0). An unstable point is located at (0, b), where the sharp corners of the two
eggshaped forms coincide.
When the initial values are x
0
= 0 and y
0
= −b, a ﬁgure eight shape appears. This
is a characteristic level curve marked by a heavy line in Figure 7.3(b). The integration
constant has the value h = −
b
4
12
+
ab
3
6
for this particular case. Inside this form the
trajectories follow closed loops, and the same holds outside this characteristic level.
7.3.3 A double eggshaped form with an envelope
A double eggshaped form enclosed in an external envelope is provided by the
following system:
˙ x = y(y
2
− b
2
)(y + a)
˙ y = x
(7.20)
7
The constant was again chosen, so that the Hamiltonian is zero at the unstable point.
NonChaotic Systems 147
The Hamiltonian of this system is given by:
H(x, y) =
y
5
5
+ a
y
4
4
− b
2
y
3
3
− ab
2
y
2
2
−
x
2
2
where we’ve set the integration constant to zero for simplicity. Figure 7.4 illustrates
this case where a = 1 and b = 0.6. The ﬁrst eggshaped form at the top of the ﬁgure
has a stable equilibrium point at (0, 0), whereas the second eggshaped form has a
stable equilibrium point at (0, −a). There are two unstable points, one at (0, −b),
where the sharp corners of the two eggshaped forms coincide, and one at (0, b),
where the envelope is formed. The level curves through those two unstable points
are marked with a heavy line in the graph. For this particular case, corresponding
values of the Hamiltonian on these curves are
h
b−
= −
2
15
b
5
−
1
4
ab
4
h
b+
=
2
15
b
5
−
1
4
ab
4
(7.21)
Inside the envelope, the trajectories follow closed loops, whereas, outside of it, the
trajectories diverge to inﬁnity.
−0.4 −0.2 0.0 0.2 0.4
−1.0
−0.5
0.0
0.5
1.0
FIGURE 7.4: A double eggshaped form with an envelope
148 Chaotic Modelling and Simulation
7.4 Symmetric Forms
We consider nowcertain forms that exhibit symmetry. Asimple symmetric Hamil
tonian form is based on the system:
˙ x = y(y − b)
˙ y = x(x − b)
(7.22)
The corresponding Hamiltonian is
H(x, y) =
y
3
− x
3
3
− b
y
2
− x
2
2
The level curves at level 0 are marked by a thick line in Figure 7.5(a). There are two
stable points, at (b, 0) and (0, b), and two unstable points, at (0, 0) and (b, b).
A slight change to (7.22) gives rise to another doubleshaped form, with symmetry
with respect to the y = x axis. The equations are:
˙ x = −y(y − b)
˙ y = x(x − b)
(7.23)
The Hamiltonian is
H(x, y) =
y
3
+ x
3
3
− b
y
2
+ x
2
2
The level curve for the outer periphery has level −
b
3
6
, and is illustrated in Fig
ure 7.5(b). The stable equilibrium points are located at (0, 0) and (b, b), whereas
the two unstable points are located at (0, b) and (b, 0). The three points of the level
curve located on the line x = y are: the point
_
b
2
,
b
2
_
, and the other two solutions of
the equation 4x
3
− 6bx
2
+ b
3
= 0.
A simple nonsymmetric form is expressed by the following two equations:
˙ x = −y(y − b)
˙ y = x(x − a)
(7.24)
The Hamiltonian in this case is
H(x, y) =
y
3
+ x
3
3
−
by
2
+ ax
2
2
There are two characteristic level curves, passing through the unstable points (a, 0)
and (0, b) respectively, and with corresponding levels h
a
= −
1
6
a
3
and h
b
= −
1
6
b
3
respectively. These level curves are shown in Figure 7.6, marked with a thick line.
There are two stable equilibrium points, at (0, 0) and (a, b).
An interesting threefold shape is given by the system
˙ x = y(y
2
− b
2
)
˙ y = x(x − b)
(7.25)
NonChaotic Systems 149
−0.4 −0.2 0.0 0.2 0.4 0.6 0.8
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
(a)
−0.4 −0.2 0.0 0.2 0.4 0.6 0.8
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
(b)
FIGURE 7.5: Symmetric forms
−0.5 0.0 0.5 1.0
−0.5
0.0
0.5
1.0
FIGURE 7.6: A simple nonsymmetric form
150 Chaotic Modelling and Simulation
The corresponding level curves of the Hamiltonian
y
4
4
− b
2
y
2
2
−
x
3
3
+ b
x
2
2
are shown in Figure 7.7(a) (b = 0.85).
−0.5 0.0 0.5 1.0 1.5
−1.0
−0.5
0.0
0.5
1.0
(a) b = 0.85 >
2
3
−0.5 0.0 0.5 1.0
−1.0
−0.5
0.0
0.5
1.0
(b) b = 0.6
−0.2 0.0 0.2 0.4 0.6
−0.6
−0.4
−0.2
0.0
0.2
0.4
0.6
(c) b = 0.3 <
2
3
FIGURE 7.7: A symmetric threefold form
There are two characteristic levels for H, which provide the level curves marked
by a heavy line. The ﬁrst value, h = 0, corresponds to the equilibrium point (0, 0),
and gives the central formation. This curve also passes through the point (
3
2
b, 0)
where x takes its maximum value in the bounded part of this curve.
The second characteristic level is h =
b
3
6
−
b
4
4
, for which the level curve passes
through the two unstable equilibrium points (b, ±b). There are three stable equilib
rium points, two at (0, ±b), where H = −
b
4
4
, and one at (b, 0), where H =
b
3
6
.
The shape varies a lot depending on the parameter b. When b >
2
3
, we get a ﬁgure
similar to Figure 7.7(a). When b =
2
3
, the two characteristic curves coincide, and the
corresponding graph is shown in Figure 7.7(b). When b <
2
3
, the two curves are not
so interrelated (Figure 7.7(c)).
7.5 More Complex Forms
More complex forms appear by adding new terms in the righthand side of the
Hamiltonian equations of the twodimensional system. The fourtype symmetric
form illustrated in Figure 7.8(a) corresponds to the system
˙ x = −y(y
2
− b
2
)
˙ y = x(x
2
− b
2
)
(7.26)
NonChaotic Systems 151
with Hamiltonian:
x
4
+ y
4
4
− b
2
x
2
+ y
2
2
There are ﬁve stable equilibrium points, one at the origin (0, 0), and the other four
located at the symmetric points (b, b), (b, −b), (−b, b) and (−b, −b). There are also
four unstable equilibrium points located at (0, b), (0, −b), (b, 0) and (−b, 0). The
level curve passing through those equilibrium points has level −
b
4
4
. In Figure 7.8(a),
a heavy line marks this critical level curve.
Introducing a new parameter a into (7.26), the following system arises:
˙ x = −y(y
2
− b
2
)
˙ y = x(x
2
− a
2
)
(7.27)
Its Hamiltonian is
x
4
+ y
4
4
−
a
2
x
2
+ b
2
y
2
2
This system has a richer structure, as illustrated in Figure 7.8(b) (a = 0.7, b = 0.5).
Two distinct level curves appear, for values H = −
1
4
a
4
and H = −
1
4
b
4
respectively.
The ﬁrst curve is associated with two 8shape forms of Figure 7.8(b). The second
curve is associated with the other more complicated form, also marked with a heavy
line. The four points located on the x axis (y = 0) are given by:
x = ±
_
a
2
±
√
a
4
− b
4
If the parameter a is introduced in both equations, then an interesting system
arises:
˙ x = −y(y − a)(y − b)
˙ y = x(x − a)(x − b)
(7.28)
Its Hamiltonian is:
H(x, y) =
x
4
+ y
4
4
− (a + b)
x
3
+ y
3
3
+ ab
x
2
+ y
2
2
The model, with parameters equal to a = 0.8 and b = 0.3, is illustrated in Fig
ure 7.8(c).
There is an outer characteristic level curve, with level h
b
= −
1
12
b
4
+
1
6
ab
3
, and
an inner characteristic level curve, with level h
a
= −
1
12
a
4
+
1
6
a
3
b. There are ﬁve
stable points located at (0, 0), (a, a), (b, b), (a, 0) and (0, a), and four unstable points
at (b, 0), (0, b), (a, b) and (b, a).
When b =
1
2
a, then h
b
= 0, and the system leads to a perfect symmetry, illustrated
in Figure 7.8(d) when the parameters are a = 0.8 and b =
a
2
= 0.4.
152 Chaotic Modelling and Simulation
−1.0 −0.5 0.0 0.5
−1.0
−0.5
0.0
0.5
(a) A 4type symmetric form
−1.0 −0.5 0.0 0.5 1.0
−0.5
0.0
0.5
(b) Two 8shape forms with an envelope
−0.2 0.0 0.2 0.4 0.6 0.8 1.0
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
(c) A complicated form
−0.2 0.0 0.2 0.4 0.6 0.8 1.0
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
(d) A perfect symmetry
FIGURE 7.8: More complex forms
NonChaotic Systems 153
7.6 HigherOrder Forms
We conclude this section with a couple of much more complicated examples, aris
ing from systems whose equations involve higherorder polynomials. One such ex
ample, illustrated in Figure 7.9(a), is the system:
˙ x = −(y − a)(y − b)(y − c)(y − e)(y − f )
˙ y = (x − a)(x − b)(x − c)(x − e)(x − f )
(7.29)
The Hamiltonian can be easily computed from the above equations, but we will omit
the tedious task here.
The parameters were set to a = 0.1, b = 0.6, c = 0.9, e = 1.5 and f = 2. There
are 13 stable equilibrium points, and 12 unstable equilibrium points. Equilibrium
regions and regions with high movement (ﬂows) appear. As in the simpler cases,
(a) Symmetry only along y = x axis (b) Multiple symmetries
FIGURE 7.9: Higherorder forms
appropriate values for the parameters lead to symmetric forms, as in Figure 7.9(b),
where the parameters are a = 0, b = 0.5, c = 1, e = 1.5 and f = 2.
154 Chaotic Modelling and Simulation
Questions and Exercises
1. Consider the system
˙ x = −y
˙ y = x(1 − x)
(a) Find its equilibrium points.
(b) Find the eigenvalues and eigenvectors at the equilibriumpoints, and char
acterise the points according to their stability.
(c) Linearise the system, and ﬁnd the type of stability at the equilibrium
points.
(d) Find a ﬁrst integral of motion for the system.
(e) Draw various trajectories of the system in the (x, y) plane.
(f) Draw the trajectories that pass through equilibrium points.
(g) Find the maximum speed of the ﬂow at the y direction (˙ y).
2. Consider the system
˙ x = 1 − y
˙ y = x + ay
(a) Find the equilibrium points for the system.
(b) Find the characteristic equation at the equilibrium points.
(c) Characterise the equilibrium points for various values of a.
(d) Examine in particular the case a = 2.
3. For each of the following systems, (a) show that it is Hamiltonian, (b) com
pute the Hamiltonian function, (c) determine the ﬁxed points, and classify
them, and (d) plot, for various values of the parameters, the level curves of the
Hamiltonian, including the curves passing through the ﬁxed points.
(a)
˙ x = y(y − b)
˙ y = ax
(b)
˙ x = y
2
− b
2
˙ y = x
(c)
˙ x = y
2
− b
2
˙ y = x
2
− a
2
4. Show that the system
˙ x = 2xy
˙ y = x
2
− y
2
is conservative, but not Hamiltonian. Find a system equivalent to it that is
Hamiltonian.
NonChaotic Systems 155
5. Show that the system
˙ x =
1
x
˙ y = −
1
y
is conservative, but not Hamiltonian. Find a system equivalent to it that is
Hamiltonian.
6. Show that each of the transformations described in section 7.2.1 have the indi
cated eﬀect on the matrix A and its eigenvalues.
Chapter 8
Rotations
In this chapter we examine the eﬀects of introducing aﬃne transformations to mod
els, in particular the eﬀect of introducing rotation. A wealth of very interesting
models arise from this seemingly simple change.
8.1 Introduction
The location (x
n
, y
n
) of a system is given in parametric form (r
n
, θ
n
) by:
x
n
= r
n
cos(θ
n
)
y
n
= r
n
sin(θ
n
)
(8.1)
where r =
_
x
2
+ y
2
and θ
n
is the rotation angle.
If we assume that transitioning from the nth state to the n + 1st state amounts to
a rotation by an angle ∆θ, then the new coordinates would be given by:
x
n+1
= r
n
cos(θ
n
+ ∆θ)
y
n+1
= r
n
sin(θ
n
+ ∆θ)
(8.2)
or after using the familiar trigonometric identities for the sum of two angles:
x
n+1
= r
n
_
cos(θ
n
) sin(∆θ) − sin(θ
n
) sin(∆θ)
_
y
n+1
= r
n
_
cos(θ
n
) sin(∆θ) + sin(θ
n
) cos(∆θ)
_ (8.3)
Therefore, the eﬀect of rotation can be written directly using the following, which
may be familiar to you already:
x
n+1
= x
n
cos(∆θ) − y
n
sin(∆θ)
y
n+1
= x
n
sin(∆θ) + y
n
cos(∆θ)
(8.4)
When the rotation is followed by a translation equal to a and parallel to the x axis,
then the tranformation (8.4) is replaced by:
x
n+1
= a + x
n
cos(∆θ) − y
n
sin(∆θ)
y
n+1
= x
n
sin(∆θ) + y
n
cos(∆θ)
(8.5)
157
158 Chaotic Modelling and Simulation
We can consider the system (8.5) as a simple system of diﬀerence equations. Let us
work out an analogous system of diﬀerential equations. The time elapsed between
two successive iterations is ∆t and is usually set to 1. Then, taking into account that
˙ x =
dx
dt
≈
∆x
∆t
= x
n
− x
n−1
and similarly for ˙ y, we obtain the diﬀerential equations:
˙ x = a + x
_
cos(∆θ) − 1
_
− y sin(∆θ)
˙ y = x sin(∆θ) + y
_
cos(∆θ) − 1
_ (8.6)
8.2 A Simple RotationTranslation System of Diﬀeren
tial Equations
If the rotation angle is small, ∆θ ≪ 1, the system (8.6), using ﬁrst order Taylor
approximations for cos and sin, can be simpliﬁed to:
˙ x = a − y∆θ
˙ y = x∆θ
(8.7)
This is a Hamiltonian system, whose ﬁrst integral of motion can be computed from
the diﬀerential equation:
∂y
∂x
=
x∆θ
a − y∆θ
This equation leads to the form:
1
ady = (∆θ)rdr (8.8)
The solution of equation (8.8) depends on the form of the function ∆θ. A reason
able assumption would be that ∆θ is “radial,” i.e. depends only on the distance r from
the origin, and not of the direction. A standard choice in mechanics, based on the
transverse component of the acceleration, is that the rotational change is inversely
proportional to the square distance, i.e. that:
∆θ ≈
h
1
r
2
Another important case is that of a mass M rotating in a circular orbit under a
central force
f (r) =
MG
r
2
1
We have used here that r
2
= x
2
+ y
2
, and consequently rdr = xdx + ydy.
Rotations 159
where G is the gravity constant. The equations of motion then result in
∆θ ≈
˙
θ =
_
MG
r
3
In this case, equation (8.8) becomes:
ady =
_
MG
r
dr
Its solution is
r =
(ay + h)
2
4MG
where h is an integration constant. This can also be rewritten as
x
2
=
_
(ay + h)
2
4MG
− y
_ _
(ay + h)
2
4MG
+ y
_
(8.9)
The ﬁxed point of the system is at
(x, y) =
_
0,
MG
a
2
_
at which point the value of h is:
h =
MG
a
The path of the trajectory in the (x, y) plane for this value of h divides the plane in
two segments. When h >
MG
a
, the trajectories diverge and the rotating object heads
oﬀ to inﬁnity. When h <
MG
a
the rotating mass moves inside the space bounded by
the above trajectory.
FIGURE 8.1: The limiting trajectory
160 Chaotic Modelling and Simulation
Some very interesting properties of equation (8.8) are illustrated in Figure 8.1.
The trajectory in the limit of escape has an egg shape, similar to but diﬀerent from
those discussed in Chapter 7. The sharp corner is at the maximum value of y =
MG
a
2
,
where x = 0. This is the ﬁxed point for the system, which is not stable.
The minimum value of y is that where the maximum rotation speed is achieved.
This can be found by setting x = 0 in equation (8.9). This gives us two solutions, the
ﬁxed point at y =
MG
a
2
as already discussed, and the solution to the equation:
(ay + h)
2
= −4ahy
The largest solution of this equation is:
y =
_
2
√
2 − 3
_
MG
a
2
≈ −0.1716
MG
a
2
(8.10)
The maximum value of x is given by:
x
max
=
MG
a
2
_
5
√
5 − 11
2
≈ 0.3
MG
a
2
This is computed by equating to zero the ﬁrst derivative of x with respect to y, that
is:
∂x
∂y
=
˙ x
˙ y
=
1
x
_
a
8(MG)
2
_
ay +
MG
a
_
3
− y
_
= 0 (8.11)
Setting z =
a
h
y, equation (8.11) becomes:
(z + 1)
3
= 8z
This can be easily seen to have the three roots:
z = 1, z =
√
5 − 2, z = −
√
5 − 2
The ﬁrst of these corresponds to the ﬁxed point
_
0,
MG
a
2
_
. The second is the one we
are after, and the corresponding y value is:
y =
MG
a
2
(
√
5 − 2)
When y = 0, then
x =
MG
4a
2
=
h
4a
, r =
h
2
4MG
=
h
4a
and ∆θ =
8a
2
h
so from (8.7) we get:
dy
dx
=
˙ y
˙ x
=
x∆θ
a − y∆θ
=
h
4a
·
8a
2
h
a
= 2
Rotations 161
This tangent appears in Figure 8.1.
The tangent at the top sharp corner of the eggshaped path can be seen by writing
equation (8.8) in a coordinate system centered at this point. For this reason, let:
z = y −
MG
a
2
= y −
h
a
Then, in terms of (x, z), equation (8.8) can be written as:
0 = −x
2
+
a
4h
z
2
·
_
2h
a
+ 2z +
a
4h
z
2
_
= −x
2
+
1
2
z
2
+ · · ·
where the omitted terms are of higher order in z. This means that near (x, z) = (0, 0),
our orbit behaves exactly like the function:
0 = −x
2
+
1
2
z
2
=
_
1
√
2
z − x
_ _
1
√
2
z + x
_
So the two tangent lines are given by the equations:
y =
h
a
±
√
2x
(a) Rotating paths (b) Paths in the plane
FIGURE 8.2: Rotating paths
Figure 8.2(a) illustrates the path of a moving particle. The particle follows an anti
clockwise direction. It starts from a point near zero and rotates away from the centre
until the highest permitted place. Then, it escapes to inﬁnity following the tangent on
the top of the eggshaped path. In Figure 8.2(b), a large number of paths are drawn.
Every path has the same value for h =
MG
a
2
. The paths outside the eggshaped limits
162 Chaotic Modelling and Simulation
have direction from left to right and they do not transverse the eggshaped region of
the plane. The iterative formula used for the above simulations is:
x
n+1
= x
n
+
_
a − y
n
√
MGr
3
n
_
d
y
n+1
= y
n
+ x
n
√
MGr
3
n
d
r
n
=
_
x
2
n
+ y
2
n
(8.12)
where, for the above ﬁgures, a = 2, MG = 100 and d = 0.05.
Other interesting properties of the aforementioned eggshaped forms arise by us
ing elements from mechanics. From the equations of motion, it is clear that the para
meter a is a constant speed with direction parallel to the x axis. The rotational speed
has two components: ˙ x, ˙ y. However, at the top and at the bottom of the eggshaped
form only the x component of the velocity is present, since x = 0. The rotational
speed, or the transverse component of the velocity, is v =
_
MG
y
. This speed must be
equal to a. Thus, without using the equations for the trajectories, the value of y at
the top, x = 0, is estimated from the equality v − a = 0. Thus,
_
MG
y
= a
and
y =
MG
a
2
To obtain the equation for the total speed at the lower point of the curve, we must
take into account that this speed must be equal to the escape speed:
v + a = v
esc
(8.13)
The escape speed for a mass rotating around a large body is known from mechanics
to be:
v
esc
=
_
2MG
y
Thus, equation (8.13) becomes:
_
MG
y
+ a =
_
2MG
y
Finally, from this last equation the value of y is:
y =
MG
a
2
(
√
2 − 1)
2
This is, up to a sign, the same result obtained in equation (8.10). It is obvious
that the system in these particular cases obeys the laws of classical mechanics. This
egglike form is a set of vortex curves with relative stability and strength.
Rotations 163
8.3 A Discrete RotationTranslation Model
The analysis above is signiﬁcant in helping us understand the dynamics under
lying the rotation analogue in the discrete case. The iterative scheme based on the
diﬀerence equations for rotationtranslation is more diﬃcult to handle analytically,
but, on the other hand, it is closer to a real life situation. No approximation is needed,
whereas, the eggshaped scheme and the trajectories are retained. Moreover, the tra
jectories inside the eggshaped boundary are perfectly closed loops. Additionally, the
chaotic nature of the rotationtranslation procedure near the centre becomes appar
ent. In the equations (8.4), we introduce an area contracting parameter b (0 < b < 1),
and the ﬁnal equations used are the following
x
n+1
= a + b (x
n
cos ∆θ − y
n
sin ∆θ)
y
n+1
= b (x
n
sin ∆θ + y
n
cos ∆θ)
(8.14)
where the same approximation
∆θ
n
≈
_
MG
r
3
n
is used.
A convenient formulation of the last iterative map expressing rotation and transla
tion is achieved by using the following diﬀerence equation
f
n+1
= a + b f
n
e
i∆θ
(8.15)
where f
n
= x
n
+ iy
n
.
The Jacobian determinant of this map is J = b
2
= λ
1
λ
2
. This is an area contracting
map if the Jacobian J is less than 1. The eigenvalues of the Jacobian are λ
1
and λ
2
,
and the translation parameter is a.
A very interesting property of the map (8.14) is the existence of a disk F of radius
ab
1−b
centered at (a, 0) in the (x, y) plane. The points inside the disk remain inside it
under the map.
2
To locate the ﬁxed points of this transformation, let us change coordinates by
setting:
z
n
= f
n
−
a
1 − b
2
Then the transformation can be rewritten as:
z
n+1
− z
n
=
_
be
i∆θ
− 1
_
z
n
+ ab
_
e
i∆θ
− b
1 − b
2
_
2
This fact is left as an exercise for the reader.
164 Chaotic Modelling and Simulation
So for a ﬁxed point we would have:
z =
ab
1 − b
2
·
_
e
i∆θ
− b
1 − be
i∆θ
_
(8.16)
It can easily be veriﬁed that
¸
¸
¸
¸
¸
¸
e
i∆θ
− b
1 − be
i∆θ
¸
¸
¸
¸
¸
¸
= 1
and consequently all ﬁxed points lie on the circle centered at
_
a
1−b
2
, 0
_
and with radius
ab
1−b
2
, in other words they satisfy:
_
x −
a
1 − b
2
_
2
+ y
2
=
a
2
b
2
(1 − b
2
)
2
Further, if we transform equation (8.16) back to an equation for f = f
n
, we get:
f =
a
1 − b
2
_
1 + b
e
i∆θ
− b
1 − be
i∆θ
_
=
a
1 − be
i∆θ
Therefore, the ﬁxed points f further satisfy:
3
cos(∆θ) =
1
2b
_
1 + b
2
−
a
2
r
2
_
(8.17)
This equation can easily be solved provided that the angle ∆θ is a function of r.
Be begin our analysis with the case where no area contraction takes place, so that
b = 1. In this case the disk F is replaced by the entire plane. An interesting property
of the system (8.5) in this case is that the resulting map often exhibits a symmetry
along the line x =
a
2
, particularly when chaotic behaviour is present. This is diﬀerent
from the symmetry axis of the diﬀerential equation analogue where the symmetry
axis is the line x = 0. In the diﬀerence equation case, the eggshaped form is shifted
to the right of the original positions of the system of coordinates.
As a possible explanation for this symmetry, let us consider the map in the form:
f
n+1
= a + f
n
e
i∆θ
For any complex number z, it can be easily seen that its reﬂection with respect to the
axis x =
a
2
is a − ¯ z, where ¯ z is as usual the complex conjugate of a complex number.
If we denote the reﬂection of f
n
by g
n
, then we see that:
g
n+1
= a −
¯
f
n+1
= a −
_
a +
¯
f
n
e
−i∆θ
_
= (−a + g
n
)e
−i∆θ
3
The derivation of this is also left as an exercise to the interested reader.
Rotations 165
Multiplying through with e
i∆θ
, we see that the reﬂections satisfy a very similar recur
sive relation, except that it goes “backwards”:
4
g
n
= a + g
n+1
e
i∆θ
Assuming that at some time the map passes very close to the x =
a
2
line, the corre
sponding point will be very close to its reﬂection, and the evolution of the system
from that point on will be very close to the reﬂection of the history of the system up
to that point. This would cause the map to seem symmetric, especially when it ex
hibits chaotic behaviour. Similarly, equilibrium points will tend to exhibit symmetry
around this axis.
We proceed now to discuss the equilibrium points of the system. We are therefore
searching for the points where x
n+1
= x
n
and y
n+1
= y
n
. This leads to x =
a
2
and
y =
a
2
cot
_
∆θ
2
_
. It is clear that there is no symmetry axis in the x direction, as y
does not show stable magnitude since it is a function of the angle ∆θ. However, an
approximation of y is achieved when the angle ∆θ is small. Then
y ≈
a
∆θ
=
MG
a
2
The point
_
x =
a
2
, y =
MG
a
2
_
is therefore an equilibrium point. However, it is not a
stable equilibrium point. The system may easily escape to inﬁnity when y is higher
than
MG
a
2
.
(a) Rotationtranslation (b) The central chaotic bulge
FIGURE 8.3: Rotationtranslation, and the chaotic bulge
4
An important detail here is that ∆θ depends on n, so the formulas don’t quite match up, unless ∆θ
n+1
is close to ∆θ
n
.
166 Chaotic Modelling and Simulation
Figures 8.3(a) and 8.3(b) illustrate a rotationtranslation case with parameters
a = 2 and MG = 100. In Figure 8.3(a), a number of paths inside the eggshaped
formation are drawn. All the paths are analogous to those obtained by the diﬀeren
tial equation analogue studied above, but they are moved to the right at a distance
x =
a
2
. A chaotic attractor that resembles a bulge appears around the centre of co
ordinates. An enlargement of this attractor is illustrated in Figure 8.3(b). This is a
symmetric chaotic formation with the line x =
a
2
as an axis of symmetry.
(a) a = 0.8 (b) a = 1
FIGURE 8.4: The chaotic bulge
Figures 8.4(a) and 8.4(b) present an illustration of the chaotic bulge of the rotation
translation model. The parameters are a = 0.8 for Figure 8.4(a) and a = 1 for
Figure 8.4(b). In both cases, MG = 100 and the rotation angle is ∆θ =
_
MG
r
3
.
In Figure 8.4(b), third and fourth order equilibrium points appear, indicated by a
small cross and a bigger cross respectively. In this case, the following relations hold:
x
n+3
= x
n
and y
n+3
= y
n
, and x
n+4
= x
n
and y
n+4
= y
n
respectively.
Accordingly, in Figure 8.4(a), two equilibrium cases appear, one of second order
(indicated by small cross) where x
n+2
= x
n
and y
n+2
= y
n
and y
n+2
= y
n
and one of
the third order where x
n+3
= x
n
and y
n+3
= y
n
.
In both cases, an algorithm is introduced for the estimation of the equilibrium
points. The convergence is quite good provided that the starting values are in the
vicinity of the equilibrium points.
As the parameter a takes higher values, higher order equilibrium points appear.
A ﬁfth order equilibrium point is added to those presented before, as illustrated in
Figure 8.5(b), when a = 1.2. The relations for the new equilibrium point are x
n+5
=
x
n
and y
n+5
= y
n
. Figure 8.5(a) illustrates the case where a = 0.2. Only a ﬁrst order
equilibrium point is present.
Rotations 167
(a) a = 0.2 (b) a = 1.2
FIGURE 8.5: The chaotic bulge
Higher order bifurcation is present in Figures 8.6(a) and 8.6(b). In Figure 8.6(a),
a tenth order bifurcation form appears in the periphery of the chaotic attractor. The
magnitude of the translation parameter is a = 1.5. In Figure 8.6(b), the translation
parameter is a = 1.9. At this value, the bifurcation is so high that the chaotic region
covers all the space bounded by the eggshaped form. The point masses following
chaotic trajectories eventually escape to inﬁnity.
(a) a = 1.5 (b) Complete chaos (a = 1.9)
FIGURE 8.6: The chaotic bulge
It is diﬃcult to ﬁnd the coordinates of equilibrium points that are located in the
centre of the islands that are in the middle of the chaotic sea of the attractor, though
168 Chaotic Modelling and Simulation
as discussed earlier those are going to be symmetric with respect to the axis x =
a
2
.
A ﬁrst order equilibrium point is characterised by the simple relations x
n+1
= x
n
and y
n+1
= y
n
, which lead to the following relation for the radius of the circle on the
periphery of which the equilibrium point must be located
R
k
≈
3
_
MG
(2kπ)
2
, k = 1, 2, . . .
(a) a = 0.1 (b) a = 0.5
FIGURE 8.7: Equilibrium points
The circles R
k
introduce a set of decreasing order with regards to the magnitude
of the radius, as illustrated in Figure 8.7(a) (a = 0.1). The ﬁrst point located on
the circle with the larger radius (k = 1) is indicated by a cross. A few isoclines
indicating the limits of the island are drawn around this point. Subsequent points
follow in decreasing order. They are located in the middle of the isoclines.
The equilibrium points of higher order follow by using diﬀerent relations. A sim
ple case is that of the estimation of the equilibrium points of the second order, that
is, when the coordinates are x
n+2
= x
n
and y
n+2
= y
n
. This assumption leads to the
following equation:
(x
2
+ y
2
) cos
2
(∆θ) + 2x
2
cos(∆θ) + x
2
− y
2
= 0
This equation is fulﬁlled if the rotation angle obeys the relation ∆θ = (2k + 1)π.
Then, this result leads to the following relation for the radius of the circle, on the
periphery of which the equilibrium points must be located:
R
k
≈
3
_
MG
((2k + 1)π)
2
, k = 1, 2, . . .
Rotations 169
The simulation results are illustrated in Figure 8.7(b). The translation parameter
is a = 0.5. The concentric circles are distributed in decreasing order according to the
magnitude of the radius R
k
. The ﬁrst two equilibrium points are distributed in the
outer circle. They are indicated by a cross and are located on the axis of symmetry
x =
a
2
, but in opposite directions: one to the positive and one to the negative part
of the semiplane for y. Three pairs of order two equilibrium points are illustrated
in Figure 8.8(a). The equilibrium pairs are indicated by a cross. Also, the three
equilibrium islands of the ﬁrst order are presented. The parameter is a = 0.04. As
this parameter has a very low value, the central chaotic bulge has a small radius.
(a) Equilibrium pairs (a = 0.04) (b) Order3 bifurcation (a = 0.7)
FIGURE 8.8: The chaotic bulge
The order three equilibrium points are presented in Figure 8.8(b). They are located
in the middle of the three islands that are in the outer part of the chaotic bulge.
The translation parameter is a = 0.7. The order three equilibrium points are in the
periphery of a circle centred at (
a
2
,
a
5
). The radius of this circle is approximated by
R ≈
3
_
MG
(2π/3)
2
−
_
a
4
_
2
8.4 A General RotationTranslation Model
The solution of the diﬀerential equation of the model (8.7) leads to the form
ay + h =
_
(∆θ)rdr
170 Chaotic Modelling and Simulation
A simple approximation of the quantity ∆θ is:
∆θ ≈
_
MG
r
β
Then, the solution has the form
ay + h =
r
2−
β
2
2 −
β
2
√
MG (8.18)
The special case β = 2 leads to the following equation form
ay + h = r
√
MG
or:
(x
2
+ y
2
)GM = (ay + h)
2
This equation expresses a family of ellipses (Figure 8.9(a)). The parameter a is
6, and the constant of integration h takes several values. The axis of symmetry is at
x =
a
2
. Figure 8.9(b) illustrates a chaotic central bulge. The outer part of this chaotic
attractor is approximated by an ellipse with h = MG + a
2
.
(a) A family of ellipses (b) The chaotic bulge
FIGURE 8.9: Elliptic forms and the chaotic bulge
When β = 4, the resulting equation for (x, y) is:
√
MGln(r) = ay + h
The rotating forms are eggshaped and similar to those provided in earlier rotation
forms. With the exception of β = 2, where the rotation paths are elliptic, the cases
with β > 2 provide eggshaped rotation forms.
Rotations 171
Another very important property of the elliptic case (β = 2) is that the rotating par
ticles remain in the elliptic paths even if high values for the parameter h are selected.
In the limit when the parameter a approaches zero the ellipses turn to concentric
circles.
8.5 Rotating Particles inside the EggShaped Form
Two cases are of particular importance. In the ﬁrst case, the translation parameter
a is quite small, and the particles are trapped inside the eggshaped form and remain
there, following the trajectories proposed by the theory discussed so far. However, a
smaller region inside the trapping region exhibits chaotic behaviour. In this region,
the particles follow chaotic paths that form the attractors presented above. In Fig
ure 8.10(a), the outer limits of the eggshaped form are drawn, and a disk of rotating
particles of equal mass is centered at (0, 0). The particles are distributed by follow
ing the inverse law for the density ρ given by ρ =
c
1
r
3
. The diameter of the disk is
chosen as to be exactly within the limits of the eggshaped form. The parameters are
a = 0.25 and GM
0
= 0.45. The rotation angle is given by
∆θ =
_
GM
0
r
3
The resulting form that the disk of rotating particles takes after time t = 10 appears
in Figure 8.10(b). The original cyclic formhas nowchanged, providing an outer form
of rotation and an inner chaotic attractorlike object.
Figures 8.10(c) and 8.10(d) illustrate the resulting picture after time t = 20 and
t = 100 respectively. The distinct inner attractor is more clearly formed. In the case
presented in Figure 8.10(d), the outer part of the rotating object extends to almost the
entire space of the shape, but by following very speciﬁc characteristic paths.
In the second case, the translation parameter is high enough so that all of the
eggshaped space is contained in the chaotic region. The system is unstable and
the rotating particles are not retained inside the eggshaped region. On the con
trary, they escape by following the escape trajectories and move away from the egg
shaped region. After a while, the majority of the particles will leave the region. Fig
ures 8.11(a), 8.11(b) and 8.11(c) illustrate three instances, in times t = 2, t = 3 and
t = 4 respectively. The translation parameter is a = 0.6. The cross in Figure 8.11(a)
is at (x, y) = (a, 0). A cross indicates the characteristic centres of the chaotic forms.
These coordinates are calculated by using as starting values in a repeated procedure
x = a and y = 0. The iterative procedure is based on the diﬀerence equations for x
and y.
The case when t = 10 appears in Figure 8.12. The rotating system of particles
is in an intermediate stage. It escapes from the eggshaped form, but a part of the
system remains around the location (x, y) = (a, 0) inside the eggshaped form. When
172 Chaotic Modelling and Simulation
(a) t = 0 (b) t = 10
(c) t = 20 (d) t = 100
FIGURE 8.10: Rotating particles
(a) t = 2 (b) t = 3 (c) t = 4
FIGURE 8.11: Development of rotating particles over time (a = 0.6).
Rotations 173
the magnitude of the translation parameter takes higher values, all the particles es
cape after a short interval (Figure 8.12(a)). An intermediate stage appears in the
Figure 8.12(b), with parameter a = 0.7 and time t = 10. The leaflike structure starts
to disappear inside the eggshaped form when close to (x, y) = (a, 0). These jets eject
material through the trajectories of the model. The speed of these jets depends on the
magnitude of the parameter a and on the original rotation speed. The speed and the
direction of the jets of material coincides to a after a large enough amount of time t.
Figure 8.12(c) illustrates the case when a = 0.8. Only the last part of the jet remains
connected to the source of the chaotic sea inside the eggshaped pattern.
(a) a = 0.6 (b) a = 0.7 (c) a = 0.8
FIGURE 8.12: Development of rotating particles according to the magnitude
of the translation parameter. The elapsed time is t = 10.
When the translation parameter is a = 1 (t = 10), all the rotating material escapes
outside the eggshaped pattern. In Figure 8.13(a), the original cyclic disk is quite
large, and a part of this disk lies outside the eggshaped formation. However, all
the material follows the escape route as a compact formation. On the other hand,
in Figure 8.13(b), the inﬂuence of a small translation parameter (a = 0.25) to an
original large cyclic cloud has a critical impact to the cloud formation. After t = 100,
the cloud is separated into two formations, the ﬁrst inside the eggshaped form with
the chaotic bulgeform in the middle, and the second outside.
8.6 Rotations Following an Inverse Square Law
We return to the original system of diﬀerential equations
˙ x = a − y∆θ
˙ y = x∆θ
(8.19)
174 Chaotic Modelling and Simulation
(a) a = 1, t = 10 (b) a = 0.25, t = 100
FIGURE 8.13: Rotations
and the resulting diﬀerential equation for x, y:
dy
dx
=
x∆θ
a − y∆θ
This equation leads to the following form
ady = (∆θ)rdr
The solution of this diﬀerential equation depends on the form of the function ∆θ. In
this section, we use a function known from mechanics, that deals with the rotation
angle. This function arises from the law related to the transverse component of the
acceleration in a circular movement. The function is expressed by the formula
˙
θ =
c
1
r
2
,
where c
1
is a constant. Provided that ∆t = 1, the following approximation for ∆θ is
obtained
∆θ ≈
c
1
r
2
Finally, we proceed to the solution of the equation:
ady =
c
1
r
dr
The solution is
c
1
ln(r) = ay + h (8.20)
where h is an integration constant.
This last equation can be transformed in the following form:
c
1
2
ln(x
2
+ y
2
) = ay + h (8.21)
Rotations 175
Exploring the properties of this last equation, we set x = 0 when y = r. In this
case the maximum or minimum values of y are obtained. The resulting equation for
y is
c
1
ln(y) = ay + h
The maximum value for y is achieved when y =
c
1
a
. Then, the value of the para
meter h at this limit is
h
crit
= c
1
_
ln
c
1
a
− 1
_
This value of h
crit
, applied to equation (8.21), gives an equation for the path of a
trajectory in the (x, y) plane, which divides the plane in two segments. This trajectory
is the outer limit of the vortex region of the rotation. When h > h
crit
, the trajectories
diverge and the rotating object heads oﬀ to inﬁnity. When h < h
crit
, the object rotates
inside the limits set by the above trajectory.
FIGURE 8.14: The characteristic trajectory
Some very interesting properties of the twodimensional function are illustrated in
Figure 8.14. The trajectory in the limit of escape is shaped as an egg. The sharper
corner is at the maximum value of y =
c
1
a
where x = 0. These values of x, y set the
derivatives ˙ x, ˙ y equal to zero. However, this point is not stable. This maximum value
is obtained by solving the following equation for y:
ln
_
ay
c
1
_
−
ay
c
1
+ 1 = 0
The minimum value of y is obtained when the rotation speed is maximum. The
resulting equation for y is:
ln
_
ay
c
1
_
+
ay
c
1
+ 1 = 0
A numerical solution gives y = 0.278
c
1
a
.
176 Chaotic Modelling and Simulation
The maximum x
max
is estimated by equating to zero the ﬁrst derivative of the
following equation for x, y:
c
1
2
ln(x
2
+ y
2
) = ay + c
1
ln
_
c
1
a
_
− c
1
After appropriate diﬀerentiation, a numerical solution of the resulting equation gives
x
max
= 0.402
c
1
a
. This is achieved at y = 0.203
c
1
a
.
When y = 0, then x =
c
1
ae
. The tangent at this point is
dy
dx
= e. This tangent appears
in Figure 8.14.
The tangent at the top sharp corner of the eggshaped form can be computed as
before, using Taylor approximation of the curves in a neighbourhood of the point
_
0,
c
1
a
_
. Setting z = y −
c
1
a
, equation (8.21) becomes:
c
1
a
ln
_
¸
¸
¸
¸
_
x
2
+ z
2
+ 2
c
1
a
z +
c
2
1
a
2
_
¸
¸
¸
¸
_
= az + c
1
+ h (8.22)
Using the expansion
ln (a + x) = ln(a) +
x
a
−
1
2
x
2
a
2
+ · · ·
where a =
c
2
1
a
2
and x = x
2
+ z
2
+ 2
c
1
a
z, we see that equation (8.22) becomes:
az + c
1
+ h =
c
1
2
ln
_
¸
¸
¸
¸
_
c
2
1
a
2
_
¸
¸
¸
¸
_
+
c
1
2
a
2
c
2
1
_
x
2
+ z
2
+ 2
c
1
a
z
_
−
c
1
2
a
4
2c
4
1
_
x
2
+ z
2
+ 2
c
1
a
z
_
2
+ · · ·
Finally, after simpliﬁcation, this becomes:
0 =
c
1
2
a
2
c
2
1
_
x
2
− 2z
2
+ · · ·
_
In other words, the two tangent lines at the point (x, z) = (0, 0) are given by the
equation:
x
2
= 2z
2
In Figures 8.15(a) and 8.15(b), important properties of the model are presented.
In Figure 8.15(a), the continuous model presented above is simulated. On the other
hand, in Figure 8.15(b), the discrete analogue of the model appears. Figure 8.15(b)
illustrates the central bulge in the case of a = 2.7 and c
1
= 100. The value of
the translation parameter is quite high. This causes the appearance of an enormous
chaotic attractor in the middle of the eggshaped area. As in the case discussed
previously, strong bifurcation is present.
As is illustrated in Figure 8.15(a), the particle follows an anticlockwise direction.
It starts from a point near zero, and it rotates away from the centre until the highest
Rotations 177
(a) The continuous case (b) The central chaotic bulge
FIGURE 8.15: Comparison of the continuous and the discrete models
permitted place. Then, it escapes to inﬁnity following the tangent on the top of the
eggshaped path. The iterative formula for the above simulations is the following:
x
n+1
= x
n
+
_
a − y
n
c
1
r
2
_
d
y
n+1
= y
n
+ x
n
c
1
r
2
d
r
n
=
_
x
2
n
+ y
2
n
(8.23)
The parameters in both ﬁgures are a = 2.7, c
1
= 100 and d = 0.0001.
178 Chaotic Modelling and Simulation
Questions and Exercises
1. Show that the maximum speed in the rotating mass of system (8.7) is obtained
exactly when
y = −
MG
a
2
, x = 0
2. Show that for h >
MG
a
the orbits of system (8.7) go to inﬁnity.
3. Consider the system (8.14), expressed via the simpler form (8.15). Show, by
considering the resulting system for z
n
=
f
n
−a
ab/(1−b)
, that if f
n
is within the circle
centred at (a, 0) and with radius
ab
1−b
, then f
n+1
is also within that circle. What
happens with point outside this circle?
4. Derive equation (8.17) for the ﬁxed points of the system (8.14).
5. Show that if z is a complex number, its reﬂection along the line x =
a
2
will be
a − ¯ z.
6. Construct the level curves in the x, y plane of equation (8.9) for h = 0, and
varying values of a. What is the form of the curves when a = 0?
7. Use the equations (8.7) where
∆θ =
_
MG
a
3
to determine the rotation speed in the x and y directions for the level curves
examined in the previous question.
8. Construct the levels curves in the x, y plane of equation (8.9) when h =
2MG
a
,
for various values of a.
9. Draw the level curves (x, y) of equation (8.18) for various values of β, and
especially for β = 1.
10. Following the methodology of section 8.6, draw escape paths of particles
through an eggshaped form.
11. Construct a rotating model providing an eggshaped form, and draw a chaotic
bulge and escape paths for both small and large values of the parameter a.
12. Explain why the continuous case of Figure 8.15(a) does not show any chaotic
behaviour.
13. Draw trajectories for the discrete case of Figure 8.15(a).
Chapter 9
Shape and Form
9.1 Introduction
About 2300 years ago, Euclid of Alexandria wrote his famous book on geometry
called The Elements. This book, and its famous ﬁve postulates, became the basis of
what is now known as Euclidean geometry. The usual notion of distance in space,
essential when we design geometric forms in two or three dimensions, carries his
name (the “Euclidean metric”).
The development of science throughout the centuries brought new theories and
improvements in all scientiﬁc ﬁelds. Nevertheless, the Euclidean metric remains
unchanged, and still plays an essential part in a bevy of topics. The theory of the
simulation of chaotic shapes included in this book is based on an analytic approach
that for the most part uses the fundamental notions of traditional Euclidean geom
etry. The essential components of this theory, namely the notions of translation,
rotation and reﬂection, are deﬁned and explored in this chapter, along with applica
tions related to the development of chaotic forms and shapes.
As one can readily verify, a large number of chaotic objects that appear in the
literature may be classiﬁed as geometric objects produced by following simple geo
metric rules. Even nonchaotic objects are similarly simulated by following the same
geometric rules.
Translation, rotation and reﬂection are not only tools essential in forming geomet
ric objects, but can also be used as a basis of understanding how a process, chaotic
or not, can generate shapes and forms. The linear movements of physical objects
are easily modelled geometrically by using translation. Rotation and translation are
present in stellar systems, galaxies, tornados and vortex movements. Reﬂection is
generally associated with light refraction, galactic object formation, as well as vor
tex formation.
This chapter includes a brief introduction to the elementary rules of analytical ge
ometry that are used throughout this book, along with several applications on the
shape and form of chaotic and nonchaotic objects. This theoretical treatment is
combined with applications to special relativity and Penrose’s tiling theory, via sim
ulations based on relatively simple aﬃne transformations. Interested readers may
ﬁnd related topics in books on hyperbolic geometry.
179
180 Chaotic Modelling and Simulation
9.1.1 Symmetry and plane isometries
Any subset of the plane is called a ﬁgure. It can be easily observed that some
ﬁgures are more interesting than others. Figures with a high degree of symmetry
are most interesting, not only because of interesting forms arising in simulations, but
also because they often occur in nature. Spiral galaxies, snowﬂakes, molecules and
crystals are examples of objects with symmetric cross sections.
By symmetry for a ﬁgure A, we mean simply a transformation that preserves the
ﬁgure. The simplest kind of symmetry in a plane ﬁgure is that of symmetry with
respect to a line. Two points X and X
′
are symmetrical with respect to a line l, if the
line l bisects and is perpendicular to the segment XX
′
. We will also say that X
′
is the
reﬂection of X along ℓ (Figure 9.1).
O
ℓ
X
X
′
FIGURE 9.1: Symmetry/reﬂection along a line
An isometry is a transformation which preserves distances; i.e. it is a transforma
tion
T : R
2
→ R
2
with the property that the distance from T(A) to T(B) is the same as the distance
from A to B. A standard theorem in Euclidean geometry is that
every planar isometry is completely determined by its eﬀect on three
noncollinear points.
The proof is fairly straightforward and instructive: Since distances are preserved,
the circle centred at A and with a given radius is mapped to the circle centred at
TA, and with the same radius. If we consider the three circles through D, and with
centres A, B, C respectively, and look at their images, then TD must be the point of
intersection of these three circles (Figure 9.2).
Isometries respect the two fundamental aspects of geometry: the incidence aspect
based on the notion of collinearity, and the metric aspect based on the notion of
distance. The fact that isometries respect collinearity is left as an exercise.
There are four basic kinds of planar isometries:
Shape and Form 181
A TA
B
TB
C
TC D
TD
FIGURE 9.2: Planar isometries are determined by their eﬀect on three points
Translations These are fully described by a single vector v. Such a transformation
is given simply by:
1
T w = w +v
Rotations These correspond to a rotation with a given angle and around a speciﬁc
point. Assuming the rotation is around the origin, the transformation is given
simply by the matrix:
rot θ =
_
cos θ − sin θ
sin θ cos θ
_
in other words:
T w = rot(θ)
_
x
y
_
=
_
x cos θ − y sin θ
x sin θ + y cos θ
_
Reﬂections A reﬂection across a line ℓ passing through the origin and forming angle
θ with the x axis is given by the matrix
ref θ =
_
cos 2θ sin 2θ
sin 2θ −cos 2θ
_
and therefore it has the form:
T w = ref(θ)
_
x
y
_
=
_
x cos 2θ + y sin 2θ
x sin 2θ − y cos 2θ
_
Glide Reﬂections A glide reﬂection is a translation followed by a rotation along
a line parallel to the translation direction. The translation and rotation can
be performed in either order. An example of a glide reﬂection is given in
ﬁgures 9.3(a) and 9.3(b). You can see that a glide reﬂection can be thought of
also as the composition of a rotation followed by a reﬂection (or equivalently,
three reﬂections along lines that do not all have a point in common).
182 Chaotic Modelling and Simulation
ℓ
A
B
O
TO
TA
TB
(a) Glide reﬂection as a rotation, fol
lowed by a reﬂection
ℓ
′
A
B
O
TO
TA
TB
(b) Glide reﬂection as a translation, followed
by a reﬂection
FIGURE 9.3: Glide reﬂection seen in two diﬀerent ways
The matrices rot θ, ref θ have a series of useful properties, which we summarise
here:
rotθrotφ = rot(θ + φ) (9.1)
rot(0) = I (9.2)
refθrefφ = rot(2(θ − φ)) (9.3)
refθrotφ = ref(θ − φ/2) (9.4)
rotφrefθ = ref(θ + φ/2) (9.5)
refθrefφrefy = ref(θ − φ + y) (9.6)
We leave the veriﬁcation of these identities to the reader.
It is worth pointing out that the composition of two reﬂections with the same centre
is a rotation with angle twice the angle between the two reﬂection lines:
ref θ ref φ =
_
cos 2(θ − φ) −sin 2(θ − φ)
sin 2(θ − φ) cos 2(θ − φ)
_
= rot 2(θ − φ)
For a geometric proof of this fact, consider the eﬀect of the combined transformation
on the points O, A, B in Figure 9.4(a). The ﬁrst reﬂection, along the line ℓ
1
, takes the
points O, A, B to the points O, A, B
′
,
2
and the second, along the line ℓ
2
, takes those
1
Here, as well as in the rest, we will identify a point P(x, y) with the vector w starting at the origin
and ending at (x, y).
2
Since a reﬂection ﬁxes all points on the reﬂection line.
Shape and Form 183
points to the points O, A
′
, B
′
. The same eﬀect would be obtained by the aforemen
tioned rotation.
At this point, you might be wondering, what happens when we take the composi
tion of two reﬂections along parallel lines. The result is a translation with a vector
twice the distance between the two lines (Figure 9.4(b)).
O
ℓ
1
ℓ
2
A
B
′
B
A
′
(a) The composition of two re
ﬂections along intersecting lines
is a rotation around the intersec
tion
ℓ
1
ℓ
2
A
B
C
A
′′
= A
′
B
′
B
′′
C
′′
= C
′
(b) The composition of two reﬂections
along parallel lines is a translation
FIGURE 9.4: The composition of two reﬂections
A very striking result is that these four types, along with the identity, are all the
plane isometries, namely:
1. Every plane isometry is a composition of reﬂections.
2. Every composition of reﬂections in the plane can be performed
with at most three reﬂections.
The relation between isometry groups and symmetries is given by the following
theorem, already known to Leonardo da Vinci:
All ﬁnite isometry groups can occur as symmetry groups of ﬁgures, and
cannot contain translations or glide reﬂections.
We ﬁnish this section with a brief discussion of the points and shapes left invariant
by the various isometries. A shape/set S is said to be left invariant by a transforma
tion T, if the sets T(S ) and S coincide (i.e. if the transformation preserves S as a
whole, though it might move the points of S around). We omit the identity, which
leaves everything ﬁxed.
Points Translations and glide reﬂections have no ﬁxed points. A rotation ﬁxes its
centre, while a reﬂection ﬁxes every point on the line of reﬂection.
184 Chaotic Modelling and Simulation
Lines A translation along a line ℓ leaves that line, as well as any line parallel to it,
invariant. A glide reﬂection leaves its line of reﬂection invariant. A rotation
has no ﬁxed lines, unless it is a rotation by 180
◦
, in which case it leaves any
line passing through the centre of rotation invariant. Finally, a reﬂection leaves
invariant the line of reﬂection, as well as any line perpendicular to it.
9.2 Isometries in Modelling
We now focus our attention on maps that make heavy use of isometries in their
construction.
9.2.1 Twodimensional rotation
Rotation in a plane is the most frequently arising case in chaotic modelling. Even
the three dimensional case of rotation, in many cases, is reduced to rotation in the
(x, y) plane around an axis of rotation perpendicular to this plane, after an appropriate
change of coordinates.
A map arising from rotation in the plane is expressed by:
_
x
n+1
y
n+1
_
= rot θ
n
_
x
n
y
n
_
=
_
x
n
cos θ
n
− y
n
sin θ
n
x
n
sin θ
n
+ y
n
cos θ
n
_
(9.7)
Orbits of the above map stay in a circle of radius r =
_
x
2
0
+ y
2
0
, where (x
0
, y
0
) is
the set of initial values. This can be easily veriﬁed from the relation r
n+1
= r
n
, and
is independent of the form of the rotation angle θ
n
. Every system of this type will
continue with circular movements without any change in the circular path.
When a translation is added to rotation, things change radically. Without loss of
generality, we will consider the translation as a single parameter a added in the x
direction. The resulting map is given by:
_
x
n+1
y
n+1
_
= rot θ
n
_
x
n
y
n
_
+
_
a
0
_
=
_
a + x cos θ
n
− y sin θ
n
x sin θ
n
+ y cos θ
n
_
(9.8)
By the addition of a translation parameter, the radius r
n+1
is not generally equal to
its previous value r
n
, as is easily veriﬁed by the relation:
r
2
n+1
= r
2
n
+ a
2
+ 2a(x
n
cos θ
n
− y
n
sin θ
n
)
It is clear that r
n+1
is also a function of the rotation angle θ
n
. Also, another relation,
not directly containing the rotation angle, holds:
r
2
n+1
− 2ax
n+1
+ a
2
= r
2
n
Shape and Form 185
This relation, after summation, yields
n
1
(r
2
i+1
− r
2
i
) + na
2
= 2a
n
1
x
i
where i = 1, 2, . . . , n. Setting ¯ x =
1
n
_
n
1
x
i
, and after the appropriate cancellations, we
obtain:
2ax =
(r
2
n
− r
2
0
)
n
+ a
2
In the limit (n → ∞), assuming the radius r
n
is bounded, the following simple
relation arises:
x =
a
2
This is suggesting the possibility of symmetry around this axis in a rotation
translation iterative procedure (see for example Figure 9.5). This subject was dis
cussed more extensively in Chapter 8.
One can easily obtain various symmetric forms by choosing a function for the
rotation angle. Selecting nonlinear functions for the rotation angle is essential for
producing forms of high complexity, as well as chaotic forms. The ﬁgures in the
subsequent sections are chaotic and nonchaotic cases of rotationtranslation iterative
schemes.
FIGURE 9.5: A symmetric graph
186 Chaotic Modelling and Simulation
9.3 Reﬂection and Translation
Movement that follows reﬂection with translation is modelled as in the previous
rotationtranslation case. The map that expresses translationreﬂection has the gen
eral form:
_
x
n+1
y
n+1
_
= ref θ
n
_
x
n
y
n
_
+
_
a
0
_
=
_
a + x cos 2θ
n
+ y sin 2θ
n
x sin 2θ
n
− y cos 2θ
n
_
(9.9)
It is easily veriﬁed that the reﬂectiontranslation map (9.9) provides symmetric
iterative forms with a symmetry axis perpendicular to the x axis at a distance equal
to a/2 from the origin. Some illustrative examples follow.
9.3.1 Space contraction
The iterative procedures presented above using rotation, reﬂection and translation
are based on a stable distance and space metric. At a ﬁrst glance, space contraction
or expansion could be a case of speciﬁc interest in relativistic problems in cosmol
ogy and elsewhere. However, several other very interesting cases appear. Many
chaotic objects are formed by applying space contraction rules along with rotation,
translation and reﬂection.
Space contraction or expansion appears in vortex formation in chaotic advection or
in chemical reactions. The simplest model uses a parameter b by which the distance
equations are multiplied. In other words, in every iteration we would have:
r
n+1
= br
n
In the case of rotationtranslation, the iterative formula takes the form
_
x
n+1
y
n+1
_
= b rot θ
n
_
x
n
y
n
_
+
_
a
0
_
=
_
a + bx
n
cos θ
n
− by
n
sin θ
n
bx
n
sin θ
n
+ by
n
cos θ
n
_
(9.10)
The system (9.10) has a Jacobian determinant of J = b
2
, which means a space
contraction or expansion proportional to b
2
. Space expansion is present for example
in supernova explosions, but also in weather modelling and wave propagation. On
the other hand, space contraction is more interesting from the point of view of sta
bility and form generation. Contraction is counterbalanced by the presence of either
rotation or translation. In the case of contraction, the iterative spiraling sink towards
a point of attraction is replaced, due to rotation and other balancing mechanisms, to
a movement inside an attracting space, the “attractor.”
It is interesting that the relativistic space contraction may lead to the same iterative
scheme as (9.10). This is easily demonstrated by observing the relativistic space
equations, that is
x
t+1
= x
t
_
1 −
v
2
c
2
= bx
t
y
t+1
= y
t
_
1 −
v
2
c
2
= by
t
(9.11)
Shape and Form 187
where
b =
_
1 −
v
2
c
2
It is evident in the previous formulae, that relativistic space contraction can be ob
served for relatively high speeds. However, as demonstrated in Chapter 12, changes
can appear in chaotic images even in lower speeds.
9.4 Application in the Ikeda Attractor
A very interesting approach regarding rotation and reﬂection along with transla
tion and space contraction appears in the ﬁgures below. The socalled Ikeda model
and the famous associated attractor express the rotationtranslationspace contraction
example (Ikeda, 1979; Ikeda et al., 1980). The iterative scheme for rotation is:
_
x
n+1
y
n+1
_
= b rot θ
n
_
x
n
y
n
_
+
_
a
0
_
Ikeda proposed the following function for the rotation angle:
θ
n
= c −
d
1 + x
2
n
+ y
2
n
(9.12)
where c and d are rotation parameters. The rotation angle θ is a function of the square
of the distance from the origin. At large distances, the system rotates with an angle
θ ≈ c, whereas, close to the origin, the rotation angle is θ ≈ c − d. The classical
Ikeda attractor is obtained when a = 1, b = 0.83, c = 0.4 and d = 6 (Figure 9.6).
FIGURE 9.6: The Ikeda attractor
188 Chaotic Modelling and Simulation
The reﬂectiontranslation map with a similar function to (9.12) for the reﬂection
angle is expressed by the iterative scheme:
_
x
n+1
y
n+1
_
= b ref θ
n
_
x
n
y
n
_
+
_
a
0
_
where the reﬂection angle is given by:
2θ
n
= c +
d
1 + x
2
n
+ y
2
n
The Jacobian determinant in this case is negative, J = −b
2
. The reﬂection attrac
tor, presented in Figure 9.7, has two separate images. The ﬁgure on the right looks
like a mirror image of the Ikeda attractor. The parameters for this simulation were
a = 2.2, b = 0.9, c = 1.5 and d = 2.15.
FIGURE 9.7: The reﬂection Ikeda attractor
9.5 Chaotic Attractors and RotationReﬂection
Both rotation and reﬂection generate chaotic attractors for appropriate values of
the parameters a, b, c, and d. Two important questions are:
a) Are there parameter values that generate chaotic attractors in both cases?
b) If there are, then what are the interrelations between rotation and reﬂection at
tractors?
Shape and Form 189
FIGURE 9.8: Comparing rotationreﬂection cases
In Figure 9.8, we see four combinations for rotation, A(α, θ), B(−α, θ), C(α, −θ)
and D(−α, −θ), compared to the four cases for reﬂection, E(α, θ), F(−α, θ), G(−α, θ)
and H(−α, −θ). A rotationreﬂection iterative scheme was used, just as in the pre
ceding section, but with a rotationreﬂection angle given by the more simple formula
θ
n
= c
_
x
2
n
+ y
2
n
.
In reﬂection, the parameter 2θ
n
was set by the same formula. The parameters were
a = 1, b = 0.68 and c = 10/3. Figure 9.9 illustrates the ﬁrst two graphs to the left of
Figure 9.8.
FIGURE 9.9: Comparing rotationreﬂection case
Another reﬂectiontranslation chaotic attractor is formed by selecting a simple
quadratic function for the reﬂection parameter θ, that is
190 Chaotic Modelling and Simulation
2θ
n
= c
_
x
2
n
+ y
2
n
.
The parameters for the simulation in Figure 9.10 were a = 1, b = 0.68 and c = 10/3.
FIGURE 9.10: A reﬂection chaotic attractor resulting from a quadratic
rotation angle
The same reﬂection model provides two other attractors, as illustrated in Fig
ure 9.11. Both ﬁgures are obtained for a = 1 and b = 0.68, whereas the other
parameter is c = 1 for the image on the left and c = 2 for the image on the right.
FIGURE 9.11: Two other reﬂection attractors from a quadratic rotation angle
Shape and Form 191
9.6 Experimenting with Rotation and Reﬂection
9.6.1 The eﬀect of space contraction on rotationtranslation
In order to explore the eﬀect of space contraction due to the introduction of a
parameter b (b = 0.8 < 1), consider the following rotationtranslation scheme:
x
n+1
= a + b(x
n
cos θ
n
− y
n
sin θ
n
)
y
n+1
= b(x
n
sin θ
n
+ y
n
cos θ
n
)
(9.13)
where the rotation angle is
θ
n
= c +
d
x
2
n
+ y
2
n
The Jacobian determinant in this case is J = b
2
= 0.64. The translation parameter
is a = 6, the space contraction parameter is b = 0.8 and the rotation parameters
are c = a/2 and d = a. The original particles are located in a circle of radius 6
centred at (x, y) = (a/2, 0). The particles are attracted into the two chaotic attractors
(Figure 9.12(b)) inside the circle. The time development of the process is illustrated
in Figure 9.12(a). The outer circle represents the original particles at time t = 0. At
time t = 1, the particles form an ellipsoid form, whereas at time t = 2, the outer
part of the ﬁrst chaotic attractor appears. According to this computer experiment the
approach to the chaotic state is quite rapid. However, the parameters of the model
used are high and the model expresses a strong chaotic process.
(a) The space contraction eﬀect (b) Space contraction and chaotic images
FIGURE 9.12: Space contraction
192 Chaotic Modelling and Simulation
9.6.2 The eﬀect of space contraction and change of reﬂection
angle on translationreﬂection
A translationreﬂection model is introduced and applied in the following computer
simulation. The set of equations expressing this model are
x
n+1
= a + b(x
n
cos θ
n
+ y
n
sin θ
n
)
y
n+1
= b(x
n
sin θ
n
− y
n
cos θ
n
)
(9.14)
where the rotation angle is:
θ
n
= c +
d
1 + x
2
n
+ y
2
n
The original set of parameters is a = 2.2, b = 0.9, c = 3. The parameter d varies,
taking the following values: 32, 16, 8, 6, 4.78 and 4.30 respectively. Six chaotic
forms of the reﬂectiontranslation scheme are presented here. The original form
appears on the top left of Figure 9.13, and the other images follow from left to right
in the top row and from left to right in the second row. We see that, as the parameter
d decreases, the original object is gradually divided into two distinct objects.
FIGURE 9.13: The eﬀect of the reﬂection parameter on chaotic images
The inﬂuence that a change in the space parameter b has on a chaotic system is
examined in the following reﬂectiontranslation scheme. The original object results
when a = 2.2, c = 3, d = 4.3 and b = 1, whereas for the other ﬁve objects the space
contracting parameter decreases, taking the values 0.99, 0.975, 0.94, 0.93 and 0.9
respectively. The original form appears on the top left of Figure 9.14, and the other
images followfromleft to right in the top rowand fromleft to right in the second row.
Once again, as b decreases, the original object is divided into two distinct objects,
but in this case following a process diﬀerent from that illustrated in Figure 9.13.
Shape and Form 193
FIGURE 9.14: The eﬀect of space contraction on chaotic images
9.6.3 Complicated rotation angle forms
More complicated chaotic forms arise by selecting a rotationtranslation scheme
with rotation angle that follows a more complicated equation. In this case, the ro
tation angle is a function of the inverse of the radius r =
_
x
2
+ y
2
plus a distance
parameter c and another rotation angle parameter m. The area contraction parameter
is set to b = 0.93. The function for the rotation angle is
θ
n
= −
1
_
c
2
+ x
2
+ y
2
−
m
_
c
2
+ x
2
+ dy
2
The parameters for the simulation were set to a = 0.74, c = 0.5, d = 1.6 and
m = 3.1. The resulting chaotic attractor is illustrated in Figure 9.15. This is a
more complicated chaotic attractor that could result in chaotic ﬂows and in galactic
formations.
FIGURE 9.15: The eﬀect of a complicated rotation angle on chaotic attractors
194 Chaotic Modelling and Simulation
A simpler form is produced by assuming that in the function for the rotation angle
the parameter c is 0. The resulting chaotic system provides an attractor expressing
ﬂows appearing in the mixing of ﬂuids but, also, in other rotationtranslation systems.
The parameters in Figure 9.16 were a = 1, b = 0.87, d = 1 and m = 3.1.
FIGURE 9.16: The eﬀect of a complicated rotation angle on chaotic attractors
A twopiece chaotic attractor is formed by using a rotationtranslation formula
with a special rotation angle. The rotation angle has a constant term denoted by c and
a varying term related to a transformed distance. The model is a space contracting
one (b = 0.9). It also has a slightly diﬀerent form for the rotationtranslation
x
n+1
= a + b(x
n
cos θ
n
− y
n
sin θ
n
)
y
n+1
= −a + b(x
n
sin θ
n
+ y
n
cos θ
n
)
(9.15)
where the rotation angle is
θ
n
= c +
d
_
x−a
b
_
2
+
_
y+a
b
_
2
The other parameters were set to a = 9, c = a and d = 6. The twopiece chaotic
attractor is presented in Figure 9.17(a).
9.6.4 Comparing rotationreﬂection
Figure 9.17(b) illustrates a twopiece chaotic attractor, resulting froma set of equa
tions expressing rotation and translation without space contraction (b = 1). The
rotation angle is a function of r, that is
θ
n
= c +
d
_
x
2
n
+ y
2
n
.
Shape and Form 195
(a) With space contraction (b) Without space contraction
FIGURE 9.17: Twopiece chaotic attractors
The process shows how an original rotating disk of particles can form two distinct
symmetric chaotic forms. The parameters were set to a = 37, c = 2.85 and d = 15.5.
FIGURE 9.18: Reﬂection: a twopiece chaotic attractor
The images presented in Figure 9.18 are the reﬂection analogue of this system.
The reﬂectiontranslation equations without space contraction (b = 1) are used. The
reﬂection angle (θ
re f
) is as in the previous rotationtranslation case but it is related
to the rotation angle (θ
rot
) by the relation θ
re f
= 2θ
rot
. When the parameters take the
values a = 80, c = 5 and d = −20, the reﬂectiontranslation model provides two
objects with mirror symmetry, as illustrated in Figure 9.18.
196 Chaotic Modelling and Simulation
9.6.5 A simple rotationtranslation model
FIGURE 9.19: A simple area preserving rotationtranslation model
Asimple space preserving (b = 1) rotationtranslation model with a rotation angle,
varying according to the square of the distance r from the origin (θ
n
= c(x
2
n
+ y
2
n
) =
cr
2
), also gives an interesting rotationtranslation image, illustrated in Figure 9.19.
There is an island with equilibrium points achieved when (x
n+1
= x
n
, y
n+1
= y
n
),
centred on the symmetry axis (x = a/2), and four symmetric islands (in two pairs)
with equilibrium points that follow the rule x
n+4
= x
n
and y
n+4
= y
n
. The parameters
were set to a = 1 and c = 0.65.
9.7 Chaotic Circular Forms
The study of a rotation process in which the rotating particles move at circular
or other paths from the centre following a speciﬁc formula is of particular interest.
The simulations arising out of these cases start by forming concentric circles with
diﬀerent particle density and can lead to more complicated forms, depending on the
selected set of nonlinear functions. The iterations follow the simple scheme:
x
n
= r
n
cos θ
n
y
n
= r
n
sin θ
n
(9.16)
where θ
n
increases in a constant way:
θ
n+1
= θ
n
+ θ
Shape and Form 197
Depending on the iterative scheme for the radius r
n
, interesting patterns arise, along
with chaotic behaviour. As a simple example, consider the iterative formula:
r
n+1
= b −
a
r
n

The (a, r) bifurcation diagram for this model is shown in Figure 9.20(a). Second,
third, fourth and higher order bifurcations appear. The parameter b is set to 2.4,
while the parameter a varies in 1.2 < a < 6.9.
Figure 9.20(b) illustrates the bifurcation diagram (a, r) of the model expressed by
the formula:
r
n+1
= b −
a
r
2
n
The parameter b is set to 2.4, while the parameter a varies in 0.2 < a < 60.
Figure 9.20(c) illustrates the (a, r) bifurcation diagram of the model:
r
n+1
= b −
a
r
4
n
The parameter b is set to 2.4 and the parameter a varies in 0.2 < a < 60.
The development of the theory of rotation with varying rotation distance from the
origin, or with varying rotation angle related to the distance r =
_
x
2
+ y
2
, is partly
explained by using Figure 9.21(a), where OC = r.
A very interesting rotation scheme is presented in Figure 9.21(b). There is only
one parameter b = 1.195. The radius r follows an inverse square law:
r
n+1
=
1
(r
n
− b)
2
(9.17)
Chaotic behaviour appears when the parameter b varies. A small change in b leads
to the onset of chaos. For b = 1.204, the bifurcation has already started to divide the
outer ring in two rings (Figure 9.21(c)). The bifurcation process leads to chaos when
higher values for b are selected.
In Figure 9.21(d), there appear the distinct circles. By giving to the chaotic
parameter b the value 1.255, the particles are distributed in a particular chaotic
way. There are cyclical regions with higher or lower densities, as presented in Fig
ure 9.21(e).
Figure 9.21(f) presents the bifurcation diagram (b, r) for the model (9.17). The
parameter b varies between 0 < b < 2.2. The model gives two values for r in the
region of 0 < b < 0.629 · · ·. At 0.629 · · ·, as b increases, a second bifurcation starts,
leading to a chaotic region. At the end of this region, three values for r appear,
making up a new chaotic region, then four values, and so on until a totally chaotic
region at 1.88 · · ·.
198 Chaotic Modelling and Simulation
(a) r
n+1
= b −
a
r
n

(b) r
n+1
= b −
a
r
2
n
(c) r
n+1
= b −
a
r
4
n
FIGURE 9.20: Bifurcation diagrams
Shape and Form 199
(a) The geometry of circular chaotic forms (b) A circular form
(c) The ﬁrst bifurcation stage (b = 1.204) (d) b = 1.21
(e) b = 1.255 (f) Bifurcation diagram for the model (9.17)
FIGURE 9.21: Circular chaotic forms
200 Chaotic Modelling and Simulation
9.8 Further Analysis
Consider a rotationtranslation case expressed by the relations
x
n+1
= a + x
n
cos θ
n
− y
n
sin θ
n
y
n+1
= x
n
sin θ
n
+ y
n
cos θ
n
(9.18)
or in the equivalent form
z
n+1
= a + z
n
e
iθ
n
where z
n
= x
n
+ iy
n
.
This map is symmetric with the line x = a/2 as the axis of symmetry. This is easily
demonstrated if we consider the ﬁxed points, i.e. solutions of the form z
n+1
= z
n
= z.
In other words, when (x
n+1
, y
n+1
) = (x
n
, y
n
) = (x, y). Now the following relation
holds:
z − a = z
or
(x − a)
2
+ y
2
= x
2
+ y
2
and, ﬁnally,
x =
a
2
whereas for y
y =
a
2
cot
θ
2
FIGURE 9.22: Geometric analogue of rotationtranslation (θ = 2)
A geometric analogue of this rotationtranslation case is illustrated in Figure 9.22
where the translation parameter is a = 2, the rotation angle is θ = 2 and the initial
Shape and Form 201
values are (x, y) = (0.9, 0.2). Figure 9.23(a) illustrates the same case, but now the
rotation angle is θ = π. In this case the point (x, y) = (a/2, 0) indicated by two
concentric small circles is the equilibrium point. An equilibrium point located at
(x, y) = (a/2, a/2) is presented in Figure 9.23(b) where the value of the rotation
angle is θ = π/2.
(a) θ = π (b) θ = π/2
FIGURE 9.23: Geometric analogues of rotationtranslation
The Jacobian determinant for the rotationtranslation case above is
J = 1 + x
n
∂θ
n
∂y
− y
n
∂θ
n
∂x
= 1 + x
n
θ
y
− y
n
θ
x
where
θ
x
=
∂θ
n
∂x
θ
y
=
∂θ
n
∂y
Analogously, the Jacobian determinant for the reﬂection case is of the form
J = −1 + xθ
y
− yθ
x
Clearly, the Jacobian for the rotationtranslation case is J = 1 when:
x
n
θ
y
= y
n
θ
x
This is the areapreserving case resulting when the rotation angle is expressed by any
function of the form
θ
n
= f
_
x
2
n
+ y
2
n
_
= f
_
r
2
n
_
202 Chaotic Modelling and Simulation
or by functions of the radius r
n
=
_
x
2
n
+ y
2
n
.
Another issue is to explore relations regarding the argument z
n+2
= z
n
or:
(x
n+2
, y
n+2
) = (x
n
, y
n
)
By applying the rotationtranslation equation above and the last relation yields:
x
n+1
+ x
n
= a
Clearly, when x
n
= a/2 then from the last relation x
n+1
= a/2. Both points coincide
as the relations (x
n+1
= a/2, y
n+1
) and (x
n
= a/2, y
n
) imply that y
n+1
= y
n
. The more
general case z
n+m
= z
n
yields a relation of the form:
x
n+m−1
+ x
n+m−2
+ · · · + x
n
= m
a
2
9.8.1 The space contraction rotationtranslation case
This case is expressed by the wellknown function
z
n+1
= a + bz
n
e
iθ
n
where 0 < b < 1 is the space contraction parameter. The Jacobian determinant of the
space contraction rotationtranslation case is:
J = b
2
(1 + x
n
θ
y
− y
n
θ
x
)
The same limit argument
(x
n+1
, y
n+1
) = (x
n
, y
n
) = (x, y)
will give the relation
(x − a)
2
+ y
2
= b
2
(x
2
+ y
2
)
or, after some algebra:
_
x −
a
1 − b
2
_
2
+ y
2
=
_
ab
1 − b
2
_
2
This is the equation of a circle with radius R =
ab
1−b
2
centred at (x, y) =
_
a
1−b
2
, 0
_
.
The resulting relations for x and y are:
x =
a(1 − b cos θ)
1 + b
2
− 2b cos θ
y =
ab sin θ
1 + b
2
− 2b cos θ
The equation for cos θ is:
cos θ =
1
2b
_
1 + b
2
−
a
2
r
2
_
When the angle θ is a function of r, the last equation provides the values of θ for
which z
n+1
= z
n
.
Questions and Exercises
1. Show that plane isometries preserve collinearity: Let A, B, C be three distinct
collinear points. Then use the fact that the distances between TA, TB, TC must
be the same as those between A, B, C respectively to show that TA, TB, TC
must also be collinear, and in fact appear in the same order (possibly reversed).
2. Verify the properties of rotations and reﬂections given in (9.6).
3. What circles are left invariant by each of the diﬀerent isometries?
4. Change the system (9.8) by adding a translation parameter a
1
to the second
equation, while retaining the translation parameter a in the ﬁrst equation. Find
the axis of symmetry for this new system.
5. Change the system (9.8) by adding a translation parameter −a to the second
equation, while retaining the translation parameter a in the ﬁrst equation. Find
the axis of symmetry for this new system.
6. In the system from the previous question, use relation (9.12) to ﬁnd a chaotic
attractor of the Ikeda type. You will likely have to use several values for the
parameters.
7. Repeat the previous question, but now using reﬂection equations.
8. Find the ﬁxed points of the relation:
r
n+1
=
1
(r
n
− b)
2
What is the smallest value of b for which there are three ﬁxed points? How
does the system behave when there is only one ﬁxed point?
9. Draw chaotic simulations for the circular system described by:
r
n+1
=
1
(r
n
− b)
2
10. Repeat the same analysis as in the two previous questions, but now using the
relations
r
n+1
= b −
a
r
2
n
203
204 Chaotic Modelling and Simulation
and
r
n+1
= b −
a
r
4
n
respectively. Try several values for the parameter a, and explore if there is a
relation between a and b related to the chaotic behaviour of the system.
Chapter 10
Chaotic Advection
10.1 The Sink Problem
Questions related to chaotic advection go back to the nineteenth century and the
development of hydrodynamics, especially the introduction of the NavierStokes
equations (Claude Navier, 1821 and George Stokes, 1845). The vortex ﬂow case
and the related forms including vortexlines and ﬁlaments, vortex rings, vortex pairs
and vortex systems can be found in the classical book by Horace Lamb, ﬁrst edited
in 1879 (Lamb, 1879). However, the formulation of a theory that partially explains
the vortex problem and gives results that coincide with the real life situations has
only been achieved in recent years, thanks to progress in computer simulations. The
introduction of terms like chaotic advection and the blinking vortex system appeared
only in last decades, in order to deﬁne and analyse speciﬁc vortex ﬂow cases.
In most cases, the formulation and solution of the problem followed the diﬀer
ential equations approach, mostly directed towards the solution of a NavierStokes
boundary value problem. Few interesting cases are based on diﬀerence equations
analogues. However, the formulation and analysis of vortex ﬂow problems by means
of diﬀerence equations can be very useful for several cases if a systematic study is ap
plied. In this chapter we follow the diﬀerence equations methodology by introducing
rotationtranslation diﬀerence equations and a nonlinear rotating angle, along with
a space contraction parameter, in order to study chaotic advection problems. The
interconnections between the diﬀerence and the diﬀerential equations case is also
studied in speciﬁc cases.
1
10.1.1 Central sink
Consider a circular bath with a sink in the centre at (x, y) = (0, 0). The water inside
the bath is rotating counterclockwise. A coloured ﬂuid is injected in the periphery
of the bath. The question we propose to address is to determine the shape of the
ﬂuid ﬁlaments if the sink is open. Geometrically the problem is that of rotation with
contraction by a parameter b < 1. The rotationtranslation model (9.10) is applied,
1
The ﬁrst three sections of this chapter have previously appeared in Skiadas (2007).
205
206 Chaotic Modelling and Simulation
with the translation parameter a being zero. The equations of ﬂow would then be:
x
t+1
= b(x
t
cos φ
t
− y
t
sin φ
t
)
y
t+1
= b(x
t
sin φ
t
+ y
t
cos φ
t
)
(10.1)
The contraction in the radial direction (r =
_
x
2
+ y
2
) is found from system (10.1):
r
t+1
= b
_
x
2
t
+ y
2
t
= br
t
(10.2)
The rotation angle is assumed to follow a function of the form
φ
t
= c +
d
r
2
t
. (10.3)
The space contraction is given by estimating the Jacobian of the ﬂow J = b
2
. When
b < 1, a particle is moving from the periphery of the bath to the sink in the centre
of coordinates following spirals, as illustrated in Figure 10.1(a). The parameters are
b = 0.85, c = 0, d = 0.4, and the initial point is (x
0
, y
0
) = (1, 0).
When the same case is simulated for particles entering from the periphery of the
rotating system at time t = 0, 1, 2, . . ., Figure 10.1(b) results. The spiral forms start
from the periphery and are directed towards the central sink. It is also interesting
that full concentric circles appear while the spiraling ﬂow continues. These circles
have a smaller diameter, or even disappear completely, when the rotation parameter
d gets smaller. The parameter b also inﬂuences the spiral. Figure 10.1(c) illustrates
an advection case when b = 0.95, c = 0 and d = 0.01.
(a) Spiral particle paths (b) Spiral forms directed
to the sink
(c) Spiral formation to
ward a central sink
FIGURE 10.1: Spiral forms
Chaotic Advection 207
10.1.2 The contraction process
From the rotationcontraction equations (10.1) and (10.2), it follows that the radial
contraction is:
∆r
t
= r
t+1
− r
t
= br
t
− r
t
= −(1 − b)r
t
The diﬀerential equation for the contraction process is found by observing that:
dr
dt
≈
∆r
∆t
=
r
t+1
− r
t
(t + 1) − t
= −(1 − b)r
The resulting diﬀerential equation expressing the radial speed
˙ r = −(1 − b)r
is solved to give
r = r
0
e
−(1−b)t
where r
0
is the initial radius.
The movement is then totally determined, as the equation for the rotation angle has
been given earlier. The paths form spirals towards the centre. When the movement
covers a full circle, the new radius will be:
r = r
0
e
−(1−b)
2π
φ
10.2 NonCentral Sink
In the following case, a circular bath with a noncentral sink is examined. The
sink is located at (x, y) = (a, 0). The equations of ﬂow are:
x
t+1
= b
_
(x
t
− a) cos φ
t
− y
t
sin φ
t
_
y
t+1
= b
_
(x
t
− a) sin φ
t
+ y
t
cos φ
t
_ (10.4)
The rotation angle is assumed to follow an equation of the form
φ
t
= c +
d
r
2
t
where r
t
=
_
(x
t
− a)
2
+ y
2
t
. The ﬁxed points for this map are given by the equation
x
2
+ y
2
= b
2
_
(x − a)
2
+ y
2
_
or, after transformation:
_
x +
ab
2
1 − b
2
_
2
+ y
2
=
_
ab
1 − b
2
_
2
208 Chaotic Modelling and Simulation
This is the equation of a circle with radius R =
ab
1−b
2
centred at (x, y) =
_
ab
2
1−b
2
, 0
_
.
The ﬂow is not symmetric. The coloured ﬂuid starting from the outer periphery of
the bath approaches the sink in few time periods, as illustrated in Figure 10.2(a). The
parameters are a = 0.15, b = 0.85, c = 0 and d = 0.1. To simplify the process, it is
assumed that the coloured ﬂuid is introduced simultaneously in the periphery of the
bath. Then, the circular form of the original coloured line is gradually transformed
into a chaotic attractor located at the sink’s centre (x, y) = (a, 0). The attractor is
quite stable in terms of form and location.
The attractor appears even if the coloured particles are introduced into a small
region of the bath, as presented in Figure 10.2(b). The coloured particles are intro
duced in a square region (0.1 × 0.1) at the right end of the bath at (x, y) = (1, 0). The
parameters are a = 0.15, b = 0.85, c = 0 and d = 0.8. As the vortex parameter d is
higher than in the previous case, the chaotic attractor appears at the 6th time step of
the process. The attractor is also larger than in the previous case.
(a) (b)
FIGURE 10.2: Chaotic attractors in a noncentral sink
10.3 Two Symmetric Sinks
10.3.1 Aref ’s blinking vortex system
Chaotic mixing in open ﬂows is usually modeled by using the blinking vortexsink
system, invented by Aref (see Aref, 1983, 1984; Aref and Balachandar, 1986). Aref’s
system models the outﬂow from a large bath tub with two sinks that are opened in
Chaotic Advection 209
an alternating manner, in order for chaotic mixing to take place in the course of the
process. To model the velocity ﬁeld caused by a sink, we assume the superposition
of the potential ﬂows of a point sink and a point vortex. If z = x + iy is the complex
coordinate in the plane of ﬂow, the complex potential for a sinking vortex point is
w(z) = −(Q + iK) ln z − z
s

where z
s
= (±a, 0), 2πQ is the sink strength and 2πK the vortex strength. The imag
inary part of w(z) is the stream function:
Ψ = −K ln r − Qφ
The streamlines are logarithmic spirals deﬁned by the function:
φ = −(K/Q) ln r + const
The diﬀerential equations of motion in polar coordinates are
˙ r = −
Q
r
r
˙
φ =
K
r
(10.5)
and their solutions are:
r =
_
r
2
0
− 2Qt
φ = φ
0
−
K
Q
ln
_
r
t
r
0
_
(10.6)
The ﬂow is fully characterised by the adimensional sink strength η and the ratio of
vortex to sink strength ξ, given by:
η =
QT
a
2
, ξ =
K
Q
(10.7)
The parameter T is the ﬂow period, usually set to 1, and a is the distance of each sink
from the centre of coordinates. As indicated in the literature (K´ arolyi and T´ el, 1997;
K´ arolyi et al., 2002), chaotic ﬂow appears for parameter values η ≥ 0.5 and ξ = 10.
More precisely, when particles are injected into the ﬂow in few time periods, they
are attracted to a speciﬁc region (the attractor) of the ﬂow system. In recent years,
several studies have appeared investigating this phenomenon both theoretically and
experimentally. The theoretical studies also include simulations using large grids
(1000×1000), as well as numerical solutions of the general equations of ﬂow. These
studies suggest that the attractors are time periodic according to the time periodicity
of the ﬂow. However, if only one sink is used, a stable attractor could be present,
both theoretically and experimentally, as the simulations presented above show.
We apply Aref’s blinking vortex system ﬁrst of all to a rotating ﬂuid. We select a
counterclockwise rotation. The symmetric sinks are located at (x, y) = (−a, 0) and
210 Chaotic Modelling and Simulation
FIGURE 10.3: The two symmetric sinks model
(x, y) = (a, 0) and the time period is T = 1. According to this system, the ﬂow is
not stationary and there are jumps in the velocity ﬁeld at each half period T/2. In
other words, a particle located at (−a, 0) appears at (a, 0) in the next time period, as
illustrated in Figure 10.3.
We propose to analyse a discrete time system, like Aref’s system, by using the the
ory of diﬀerence equations and discrete systems. It is more straightforward to con
sider the geometry of this system. The model we search must be a rotationtranslation
one with a contraction parameter b < 1 expressing the gradual shortening of the ra
dius r that causes the particles to follow logarithmic spiral trajectories around the
sinking vortices. In accordance with the theory just discussed, a rotationtranslation
model of this type is expressed by the diﬀerence equation:
z
t+1
= a + b(z
t
− z
s
)e
iφ
t
(10.8)
Equation (10.8) can be written as
x
t+1
+ iy
t+1
= a + b
_
(x
t
+ a) + iy
t
_
(cos φ
t
+ i sin φ
t
) (10.9)
The system of iterative diﬀerence relations for x and y is obtained by equating the
real and imaginary parts of (10.9):
x
t+1
= a + b(x
t
+ a) cos φ
t
− by
t
sin φ
t
y
t+1
= b(x
t
+ a) sin φ
t
+ by
t
cos φ
t
(10.10)
If a particle is located at (x
0
, y
0
) = (−a, 0), then after time t = 1 it will be located
at (x
1
, y
1
) = (a, 0). We next need to specify the form of the angle φ
t
. We will use the
form:
φ
t
= φ
0
+
ηξ
r
2
= c +
d
r
2
(10.11)
Chaotic Advection 211
where c = φ
0
= π is chosen to account for the halfcycle rotation, d = ηξ is the
vortex strength, and r = r
t
is the distance from the vortex (−a, 0), namely:
r =
_
(x + a)
2
+ y
2
For the experiments presented in the literature, η = 0.5, ξ = 10 and thus d = 5.
However, the chaotic region is wider, as illustrated in the following ﬁgures.
FIGURE 10.4: Chaotic attractor in the twosink problem
The chaotic attractor in Figure 10.4 illustrates the twosink case for parameter
values a = 1, b = 0.8, c = π and d = 3. There are two main vortex forms
counterbalancing each other. The ﬁrst form is located at the righthandside sink at
(x, y) = (a, 0). The second vortex form is centred at (x, y) = (a+2ab cos φ, 2ab sin φ),
where φ = d/(4a
2
). The two main vortex forms can be separated when the parameter
d expressing the vortex strength is relatively small. Such a case is presented in Fig
ure 10.5. The parameter d here is set to 1, while the other parameters remain the
same as in the previous example. The attractor is now completely separated into two
chaotic vortex forms (attractors).
Another possibility is to give high values to the parameter d expressing the vortex
strength. The value d = 2π leads to a more complicated vortex form, as presented
in Figure 10.6. There are three equilibrium points, for times t = 1, 2, 3. The ﬁrst of
these points is the centre of the righthandside sink.
One can also ﬁnd the form of the vortices in the case of three sinks located in an
equilateral triangle. The simulation of this situation is achieved by selecting a value
c =
2π
3
for the sink strength. The other parameters are b = 0.95, d = π and a = 5.
Figure 10.7(a) illustrates this case.
A form with four vortices is presented in Figure 10.7(b). This is achieved by as
suming a special value, c =
2π
4
, for the parameter c, thus dividing the total circle in
four sectors. The positions of the four sinks are located on the corners of a square.
212 Chaotic Modelling and Simulation
FIGURE 10.5: Two distinct vortex forms (d = 1)
FIGURE 10.6: The chaotic attractor with strong vortex strength parameter
d = 2π
Chaotic Advection 213
The parameters selected for the simulation are b = 0.98, d = π and a = 5. Simi
larly, a ﬁfth order vortex form results when c =
2π
5
and the other parameters remain
unchanged (Figure 10.7(c)).
(a) Third order (b) Fourth order (c) Fifth order
FIGURE 10.7: Vortex forms
10.4 Chaotic Forms without Space Contraction
Chaotic forms similar to those produced earlier in this chapter, but without any
space contraction, i.e. when b = 1, resulting in a Jacobian of J = 1, are of particular
interest. As there is no radial or space contraction, the ﬂow could be chaotic, but the
resulting chaotic attractors are not classiﬁed as main vortex forms.
Figure 10.8(a) illustrates the case where a = 1, b = 1, c = π and d = 1. The chaotic
form is symmetric. The axis y = 0 is the symmetry axis. A division of the main
chaotic form is starting. As the vortex strength parameter d is decreasing (d = 0.4 for
Figure 10.8(b) and d = 0.3 for Figure 10.8(c)) two distinct symmetric chaotic forms
arise. Figure 10.8(d) gives an enlargement of the righthandside chaotic attractor
produced when d = 0.3. Interesting chaotic details are present in this attractor.
10.5 Other Chaotic Forms
In the following simulation, we use the classical rotationtranslation model with a
rotation angle expressed by an inverse function of the square of the radius r:
θ
n
= c +
d
x
2
n
+ y
2
n
214 Chaotic Modelling and Simulation
We examine the eﬀect of varying the rotation parameter c while keeping the other
parameters ﬁxed. The space contracting parameter b takes a value close to 1 (b =
0.99), which could express low speeds directed towards the sink. When a = 10,
d = 10 and c = 1, six rotating vortexlike spiral objects appear (Figure 10.9(a)).
(a) One chaotic image (d = 1) (b) Starting division into two images (d = 0.4)
(c) Division into two images (d = 0.3) (d) Enlargement of the previous righthand
side attractor
FIGURE 10.8: Chaotic forms without space contraction
When c = 2, with the other parameters unchanged, three objects appear, as il
lustrated in Figure 10.9(b). These chaotic forms illustrate twoarmed spiral galaxies
located at
(x, y) = (0, 0), (x, y) = (a, 0) and (x, y) =
_
a + ba cos (θ
a
) , ba sin (θ
a
)
_
where:
θ
a
= c +
d
a
2
Chaotic Advection 215
(a) Sixth order (b) Third order
(c) Second order (d) Fifth order
FIGURE 10.9: Higher order vortices
216 Chaotic Modelling and Simulation
For c = 3, the chaotic form of the attractor is two onearmed spirals, illustrated in
Figure 10.9(c). The chaotic attractors are located at (x, y) = (0, 0) and (x, y) = (a, 0).
Five chaotic attractors appear when c = 5 (Figure 10.9(d)).
When c takes smaller values, for instance c = 0.5, multiarmed spirals are present,
as seen in Figure 10.10(a). The original rotating disk has radius R = 1. When R = 2,
Figure 10.10(b) appears. The number of spirals remains the same, but more details
are present in Figure 10.10(c). The original rotating disk has radius R = 3.
(a) R = 1 (b) R = 2 (c) R = 3
FIGURE 10.10: Spiralling towards the sink
10.6 Complex Sinusoidal Rotation Angle
Very interesting chaotic attractors arise by assuming that the rotation angle of the
rotationtranslation model is a complex sinusoidal function. For the ﬁrst application,
the following function for the rotation angle is considered:
θ
n
=
1
sin
_
c −
d
1+x
2
n
+y
2
n
_
The parameters are set to a = 1, b = 0.83, c = 0.4 and d = 6. Figure 10.11
illustrates the resulting chaotic attractor. The image shows a circular ring along with
an internal vortex form and an external part consisting of three main vortex forms
and a smaller fourth one.
A change in the function for the rotation angle results in the interesting chaotic
rotation image presented in Figure 10.12(a). The parameters remain the same as in
Chaotic Advection 217
FIGURE 10.11: A sinusoidal rotation angle chaotic attractor (c = 0.4, d = 6)
the preceding example, while the rotation angle function now is:
θ
n
= sin
_
¸
¸
¸
¸
¸
_
c −
d
_
x
2
n
+ y
2
n
_
¸
¸
¸
¸
¸
_
−1
= sin
_
c −
d
r
n
_
−1
The resulting image has a circular ring with two main vortex forms, and a smaller
one in the outside region, as well as vortex forms inside, interconnected with a central
rotating part.
(a) θ
n
= 1/ sin (c − d/r
n
) (b) d = 16
FIGURE 10.12: Sinusoidal rotation angle chaotic attractors
218 Chaotic Modelling and Simulation
A relatively simpler function for the rotation angle, with only one parameter, is:
θ
n
= sin
_
d
1 + x
2
n
+ y
2
n
_
−1
However, the resulting chaotic image in Figure 10.12(b) is also quite complicated.
As before, there is a circular ring and three connected vortex forms outside of the
ring, while several vortex forms rotate clockwise and counterclockwise inside the
ring. The parameters are a = 1.2, b = 0.85 and d = 16.
(a) θ = 1/ sin
_
dr
2
_
, b = 0.7 (b) θ = 1/ sin
_
dr
2
_
, a = 0.85, b = 0.8
FIGURE 10.13: Sinusoidal rotation angle chaotic attractors
A completely diﬀerent image results by introducing another function for the rota
tion angle:
θ
n
=
1
sin
_
d
_
x
2
n
+ y
2
n
__ =
1
sin
_
dr
2
_
The same rotationtranslation case with parameters a = 1.2, b = 0.85 and d = 16
provides also a circular ring image (Figure 10.13(a)) with surrounding vortex forms,
but in this case the outside ring vortex forms are small, whereas the inside part is
larger, including a central circular bulge and vortex forms rotating clockwise and
counterclockwise.
The standard rotationtranslation model is also applied with parameters a = 1.1,
b = 0.8 and d = 16 (Figure 10.13(b)). A large number of concentric rings appear.
Chaotic Advection 219
10.7 A Special RotationTranslation Model
A special rotationtranslation model is given by the following set of relations
_
x
n+1
y
n+1
_
= rot θ
n
_
b
_
x
2
n
− y
2
n
_
2bx
n
y
n
_
+
_
a
0
_
=
_
¸
¸
¸
¸
¸
_
a + b
_
x
2
n
− y
2
n
_
cos θ
n
− 2bx
n
y
n
sin θ
n
b
_
x
2
n
− y
2
n
_
sin θ
n
+ 2bx
n
y
n
cos θ
n
_
¸
¸
¸
¸
¸
_
where the rotation angle is given by the relation
θ
n
=
d
r
n
This map deserves special attention, as the Jacobian is J = 4b
2
r
2
. When r =
1/(2b), the map is area preserving, whereas when r < 1/(2b) a space contraction
takes place. For r > 1/(2b) area expansion takes place. The parameter values se
lected for the application are a = 0.8 for the translation parameter, b = 0.6 for the
area contraction parameter and d = 5 for the rotation parameter. Figure 10.14(a)
illustrates the simulation results. There is a rotation centre and vortexlike forms.
10.8 Other RotationTranslation Models
10.8.1 Elliptic rotationtranslation
The following system of two diﬀerence equations provides elliptic forms when the
ellipticity parameter h is not 1.
x
n+1
= a + b
_
x
n
cos θ
n
− hy
n
sin θ
n
_
y
n+1
= b
_
1
h
x
n
sin θ
n
+ y
n
cos θ
n
_ (10.12)
This system results is an analogue of the classical rotationtranslation system,
where the the rotation takes place on ellipses where the ratio of the one side to the
other is h. They arise from the rotationtranslation equations, after a rescaling on the
x direction by h. To see this, denote by ( ¯ x
n
, ¯ y
n
) the new coordinates. Then ¯ x
n
= hx
n
,
and ¯ y
n
= y
n
, hence we have:
¯ x
n+1
= hx
n+1
= hx
n
cos(θ) − hy
n
sin(θ)
= ¯ x
n
cos(θ) − h¯ y
n
sin(θ)
220 Chaotic Modelling and Simulation
(a) A special map (b) An elliptic map
FIGURE 10.14: Chaotic images of rotationtranslation maps
and
¯ y
n+1
= y
n+1
= x
n
sin(θ) + y
n
sin(θ)
=
1
h
¯ x
n
cos(θ) − ¯ y
n
sin(θ)
Adding the translation and contraction parameters results in (10.12). This case is
illustrated in Figure 10.14(b), where the rotation angle has the form θ = dr
2
, and
the parameters are a = 0.5, b = 0.9, d = 3 and h = 0.8. The elliptic rotation
equations give a somewhat deformed image, compared to the previous circular type
cases presented in this chapter.
10.8.2 Rotationtranslation with special rotation angle
This case is modelled by using the classical rotationtranslation model with space
contraction, but now the rotation angle obeys the function
θ
n
= c +
d
r
2
n
− e
where, for our example, e = 4.5 and the other parameters are a = 3, b = 0.9, c = 1.52
and d = 1.014. A space contraction parameter e is inserted in the function expressing
the rotation angle. The resulting image is illustrated in Figure 10.15(a). This image,
and the ones following it, are of particular interest when studying chaotic advection
cases.
The main part of the image is a circular ring centred at (x, y) = (a, 0). Three vortex
forms appear outside the circular ring and two smaller vortex forms appear inside the
Chaotic Advection 221
ring. All the vortex forms originate from the periphery of the circular ring. Smaller
vortex forms appear inside every main vortex form. In the same ﬁgure, the two axes
of coordinates and the part of the circle expressing the space curvature due to the
space contraction parameter b and the translation parameter a are illustrated. This
circle is of radius R =
ab
1−b
2
centred at (x, y) =
_
a
1−b
2
, 0
_
.
As the parameter c changes, the resulting image shows the same circular ring
located at the same centre and having similar magnitude, but changes appear in the
vortex forms. Figure 10.15(b) illustrates the case where c = 4.3 and d = 1, whereas
the other parameters remain unchanged. The two vortex forms inside the circular
ring take up more space, while the outer vortex forms create a main vortex image
with vortex forms inside.
Figure 10.15(c) has the same parameters as the previous case, except for c = 8.5.
There are now two outer vortex forms. However the image is similar to that of
Figure 10.15(a). Figure 10.15(d) is similar, but now with c = 17.
(a) c = 1.52, d = 1.014 (b) c = 4.3, d = 1
(c) c = 8.5 (d) c = 17
FIGURE 10.15: Chaotic images with area contraction rotation angles
222 Chaotic Modelling and Simulation
Questions and Exercises
1. Find a symmetry axis for the system (10.10).
2. Consider the system (10.10). Add a translation parameter “k” at the second
equation, and ﬁnd a symmetry axis for this new system. Use the rotation angle
φ
t
= c + d/r
2
t
, and draw chaotic images when c = π/2 and d is varying.
3. Chaotic advection is closely related to chaotic mixing. Discuss the mixing of
ﬂuids when the chaotic region is reached. What is the disadvantage with the
presence of chaos in mixing? How can we avoid chaotic phenomena in ﬂuid
mixing?
4. Change the rotation equations (10.1) to their reﬂection analogues, and ﬁnd the
resulting chaotic and nonchaotic forms for various values of the parameters
b, c and d.
5. Change the rotation equations (10.10) to their reﬂection analogues, and ﬁnd the
resulting chaotic and nonchaotic forms for various values of the parameters
b, c and d.
Chapter 11
Chaos in Galaxies and Related
Simulations
11.1 Introduction
A large variety of chaotic forms appears in stellar observations. These forms ex
hibit astonishingly large varieties of shapes. At ﬁrst glance, there is not a simple
method to handle and organise the entire material of stellar observations based on a
few rules. The related cosmology theories for star and galaxy creation and formation
are of considerable interest. Related laws help us describe the evolution of stellar
shape over time.
Already from the earliest observations, it was evident that chaos was the main
characteristic of the shape and form of many stellar formations, including galaxies
and chaotic nebulae. A great number of stellar formations develop under high tem
peratures, the socalled ‘ﬁre’ that the ancient Greek philosopher, Heracleitos of Eph
esus (Heraclitus), had emphasised as the basis of genesis. All these “ﬁre” formations
are mainly chaotic forms.
A fundamental problem in nonlinear dynamics is that of exploring how the prop
erties of orbits change and evolve as one or more parameters of a dynamical system
change, in order that the system exhibit chaotic behaviour. Recent advances in non
linear physics have resulted from such an approach. The transition to chaos and the
routes to chaos are extensively explored in the case of onedimensional maps and,
mainly, for the logistic map. Of considerable interest however are also the chaotic
dynamics of twodimensional maps, such as the catmap and the H´ enon map.
Another very important transition, related to physics, is one in which the system
is in a chaotic state and, as parameters change, the system transitions from one type
of chaos to another. Of particular interest is the case where the chaotic attractors that
correspond to a system change as a parameter gradually takes higher or lower values.
Many stellar and galactic systems may ﬁt to chaotic attractor phenomena. Chaotic at
tractors in twodimensional and threedimensional space are quite stable objects and
can be useful in exploring and simulating the forms of galaxies, clusters of galaxies,
nebulae, black holes and other stellar objects. There are several approaches regard
ing the handling of chaotic situations occurring under the inﬂuence of a central force
as in the case of stellar systems. An approach is to consider a system of masses
moving in diﬀerent orbits under gravity. Early studies of the subject have been
223
224 Chaotic Modelling and Simulation
undertaken by Contopoulos and Bozis (1964); LyndenBell (1969); Toomre (1963,
1964); Hunter and Toomre (1969); Contopoulos et al. (1973). A detailed approach
of chaotic dynamics in astronomy can be found in Contopoulos (2002). Ostriker and
Peebles (1973) explore the stability of ﬂattened galaxies while Raha et al. (1991)
study the dynamical instability of bars in disk galaxies. The box and peanut shapes
generated by stellar bars are discussed in Combes et al. (1990). Extensive work on
how one can separate chaotic from ordered domains is found in the works of Con
topoulos and Voglis (Patsis et al., 1997). Objects of very complicated morphology
are found and simulated in Xray clusters (Navarro et al., 1995). The discovery of
new ultraluminous IRAS galaxies give rise to new studies on how galactic systems
interact.
A number of works regarding chaos and chaotic attractors and the interaction with
chaotic dynamics in astronomy are of interest. Most notable among them are the
work of White et al. (1998) on the anomalous transport near threshold of the stan
dard map and the approach by Arrowsmith and Place (1990); Acheson (1997) in
their books on dynamical systems and calculus and chaos. H´ enon and Heiles (1964)
developed a Hamiltonian system which developed into a prototype for computer ex
periments of chaotic dynamics in astronomy. Of considerable interest is also the
areapreserving quadratic map developed by H´ enon (1969). Period doubling bifur
cations of this map appear in recent studies (Murakami et al., 2002).
It is very complicated to express chaos analytically, namely so that every stage is
explained and deﬁned in a deterministic way. However, the forms resulting from bil
lions of chaotic intermediate steps are frequently of considerable ‘attracting’ shape.
These attractors are in many cases reproduced in the laboratory by using very simple
rules.
Following the studies on chaos in the last decades, it is evident very few main char
acteristics or parameters of the chaotic system play a critical role in the formation
of attractors. In stellar systems the main characteristics are: 1) gravity, the attract
ing force between bodies (masses), and 2) the ﬁrst BigBang. Geometrically, stellar
systems move, grow or decline by: a) rotating and b) translating (following linear
movements). According to the established theories, expansion of the universe (trans
lation) follows the original BigBang, whereas gravity, the attracting force, causes
translation in the opposite direction. Attracting forces in diﬀerent directions cause
rotations and translations in stellar formations.
In chaotic dynamics, the main eﬀort is usually centred on the formulation of rela
tively simple models with only a few chaotic parameters. If the main characteristics
of the phenomenon are included in the ﬁnal model, the chaotic character of many
phenomena in nature could be well expressed. Chaotic formations in galaxies could
appear when rotations, reﬂections and translations (axial movements) are present
along with the inﬂuence of gravitational forces that play an important role. In this
chapter, simple iterative models are presented. They are the simplest possible mod
els, including rotation, reﬂection, translation and the inﬂuence of gravitational forces
on the angle of rotation. These models present interesting aspects that are studied an
alytically and are presented in several graphs following the simulation. Several spiral
forms of various formations can be modelled by changing the parameters of the mod
Chaos in Galaxies and Related Simulations 225
els. The study of galactic forms by using diﬀerential equations and Hamiltonians is
presented in Chapter 12.
11.2 Chaos in the Solar System
Complicated and even chaotic orbits and paths of the solar system are studied and
simulated in related chapters of this book. The bodies in the solar system mainly
follow cyclic or elliptic orbits around a centre of mass. In this case, the simulation
can follow a purely geometric approach. To simplify the problem, the centre of mass
is located in the centre of the large mass. The attracting force is, of course, gravity,
and it is presented by an inverse square law for the distance. The nondimensional
equations of motion for a body with small mass m rotating around a large body
with mass M, which we consider to stay in equilibrium position, are, in Cartesian
coordinates:
¨ x = −
x
_
x
2
+ y
2
_
3/2
¨ y = −
y
_
x
2
+ y
2
_
3/2
(11.1)
The initial conditions are x = 1, y = 0, ˙ x = 0 and ˙ y = ν. The escape velocity is
ν =
√
2, and the circular orbit is obtained when ν = 1. The circular orbit is illustrated
in Figure 11.1(a). In the same ﬁgure an elliptic orbit is presented, where the initial
velocity is 1 < ν = 1.2 <
√
2. A hyperbolic orbit is also presented. The initial
velocity is higher than the escape limit (ν = 1.45 >
√
2) and the particle ﬂies oﬀ to
inﬁnity. Figure 11.1(b) illustrates a large number of elliptic paths for the same value
of the initial velocity as before. The paths remain in the same plane, but deviate from
the original position, forming the shape presented in the ﬁgure.
The motions of two attracting point masses are relatively simple. The equations
of motion for the point mass m
1
in Cartesian coordinates are:
¨ x
1
= −
(x
1
− x
2
)m
2
_
(x
1
− x
2
)
2
+ (y
1
− y
2
)
2
_
3/2
¨ y
1
= −
(y
1
− y
2
)m
2
_
(x
1
− x
2
)
2
+ (y
1
− y
2
)
2
_
3/2
(11.2)
The case for the point mass m
2
is analogous.
Without loss of generality, the gravity constant is G = 1 and the total mass of the
rotating system is M = m
1
+ m
2
. If we give the two masses initial positions (x
1
, y
1
)
and (x
2
, y
2
), and initial velocities ˙ x
1
, ˙ y
1
, ˙ x
2
and ˙ y
2
, the masses revolve around each
another while drifting together, as a pair, in space. Their centre of mass drifts at
a constant velocity. Relative to an observer moving with the centre of mass, their
orbits are ellipses and, in special cases, circles.
226 Chaotic Modelling and Simulation
(a) Circular and elliptic orbits (b) Elliptic paths
FIGURE 11.1: Elliptic orbits
The motion of two attracting points in space is presented in Figure 11.2(a). The
heavy line represents the movement of the centre of mass. The masses revolve around
and drift together. The points have masses m
1
= 0.4, m
2
= 0.6, and initial velocities
˙ x
1
= −0.15, ˙ x
2
= 0, ˙ y
1
= −0.45 and ˙ y
2
= 0.25. The same motion, viewed relative
to the centre of mass, is illustrated in Figure 11.2(b). Each pointmass follows an
elliptic path.
(a) Attracting bodies in space (b) Relative movement
FIGURE 11.2: Attracting bodies in space and relative movement
The threebody problem reduced in the plane still gives very complicated and
chaotic orbits. To simplify the problem, we assume that the third body, a satellite,
Chaos in Galaxies and Related Simulations 227
is of negligible mass related to the other two bodies. The other two, the planets
(primaries), travel about the centre of the combined mass in circular orbits in the
same plane. The satellite rotates in the same plane. The total mass of the planets is
m
1
+ m
2
= 1. The equations of motion for the satellite are
¨ x = −
m
1
(x − x
1
)
r
3/2
1
−
m
2
(x − x
2
)
r
3/2
2
¨ y = −
m
1
(y − y
1
)
r
3/2
1
−
m
2
(y − y
2
)
r
3/2
2
(11.3)
where (x, y) is the position of the satellite, r
1
, r
2
are the distances of the satellite from
the two planets m
1
, m
2
respectively, and (x
1
, y
1
) and (x
2
, y
2
) are the positions of the
planets at the same time t.
In the following example, the primaries have masses m
1
= 0.9 and m
2
= 0.1, and
the position and initial velocity of the satellite are (−1.033, 0) and (0, 0.35) respec
tively. The complicated and chaotic paths that arise are illustrated in the following
ﬁgures. In Figure 11.3(a), the movement is viewed from the outside of the system,
from “space.” It shows that the satellite can move between the two revolving pri
maries. Figure 11.3(b) illustrates the orbit of the satellite viewed in a coordinate
system that rotates with the revolution of the planets (rotating frame), with the xaxis
always passing through the planets, according to the rotation formula:
x
new
= x cos t + y sin t
y
new
= −x sin t + y cos t
(11.4)
(a) Motion in space (b) Rotating frame
FIGURE 11.3: Motions in space
By using diﬀerence equations to model solar systems, it is possible to simulate
228 Chaotic Modelling and Simulation
chaotic patterns like, for instance, the rings of Saturn and of other planets. The
simplest approach is to use an equation such as the logistic, and to assume that the
system rotates following elliptic orbits. Two realisations appear in the next ﬁgures.
The parameter of the logistic model is b = 3.6 for Figure 11.4(a) and b = 3.68 for
Figure 11.4(b). The rotation formula is given by:
x
t
= 2r
t
cos t
y
t
= 2r
t
sin t
(11.5)
where r
t+1
= br
t
(1 − r
t
).
(a) b = 3.6 (b) b = 3.68
FIGURE 11.4: Ring systems
11.3 Galaxy Models and Modelling
Studies on chaotic dynamics in galaxies and in other complicated stellar objects
indicate that order and chaos coexist in these structures. It is also quite diﬃcult to
simulate complicated stellar objects, especially in the limits where chaos is present.
Experience on chaotic simulations during the last decades indicates that the simpler
the chaotic model, the more successful it is. Remarkable solutions of the ﬂow equa
tions were provided by Lorenz (1963). He transformed a highdimensional system
into a threedimensional one, retaining the chaotic character of the system. R¨ ossler
(1976d) proposed the simplest model of threedimensional ﬂow. Since, even nowa
days, the subject of chaos in astronomy has many unexplored aspects, a convenient
way of dealing with the problem is to explore the qualitative and quantitative aspects
Chaos in Galaxies and Related Simulations 229
by considering simple chaotic models, in which only the main characteristics of the
system under consideration are retained. Such characteristics in galaxies and galactic
formations are of course: 1) rotationreﬂection, 2) translation, and 3) area contrac
tion due to gravitation (expressed by a parameter b < 1). Rotation is connected with
the theory of a central force by assuming that the rotation angle obeys a function of
the distance r from the origin. Translation is assumed to take place when a strong
gravitation source is located in a large distance from the origin or, even, when gravity
waves are present. To simplify the problem, only the twodimensional approach in
the (x, y) plane is presented here.
A convenient formulation of the iterative map expressing rotationreﬂection and
translation is given by the following diﬀerence equation:
f
n+1
= a + b f
n
e
iθ
where:
f
n
= x
n
+ iy
n
(for rotation)
f
n
= x
n
− iy
n
(for reﬂection)
For rotation, the determinant of this map is det J = b
2
= λ
1
λ
2
. This is an area
contracting map if the Jacobian J is less than 1. The eigenvalues of the Jacobian are
λ
1
and λ
2
. For reﬂection the Jacobian gives: det J = −b
2
. The translation parameter
is a in both cases.
In the cases discussed above, rotation and reﬂection are followed by a translation
along the xaxis. As discussed earlier, this has the advantage, over a general trans
lation, of reducing the number of parameters by one, without aﬀecting the results in
the majority of the cases studied.
The angle θ is assumed to be a function of the distance r from the origin. In the
cases studied here, r =
_
x
2
+ y
2
, the Euclidean metric. The general forms studied
can be classiﬁed as follows: θ = g(r
m
), where m = 1, 2, −1, −2, −3/2, are the most
convenient values for the exponent. By expanding the function g(r
m
) in a Taylor
series and retaining the ﬁrst terms to the right the following forms for the angle θ for
reﬂection and rotation cases arise:
θ
n
= cr
n
θ
n
= cr
2
n
θ
n
= c +
d
r
n
θ
n
= c +
d
r
2
n
θ
n
= c +
d
r
2/3
n
Both rotation and reﬂection generate chaotic attractors for appropriate values of the
parameters a, b, c, and d.
230 Chaotic Modelling and Simulation
A discussed in Chapter 8, a very interesting property of the map is the existence of
a disk F of radius
ab
1−b
, centred at (a, 0) in the (x, y) plane, the points of which remain
inside of it under the map. Further, the ﬁxed points obey the equations
_
x
n
−
a
1 − b
2
_
2
+ y
2
n
=
a
2
b
2
(1 − b
2
)
2
and, for rotation:
cos θ
n
=
1
2b
_
1 + b
2
−
a
2
r
2
n
_
So the ﬁxed points lie on a circle K with radius R =
ab
1−b
2
, centred at (
a
1−b
2
, 0) in the
(x, y) plane.
For the ﬁrst application we choose the following function for the rotation angle:
θ
n
= c +
d
r
2
The simulation results are illustrated in Figures 11.5(a) through 11.5(f). In these
ﬁgures, the parameters of the model are b = 0.8, c = 3 and d = 6. This is an example
where an area contraction takes place, equal to b
2
= 0.64. The other parameter, a,
varies, starting from very low values, (a = 0.01 in Figure 11.5(a)) and taking the
values a = 0.45, a = 1.6, a = 2.4, a = 4 and a = 5 in the subsequent ﬁgures.
In Figure 11.5(a), the parameter a seems to have very little eﬀect in the devel
opment of the phenomenon. The point masses are distributed around the centre of
coordinates and their distribution follows an inverse law. In the second case (Fig
ure 11.5(b)), a rotating form appears. This chaotic form is restricted inside the two
circles F and K that were discussed earlier. Higher values of a, as in Figure 11.5(d),
produce chaotic forms in which a centre of revolution appears in (a, 0), but further a
quite large rotating form appears which, when a takes higher values (Figure 11.5(c)
and Figure 11.5(e)), leads to the formation of two distinct attractors. In the case of
Figure 11.5(f), these attractors are ﬁnally separated in two forms. The centres of the
two attractors are located at (a, 0) for the attractor on the right side of Figure 11.5(f),
and at
_
a + a cos
_
c +
d
a
2
_
, a sin
_
c +
d
a
2
__
for the attractor on the left side. An interesting property of the above simulation is
that the original single simple rotating form splits into two forms, that show the same
direction of rotation and are located at places depending on the displacement a.
More complicated chaotic forms arise for appropriate values of the parameters a,
b, c and d. When a = c = d = 4 and b = 0.9, three attractors appear, one of which is
located at (a, 0). This is illustrated in Figure 11.6(a).
Another example appears in Figure 11.6(b), presenting a cluster of chaotic attrac
tors. The ﬁrst attractor, located at (a, 0), is at the top of the ﬁgure. The second
attractor is the one on the right side of the ﬁgure, while the other attractors continue
in a clockwise direction until the ﬁnal point, located at the circle K. The parameters
are: a = c = d = 5.25 and b = 0.8.
Chaos in Galaxies and Related Simulations 231
(a) Initial stage (a = 0.01) (b) Starting rotation (a = 0.45)
(c) Middle stage (a = 1.6) (d) Second phase (a = 2.4)
(e) Two interconnected images (a = 4) (f) Two separate images (a = 5)
FIGURE 11.5: Chaotic forms of a simple rotationtranslation model
232 Chaotic Modelling and Simulation
(a) Three chaotic images (b) Multiple chaotic images
FIGURE 11.6: Chaotic forms of a simple rotationtranslation model
Rotating forms can also be formed by using rotation equations with a linear ex
pression for the rotation angle:
θ
n
= c + dr
n
The translation parameter is a, and an area contracting parameter b < 1 is essential
to keep the system into a limit included within a circle F of radius
ab
1−b
centred at
(a, 0). A number of rotating point masses are equally distributed within the circle F,
and the question is how these points will be distributed after time t corresponding
to n iterations. The results are summarised in Figures 11.7(a) to 11.7(d). A two
armed spiral structure appears, located at
_
a
2
, lim(y
n
= y
n+1
)
_
, and four satellites are
located at (lim(x
n
), lim(y
n
)). Two of the satellites have their location centre at the
circumference of the circle K of radius
ab
1−b
2
. The centre of this circle is at
_
a
1−b
2
, 0
_
.
In Figures 11.7(a), 11.7(b) and 11.7(c) the parameters are a = 4.25, b = 0.9, c = 1.5
and d = 0.3. In Figure 11.7(d) the parameters are a = 2.3, b = 0.9, c = 2.03 and
d = 0.3. In these cases there are attracting points and no other attractors. The point
masses are trapped in trajectories and are led to the attracting points as when a black
hole is present. The attracting points are solutions of the equations that satisfy the
special conditions set for the general rotation equation above.
Chaotic attractors and attracting points appear in an example of rotation (Fig
ure 11.8(a)) where the rotation angle follows an inverse law regarding the distance r
from the origin:
θ
n
= c +
d
r
n
In the ﬁrst case, the translation parameter is a = 0.6 and the other parameters are
b = 0.9, c = 3 and d = 6. The chaotic attractor is located at (a, 0) and the attracting
point is located in the periphery of the circle K. The rotating points inside the circle
F are trapped in the attractor region, or they disappear into the attracting point. The
points that oscillate inside the attracting region escape from this region only if they
Chaos in Galaxies and Related Simulations 233
(a) Chaotic object with satellites (b) Enlargement of chaotic objects
(c) A twoarmed spiral with satellites (d) A twoarmed spiral galactic form
FIGURE 11.7: Chaotic forms of a simple rotationtranslation model
234 Chaotic Modelling and Simulation
reach the periphery of the attractor, and then they follow the trajectory that leads to
the attracting point.
The same behaviour is seen in the next example (Figure 11.8(b)), where the para
meters are a = 5 and c = 2, while the other parameters remain unchanged. Now two
attractors appear, and each attractor is connected to the attracting point with trajec
tories. In Figure 11.8(b), the attracting points are the points where x
n+2
= x
n
and
y
n+2
= y
n
, while, in Figure 11.8(a), the attracting point is the point where x
n+1
= x
n
and y
n+1
= y
n
.
(a) Attractor and attracting point (b) Two attractors and the accompanying
attracting points
FIGURE 11.8: Attractors and attracting points in rotationtranslation model
11.3.1 A special rotationtranslation image
This example comes from Contopoulos and Bozis (1964). The situation exam
ined was that of two galaxies approaching each other. Here, we apply a rotation
translation model with space contraction. The rotation angle and the other para
meters, except for the space contraction parameter, are the same as in Contopoulos
and Bozis (1964). The rotation angle has the form:
θ = −d
_
1
6
ln
r
2
− cr + c
2
(r + c)
2
−
1
3
tan
2r − c
c
√
3
_
The parameters for the system are a = 4 for translation, and c =
1
0.012
1/3
, d =
0.0055
0.012
2/3
and b = 0.9 for space contraction. The resulting shape is illustrated in Figure 11.9.
A number of particles are introduced into a small circle near the centre of coordi
nates. The particles follow characteristic trajectories over time. The image is quite
complicated. There are several attracting points, as well as places with high and low
density.
Chaos in Galaxies and Related Simulations 235
FIGURE 11.9: A rotation image following a ContopoulosBozis paper
(approaching galaxies)
11.4 RotationReﬂection
It is interesting to explore the relationships between chaotic attractors in the pres
ence of rotation and reﬂection respectively, for the same set of parameters. One
would expect that the attractors for the reﬂection case would be a mirror image of
the attractors for the rotation case. This can be seen in some particular cases, for in
stance in the following, where the rotation or reﬂection angle is of the form θ
n
= dr
2
.
The parameters in both cases (Figure 11.10(a) illustrates rotation and Figure 11.10(b)
illustrates reﬂection) are, a = 2, b = 0.5 and d = 0.65. The chaotic forms of the at
tractors verify the assumption that there is a mirrorimage relationship between the
two.
The chaotic geometric modelling approach presented in this section could be use
ful in galaxy simulations. Chaotic attractors and other forms that appear in real space
(x, y) simulations indicate that elements of any form can be trapped inside these at
tractors. When the space contraction parameter b < 1 three cases appear, according
to the magnitude of the rotation angle θ:
1. the particles lead to attracting points where they disappear as if in a blackhole,
2. in the presence of a chaotic attractor, the particles will stay or oscillate inside
this attractor, and
3. the particles follow trajectories leading away from the rotating system if the
translation parameter is relatively high.
236 Chaotic Modelling and Simulation
(a) Rotation image (b) Reﬂection image
FIGURE 11.10: Comparing chaotic forms: rotationreﬂection
Various chaotic geometric forms appear. Special characteristics of these forms,
such as the coordinates of the centre and the geometric interactions of these forms
between each other, can, in some cases, be estimated. Some chaotic forms form
groups or clusters in real space. Some chaotic forms are divided into other similar
forms when the parameters of the model take particular values. These chaotic forms
specify places in space that attract masses (particles) from the space around.
This kind of chaotic simulation of galaxy forms has the advantage that it needs
limited computing power, and it mainly models “space,” rather than “elements” in
that space. The explanatory ability of chaotic models is high and could be useful in
many cases.
11.5 Relativity in RotationTranslation Systems
In rotationtranslation systems, a number of particles follow circular paths, while,
at the same time, a translation movement takes place. Numerous chaotic and non
chaotic forms arise. As we derived earlier, a twodimensional rotation map is area
preserving if the Jacobian determinant of the map is J = 1. The question we would
like to address in this section is how the shape of the attractors changes when the
speed of the rotating system relative to another system, i.e. the relativistic speed,
inﬂuences the system parameters. In other words, how do the equations of special
relativity enter into the rotationtranslation equation system? The introduction of the
relativity equations into a rotationtranslation map will change the Jacobian of the
system, and can thus produce an areacontracting map. The usual obstacle in appli
cations is that high relativistic speeds are not possible in real situations. However,
computer simulations show that even low relativistic speeds have a considerable in
Chaos in Galaxies and Related Simulations 237
ﬂuence on the chaotic forms.
A simple example is presented in the following ﬁgures. Figure 11.11 illustrates the
chaotic image from the rotationtranslation map without entering relativistic equa
tions. A chaotic bulge appears.
FIGURE 11.11: A rotationtranslation chaotic image
The rotationtranslation chaotic bulge presented in Figure 11.11 is formed when
the space parameter is b = 1. The bulge is symmetric. The translation parameter is
a =
π
6
: The other parameters are: c =
π
12
and d =
π
3
. The equation for the rotation
angle is:
θ
n
= c +
d
r
n
where r
n
=
_
x
2
n
+ y
2
n
.
The relativisticlike approach to the rotationtranslation equations takes into ac
count the length contraction governed by the equations:
x
rel
= x
_
1 −
v
2
c
2
y
rel
= y
_
1 −
v
2
c
2
where v is the relativistic speed and c is the speed of light.
Inserting these relativistic forms into the rotationtranslation equations
x
n+1
= a + b(x
n
cos 2θ
n
− y
n
sin 2θ
n
)
y
n+1
= b(y
n
sin 2θ
n
+ y
n
cos 2θ
n
)
(11.6)
a new form is obtained. After rearrangement, the relativisticlike coeﬃcient appears
as a change in the parameter b, which now becomes:
b
∗
= b
_
1 −
v
2
c
2
238 Chaotic Modelling and Simulation
When v = 0, the parameter b is set equal to 1. When v = c, then b = 0. In reality,
v ≪ c and the values of b are close to 1. However, even for small relativistic speeds,
the inﬂuence on the chaotic bulge and on the space around the bulge is quite evident.
Figure 11.12(a) illustrates the chaotic bulge and the space outside the central chaotic
space. We launch particles at x = (−1.5, −1.4, . . . , 0) and y = 0.1. The space outside
the chaotic bulge is characterised by closed curves and islands.
A relativistic speed equal to 1000km/sec corresponds to b = 0.9999945 for the
space contraction parameter b. The inﬂuence of this change of parameter b in the
shape of the previous object is illustrated in Figure 11.12(b). First of all, we observe
that the nonchaotic lines outside the chaotic bulge disappear. On the other hand,
the lines expand, covering more space and joining each other. Over time, the space
outside the chaotic image is covered by particles that are directed towards the chaotic
bulge. The nonchaotic islands are shorter. This is more clear in Figure 11.12(c), in
which the relativistic speed is v = 2500km/sec, corresponding to a shape contraction
parameter b = 0.999965. When v = 2760km/sec, a chaotic limit is present. The
islands in the chaotic sea are transformed into attracting regions, as illustrated in
Figure 11.12(d). The particles from the chaotic bulge are directed into these regions,
and the equilibrium points in these regions become attracting sinks.
When the relativistic speed v is between 2760km/sec and 89000km/sec, the par
ticles are guided to the points of attraction. However, a change in the images takes
place as the relativistic speed gets higher. This is shown in Figures 11.13(a) through
11.13(f). After the limiting speed v ≈ 2760km/sec, symmetry breaks down. This
is very clear when v = 10000km/sec and b = 0.99944 (Figure 11.13(b)). In Fig
ures 11.13(c) through 11.13(e), the relativistic speed takes higher values (those val
ues being v = 20000km/sec and b = 0.99778, 50000km/sec and b = 0.9860, and
70, 000km/sec and b = 0.9724 respectively), and the images take a rotation form
with three main arms and a smaller one. When v = 89, 000km/sec and b = 0.954981,
the rotating image has the form of a chaotic attractor where there are no attracting
points or sinks, and the particles are trapped in the space covered by the chaotic
attractor. The attractor is a totally nonsymmetric rotation object with two main
rotation arms and a smaller one (Figure 11.13(f)). A circular disk is centred at
(x, y) = (a, 0).
The chaotic attractor becomes sharper for high relativistic speeds, namely values
higher than v = 89, 000km/sec, as illustrated in the following two ﬁgures. In Fig
ure 11.14(a) the chaotic image is simulated for v = 120, 000km/sec and b = 0.9165,
whereas in Figure 11.14(b) the relativistic speed is in the upper limit for attractor
formation, at v ≈ 170000km/sec (b = 0.8239471).
A very surprising property of the rotationtranslation system is that the chaotic
image produced under the relativistic inﬂuence in large relativistic speeds is present
at the beginning of the process, when time t is very small, if the area contracting
parameter b = 1. In Figures 11.15(a) through 11.15(f), illustrations of the chaotic
evolution during the ﬁrst time periods appear.
The particles introduced at time t = 0 are distributed in a small circle centred at the
origin. The radius of this circle is r
intr
= 0.01. At the beginning of the process, at time
t = 10, the image is similar to that appearing in Figure 11.14(b), where the relativistic
Chaos in Galaxies and Related Simulations 239
(a) Nonrelativistic rotationtranslation
for v = 0 (b = 1)
(b) Relativisticlike rotationtranslation
for v = 1000 (b = 0.9999945)
(c) Chaotic forms for speed v =
2500km/sec
(d) Chaotic limit at speed
v = 2760km/sec
FIGURE 11.12: Chaotic forms for low relativistic speeds
240 Chaotic Modelling and Simulation
(a) v = 5000km/sec (b) v = 10000km/sec
(c) v = 20000km/sec (d) v = 50000km/sec
(e) v = 70000km/sec (f) v = 89000km/sec
FIGURE 11.13: Relativistic rotationtranslation forms for medium and high
relativistic speeds
Chaos in Galaxies and Related Simulations 241
(a) v = 120000km/sec (b) v = 170000km/sec
FIGURE 11.14: Relativistic rotationtranslation forms for high relativistic
speeds
speed is v = 170000km/sec and the space contracting parameter is b = 0.8239471.
Instead a similar image is now produced at time t = 10, but with parameter b = 1.
The image is totally nonsymmetric. As the time t increases, the chaotic images
become more symmetric, and at time t = 500 the resulting Figure 11.15(e) is a
symmetric object almost identical to the equilibrium stage image (Figure 11.15(f))
resulting when t → ∞.
11.6 Other Relativistic Forms
More rotationtranslation relativistic objects are based on the above equation forms
but with diﬀerent values for the parameters: a = 0.8π, c = 0.4π and d = 1.6π. The
resulting relativistic forms are illustrated in Figures 11.16(a) through 11.16(d).
A rotationtranslation relativistic model based on the above equation forms, with
parameters a =
4π
3
, c =
4π
3
and d =
4π
3
, provides the relativistic images presented in
Figures 11.17(a) and 11.17(b).
A rotationtranslation relativistic object based on the above equation forms, with
parameters a = c = d =
2π
0.3
, results in the relativistic forms illustrated in Fig
ures 11.18(a) through 11.18(c).
Figure 11.18(c) illustrates a pair of galaxylike objects formed for very high rel
ativistic speed v =
c
3
= 100000km/sec. However, similar objects appear in the ﬁrst
time periods of the same rotationtranslation procedure, but for b = 1, as illustrated
in Figures 11.18(d) through 11.18(h). In this case, the original circular disk of par
ticles has radius r = 5. Very early, at times t = 4 and t = 6, a pair of twoarmed
spiral galaxies appears. Later on, at time t = 10, a circular arm appears, and at time
t = 20, the two spiral galaxies are connected through two connecting circular arms.
242 Chaotic Modelling and Simulation
(a) t = 10 (b) t = 20
(c) t = 50 (d) t = 100
(e) t = 500 (f) t → ∞
FIGURE 11.15: Nonrelativistic chaotic images in the early period of a
rotationtranslation process with b = 1
Chaos in Galaxies and Related Simulations 243
(a) v = 0 (b) v = 20.000km/sec
(c) v = 50.000km/sec (d) v = 100.000km/sec
FIGURE 11.16: Other relativistic chaotic images
(a) Chaotic form for speed v = 0 (b) Chaotic form at speed
v = 30000km/sec
FIGURE 11.17: Various relativistic chaotic forms
244 Chaotic Modelling and Simulation
(a) v = 0 (b) v = 60.000km/sec
(c) v = 100.000km/sec (d) t = 4
(e) t = 6 (f) t = 10
(g) t = 20 (h) t = 50
FIGURE 11.18: Relativistic and initial chaotic images of a rotationtranslation
process
Chaos in Galaxies and Related Simulations 245
Even later, at time t = 50, the twoarmed spiral galaxies are interconnected and start
to form the equilibrium shape that is illustrated in Figure 11.18(a).
Similar galaxylike objects as above arise from the following set of parameters:
a = c = d =
2π
3
. The illustrations for b = 0.9 are quite clear, perhaps better than in
the previous case (Figure 11.19).
FIGURE 11.19: Two galaxylike chaotic images
Three galaxy forms are illustrated in Figure 11.20(a). These forms arise from the
rotationtranslation equations for b = 0.9. The other parameters are a = π for the
translation parameter and c = d = π for the rotation angle parameters.
Figure 11.20(d) arises when the relativistic speed is v = 25000km/sec and b =
0.9965217. The three galaxylike objects are already clearly formed. More clear
forms appear for higher relativistic speeds, where b = 0.99 (Figure 11.20(b)). The
equilibrium form object (b = 1) is illustrated in Figure 11.20(c).
A rotationtranslation relativistic object based on the above equation forms, but
with parameters a = 1.8, b = 0.9, c = 3a and d = 3a, has the relativistic forms
presented in Figure 11.21
The particles are introduced into the rotating system on the circle with radius r =
0.1 centred at (x, y) = (0, 0). In Figure 11.21, this circle appears as a small dot in
the centre of coordinates. An important point of reference is (x, y) = (a, 0); and it is
labelled by a cross. In the lower part of the same ﬁgure a cross indicates the point
where all the particles arrive after a suﬃciently large amount of time. This is simply
the ﬁxed point of the rotationtranslation equations. An iterative procedure estimates
this point during computer simulation.
A galaxylike object is illustrated in Figure 11.22. The rotationtranslation equa
tions and the equation for the rotation angle are the same as above. The parameter
values are a = 5.8, c = 2.25, d = 2.5 and b = 0.9. The particles are originally intro
duced in a circle with radius r = 3 and centred at the centre of coordinates (indicated
by the small cross on the left of the ﬁgure). The cluster of points on the right side of
the ﬁgure is centred at (x, y) = (a, 0).
246 Chaotic Modelling and Simulation
(a) b = 0.9 (b) b = 0.99
(c) b = 1 (d) v = 25000(b = 0.9965217)
FIGURE 11.20: Other relativistic chaotic images
FIGURE 11.21: A relativistic chaotic image when b = 0.9
Chaos in Galaxies and Related Simulations 247
FIGURE 11.22: A relativistic chaotic image when b = 0.9
11.7 Galactic Clusters
Figure 11.23 illustrates the simulation of a galactic cluster. Seven galaxylike
objects appear, while the particles originated in a disk with radius R = 10. A three
armed spiral is located in the middle, three twoarmed spirals are connected to the
threearmed spiral, and three elliptical galaxies are in the outer part. The model is a
rotationtranslation one expressed by the following set of equations
x
n+1
= a + b(x
n
cos θ
n
− y
n
sin θ
n
)
y
n+1
= −a + b(x
n
sin θ
n
+ y
n
cos θ
n
)
where the rotation angle has the form
θ
n
= c +
d
x
2
+ y
2
The parameters for the simulation are a = 2.6, b = 0.98, c = 3.2π and d = 6.
FIGURE 11.23: A galacticlike cluster
248 Chaotic Modelling and Simulation
11.8 Relativistic ReﬂectionTranslation
The reﬂectiontranslation relativistic case is similar to the rotationtranslation case,
but the resulting objects are quite diﬀerent. In the following, a simple example is pre
sented. The parameter values are a = 0.2 for the translation parameter, and c = 5.1
and d = −35 for the reﬂection parameters. The reﬂection angle function is
θ
n
= c +
d
r
2
n
and the reﬂectiontranslation iterative map is given by
x
n+1
= a + b(x
n
cos 2θ
n
+ y
n
sin 2θ
n
)
y
n+1
= b(y
n
sin 2θ
n
− y
n
cos 2θ
n
)
(11.7)
where:
b =
_
1 −
v
2
c
2
Figure 11.24(a) illustrates the simplest case, when the relativistic speed is v = 0.
A central circular chaotic bulge is present, located at (x, y) = (
a
2
, 0). The rest of the
object has a mirrorimage symmetry, islands are present, as well as two characteristic
points located at the solutions of (x
n+2
, y
n+2
) = (x
n
, y
n
). In the case of relativistic
speed, these are attracting points.
Figure 11.24(b) presents a relativistic object when the speed is much higher, v =
1000km/sec. A central chaotic bulge is present, but there are two attracting points
towards which the particles are directed, forming two rotating arms and creating the
form of a twoarmed spiral galaxy. A similar case with more clear arm structure is
illustrated in Figure 11.24(c), where the relativistic speed is v = 2000km/sec.
The reﬂectiontranslation at speed level v ≥ 23500km/sec has special characteris
tics, as presented in Figure 11.24(d). The trajectories of the lower part of the plane
guide the particles to an attracting point. The trajectories in the middle part of the
plane guide the particles into the central bulge, whereas the trajectories of the upper
part of the plane guide the particles into the two attracting arms of the galacticlike
object.
11.9 Rotating Disks of Particles
11.9.1 A circular rotating disk
A number of galaxy simulations are based on the idea that particles are originally
situated in a rotating disk. The simplest case is to consider a circle of some radius
Chaos in Galaxies and Related Simulations 249
(a) v = 0 (b) v = 1000km/sec
(c) v = 2000km/sec (d) v = 23500km/sec
FIGURE 11.24: Relativistic reﬂectiontranslation
250 Chaotic Modelling and Simulation
R (i.e. a ﬂat disk). The particles at time t = 0 are distributed according to a cer
tain distribution. For galaxy simulations, we assume that the collisions between the
particles are negligible. The governing force is a central force due to gravity. In the
application a cutoﬀ radius has been used, to prevent the denominator becoming 0.
Figure 11.25(a) illustrates the rotating disk of particles in its original position. The
upper left part of the ﬁgure shows the uniform distribution generating the particles,
whereas the lower left part shows the density distribution of the distance of the par
ticles from the centre. This density follows an inverse power law (∼ 1/r
e
). The
particles are placed according to a joint distribution on the polar coordinate space
(r, θ), that is uniform in θ and follows an inverse power law in r (α ≥ 0):
f (r, θ) = cr
−α
, θ ∈ [0, 2π], r ∈ [r
0
, R] (11.8)
Then, the density of the particles on a small annulus of radius r would follow an
inverse power law (∼ r
−α−1
).
11.9.2 The rotating ellipsoid
Another important case is that of the formation of an ellipsoidal disk of rotating
particles. In several cases, rotating bodies take the shape of a rotating ellipsoid. A
disk of particles can form an ellipsoid when external forces are present. When an
ellipsoidal disk is rotated, twoarmed spirals appear. Similar to the aforementioned
circular disk formation, an ellipsoidal disk with gravitational particles is assumed to
be denser in the centre, and then the density gradually decreases from the centre to
the outer part of the ellipsoid, as illustrated in Figure 11.25(b).
(a) A rotating circular disk (b) A rotating ellipsoidal disk
FIGURE 11.25: Rotating disks
Chaos in Galaxies and Related Simulations 251
The set of equations that provide the twodimensional ellipsoid is
x
t
= a cos φ
t
y
t
= b sin φ
t
z
t
= z
0
(11.9)
or:
_
x
a
_
2
+
_
y
b
_
2
= 1
When the system turns by an angle of θ
t
from time t to time t + 1, the equations
become:
x
t+1
= a cos(φ
t
+ θ
t
)
y
t+1
= b sin(φ
t
+ θ
t
)
z
t+1
= z
0
(11.10)
Using the trigonometric sum formulas, the following iterative formula can be ob
tained:
x
t+1
= x
t
cos θ
t
−
a
b
y
t
sin θ
t
y
t+1
=
b
a
x
t
sin θ
t
+ y
t
cos θ
t
z
t+1
= z
0
(11.11)
By setting h =
b
a
, assuming that 0 < b < a (0 < h < 1), the system (11.11) takes the
form:
x
t+1
= x
t
cos θ
t
−
1
h
y
t
sin θ
t
y
t+1
= hx
t
sin θ
t
+ y
t
cos θ
t
z
t+1
= z
0
(11.12)
In Figure 11.26(a) a twoarmed spiral galaxy is formed, based on a rotating ﬂat
ellipsoid with eccentricity parameter h =
b
a
=
1
3
. The system (11.12) is used, with
the rotation angle being
θ
t
=
d
_
r
0
+ (30r)
7
where the cutoﬀ radius is r
0
= 0.01 and the rotation parameter is d = 6.7.
The elapsed time is t = 10 time units. When t = 1, the form of a twoarmed
spiral results (Figure 11.26(b)). By observing the two images, it becomes clear that
the original twoarmed spiral at time t = 1 is transforming into a twoarmed bar
formation after time t = 10.
The ellipsoidal disk of particles is rotated counterclockwise. After t = 5 time units,
a twoarmed spiral is present, located at the centre of coordinates (Figure 11.27(a)).
The rotation follows an angle equal to θ =
0.01
0.001+r
2
, whereas b = 1, resulting in an
areapreserving map. A central chaotic bulge leads to two spiral arms.
252 Chaotic Modelling and Simulation
(a) A twoarmed bar galaxy (t = 10) (b) A twoarmed spiral galaxy (t = 1)
FIGURE 11.26: Twoarmed galaxies
(a) A twoarmed spiral galaxy (t = 5) (b) A twoarmed spiral galaxy (t = 5)
with space contraction b = 0.9
FIGURE 11.27: Twoarmed galaxies
Chaos in Galaxies and Related Simulations 253
A similar case, but with b = 0.9, is presented in Figure 11.27(b). After time
t = 5, the original ellipsoid is smaller and the resulting twoarmed spiral has a central
region resembling a bar, rather than a simple bulge. The space contraction relations
for the ellipsoid are
x
t+1
= b
_
x
t
cos θ
t
−
1
h
y
t
sin θ
t
_
y
t+1
= b
_
hx
t
sin θ
t
+ y
t
cos θ
t
_
z
t+1
= z
0
(11.13)
11.10 Rotating Particles under Distant Attracting Masses
11.10.1 One attracting mass
We consider a number of rotating particles in a disk of radius r = 1, attracted by an
outside mass located on the xaxis, at distance h from the origin. The density of the
disk of the initial rotating particles follows an inverse power law as previously. The
particles rotate with the same angular velocity. The issue at hand is to explore the
shape of this disk of particles after the outside mass has exerted its inﬂuence for time
t. The attracting force is of course the gravitational force. If the rotation velocity of
the particles on the disk is very small, or equal to zero, the particles will be attracted
by the mass and will be directed to it. But, as the angular velocity is higher, a number
of particles are trapped inside the rotating disk. The distribution of these particles
inside the disk is not uniform. A central chaotic region can be seen. The paths of the
particles are chaotic in this region. The particles are trapped in this chaotic region,
and aside from their chaotic movement, they tend to keep their position in the chaotic
region. In other words the chaotic movement is responsible for a space stability, in
that the particles move chaotically in the ﬁxed space. On the other hand, the particles
moving outside this region are following nonchaotic paths, and are mainly attracted
by the outside mass.
The rotationattraction model results from the rotation model after a transforma
tion that takes into account the inﬂuence of the attracting mass. The twodimensional
equations have the form
x
t+1
=
d(h − x)
_
(h − x)
2
+ y
2
_
3/2
+ x
t
cos θ
t
− y
t
sin θ
t
y
t+1
= −
dy
_
(h − x)
2
+ y
2
_
3/2
+ x
t
sin θ
t
+ y
t
cos θ
t
(11.14)
where h and d are parameters and the rotation angle is:
θ
t
=
c
_
r
2
0
+ r
2
t
_
3/2
254 Chaotic Modelling and Simulation
The cutoﬀ radius is r
0
, c is a rotation parameter and r
t
=
_
x
2
t
+ y
2
t
. As is obvious
from (11.14), the inﬂuence of the attracting mass is expressed by the ﬁrst term on the
right hand side of the relations. These terms arise by considering the gravitational
force acting on the small mass m of a rotating particle located at (x, y), from the
attracting mass M located on the xaxis at (x, y) = (h, 0). Therefore the Euclidean
distance between the particle m and the mass M is
r
(m,M)
=
_
(h − x)
2
+ y
2
and the attracting force is
F =
GMm
r
2
(m,M)
where G is the gravitational constant. The acceleration of the particle therefore is:
γ
m
=
GM
r
2
(m,M)
The xcomponent of the acceleration is
γ
x
=
γ
m
(h − x)
r
(m,M)
=
GM(h − x)
r
3
(m,M)
or
γ
x
=
d(h − x)
_
(h − x)
2
+ y
2
_
3/2
where d = GM.
Similarly, the ycomponent of the acceleration is
γ
y
= −
γ
m
y
r
(m,M)
= −
GMy
r
3
(m,M)
or:
γ
y
= −
dy
_
(h − x)
2
+ y
2
_
3/2
In other words, the xcomponent of the acceleration moves the particle towards the
positive xdirection, whereas the ycomponent of the acceleration directs the particle
towards the centre of the rotating disk.
Figure 11.28(a) illustrates the rotating disk at time t = 0, before the attracting mass
is introduced. Small circles in the periphery, at r = 1, indicate a number of rotating
particles. Figure 11.28(b) shows the eﬀect of the introduction of an attracting mass at
distance h = 3 from the origin. The parameters are r
0
= 0.001, c = 0.4 and d = 0.8,
and the elapsed time is t = 5.
In Figure 11.28(b), the original outer (peripheral) particles change from a circu
lar form into an eggshaped image. However, several particles that originated from
Chaos in Galaxies and Related Simulations 255
(a) The rotating disk, r = 1. Outer parti
cles at r = 1 rotate before the attracting
mass is introduced.
(b) The rotating particles after time t =
5 from the introduction of the attracting
mass
FIGURE 11.28: Rotating particles under the inﬂuence of an attracting mass
within the disc, with r < 1, are attracted from the mass M, escape outside the egg
shaped form and approach the attracting mass. These particles have an energy level
higher than the escape energy level. The chaotic part of this ﬁgure is located in the
central area of the rotating disk. The particles in the chaotic region at time t = 5 form
a twoarmed spiral rotating image. A chaotic path of such a particle is presented in
Figure 11.29(a). The particle follows a boxlike path. The elapsed time is t = 100
and the parameters are c = 0.2, d = 0.3. The original position of the particle is at
(x, y) = (0.2, 0.6).
Figure 11.29(b) illustrates a view of the twoarmed spiral galaxy and the surround
ing cloud at time t = 4. The original disk radius is r = 1, the cutoﬀ radius is
r
0
= 0.001, and the density follows an inverse power law ( ∼ 1/r
5
). The parameters
are c = 0.567, d = 0.1, and the distance of the attracting mass is h = 3.
11.10.2 The area of the chaotic region in galaxy simulations
Computer experiments show that the chaotic region in the above simulations is
larger when the rotation velocity is higher. However, it is hard to say that the con
struction of a chaotic area equation is possible. Another important point is that the
chaotic area changes as the time t varies. If the chaotic area is stabilised after enough
time has elapsed, then we say that the stochastic area is asymptotically stable, and we
can search for possible estimation methods. One approach is to construct a multidi
mensional graph related to the chaotic area, with the rotation velocity, the attracting
mass M and the distance of the attracting mass h as parameters.
Such an asymptotically stable chaotic attractor is illustrated in Figure 11.30. The
circular central part and the rotating arms appear. The attracting mass M is located
at distance h = 3 to the right of the attractor, at (x, y) = (3, 0). The parameters are
c = 0.5 and d = 1. The centre of the circular central part has moved to the right, at
256 Chaotic Modelling and Simulation
(a) A particle trapped in the chaotic re
gion, that oscillates between the spiral
arms
(b) Aviewof the twoarmed spiral galaxy
and the surrounding cloud at time t = 4
FIGURE 11.29: Boxlike paths of a twoarmed spiral galaxy
FIGURE 11.30: An asymptotically stable chaotic attractor
Chaos in Galaxies and Related Simulations 257
(x, y) = (0.117, 0). The outer parts of the attractor extend to x ≈ 0.61 to the right and
x ≈ −0.52 to the left. The larger y coordinate is at y ≈ 0.62, and the smaller is at
y ≈ −0.57.
11.10.3 The speed of particles
There are several theories for the distribution of the speed of particles in rotating
disks. Furthermore, observations of galaxies show that the mean rotation speed in
the periphery of the disk is higher than the speed predicted from theory. This urged
astronomers to conclude the existence of nonobserved mass in the galaxies, the
wellknown dark matter. Every simulation based on a rotating disk of particles has
to take into account the gradual change of the particle density, an inverse power law
of ∼ 1/r
5
seems to be acceptable, and results in a rotation velocity distribution that
is closer to the results of dark matter theory. As it is not, even nowadays, completely
clear how to ﬁnd a global model for the rotating disks of particles that would be quite
close to real situations, a test of our model regarding the rotation speed is called for.
The simple assumption of a stable angular rotation speed will lead to very high
rotation speeds for large r. In the proposed models, the rotation angle is usually
equal to an inverse function of the radius r from the origin, giving a function for
the rotation speed that is close to some real situations. In the following computer
experiment, the rotation speed is estimated and presented in Figure 11.31.
The rotation disk has radius r = 2. The rotation equations (11.14) are those pre
sented above. Estimates of the mean speed of particles at concentric areas of radius
diﬀering by 0.1 are presented in Figure 11.31. The graph veriﬁes the results known
from real situations. The rotation speed is initially growing and reaching an upper
limit; thereafter it gradually declines. A divergence is estimated at radius r = 0.40
and r = 0.45.
FIGURE 11.31: Rotation speed vs. radius r
258 Chaotic Modelling and Simulation
11.11 Two Equal Attracting Masses in Opposite Direc
tions
We now consider the case where the rotating particles in a disk of radius r = 1
are attracted by two equal outside masses, located on the xaxis at distances h and
−h from the origin. The density of the disk of rotating particles follows an inverse
power law, of the order 1/r
n
, where n = 3, 4, 5, . . .. The particles rotate with the
same angular velocity. As before, we wish to explore the shape of the original disk
of particles after the outside masses have exerted their inﬂuence for time t. The
attracting force is again the gravitational force.
The situation is in many ways similar to that of one attracting mass. If the rotation
velocity of the particles on the disk is very small, or equal to zero, the particles
on the disk will be attracted from the masses and will be separated in two parts
and directed towards the masses. On the other hand, when the angular velocity is
higher, a number of particles are trapped inside the rotating disk. The space inside
this disk is not uniform. A central chaotic region is again visible. The paths of the
particles are chaotic in this region, exhibiting the same kind of space stability as
before. The particles moving outside this region are following nonchaotic paths,
mainly inﬂuenced and attracted by the outside masses. As the two equal masses act
in opposite directions, the resulting forms will be symmetric and the galactic forms
would be twoarmed spirals.
The twomassrotationattraction model results from (11.14) by taking into ac
count the inﬂuence of the two opposite attracting masses. The twodimensional
equations have the form
x
t+1
= A(h − x
t
) + x
t
cos θ
t
− y
t
sin θ
t
y
t+1
= −Ay
t
+ x
t
sin θ
t
+ y
t
cos θ
t
(11.15)
where A is given by
A =
d
_
(h − x)
2
+ y
2
_
3/2
+
d
_
(−h − x)
2
+ y
2
_
3/2
h and d are parameters, and the rotation angle is:
θ
t
=
c
_
r
2
0
+ r
2
t
_
3/2
The cutoﬀ radius is r
0
, c is a rotation parameter and r
t
=
_
x
2
t
+ y
2
t
.
As is clear from the above map, the inﬂuence of the attracting masses is expressed
by the term A. This term arises by considering the gravitational force acting on the
small mass m of a rotating particle located at (x, y) from the attracting masses M
located at (x, y) = (h, 0) and (x, y) = (−h, 0) on the xaxis, similar to the case of one
Chaos in Galaxies and Related Simulations 259
FIGURE 11.32: Two attracting masses and the disk of rotating particles at
time t = 0
attracting mass. Figure 11.32 illustrates this situation at time t = 0. The attracting
masses are located on the xaxis at distances h
1
= 3 and h
2
= −3 from the origin.
Figure 11.33(a) shows the results of the simulation. The parameters are d = GM
and h = 3. The resulting rotationattraction form after time t = 1 is a perfect sym
metric twoarmed spiral.
(a) A twoarmed spiral after time t = 1 (b) Rotationattraction image at t = 2
FIGURE 11.33: Rotating disks with two attracting masses
At time t = 2, the twoarmed spiral starts to change to a more complicated form
(Figure 11.33(b)), whereas at time t = 5 the resulting form is closer to an ellipsoidal
centre with outer circular rings (Figure 11.34(a)). The asymptotically stable rotation
attraction form for large time t is illustrated in Figure 11.34(b).
Another example is related to two equal masses located on the xaxis, at h
1
= 3
and h
2
= −3, and inﬂuencing the original rotating disk of particles. The parameters
are c = 2.2 and d = d
1
= 2. For suﬃciently large time t, the circular rotating disk
takes the form presented in Figure 11.34(c). The centre of the circlelike central
260 Chaotic Modelling and Simulation
(a) Rotationattraction for
t = 5
(b) Asymptotically stable
rotationattraction when
t ≫ 1000
(c) Asymptotically stable
rotationattraction over a
long time
FIGURE 11.34: Rotating disks with two attracting masses over a long time
disk (bulge) is at the centre of coordinates. Symmetric arms and eyelike forms are
extended close to the central disk.
11.11.1 Symmetric unequal attracting masses
In the above example the two equal attracting masses act from opposite directions
in the xaxis on the rotating disk of particles. The centre of this disk was originally
located at the centre of coordinates. In this section we examine the case where the
two masses located at (x, y) = (h = 3, 0) and (x, y) = (h1 = −3, 0) have unequal
masses. Then, equations (11.15) still hold, where A now is:
A =
d
1
_
(h − x)
2
+ y
2
_
3/2
+
d
2
_
(−h − x)
2
+ y
2
_
3/2
We will take those masses to be M for the ﬁrst and M/4 for the second. Now, as the
opposing attracting forces are not equal, the disk of particles moves to the right and
the galaxy arms are not symmetric. In Figure 11.35(a) the parameters are c = 0.5,
d
1
= 2 for the large mass M and d
2
= d
1
/4 for the small mass M/4. The object is a
chaotic attractor. This is the attracting space for every original point mass left in the
vicinity of the attractor and having enough time to enter the attracting space. Two
main arms appear, as well as a smaller third arm originating from the second arm.
Figure 11.35(b) illustrates the case where the parameter expressing rotation has
a higher value, c = 0.8. Two nonsymmetric arms appear. These are connected
to the central disk with smaller circular parts. The eﬀect of a varying parameter c
is illustrated in Figures 11.35(c) and 11.35(d). The chaotic arms are larger, as the
parameter is c = 1.1 in the ﬁrst ﬁgure and c = 2 in the second.
Another interesting point is the form that the initial disk takes at the very early
time periods. The parameters are c = 2, d
1
= 2, d
2
= d
1
/4. At time t = 1, a two
armed spiral is formed (Figure 11.36(a)). The two arms are diﬀerent. The larger one
is directed towards the bigger attracting mass to the right of the rotating disk, while
Chaos in Galaxies and Related Simulations 261
(a) c = 0.5 (b) c = 0.8
(c) A form of the chaotic attractor, when
c = 1.1
(d) Another form of the chaotic attractor,
when c = 2
FIGURE 11.35: Chaotic attractors in the case of two unequal symmetric
attracting masses
262 Chaotic Modelling and Simulation
the smaller is directed towards the small attracting mass to the left of the centre of
coordinates.
At time t = 2, both spiral arms are directed to the big attracting mass on the right
of the coordinate centre (Figure 11.36(b)). At time t = 3 a new spiral arm is added
(Figure 11.36(c)). All three arms are directed towards the big attracting mass located
on the righthand side. At time t = 4 the total number of the spiral arms is now 4
and all are directed towards the big attracting mass on the right (Figure 11.36(d)).
More complicated forms arise at times t = 5, t = 6 and t = 8 as illustrated in
Figures 11.37(a) through 11.37(c).
(a) A twoarmed spiral, t = 1 (b) A twoarmed spiral, t = 2
(c) A threearmed spiral, t = 3 (d) A fourarmed spiral, t = 4
FIGURE 11.36: Rotating disc at early times
Chaos in Galaxies and Related Simulations 263
(a) t = 5 (b) t = 6 (c) t = 8
FIGURE 11.37: Multiarm images
11.12 Two Attracting Equal Nonsymmetric Masses
We consider now the case where there are two equal attracting masses, located on
the xaxis and yaxis respectively, at equal distance h from the origin. Once again,
the density of the disk of rotating particles follows an inverse power law. As in the
previous cases, the space inside this disk is not uniform. A central chaotic region is
again visible, and paths of the particles in this region are chaotic. As the two equal
masses act on the rotating particles fromperpendicular directions, the resulting forms
must be nonsymmetric, and the galactic forms would be nonsymmetric twoarmed
spirals.
The rotationattraction model in this case takes the form
x
t+1
= A
1
(h − x) − A
2
y + x
t
cos θ
t
− y
t
sin θ
t
y
t+1
= −A
1
y + A
2
(h − y) + x
t
sin θ
t
+ y
t
cos θ
t
(11.16)
where
A
1
=
d
_
(h − x)
2
+ y
2
_
3/2
A
2
=
d
_
x
2
+ (h − y)
2
_
3/2
h and d are parameters, and the rotation angle is
θ
t
=
c
(r
2
0
+ r
2
t
)
3/2
r
0
is the cutoﬀ radius, c is a rotation parameter and r
t
=
_
x
2
t
+ y
2
t
.
As is obvious from the above map, the inﬂuence of the attracting masses is ex
pressed by terms A
1
, A
2
respectively. These terms arise by similar considerations as
above. Figure 11.38(a) illustrates the situation at time t = 1. The distance is h = 3,
and the rotating disk radius is r = 1.5.
264 Chaotic Modelling and Simulation
(a) t = 1 (b) Large time t
FIGURE 11.38: The eﬀect of two perpendicularly attracting masses
The two spiral arms are directed towards the attracting masses. The central bulge
has a circular form. Another circular ring outside the central bulge is the starting
point of the spiral arms. The parameters are c = 1.3, d = 1 and h = 3. Two non
symmetric spiral arms are formed for large time t as presented in Figure 11.38(b).
Chaos in Galaxies and Related Simulations 265
Questions and Exercises
1. Show that equations (11.1) imply the following relation for the motion of a
mass m under the inﬂuence of one large mass M, with MG = 1:
1
2
_
x
′2
+ y
′2
_
−
1
r
= const.
2. From the relation in the previous exercise, determine the escape velocity, by
considering that for a mass to escape, its speed at r = ∞ should be “non
negative.”
3. Letting MG = 1 for simplicity, the equation
mv
2
cycl
r
=
mMG
r
2
=
m
r
2
gives us the velocity of a particle that moves in a circular orbit of radius r
around an attracting mass M. Use this equation, in combination with the re
sults in the previous questions, to show that the escape velocity of a particle at
some distance from the mass, and the rotating velocity of that particle at the
same distance, are related via:
v
esc
=
√
2v
cycl
4. Consider the system
x
′
1
= x
3
x
′
2
= x
4
x
′
3
= −
1
r
c
x
1
x
′
4
= −
1
r
c
x
2
where r =
_
x
2
1
+ x
2
2
. Find the trajectories of this system in the (x
1
, x
2
) plane
when c =
√
2. Do the same when c = 1 and c = 1/2. (Use a fourth order
RungeKutta algorithm and initial values x
1
= 1, x
2
= 0, x
3
= 0 and x
4
= 1.3.)
5. Use the translation reﬂection equations
x
t+1
= a + x
t
cos(2θ
t
) + y
t
sin(2θ
t
)
y
t+1
= x
t
sin(2θ
t
) − y
t
cos(2θ
t
)
to create images with outer and inner formations. The rotation angle should
have the form θ
t
= c + d/r. One interesting choice of parameters is: a = 1,
c = 5.1 and d = −20, with initial values x = a and y = 0. Try other parameter
values as well.
266 Chaotic Modelling and Simulation
6. Use the parametric equations
x
t
= r
t
cos(t)
y
t
= r
t
sin(t)
where
r
t
=
ep
1 − e cos(t)
to obtain ellipses. Start with parameter values e = 0.5 and p = 2.
7. Use the parametric equations
x
t
= r
t
cos(tr
t
)
y
t
= r
t
sin(tr
t
)
where
r
t
=
ep
1 − e cos(t)
to draw trajectories in the (x
t
, y
t
) plane. Start with parameter values e = 0.5
and p = 2, and try several values for e. What do you observe?
8. Draw the (x
t
, y
t
) trajectories for the iterative process:
x
t+1
= 0.15 + 0.5r
t
cos(t)
y
t+1
= r
t
sin(t)
where
r
t
= 0.5 −
1
_
x
2
t
+ y
2
t
9. For the ContopoulosBozis simulation described in Section 11.3.1, try other
values for the space contracting parameter b and observe the resulting (x
t
, y
t
)
diagrams.
10. Show that if particles are distributed on a disk according to the probability
density function (11.8), then the density of particles on any thin annulus at
radius r will be proportional to r
−α−1
.
Chapter 12
GalacticType Potentials and the
H´enonHeiles System
12.1 Introduction
The work of H´ enon and Heiles on galactic motion (H´ enon and Heiles, 1964) is
the ﬁrst complete work proving the existence of chaos in galaxy formations. From
its ﬁrst appearance in the Astronomical Journal (1964), it provided considerable evi
dence of the existence of chaotic motions in galaxies. The work by H´ enon and Heiles
started as a computer experiment in order to explore the existence and applicability
of the third integral of galactic motion. Earlier, in 1956, George Contopoulos (Con
topoulos, 1956) started a pioneering work on galactic motion and renewed the in
terest on the third integral of motion (Contopoulos, 1960). Contopoulos explored
the boxlike paths of a body in a galactic potential and found that the orbits in the
meridian plane of an axisymmetric galaxy are like Lissajous
1
ﬁgures (Contopoulos,
1965). Contopoulos expected the orbits to be ergodic and ﬁll all the space inside the
energy surface (Contopoulos, 2002). Instead, he found that the orbits did not ﬁll all
the available space, but instead ﬁlled curvilinear parallelograms, like deformed Lis
sajous ﬁgures. According to his writings he could prove later (Contopoulos, 1960)
that such orbits can be explained qualitatively and quantitatively by a formal third
integral of motion. The work of Contopoulos and H´ enonHeiles was a result of what
we call computer experiments. This new type of experiments gave new directions
to various scientiﬁc ﬁelds, including astronomy. The computer results were some
times surprising, and often contradicting existing theories. Contopoulos (1958) used
a computer in Stockholm Observatory in 1958, and H´ enon and Heiles (1964) used a
computer at Princeton University from 1962 to 1963.
1
Lissajous curves are given in parametric form by equations that describe complex harmonic motion.
267
268 Chaotic Modelling and Simulation
12.2 The H´enonHeiles System
The Hamiltonian applied by H´ enon and Heiles (1964) is of the form:
H =
1
2
( ˙ x
2
+ ˙ y
2
) +
1
2
_
x
2
+ y
2
+ 2x
2
y −
2y
3
3
_
= h (12.1)
where the potential U(x, y) is
U(x, y) =
1
2
_
x
2
+ y
2
+ 2x
2
y −
2y
3
3
_
(12.2)
or in polar coordinates:
V(r, θ) =
1
2
r
2
+
1
3
r
3
sin(3θ)
A more general form for a potential would be:
V(r, θ) = r
4
+ ar
2
+ br
3
cos(3θ)
The Hamiltonian for a generalised H´ enonHeiles potential is (p
x
= ˙ x, p
y
= ˙ y)
H =
1
2
(p
2
x
+ p
2
y
+ Ax
2
+ By
2
) + Dx
2
y −
1
3
Cy
3
FIGURE 12.1: The original H´ enonHeiles (y, ˙ y) diagram
Here, we reproduce by simulation the original ﬁgure presented in the H´ enon
Heiles paper (Figure 12.1). They estimated the paths of a particle according to the
previously mentioned Hamiltonian at an energy level E = h = 1/12, and then they
gave the (y, ˙ y) diagram when x = 0. At this low energy level mainly regular orbits
appear. At a higher energy level chaotic regions are formed.
GalacticType Potentials and the H´ enonHeiles System 269
(a) E = 1/8 (b) E = 1/6
FIGURE 12.2: (y, ˙ y) diagrams for the H´ enonHeiles system
A Poincar´ e section at x = 0 is presented in Figure 12.2(a) for the H´ enonHeiles
model. The energy level is at E = 1/8 = 0.125. Regular and chaotic trajectories are
present. Several islands appear surrounded by the chaotic sea.
In the H´ enonHeiles Hamiltonian, the upper limit for the energy is E = h = 1/6.
Figure 12.2(b) illustrates this case. The (y, ˙ y) diagram is almost totally chaotic. How
ever, small islands remain inside the chaotic sea.
The H´ enonHeiles system is given by the following set of diﬀerential equations:
˙ x = u
˙ y = v
˙ u = −x − 2xy
˙ v = −y − x
2
+ y
2
(12.3)
The Jacobian of this system is
J = 1 − 4
_
x
2
+ y
2
_
= 1 − 4r
2
This system results from the Hamiltonian (12.1), and the related potential (12.2)
that H´ enon and Heiles chose for their famous computer experiment. The selection
of this function to express the potential in the above Hamiltonian has the advantages
that it is relatively simple but, at the same time, it gives adequate information related
to chaotic and nonchaotic paths. During those early days of computer use and the
discovery of chaotic surprises in nonlinear systems, it was important to use sim
ple nonlinear functions in order to explore chaos. Many scientists of the time were
able to verify the validity of the theory and expand the results into new ﬁelds. Con
topoulos (1960) explored a ﬁeld of nonlinear dynamics in a galaxy, which was later
analysed at Princeton University in 1963 by the French astronomer Michel H´ enon
and his colleague Carl Heiles from the United States. The paper was published one
year later in the Astronomical Journal (H´ enon and Heiles, 1964). In 1963 Lorenz’s
famous work on chaotic modelling was also published (Lorenz, 1963).
270 Chaotic Modelling and Simulation
12.3 Discrete Analogues to the H´enonHeiles System
In this section we examine discrete analogues to the H´ enonHeiles model. Discrete
models are very important in studying chaotic systems, as they are usually simpler
than the continuous time systems and, quite often, they explore other aspects of the
chaotic system that are not present in continuous time models. The simulation of this
discrete model gives an image presented in Figure 12.3 at the energy level E = 1/12.
The parameter c was set to 0.1. The discrete model is based on the following set of
equations:
x
n+1
= cx
n
+ u
n
y
n+1
= cy
n
+ v
n
u
n
= −x
n
− 2x
n+1
y
n+1
v
n
= −y
n
− x
2
n+1
+ y
2
n+1
(12.4)
The Jacobian of the discrete model is:
J = 1 + 2(c − 1)y
n+1
+ 4(c − 1)
_
x
2
n+1
+ y
2
n+1
_
It is obvious that, when c = 1, J = 1 and then the spacepreserving property is sat
FIGURE 12.3: A discrete analogue to the H´ enonHeiles system (E = 1/12)
isﬁed. However, when c takes values less than 1, very interesting behaviour results,
as in Figure 12.3, where c = 0.1. The similarities with the H´ enonHeiles system are
obvious when comparing Figure 12.3 with Figure 12.1.
H´ enon and Heiles proposed and applied a diﬀerent, simple, discrete model:
x
n+1
= x
n
+ a
_
y
n
− y
3
n
_
y
n+1
= y
n
− a
_
x
n+1
− x
3
n+1
_ (12.5)
This model has only one parameter a. When a = 1.6, Figure 12.4(a) results
from a set of initial values (x, y). A very interesting property of model (12.5) arises
GalacticType Potentials and the H´ enonHeiles System 271
when a = 0.1. The form presented in Figure 12.4(b) has similarities with the above
proposed more complicated discrete form of the H´ enonHeiles model.
(a) a = 1.6 (b) a = 0.1 (c) a = 2.6
FIGURE 12.4: The discrete H´ enonHeiles model
The simple model used by H´ enon and Heiles has more interesting properties as is
illustrated in Figure 12.4(c). Here, a = 2.6, and the graph represents a barlike image
with two tails. The challenge now is to apply rotation to this simple model, in order
to explore the new features of the model. The classical rotation equation is applied,
with a stable rotation angle θ = 0.9. The graph, in this case, turns out to be a chaotic
attractor, as illustrated in Figure 12.5(a).
(a) A chaotic attractor for the discrete
rotation H´ enonHeiles model
(b) A twoarmed spiral galaxy
FIGURE 12.5: Discrete analogues to the H´ enonHeiles system
The graph in Figure 12.5(b), illustrating a twoarmed spiral galaxy, is produced by
272 Chaotic Modelling and Simulation
using the following simple rotation model:
x
n+1
= b(x
n
cos a − y
n
sin a) + x
2
n
sin a
y
n+1
= b(x
n
sin a + y
n
cos a) − x
2
n
cos a
The parameter values for the simulation were a = 0.2 and b = 0.9. As the Jacobian
is J = b
2
, the second parameter is an areacontracting parameter.
12.4 Paths of Particles in the H´enonHeiles System
The H´ enonHeiles systemexamined above also gives rise to some interesting paths
in the (x, y) plane. The paths are characterised by the level of the potential at that
stage. The form of the potential is triangular. A totally chaotic trajectory is illustrated
in Figures 12.6(a) and 12.6(b). The energy level is E = h = 1/6. This is the escape
limit, at which the equipotential triangle is drawn. All the paths of the particle are
included within the triangle. The parameters in this case were x = 0, y = 0 and
v = 0.1. The value of u, u = 0.5668627 · · ·, was determined by the condition that the
energy level of Hamiltonian should be E = h = 1/6.
(a) E = 1/6 (b) E = 1/6
FIGURE 12.6: Chaotic paths in the H´ enonHeiles system
Figure 12.7(a) illustrates the equipotential lines for the H´ enonHeiles system. The
particle is trapped inside this space, and the paths follow ordinary or chaotic paths
depending on the energy level in each case. The limits for the potential are estimated
GalacticType Potentials and the H´ enonHeiles System 273
by computing the critical points of the potential function, solving the system:
∂U
∂x
= −¨ x = x + 2xy = 0
∂U
∂y
= −¨ y = y + x
2
− y
2
= 0
This systemhas four solutions. One is (0, 0), the point with the minimumpotential.
The other three are all points of maximum potential U(x, y) = 1/6:
(x, y) = (0, 1)
(x, y) = (
√
3/2, −1/2)
(x, y) = (−
√
3/2, −1/2)
It is, then, easy to draw the equilateral triangle with corners located at these three
coordinates for (x, y). In Figure 12.7(a), several equipotential lines are drawn. The
potential U is constrained in the interval (0, 1/6). A nonchaotic orbit is computed
and presented in Figure 12.7(b) for the energy level E = 1/12. The parameter values
for the simulation were x = 0.39581, y = 0, ˙ x = u = 0.1 and ˙ y = v = 0.
(a) Equipotential curves (b) A nonchaotic orbit
FIGURE 12.7: The H´ enonHeiles system
12.5 Other Forms for the Hamiltonian
Several other forms for the Hamiltonian can be used in galaxy simulations. A
number of these models are based on Hamiltonians with a fourth power order po
tential. Such a potential, presented in this section, is closer to real situations than
274 Chaotic Modelling and Simulation
FIGURE 12.8: Boxlike orbits
the third power order potential of the H´ enonHeiles model. Boxlike orbits, as well
as orbits of other types, arise in this case. Figure 12.8 presents the eﬀect of rota
tion applied to particles following boxlike orbits (left image). The image on the
right represents a fourarmed spiral galaxy that results when the moving particles
are rotated by using the rotation equations. As the case under discussion involves
stellar bodies, the rotation angle must be selected to follow an inverse law with re
gards to the distance r of the particles from the centre of coordinates, as suggested in
chapter 11. In this application, the following equation is used for the rotation angle:
θ = 1 +
1
(r
o
+ r)
3/2
The cutoﬀ radius is equal to r
0
= 0.1, the simple rotation is expressed by the ﬁrst
term on the right, while the second term represents the inﬂuence of the attracting
forces on the rotating particles.
The Hamiltonian proposed here has the form:
H =
1
2
(u
2
+ v
2
) + U(x, y)
where the potential U(x, y) is:
U(x, y) =
x
2
2
+
y
2
2
+
x
4
4
−
y
4
4
− x
2
y
2
The parameters used in the above simulation were x = 0, y = −0.25, u = 0.23 and
v = 0.07.
The equations for ˙ u and ˙ v follow from the potential function, and are
−˙ u = x + x
3
− 2xy
2
−˙ v = y − y
3
− 2x
2
y
The maximum U(x, y) is achieved at x = 1/
√
5 and y =
√
3/
√
5, and it is h =
U(x, y) = 1/5.
GalacticType Potentials and the H´ enonHeiles System 275
12.6 The Simplest Form for the Hamiltonian
The simplest form for the Hamiltonian is:
H =
1
2
( ˙ x
2
+ ˙ y
2
) +
1
2
(x
2
+ y
2
) = h
where the potential
U(x, y) =
1
2
(x
2
+ y
2
)
This gives a family of circles centred at (x, y) = (0, 0) in a (x, y) diagram. The
equations for ¨ x and ¨ y are:
˙ u = ¨ x = −x
˙ v = ¨ y = −y
It is obvious that the minimum acceptable value for the potential is 0 at (x, y) = (0, 0),
while there is no upper limit. The paths are ellipses and, in some cases, circles.
Several paths are drawn in Figure 12.9(a) (left). The starting point is at (x, y) = (1, 0),
u was kept at 0, while the parameter v takes successively the values 0.1, 0.2, . . . , 1.0.
There appear 9 ellipses and one circle for v = 1.0 in the diagram on the left. The
diagram on the right side of the same ﬁgure illustrates the shape the orbits take after
rotation with an angle:
θ =
0.1
(0.001 + r)
3
The outer circle of the original diagram on the left remains a circle under the
rotation, but the ellipses gradually change shape, resulting in a twoarmed spiral.
The twoarmed spiral is formed at t = 1, following the original formation of the
circles and ellipses, as illustrated in Figure 12.9(a). If we follow the development
of the curves on the right side of the ﬁgures over time, we observe that there is
an anticlockwise turn of the galaxylike object, a larger central circle, and more
distinct arms. The cases for t = 5 and t = 20 are presented in Figures 12.9(b) and
Figure 12.9(c).
12.7 Gravitational Attraction
The case of a small mass m attracted by a large mass M has been widely explored.
Here, we assume that the large mass is located at the centre of coordinates, and it
remains stable. Anumber of particles are attracted by the large mass fromtheir initial
position (x, y), and directed towards the mass. The particles have initial velocities
˙ x = u and ˙ y = v at position (x, y). In this case, the energy conservation equation
276 Chaotic Modelling and Simulation
(a) t = 1
(b) t = 5
(c) t = 20
FIGURE 12.9: A simple Hamiltonian: orbits and rotation forms at various
times
GalacticType Potentials and the H´ enonHeiles System 277
holds:
E =
1
2
m
_
u
2
+ v
2
_
−
GMm
r
or
u
2
+ v
2
2
−
GM
r
=
E
m
= h
The potential is
U = −
GM
r
were G is the gravitational constant, and r =
_
x
2
+ y
2
. By assuming GM = 1, the
equations of motion are:
¨ x = ˙ u =
x
r
3
¨ y = ˙ v =
y
r
3
In this computer experiment, we place particles with mass m = 1 at position
(x, y) = (1, 0) and initial velocities u = 0 and v in the interval (0, 1). The v = 1 case
corresponds to a circular path located at (x, y) = (0, 0) and with radius r = 1. The
escape velocity is v =
√
2. In this case, the path is a hyperbola. When v ≥
√
2, the
particles escape to inﬁnity. The ten paths used have velocities v = 0.1, 0.2, . . . , 1.0.
These paths are illustrated in Figure 12.10(a) (left). The right side of Figure 12.10(a)
is the rotation image of the left side after time t = 1. The rotation angle follows the
relation:
θ =
0.5
(0.001 + r)
3
The resulting shape is that of a onearmed spiral galaxy.
Another computer experiment is based on two attracting masses, located at (x, y) =
(1, 0) and (x, y) = (−1, 0). The Hamiltonian in this case is
H =
1
2
(u
2
+ v
2
) −
1
r
1
−
1
r
2
=
E
m
= h
where r
1
=
_
(x − 1)
2
+ y
2
and r
2
=
_
(x + 1)
2
+ y
2
. Excluding the diverging paths,
the resulting image appears in the left part of Figure 12.10(b). In the right part of the
same ﬁgure, the rotation image is illustrated after time t = 1. A twoarmed spiral is
formed. The rotation angle is of the form:
θ =
0.1
(0.001 + r)
2
Another twoarmed spiral galaxy is formed after rotation for time t = 1 of the
image resulting when a moving particle is attracted by two masses located at (x, y) =
(1, 0) and (x, y) = (−1, 0), as in the previous case, but now with a slightly diﬀerent
rotation angle:
θ =
0.5
(0.001 + r)
2
278 Chaotic Modelling and Simulation
(a) A single attracting mass
(b) Two attracting masses
(c) Two attracting masses
FIGURE 12.10: Orbits and rotation forms at time t = 1
GalacticType Potentials and the H´ enonHeiles System 279
The initial parameters were x = 1.2, y = 0, u = 0 and v = 3.95. The paths are
illustrated in the left side of Figure 12.10(c), whereas in the right side the rotation
object after time t = 1 is presented.
12.8 A Logarithmic Potential
A logarithmic form for the potential is used in this section to simulate some inter
esting cases in galaxy formation. The potential has the form
U =
1
2
ln
_
R
2
+ x
2
+ c
2
y
2
_
where R is the cutoﬀ radius and c is a parameter. Figure 12.11(a) shows a box orbit
(left). The parameters are R = 0.1, c = 0.8 and the starting values are x = 0, y = 1,
u = 0.2 and v = 0. The right part of Figure 12.11(a) presents the rotation of the
boxlike orbit with a rotation angle:
θ =
0.01
(0.001 + r)
3
A chaotic orbit is illustrated in Figure 12.11(b). The parameters are R = 0.14,
c = 1.25 and the initial conditions are x = 0, y = 0.792121, u = 0.1 and v = 0. A
loop orbit is also possible by using the above potential. The loop avoids the centre, as
illustrated in the Figure 12.11(c), while the rotation image gives a twoarmed spiral.
The parameters are R = 0.14, c = 1.25 and the initial conditions are x = 0, y = 0.2,
u = 0.1 and v = 0.
12.9 Hamiltonians with a Galactic Type Potential: The
Contopoulos System
A Hamiltonian with a “galactic type” potential was ﬁrst introduced by Contopou
los (1958, 1960) in his pioneering work on galaxies. The potential is based on the
addition of two harmonic oscillators, along with higher order terms, to give the form:
U(x, y) =
1
2
_
w
2
1
x
2
+ w
2
2
y
2
_
− exy
2
The resulting Hamiltonian is:
H =
1
2
_
u
2
+ v
2
_
+ U(x, y) = h
280 Chaotic Modelling and Simulation
(a) Box orbit
(b) Chaotic orbit
(c) Loop orbit
FIGURE 12.11: Logarithmic potential: orbits and rotation forms at time t = 1
GalacticType Potentials and the H´ enonHeiles System 281
Without loss of generality this Hamiltonian can be simpliﬁed to:
2
H =
1
2
_
u
2
+ v
2
_
+
1
2
_
k
2
x
2
+ y
2
_
− exy
2
= 1/2
where k = w
1
/w
2
is the very important resonance ratio. The equations of motion are
given by:
˙ u = −
∂U
∂x
= −k
2
x − ey
2
˙ v = −
∂U
∂y
= y − 2exy
FIGURE 12.12: The Contopoulos system: orbits at resonance ratio 4/1 (left)
and 2/3 (right)
Resonant orbits characterised as 4/1 (left image) and 2/3 (right image) are illus
trated in Figure 12.12. For the case on the left, the parameters are e = 0.1 and
k = w
1
/w
2
= 4/1, and the initial conditions are x = 0.2, y = 0, u = 0 and v = 0.6.
For the case on the right, the parameters are e = 0.1 and k = w
1
/w
2
= 2/3, and the
initial conditions are x = 0.1, y = 0, u = 0 and v = 0.9977753.
12.10 Another Simple Hamiltonian System
A very simple Hamiltonian system that shows interesting properties has the fol
lowing Hamiltonian:
H =
1
2
_
u
2
+ v
2
_
+
1
2
_
x
2
+ y
2
_
−
1
2
x
2
y = h
2
Contopoulos (2002).
282 Chaotic Modelling and Simulation
The potential function is:
U(x, y) =
1
2
_
x
2
+ y
2
_
−
1
2
x
2
y
The equations of motion are
˙ u = −(x − xy)
˙ v = −(y − x
2
/2)
A number of equipotential curves are drawn based on the equation:
1
2
_
x
2
+ y
2
_
−
1
2
x
2
y = h
The analysis of this equation gives the acceptable range 0 < h < 1/2 for h. A
heavy line with equation x =
_
1 + y represents the h = 1/2 equipotential curve in
Figure 12.13(a). This curve, and the line y = 1, deﬁne the space where the closed
equipotential curves are located.
Particle orbits at high energy level for this Hamiltonian system are illustrated in
Figure 12.13(b). The parameters and initial values are: h = 1/2 − 0.001, x = 0.97,
y = 0, u = 0, and v is estimated from the Hamiltonian function.
(a) Equipotential curves (b) Orbits at high energy level
FIGURE 12.13: Another simple Hamiltonian system
GalacticType Potentials and the H´ enonHeiles System 283
Questions and Exercises
1. Consider the set of parametric equations:
x = sin(at)
y = sin(bt)
(a) Draw the curve corresponding to this system when the parameters are
a = 2 and b = 1.
(b) Determine the implicit equation for the curve in the case where a = 2
and b = 1.
(c) Examine the curves arising for other integer values of the parameters a, b.
(d) Consider the more general system:
x = sin(at + c)
y = sin(bt)
Examine the curves arising for various integer values for a, b, and for
c = π/k, k ∈ N.
2. Consider the system:
x
n+1
= x
n
+ a sin(y
n
)
y
n+1
= x
n
− a sin(x
n+1
)
(a) Determine the ﬁxed points of this system, and their stability.
(b) Compare this system to the system (12.5), and draw the characteristic
graphs of this system, especially the (x
n
, y
n
) diagrams, for various values
of the parameter a.
3. Draw the (x, y), (u, v), (x, u) and (y, v) diagrams for the discrete system (12.4)
for c = 0.1.
4. Draw the (x, y), (u, v), (x, u) and (y, v) diagrams for the HenonHeiles sys
tem (12.3).
Chapter 13
Odds and Ends
In this chapter we collect a number of interesting systems and examples, that we
did not ﬁnd an appropriate place for in the rest of this book. Topics include forced
nonlinear oscillators, the eﬀect of introducing noise in threedimensional attractors,
and the LotkaVolterra and pendulum systems. We close this chapter, and the book,
with ﬁve interesting attractors.
13.1 Forced Nonlinear Oscillators
The simple twodimensional model described by the diﬀerential equations
˙ x = −xy
2
˙ y = xy
2
− y
(13.1)
gives some very simple orbits in the (x, y) plane. If a sinusoidal forcing term
f (t) = a + b cos wt
is added in the ﬁrst equation, the system becomes
˙ x = f (t) − xy
2
˙ y = xy
2
− y
(13.2)
This new system can give interesting chaotic paths, as it is now threedimensional
(x, y, t). Figure 13.1 illustrates the (x, y) diagramfor this system, for parameter values
a = 0.999, b = 0.42 and w = 1.75. The resulting paths are chaotic in this case.
13.2 The Eﬀect of Noise in ThreeDimensional Models
We discuss here the eﬀect of adding noise to some of the models described in
Chapter 6, using computer simulations. Recall that the equations of the R¨ ossler
285
286 Chaotic Modelling and Simulation
FIGURE 13.1: Adding a sinusoidal forcing term to a twodimensional system
model are
˙ x = −y − z
˙ y = x − ez
˙ z = f + xz − mz
(13.3)
A multiplicative noise is added in every parameter, according to the formula
e
∗
= e(1 + ku
e
)
f
∗
= f (1 + ku
f
)
m
∗
= m(1 + ku
m
)
(13.4)
where u
e
, u
f
and u
m
are independent and uniformly distributed random variables in
the interval (−0.5, 0.5), and the noise parameter is k = 3.
The parameters for the simulation were set to e = 0.2, f = 0.4 and m = 5.7.
Figure 13.2(a) illustrates the threedimensional view of the R¨ ossler attractor that
arises. The main inﬂuence is related to the zaxis, where the maximum speed is
achieved. A very interesting observation is that even strong noise does not aﬀect the
general form of the attractor much. The paths on the (x, y) plane in particular show a
quite stable behaviour.
For the Lorenz model, the original equations are
˙ x = −sx + sy
˙ y = −xz + rz − y
˙ z = xy − bz
(13.5)
The form of the Lorenz attractor when multiplicative noise is added is presented in
Figure 13.2(b).
In this case, the noise is generated by a logistic process of the form:
f
n+1
= 4 f
n
(1 − f
n
)
This logistic process, when the chaotic parameter takes the highest value (4), may
generate a uniformlike distribution, giving random numbers between zero and one.
Odds and Ends 287
(a) Noise in the R¨ osler model
(threedimensional view)
(b) Noise in the Lorenz attractor (k = 20)
(c) Noise in the Lorenz attractor (k = 50) (d) Noise in the Lorenz attractor
(k = 150)
FIGURE 13.2: The eﬀect of noise on threedimensional models
288 Chaotic Modelling and Simulation
The results presented here are similar to those produced when uniform noise is used
instead.
The basic parameter values for the application were set to s = 10, b = 8/3 and
r = 28, and the noise parameter was set to k = 20. The resulting chaotic object has
a form similar to that of the Lorenz chaotic attractor. When the noise parameter k is
higher (k = 50), the resulting chaotic image retains the Lorenz attractor shape, yet,
in this case, the chaotic paths are mostly inﬂuenced by the noise term as illustrated
in Figure 13.2(c). Higher values of the noise term destroy the original shape of the
attractor as presented in Figure 13.2(d) where k = 150. At such a high noise level,
the chaotic image turns out to be a stochastic one.
13.3 The LotkaVolterra Theory for the Growth of Two
Conﬂicting Populations
A special case of the LotkaVolterra system was already discussed in section 6.2.
We expand on it somewhat in this section.
The LotkaVolterra system concerns the predatorprey problem, a problem fre
quently arising in ecology. Let N
2
be a measure of the population of a species,
which preys upon a second species whose population is measured by N
1
. Then the
population of N
1
diminishes with a factor proportional to the product of the two pop
ulations, N
1
N
2
, whereas the population of y increases with a factor proportional to
N
1
N
2
. The diﬀerential equations of growth or decline of the two populations are
therefore given by the following set of equations:
˙
N
1
= aN
1
− bN
1
N
2
˙
N
2
= −cN
2
+ dN
1
N
2
(13.6)
where the parameters a, b, c, d are positive numbers. This system may be simpliﬁed
by introducing the transformation:
N
1
=
c
d
x
N
2
=
a
b
y
The resulting system is
˙ x = a(x − xy)
˙ y = −c(y − xy)
(13.7)
Diﬀerentiating both equations, and eliminating y and ˙ y, the following nonlinear
diﬀerential equation for x is obtained:
¨ x =
1
x
˙ x
2
+ acx − c ˙ x + cx ˙ x − acx
2
Odds and Ends 289
or
¨ x =
1
x
˙ x
2
− c ˙ x(1 − x) + acx(1 − x)
The equation for the phase trajectories is
dy
dx
= −
c(y − xy)
a(x − xy)
or
c
_
˙ x −
˙ x
x
_
+ a
_
˙ y −
˙ y
y
_
= 0
Integrating this equation, we obtain
c(x − ln x) + a(y − ln y) = K
where K is the integration constant.
The limit cycles are closed curves around the equilibrium point M = (x, y) =
(1, 1). This point is indicated by a small cycle in Figure 13.3. The parameters are
a = 1 and c = 3. The initial values are (x, y) = (2, 1). The graph on the top of
Figure 13.3 illustrates the limit cycle, whereas the (t, x) (heavy line) and the (t, y)
(light line) graphs are presented in the lower part of the same ﬁgure.
FIGURE 13.3: The LotkaVolterra system: limit cycle (upper image) and (t, x)
(heavy line) and the (t, y) (light line) graphs
The phase trajectories (x, y) for the LotkaVolterra system are illustrated in Fig
ure 13.4(a). The parameters are a = 1 and c = 3. Figure 13.4(b) illustrates the
(x, ˙ x) diagram (closed curve) and the (t, x) oscillations after a numerical solution of
the nonlinear second order diﬀerential equation for x presented above as a solution
of the LotkaVolterra system. The same parameter values (a = 1 and c = 3) and
starting values for x and y are selected as in the previous case (x = 2 and y = 1). The
resulting oscillations for x are precisely the same as in the previous case as it was ex
pected. Similar results can be obtained after a numerical solution of the diﬀerential
equation for y.
290 Chaotic Modelling and Simulation
(a) (x, y) phase trajectories (a = 1, c = 3) (b) (x, ˙ x) diagram (closed curve) and the
(t, x) oscillations
FIGURE 13.4: The LotkaVolterra system
13.4 The Pendulum
Numerous applications and advances in the study of nonlinear functions originate
from the pendulum diﬀerential equations. The diﬀerential equation expressing the
vibrations of the pendulum is of the form
d
2
θ
dt
2
= −
g
L
sin θ (13.8)
where g is the gravitation constant, L is the length of the pendulum and θ is the angle
between the position of the pendulum and the vertical direction.
When the displacement of the pendulum from the equilibrium position is small,
then the angle θ is small and sin θ may be approximated by θ. Consequently, (13.8)
may be replaced with a simpler equation, expressing the behaviour for small ampli
tudes of the pendulum, that is
d
2
θ
dt
2
= −
g
L
θ
Considering that sin θ =
x
L
and θ ≈ sin θ, the pendulum equation for θ may be
replaced by a diﬀerential equation for the displacement x during time of the form
d
2
x
dt
2
= −
g
L
2
x
or
d
2
x
dt
2
+ k
2
x = 0
where k =
√
g
L
.
Odds and Ends 291
This is the classical equation expressing simple harmonic oscillations. It can be
solved explicitly, and the general solution has the form
x = Acos kt + Bsin kt
or
x = C cos (kt − D)
where A, B, C and D are constants.
The simulation of the pendulum equation is also possible by applying numerical
techniques. The second order diﬀerential equation is transformed to a system of two
ﬁrst order diﬀerential equations of the form
dθ
dt
= f
d f
dt
= −
g
L
sin θ
(13.9)
Then, the RungeKutta method will provide numerical solutions to the system.
The position of the pendulum in a Cartesian coordinate system is located by using
the expressions x = L sin θ and y = L cos θ. Figure 13.5 illustrates the movements of
the pendulum when the starting angle is θ =
π
6
. The time variation of the horizontal
displacement x is given in the lower part of the ﬁgure. The classical sinusoidal form
appears.
FIGURE 13.5: The pendulum and the (t, x) oscillations
Equation (13.8) can actually be integrated to give a ﬁrst integral satisfying
dθ
dt
=
_
2g
L
√
cos θ − cos ω
where ωis the angle of maximum displacement of the pendulum from its equilibrium
position. Figure 13.5 may also be obtained by using this formula. The ﬁnal solution
is given by means of elliptic integrals.
292 Chaotic Modelling and Simulation
13.5 A Special SecondOrder Diﬀerential Equation
Figure 13.6 illustrates the (y, ˙ y) diagram of the second order diﬀerential equation
¨ y = y
2
− 1 (13.10)
Two equilibrium points are found. One in (y, ˙ y) = (1, 0), which is unstable, and
another is (y, ˙ y) = (−1, 0), which is stable. The closedloop curves appear around
this stable ﬁxed point. The limiting curve passes through the unstable point and the
point (y, ˙ y) = (−2, 0).
FIGURE 13.6: The (y, ˙ y) curves of a special second order diﬀerential equation
system
Integration of (13.10) results in the ﬁrst order diﬀerential equation
˙ y
2
=
2
3
y
3
− 2y + c
where c is the integration constant.
13.6 Other Patterns and Chaotic Forms
This section is a smorgasbord of chaotic attractors that are particularly interesting,
but that we don’t analyse in greater detail. By using the rotation formula with b = 1
and an equation for the rotation angle which takes into account a parallel movement
of the coordinates equal to s = 4.1 according to the relation
θ
t
= c −
d
1 + (x
t
− s)
2
+ (y
t
− s)
2
Odds and Ends 293
a formation like the outer parts of a supernova explosion appears (Figure 13.7). It
can also be thought of as an artistic drawing presenting ﬁve dolphins in a circle. The
parameters are c = 2.8 and d = 13.
FIGURE 13.7: The dolphin attractor
A graph simulating a topdown view of a ship is shown in Figure 13.8(a). It is
based on a set of rotation equations of the form
x
t+1
= −a − (x
t
− a) cos θ +
1
r
t
y
t
sin θ
y
t+1
= −r
t
x
t
sin θ − y
t
cos θ
r
t
=
_
0.5
_
x
2
t
+
_
x
4
t
+ 4y
2
t
_
(13.11)
The symmetry axis is located at x =
1
2
a(cos θ − 1). The parameters are θ = 2 and
a = 2.8.
A rather startling attractor that simulates a signature is based on the following set
of diﬀerence equations.
x
t+1
= x
t
cos θ
t
− y
t
sin θ
t
+ 1 − 0.8x
t
z
t
y
t+1
= x
t
sin θ
t
+ y
t
cos θ
t
z
t+2
= 1.4z
t+1
+ 0.3z
t
(1 − z
t
)
θ
t
= 5.5 −
1
_
x
2
t
+ y
2
t
+ z
2
t
(13.12)
A twodimensional (x, y)view of the simulation results appears in Figure 13.8(b).
The Ushiki model (see Ushiki, 1982) gives a form of two separated attractors (see
Figure 13.9(a)) for parameter values a = 3.64, b = 3.1, c = 0.1, d = 0.35. The
294 Chaotic Modelling and Simulation
(a) A pattern of a ship (b) The signature attractor
FIGURE 13.8: The ship and signature attractors
equations of the model are:
x
t+1
= (a − x
t
− by
t
)x
t
y
t+1
= (c − y
t
− dx
t
)y
t
(13.13)
(a) The Ushiki attractor with
two objects
(b) The Ushiki attractor with
four objects
(c) The Tinkerbell attractor
FIGURE 13.9: The Ushiki and Tinkerbell attractors
Afourobject attractor is obtained when the parameters take the values a = 3.6, b =
3.2, c = 0.04 and d = 0.4 (Figure 13.9(b)).
The Tinkerbell model’s iterations form six chaotic attractors (Figure 13.9(c)). The
model is based on the following set of equations
x
t+1
= x
2
t
− y
2
t
+ 0.91x
t
− 0.6y
t
y
t+1
= 2x
t
y
t
+ 2.5x
t
(13.14)
Odds and Ends 295
Questions and Exercises
1. Consider the following simple threedimensional system of LotkaVolterra
type:
x
′
= a(x − xyz)
y
′
= b(−y + xyz)
z
′
= c(z − xyz)
Find the equilibrium points and the type of stability.
2. Find the line of symmetry of the ship attractor.
3. Consider the diﬀerence delay equation:
z
n+2
= azn + 1 + bz
n
(1 − z
n
)
Study this equation by ﬁnding equilibrium points, exploring chaotic behaviour
and drawing the related graphs and bifurcation diagrams for speciﬁc values of
the parameters a and b.
4. Find the equilibrium points of the Ushiki attractor.
5. Find the equilibrium points of the Tinkerbell attractor.
Chapter 14
Milestones
It is fascinating and very instructive for any scientiﬁc ﬁeld to examine the history
and milestones that were the basis for the establishment and advancement of the
ﬁeld. Some researchers were lucky to publish and disseminate their ﬁndings to a
broad audience. Others circulated their research only to a small number of friends
and colleagues. And sometimes new ﬁndings came too early to be accepted by the
scientiﬁc community.
During the preparation of any work an extensive bibliography is used, both directly
and indirectly, by inﬂuencing the approach and direction of the research. This work
was no exception. We hope the extensive bibliography, presented in the references
section, will help the reader become more acquainted with the fascinating ﬁelds of
chaotic modelling and simulation, and their historical development.
The table that follows presents an account of the groundbreaking steps in the
development of chaotic modelling and simulation, as it is perceived in our days,
through the accounts included in various books and papers. Papers written after
1980 are not included, and very few works published in the seventies appear. We
have tried to emphasise those scientiﬁc works that are considered “classical,” along
with papers and books that described chaotic phenomena or proposed theoretical or
mathematical tools that triggered the future development of the chaotic ﬁeld.
Year Name Title of Publication
1798 T. Malthus An essay on the principle of population
1823 M. Faraday On a peculiar class of acoustical ﬁgures, and on cer
tain forms assumed by groups of particles upon vi
brating elastic surfaces
1825 B. Gompertz On the nature of the function expressing of the law
of human mortality
1859 P. Riess Das Anblasen oﬀener Rohre durch eine Flamme
— P. L. Rijke Notiz ¨ uber eine neue Art, die in einer an beiden En
den oﬀenen R¨ ohre enthaltene Luft in Schwingungen
zu versetzen
1883 G. Cantor
¨
Uber unendliche, lineare Punktmannigfaltigkeiten
— Lord Rayleigh On maintained vibrations
297
298 Chaotic Modelling and Simulation
Year Name Title of Publication
— Lord Rayleigh On the crispations of ﬂuid resting upon a vibrating
support
1887 Lord Rayleigh On the maintenance of vibrations by forces of dou
ble frequency, and on the propagation of waves
through a medium endowed with a period structure
1889 S. Kowalevskaya Sur le probl` eme de la rotation d’un corps solide au
tour d’un point ﬁxe
— S. Kowalevskaya Sur une propri´ et´ e du syst` eme d’´ equations
diﬀ´ erentielles qui d´ eﬁnit la rotation d’un corps
solide autour d’un point ﬁxe
1890 H. Poincar´ e M´ emoire sur les courbes d´ eﬁnies par les ´ equations
diﬀ´ erentielles IVI, Oeuvre I
— H. Poincar´ e Sur les ´ equations de la dynamique et le probl` eme de
trois corps
1891 Lord Rayleigh On the problemof randomvibrations, and of random
ﬂights in one, two, or three dimensions
1892 M. A. Lyapunov Probl` eme g´ en´ eral de la stabilit´ e du mouvement,
(1907 translation from the 1892 Russian original)
— H. Poincar´ e Les m´ ethodes nouvelles de la m´ ecanique c´ eleste
1898 J. Hadamard Les surfaces ` a courbures oppos´ ees et leurs lignes
g´ eod´ esiques
1899 H. Poincar´ e Les m´ ethodes nouvelles de la m´ ecanique c´ eleste
1905 H. Poincar´ e Lecons de m´ ecanique
1908 H. Poincar´ e Science et methode
1910 A. Lotka Zur Theorie der periodischen Raktionen
1918 G. Duﬃng Erzwungene Schwingungen bei ver¨ anderlicher
Eigenfrequenz
— G. Julia M´ emoire sur l’it´ eration des fonctions rationelles
1919 F. Hausdorﬀ Dimension und ¨ außeres Maß
— P. Fatou Sur les ´ equations fonctionelles
1920 P. Fatou Sur les ´ equations fonctionelles
1923 M. H. Dulac Sur les cycles limites
— F. C. Ritt Permutable rational functions
1926 P. Fatou Sur l’it´ eration des fonctions transcendentes enti` eres
Milestones 299
Year Name Title of Publication
1927 B. van der Pol Forced oscillations in a circuit with nonlinear
resistance
— B. van der Pol and
J. van der Mark
Frequency demultiplication
1927 G. D. Birkhoﬀ Sur le probl` eme restreint des trois corps
1932 G. D. Birkhoﬀ Sur l’existence de r´ egions d’instabilit´ ee dynamique
— A. Denjoy Sur les courbes d´ eﬁnies par les ´ equations
diﬀ´ erentielles ` a la surface du tore
1934 G. F. Gause The struggle for existence
1935 G. D. Birkhoﬀ Sur le probl` eme restreint des trois corps
— P. O. Pedersen Subharmonics in forced oscillations in dissipative
systems
1936 A. N. Kolmogorov Sulla teoria di Volterra della lotta per l’esistenza
1937 R. A. Fisher The wave of advance of advantageous genes
— A. Kolmogorov et
al.
Etude de l’´ equation de la diﬀusion avec croissance
de la quantit´ ede mati` ere et son application ` a un
probl` eme biologique
1942 J. L. Doob The Brownian movement and stochastic equations
1942 E. Hopf Abzweigung einer periodischen L¨ osung von einer
station¨ aren L¨ osung eines Diﬀerentialsystems
1942 A. Kolmogorov Interpolation and extrapolation of stationary series
1943 S. Chandrasekhar Stochastic problems in physics and astronomy
1944 L. D. Landau On the problem of turbulence
1945 M. L. Cartwright
and J. E. Little
wood
On nonlinear diﬀerential equations of the second
order
— S. O. Rice Mathematical analysis of random noise
1947 M. Kac Random walk and the theory of Brownian motion
1948 E. Hopf A mathematical example displaying features of
turbulence
— M. L. Cartwright Forced oscillations in nearly sinusoidal systems
1949 C. E. Shannon and
W. Weaver
The mathematical theory of information
1951 G. H. Markstein Experimental and theoretical studies of ﬂamefront
stability
300 Chaotic Modelling and Simulation
Year Name Title of Publication
1952 A. L. Hodgkin and
A. F. Huxley
A quantitative description of membrane current and
its application to conduction and excitation in nerve
— A. L. Hodgkin and
A. F. Huxley
Current carried by sodium and potassium ions
through the membrane of the giant axon of loligo
1953 N. Metropolis et al. Equations of state calculations by fast computing
machine
1954 A. N. Kolmogorov Preservation of conditionally periodic movements
with small change in the Hamiltonian function
1956 W. Feller An introduction to probability theory and its
applications
1956 R. E. Kalman Nonlinear aspects of sample data control systems
— P. Landberg Vibrations caused by chip formation
— E. N. Lorenz Empirical orthonogal functions and statistical
weather prediction
— G. Contopoulos On the isophotes of ellipsoidal nebulae
1958 A. N. Kolmogorov A new invariant of transitive dynamical systems
— P. J. Myrberg Iteration der reellen Polynome zweiten Grades
— G. Contopoulos On the vertical motions of stars in a galaxy
1959 N. N. Leonov Transformations d’une droite en ellem` eme
— A. R´ enyi On the dimension and entropy of probability
distributions
1960 R. E. Kalman A new approach to linear ﬁltering and prediction
problems
— N. N. Leonov Transformation ponctuelle d’une droite en elle
m` eme discontinue, lin´ eaire par morceaux
— N. N. Leonov Th´ eorie des transformations discontinues d’une
droite en ellem` eme
— G. Contopoulos A third integral of motion in a galaxy
1961 R. A. FitzHugh Impulses and physiological states in theoretical
models of nerve membrane
1962 P. J. Myrberg Sur l’it´ eration des polynomes r´ eels quadratiques
— J. Nagumo et al. An active pulse transmission line simulating nerve
axon
1963 E. N. Lorenz Deterministic nonperiodic ﬂow
— B. Mandelbrot The variation of certain speculative prices
Milestones 301
Year Name Title of Publication
1964 E. N. Lorenz The problem of deducing the climate from the gov
erning equations
— M. H´ enon and C.
Heiles
The applicability of the third integral of motion:
Some numerical experiments
— A. N. Sarkovskii Coexistence of cycles of a continuous map of a line
into itself
— G. Contopoulos
and G. Bozis
Escape of stars during the collision of two galaxies
1965 D. Coles Transition in circular couette ﬂow
— A. N. Kolmogorov Three approaches to the quantitative deﬁnition of
information
— J. R. Pasta et al. Studies on nonlinear problems
— L. P. Shilnikov A case of the existence of a countable number of
periodic motions
— G. Contopoulos Periodic and tube orbits
1967 H. Degn Eﬀect of bromine derivatives of malonic acid on the
oscillating reaction of malonic acid, cerium ions and
bromate
1967 B. Mandelbrot The variation of some other speculative prices
— L. P. Shilnikov On the Poincar´ eBirkhoﬀ problem
— L. P. Shilnikov The existence of a denumerable set of periodic
motions in fourdimensional space in an extended
neighbourhood of a saddlefocus
1968 M. Kuczma Functional equations in a single variable
— I. Prigogine and R.
Lefever
Symmetry breaking instabilities in dissipative sys
tems. II
1969 R. FitzHugh Mathematical models of excitation and propagation
in nerve and muscle
— M. H´ enon Numerical study of quadratic areapreserving
mappings
– E. N. Lorenz Atmospheric predictability as revealed by naturally
occurring analogies
— I. Prigogine et al. Symmetry breaking instabilities in biological
systems
1976 O. E. R¨ ossler Chaotic behavior in simple reaction systems
302 Chaotic Modelling and Simulation
Year Name Title of Publication
— O. E. R¨ ossler Chemical turbulence: Chaos in a simple reaction
diﬀusion system
— O. E. R¨ ossler Diﬀerent types of chaos in two simple diﬀerential
equations
1978 M. J. Feigenbaum Quantitative universality for a class of nonlinear
transformations
1979 M. J. Feigenbaum The universal metric properties of nonlinear
transformations
— O. E. R¨ ossler An equation for hyperchaos
— O. E. R¨ ossler Continuous chaos — four prototype equations
— K. Ikeda Multiplevalued stationary state and its instability of
the transmitted light by a ring cavity system
1980 K. Ikeda et al. Optical turbulence: Chaotic behaviour of transmit
ted light from a ring cavity
References
Acheson, D. (1997). From calculus to chaos. Oxford: Oxford University Press.
Adachi, S., M. Toda, and K. Ikeda (1988a). Potential for mixing in quantum chaos.
Phys. Rev. Lett. 61, 635.
Adachi, S., M. Toda, and K. Ikeda (1988b). Quantumclassical correspondence in
manydimensional quantum chaos. Phys. Rev. Lett. 61, 659.
Adler, M. and P. van Moerbeke (1994). The Kowalevski and H´ enonHeiles motions
as Manakov geodesic ﬂows on SO(4). Comm. Math. Phys. 113, 649.
Alligood, K. T. and T. Sauer (1988). Rotation numbers of periodic orbits in the
H´ enon’s map. Commun. Math. Phys. 120, 105.
Alligood, K. T., T. D. Sauer, and J. A. Yorke (1997). Chaos; An introduction to
dynamical systems. Berlin: SpringerVerlag.
Almirantis, Y. and M. Kaufman (1992). Numerical study of travelling waves in a
reactiondiﬀusion system: response to a spatiotemporal forcing. Int. J. of Bifurca
tion and Chaos 2(1), 51–60.
Almirantis, Y. and M. Kaufman (1995). Chiral selection of rotating waves in a
reactiondiﬀusion system: The eﬀect of a circularly polarized electromagnetic
ﬁeld. Int. J. of Bifurcation and Chaos 5(2), 507–518.
Almirantis, Y. and G. Nicolis (1987). Morphogenesis in an asymmetric medium.
Bull. Math. Biol. 47, 519–530.
Alsing, P. M., A. Gavrielides, and V. Kovanis (1994a). Historydependent control of
unstable periodic orbits. Phys. Rev. E 50, 1968.
Alsing, P. M., A. Gavrielides, and V. Kovanis (1994b). Using neural networks for
controlling chaos. Phys. Rev. E 49(2), 1225–1231.
Androulakakis, S. P., B. Greenspan, and T. T. H. H. Qammar (1991). Practical con
siderations on the calculation of the uncertainty exponent and the fractal dimen
sion of basin boundaries. Int. J. of Bifurcation and Chaos 1(2), 327–333.
Antoniou, I., V. Basios, and F. Bosco (1996). Probabilistic control of chaos: The
betaadic R´ enyi map under control. Int. J. of Bifurcation and Chaos 6(8), 1563–
1573.
Aref, H. (1983). Integrable, chaotic, and turbulent vortex motion in twodimensional
ﬂows. Ann. Rev. Fluid Mech. 15, 345.
303
304 Chaotic Modelling and Simulation
Aref, H. (1984). Stirring by chaotic advection. J. Fluid mech. 143, 1.
Aref, H. and S. Balachandar (1986). Chaotic advection in a Stokes ﬂow. Phys.
Fluids 29, 3515–3521.
Argyris, J. and I. Andreadis (1998). On linearisable noisy systems. Chaos Solitons
Fractals 9(6), 895–899.
Argyris, J. and I. Andreadis (2000). On the inﬂuence of noise on the coexistence of
chaotic attractors. Chaos Solitons Fractals 11(6), 941–946.
Argyris, J. and C. Ciubotariu (1999). A new physical eﬀect modeled by an ikeda
map depending on a monotonically timevarying parameter. Int. J. of Bifurcation
and Chaos 9(6), 1111–1120.
Argyris, J., B. V., N. Ovakimyan, and M. Minasyan (1996). Chaotic vibrations of a
nonlinear viscoelastic beam. Chaos Solitons Fractals 7(2), 151–163.
Arn´ eodo, A., F. Argoul, J. Elezgaray, and P. Richetti (1993). Homoclinic chaos in
chemical systems. Physica D 62, 134–169.
Arneodo, A., P. Coullet, and C. Tresser (1979). A renormalization group with peri
odic behavior. Phys. Lett. A 70, 74.
Arneodo, A., P. Coullet, and C. Tresser (1980). Occurrence of strange attractors in
threedimensional Volterra equations. Phys. Lett. A 79, 59.
Arneodo, A., P. Coullet, and C. Tresser (1981a). A possible new mechanism for the
onset of turbulence. Phys. Lett. A 81, 197.
Arneodo, A., P. Coullet, and C. Tresser (1981b). Possible new strange attractors with
spiral structure. Commun. Math. Phys. 79, 573–579.
Arnold, L. (1990). Stochastic diﬀerential equations as dynamical systems. In M. K.
et al. (Ed.), Proceedings MTNS89 Amsterdam, Volume I, Boston, pp. 489–495.
Birkh¨ auser.
Arnold, L. (1998). Random Dynamical Systems. Berlin: SpringerVerlag.
Arnold, V. I. (1983). Geometrical methods in the theory of ordinary diﬀerential
equations. Berlin: SpringerVerlag.
Aronson, D. G. (1980). Density dependent reactiondiﬀusion system. In Dynamics
and modelling of reactive systems, pp. 161–176. New York: Academic Press.
Aronson, J. (1990). CHAOS: A SUNbased program for analyzing chaotic systems.
Computers in Physics 4(4), 408–417.
Arrowsmith, D. K. and C. M. Place (1990). An introduction to dynamic systems.
Cambridge: Cambridge University Press.
Aston, P. J. (1998). A chaotic Hopf bifurcation in coupled maps. Physica D118(34),
199–220.
References 305
Aston, P. J. (1999). Bifurcations of the horizontally forced spherical pendulum.
Comput. Method. Appl. Mech. Eng. 170(34), 343–353.
Auerbach, D., C. Grebogi, E. Ott, and J. Yorke (1992). Controlling chaos in high
dimensional systems. Phys. Rev. Lett. 69, 3479–3482.
Awrejcewicz, J. (1989). Bifurcation and chaos in simple dynamical systems. Singa
pore: World Scientiﬁc.
Bahar, S. (1996a). Further studies of bifurcations and chaotic orbits generated by
iterated function systems. Chaos Solitons Fractals 7(1), 41–47.
Bahar, S. (1996b). Patterns of bifurcation in iterated function systems. Chaos Soli
tons Fractals 7(2), 205–210.
Bahar, S. (1997). Orbits embedded in IFS attractors. Int. J. of Bifurcation and
Chaos 7(3), 741–749.
Balakrishnan, V., C. Nicolis, and G. Nicolis (1995). Extreme value distributions in
chaotic dynamics. J. Stat. Phys. 80(12), 307–336.
Balakrishnan, V., G. Nicolis, and C. Nicolis (1997). Recurrence time statistics in
chaotic dynamics. I. discrete time maps. J. Stat. Phys. 86(12), 191–212.
Balakrishnan, V., G. Nicolis, and C. Nicolis (2000). Recurrence time statistics in de
terministic and stochastic dynamical systems in continuous time: A comparison.
Phys. Rev. E 61(3), 2490–2499.
Basios, V., T. Bountis, and G. Nicolis (1999). Controlling the onset of homoclinic
chaos due to parametric noise. Phys. Lett. A 251(4), 250–258.
Bass, F. (1969). A new product growth model for consumer durables. Management
Science 15(217–231).
Belyakova, G. V. and L. A. Belyakov (1997). On bifurcations of periodic orbits in
the van der PolDuﬃng equation. J. of Bifurcation and Chaos 7(2), 459–462.
Berg´ e, P., M. Dubois, P. Manneville, and Y. Pomeau (1980). Intermittency in
RayleighB´ enard convection. J. Phys. Lett. 41, 341.
Berg´ e, P. and Y. Pomeau (1980). La turbulence. La Recherche 11, 422.
Bier, M. and T. C. Bountis (1984). Remerging Feigenbaum trees in dynamical sys
tems. Phys. Lett. A 104, 29.
Biktashev, V. N. (1989). Evolution of twist of an autowave vortex. Physica D 36,
167–172.
Birkhoﬀ, G. and G.C. Rota (1989). Ordinary diﬀerential equations. New York,
N.Y.: Wiley.
Birkhoﬀ, G. D. (1927). On the periodic motions of dynamical systems. Acta Math.
(reprinted in MacKay and Meiss 1987) 50, 359.
306 Chaotic Modelling and Simulation
Birkhoﬀ, G. D. (1932). Sur l’existence de r´ egions d’instabilit´ ee dynamique. Ann.
Inst. H. Poincar´ e 2, 369.
Birkhoﬀ, G. D. (1935). Sur le probl` eme restreint des trois corps. Ann. Scuola Norm.
Sup. Pisa 4, 267.
Bolotin, V. V., A. A. Grishko, A. N. Kounadis, C. Gantes, and J. B. Roberts (1998).
Inﬂuence of initial conditions on the postcritical behavior of a nonlinear aeroelas
tic system. Nonlinear Dyn. 15(1), 63–81.
Borland, L. (1996). Simultaneous modeling of nonlinear determistic and stochastic
dynamics. Physica D 99(23), 175–190.
Borland, L. (1998). Microscopic dynamics of the nonlinear FokkerPlanck equation:
A phenomenological model. Phys. Rev. E 57(6), 6634–6642.
Borland, L. and H. Haken (1992). Unbiased determination of forces causing ob
served processes. The case of additive and weak multiplicative noise. Z. Phys. B 
Condens. Matter 81, 95.
Borland, L. and H. Haken (1993a). Learning the dynamics of twodimensional
stochastic Markov processes. Open Syst. and Inf. Dyn. 1(3), 311.
Borland, L. and H. Haken (1993b). On the constraints necessary for macroscopic
prediction of stochastic timedependent processes. ROMP 33, 35.
Boudourides, M. A. and N. A. Fotiades (2000). Piecewise linear interval maps both
expanding and contracting. Dyn. Stab. Syst. 15(4), 343–351.
Bountis, T. (1992). Chaotic dynamics. Theory and practice. New York: Plenum
Press.
Bountis, T., L. Drossos, and I. C. Percival (1991a). Nonintegrable systems with
algebraic singularities in complex time. J. Phys. 24, 3217.
Bountis, T., L. Drossos, and I. C. Percival (1991b). On nonintegrable systems with
square root singularities in complex time. Phys. Lett. A 159, 1.
Bountis, T. and R. H. Helleman (1981). On the stability of periodic orbits of two
dimensional mappings. J. Math. Phys. 22, 1867.
Bountis, T., L. Karakatsanis, G. Papaioannou, and G. Pavlos (1993). Determinism
and noise in surface temperature time series. Ann. Geophys. 11, 947–959.
Bountis, T., V. Papageorgiou, and M. Bier (1987). On the singularity analysis of
intersecting separatrices in nearintegrable dynamical systems. Physica D 24, 292.
Bountis, T., H. Segur, and F. Vivaldi (1982). Integrable Hamiltonian systems and the
Painlev´ e property. Phys. Rev. A 25, 1257.
Bountis, T. C. (1981). Period doubling bifurcations and universality in conservative
systems. Physica D 3, 577–589.
References 307
Boyarsky, A. (1986). A functional equation for a segment of the H´ enon map unstable
manifold. Physica D 21, 415.
Braun, T. and I. A. Heisler (2000). A comparative investigation of controlling chaos
in a R¨ ossler system. Physica A 283(12), 136–139.
Briggs, K. M. (1989). How to calculate the Feigenbaum constants on your PC. Aust.
Math Soc. Gazette 16, 89–92.
Brock, W. A. and C. H. Hommes (1997, August). Heterogeneous beliefs and routes
to chaos in a simple asset pricing model. Journal of Economic Dynamics and
Control 22(89), 1235–1274.
Brock, W. A., J. Lakonishok, and B. LeBaron (1992). Simple technical trading rules
and the stochastic properties of stock returns. J. Finance 47, 1731–1764.
Brock, W. A. and B. LeBaron (1993). Using structural modelling in building statis
tical models of volatility and volume of stock market returns. Technical report,
Univ. of Wisconsin, Madison, Madison, Wisconsin.
Brock, W. A. and A. G. Malliaris (1989). Diﬀerential equations, stability and chaos
in dynamic economics. Amsterdam: North Holland.
Budinsky, N. and T. Bountis (1983). Stability of nonlinear modes and chaotic prop
erties of 1D FermiPastaUlam lattices. Physica D 8, 445–452.
Bunner, M. J. (1999). The control of highdimensional chaos in timedelay systems
to an arbitrary goal dynamics. Chaos 9(1), 233–237.
Busse, H. (1969). A spatial periodic homogeneous chemical reaction. J. Phys.
Chem. 73, 750.
Cabrera, J. L. and F. J. de la Rubia (1996). Analysis of the behavior of a random
nonlinear delay discrete equation. Int. J. of Bifurcation and Chaos 6(9), 1683–
1690.
Calogero, F. (1971). Solution of the 1D Nbody problem with quadratic and/or
inversely quartic pair potentials. J. Math. Phys. 12, 419.
Campanino, M., H. Epstein, and D. Ruelle (1982). On Feigenbaum’s functional
equation. Topology 21, 125–129.
Caranicolas, N. and C. Vizikis (1987). Chaos in a quartic dynamical model (celestial
mechanics). Celest. Mech. 40, 35.
Caratheodory, C. (1982). Calculus of variations and partial diﬀerential equations of
the ﬁrst order. New York: Chelsea.
Carroll, T. L. (1994). Synchronization of chaos. Ciencia Hoje 18, 26.
Carroll, T. L. (2002). Noiseresistant chaotic maps. Chaos 12(2), 275–278.
Carroll, T. L. and L. M. Pecora (1999). Using multiple attractor chaotic systems for
308 Chaotic Modelling and Simulation
communication. Chaos 9(2), 445–451.
Cartwright, M. L. (1948). Forced oscillations in nearly sinusoidal systems. J. Inst.
Elec. Eng. 95, 88.
Cartwright, M. L. and J. E. Littlewood (1945). On nonlinear diﬀerential equations
of the second order. J. London Math. Soc. 20, 180–189.
Casdagli, M. (1989). Nonlinear prediction of chaotic time series. Physica D 35,
335–356.
Casdagli, M. (1991). Chaos and deterministic “versus” stochastic nonlinear mod
elling. J. Roy. Statist. Soc. Ser. B 54(2), 303–328.
Casdagli, M. C. and A. S. Weigend (1993). Exploring the continuum between de
terministic and stochastic modeling. In A. S. Weigend and N. A. Gershenfeld
(Eds.), Time series prediction: Forecasting the future and understanding the past,
pp. 347–366. Reading, MA: AddisonWesley.
Catsigeras, E. (1996). Cascades of period doubling bifurcations in n dimensions.
Nonlinearity 9, 1061–1070.
Catsigeras, E. and H. Enrich (1999). Persistence of the Feigenbaum attractor in one
parameter families. Commun. Math. Phys. 207(3), 621–640.
Caurier, E. and B. Grammaticos (1986). Quantum chaos with nonergodic Hamilto
nians. Europhys. Lett. 2, 417.
Caurier, E. and B. Grammaticos (1989). Extreme level repulsion for chaotic quantum
Hamiltonians. Phys. Lett. A 136, 387–390.
Celikovsky, S. and G. R. Chen (2002). On a generalized Lorenz canonical form of
chaotic systems. Int. J. Bifurcation Chaos 12(8), 1789–1812.
Celka, P. (1996). A simple way to compute the existence region of 1D chaotic at
tractors in 2Dmaps. Physica D 90(3), 235–241.
Celka, P. (1997). Delaydiﬀerential equation versus 1Dmap: Application to chaos
control. Physica D 104(127147).
Chandrasekhar, S. (1943). Stochastic problems in physics and astronomy. Mod.
Phys. 15(1), 1–89.
Chirikov, B. (1996). Natural laws and human prediction. In P. Weingartner and
G. Schurz (Eds.), Law and prediction in the light of chaos research, Volume LNP
473, pp. 10–34. Berlin: SpringerVerlag.
Chirikov, B. V. (1979). A universal instability of manydimensional oscillator sys
tems. Phys. Rep. 52, 263.
Chirikov, B. V. (1983). Chaotic dynamics in Hamiltonian systems with divided phase
space. In Dynamical systems and chaos, Volume 179 of Lecture notes in physics,
References 309
pp. 29–46.
Chorafas, D. (1994). Chaos theory in the ﬁnancial markets. Applying fractals. Fuzzy
logic. Genetic algorithms. Probus Publishing Co.
Chua, L. O. (1980). Dynamic nonlinear networks: State of the art. IEEE Trans.
Circuit Syst. 27(11), 1059–1087.
Chua, L. O. (1992). The genesis of Chua’s circuit. Archiv f¨ ur Elektronik &
¨
U.
technik 46, 250–257.
Chua, L. O. (1993a). Global unfolding of Chua’s circuit. IEICE Trans. Fundamentals
E 76A, 704–734.
Chua, L. O. (1993b). A zoo of strange attractors from the canonical Chua’s circuit.
J. Circuit, Systems, and Computers 2.
Chua, L. O., M. Itoh, L. Kocarev, and K. Eckert (1993). Chaos synchronization in
Chua’s circuit. J. Circuits, Systems and Computers 3(1), 93–108.
Chung, J. S. and M. M. Bernitsas (1992). Dynamics of twoline ship tow
ing/mooring systems: bifurcations, singularities, instability boundaries, chaos. J.
Ship Res. 36(2).
Cladis, P. E. and P. PalﬀyMuhoray (1994). SpatioTemporal Patterns in Nonequi
librium Complex Systems (SFI Studies in the Sciences of Complexity). Reading,
MA: AddisonWesley.
Clerc, M., P. Coullet, and E. Tirapegui (1999). Lorenz bifurcation: Instabilities in
quasireversible systems. Phys. Rev. Lett. 83(19), 3820–3823.
Coleman, M. J. and A. Ruina (1998). An uncontrolled walking toy that cannot stand
still. Phys. Rev. Lett. 80(16), 3658–3661.
Coleman, S. (1993). Cycles and chaos in political party voting. Journal of Mathe
matical Sociology 18, 47–64.
Coleman, S. (1995). Dynamics in the fragmentation of political party systems.
Qualilty and Quantity 29, 141–155.
Coles, D. (1965). Transition in circular couette ﬂow. J. of Fluid Mech. 21, 385–425.
Collet, P. and J.P. Eckmann (1980a). Iterated maps on the interval as dynamical
system. Basel: Birkh¨ auser.
Collet, P. and J.P. Eckmann (1980b). On the abundance of chaotic behavior in one
dimension. In R. Helleman (Ed.), Nonlinear Dynamics, Volume 357 of Ann. N. Y.
Acad. Sci., pp. 337–342. New York: The New York Academy of Sciences.
Collet, P. and J.P. Eckmann (1983). Positive Liapunov exponents and absolute con
tinuity for maps of the interval. Ergodic Theory & Dynamical Systems 3, 13–46.
Collet, P., J.P. Eckmann, and H. Koch (1981a). On universality for areapreserving
310 Chaotic Modelling and Simulation
maps of the plane. Physica D 3, 457.
Collet, P., J.P. Eckmann, and H. Koch (1981b). Period doubling bifurcations for
famlies of maps on R
n
. J. Stat. Phys. 25, 1.
Collet, P., J.P. Eckmann, and O. E. Lanford (1980). Universal properties of maps on
an interval. Commun. Math. Phys. 76, 211–254.
Collet, P., J.P. Eckmann, and L. Thomas (1981). A note on the power spectrum of
the iterates of Feigenbaum’s function. Commun. Math. Phys. 81, 261–265.
Combes, F., F. Debbasch, D. Friedli, and D. Pfenniger (1990). Box and peanut shapes
generated by stellar bars. Astron. Astrophys. 233, 82.
Contopoulos, G. (1956). On the isophotes of ellipsoidal nebulae. Z. Astrophysic 39,
126.
Contopoulos, G. (1958). On the vertical motions of stars in a galaxy. Stockholm
Ann. 20 No5.
Contopoulos, G. (1960). A third integral of motion in a galaxy. Z. Astrophysic 49,
273–291.
Contopoulos, G. (1965). Periodic and tube orbits. Astron. J. 70, 526.
Contopoulos, G. (1981). Do successive bifurcations in Hamiltonian systems have
the same universal ratio? Lett. Nuovo Cimento 30, 498.
Contopoulos, G. (2001). The development of nonlinear dynamics in astronomy.
Found. Phys. 31(1), 89–114.
Contopoulos, G. (2002). Order and chaos in dynamical astronomy. Berlin: Springer
Verlag.
Contopoulos, G. and G. Bozis (1964). Escape of stars during the collision of two
galaxies. Astrophys. J. 139, 1239.
Contopoulos, G., M. H´ enon, and D. LyndenBell (Eds.) (1973). Dynamic structure
and evolution of stellar systems. Saas.
Contopoulos, G. and C. Polymilis (1987). Approximations of the 3particle Toda
lattice. Physica D 24, 328.
Coullet, P. and F. Plaza (1995). Excitable spiral waves in nematic liquid crystals. Int.
J. of Bifurcation and Chaos 4(5), 1173–1182.
Coullet, P., C. Tresser, and A. Arneodo (1979). Transition to stochasticity for a class
of forced oscillators. Phys. Lett. A 72, 268.
Creedy, J. (1994). Chaos and nonlinear models in economics: theory and applica
tions. Aldershot: Elgar.
Crutchﬁeld, J. P. (1983). Noisy chaos. Ph.D. dissertation, University of California,
References 311
Santa Cruz.
Crutchﬁeld, J. P. and B. A. Huberman (1980). Fluctuations and the onset of chaos.
Phys. Lett. A 77, 407–410.
Curry, J. H. (1981). On computing the entropy of the H´ enon attractor. J. Stat.
Phys. 26, 683.
Cvitanovi´ c, P. (1988). Invariant measures of strange sets in terms of cycles. Phys.
Rev. Lett. 61, 2729–2732.
Cvitanovi´ c, P. (1989). Universality in chaos. Bristol: Adam Hilger.
Cvitanovi´ c, P. (1991). Periodic orbits as the skeleton of classical and quantum chaos.
Physica D 51, 138–151.
Cvitanovi´ c, P. and B. Eckhardt (1989). Periodic orbit quantization of chaotic sys
tems. Phys. Rev. Lett. 63, 823–826.
Cvitanovi´ c, P., P. Gaspard, and T. Schreiber (1992). Investigation of the Lorentz gas
in terms of periodic orbits. Chaos 2, 85.
Cvitanovi´ c, P., G. H. Gunaratne, and M. J. Vinson (1990). On the modelocking
universality for critical circle maps. Nonlinearity 3, 873–885.
Cvitanovi´ c, P., K. Hansen, J. Rof, and G. Vattay (1998). Beyond the periodic orbit
theory. Nonlinearity 11, 1209–1232.
Cvitanovi´ c, P. and J. Myrheim (1983). Universality for period ntuplings in complex
mappings. Phys. Lett. A 94, 329–333.
Cybenko, G. (1989). Approximation by superpositions of a sigmoidal function.
Math. Control, Signals Syst 2, 303–314.
Daido, H. (1980). Analytic conditions for the appearance of homoclinic and hetero
clinic points of a 2dimensional mapping: the case of the H´ enon mapping (1831).
Prog. Theor. Phys. 63, 1190.
Daido, H. (1981). Theory of the perioddoublings of 1D mappings based on the
parameter dependence. Phys. Lett. A 83, 246.
Daido, H. (1997). Strange waves in coupledoscillator arrays: Mapping approach.
Phys. Rev. Lett. 78(9), 1583–1686.
Davies, M. (1994). Noise reduction schemes for chaotic time series. Physica D 79,
174–192.
Davies, M. E. (1992). An order N noise reduction algorithm with quadratic conver
gence. Preprint.
Davis, H. T. (1960). Introduction to nonlinear diﬀerential and integral equations.
New York: Dover.
312 Chaotic Modelling and Simulation
Davis, S. H. (1987). Coupled Lorenz oscillators. Physica D 24, 226.
Day, R. H. (1994). Complex economic dynamics. Cambridge, MA: MIT Press.
Day, R. H. and M. Zhang (1996). Classical economic growth theory: a global bifur
cation analysis. Chaos Solitons Fractals 7(12), 1969–1988.
de la Llave, R. and S. Tompaidis (1995). On the singularity structure of invariant
curves of sympletic mappings. Chaos 5(1), 227–237.
de Olivera, C. R. and C. P. Malta (1987). Bifurcations in a class of timedelay equa
tions. Phys. Rev. A 36, 3997–4001.
Degn, H. (1967). Eﬀect of bromine derivatives of malonic acid on the oscillating
reaction of malonic acid, cerium ions and bromate. Nature 213, 589–590.
Dellnitz, M., M. Field, M. Golubitsky, J. Ma, and A. Hoohmann (1995). Cycling
chaos. Int. J. of Bifurcation and Chaos 5(4), 1243–1247.
Dellnitz, M., M. Golubitsky, A. Hohmann, and I. Stewart (1995). Spirals in scalar
reactiondiﬀusion equations. Int. J. of Bifurcation and Chaos 5(6), 1487–1501.
Dendrinos, D. S. (1994). Traﬃcﬂow dynamics: a search for chaos. Chaos Solitons
Fractals 4, 605–617.
Dendrinos, D. S. and M. Sonis (1990). Chaos and socialspatial dynamics. New
York: SpringerVerlag Verlag.
Denjoy, A. (1932). Sur les courbes d´ eﬁnies par les ´ equations diﬀ´ erentielles ` a la
surface du tore. J. Math. Pures Appl. 11, 333–375.
Derrida, B. and Y. Pomeau (1980). Feigenbaum’s ratios of twodimensional area
preserving maps. Phys. Lett. A 80, 217–219.
Devaney, R. (1976). Reversible diﬀeomorphisms and ﬂows. Trans. Amer. Math.
Soc. 218, 89–113.
Devaney, R. L. (1988). Chaotic bursts in nonlinear dynamical systems. Science 235,
342–345.
Devaney, R. L. (1989). An introduction to chaotic dynamical systems. Reading, MA:
AddisonWesley.
Devaney, R. L. (1991). e
z
: Dynamics and bifurcations. Int. J. of Bifurcation and
Chaos 1(2), 287–308.
Devaney, R. L. and M. Durkin (1991). The exploding exponential and other chaotic
bursts in complex dynamics. Amer. Math. Monthly 98, 217–233.
Devaney, R. L. and X. Jarque (1997). Misiurewicz points for complex exponentials.
Int. J. of Bifurcation and Chaos 7(7), 1599–1615.
Dewar, R. L. and A. B. Khorev (1995). Rational quadraticﬂux minimizing circles
References 313
for areapreserving twist maps. Physica D 85(12), 66–78.
Diakonos, F. K., D. Pingel, and P. Schmelcher (1999). A stochastic approach to the
construction of onedimensional chaotic maps with prescribed statistical proper
ties. Phys. Lett. A 264(23), 162–170.
Diakonos, F. K., P. Schmelcher, and O. Biham (1998). Systematic computation of
the least unstable periodic orbits in chaotic attractors. Phys. Rev. Lett. 81(20),
4349–4352.
Diamond, P. (1994). Chaos in iterated fuzzy systems. J. Math. Anal. Appl. 184,
472–484.
Dimarogonas, A. D. (1996). Vibration of cracked structures: a state of the art review.
Engineering Fracture Mechanics 55(5), 831–857.
Ding, M., E. Ott, and C. Grebogi (1994). Crisis control: preventing chaosinduced
capsizing of a ship. Phys. Rev. E 50(5), 4228–4230.
Ditto, W. L. (1996). Applications of chaos in biology and medicine. AIP Conf.
Proc. 376, 175–201.
Doering, C. R. and J. D. Gibbon (1998). On the shape and dimension of the Lorenz
attractor (vol 10, pg 255, 1995). Dynam. Stabil. Syst. 13(3), 299–301.
Doherty, M. F. and J. M. Ottino (1988). Chaos in deterministic systems: strange at
tractors, turbulence, and applications in chemical engineering. Chem. Eng. Sci. 43,
139.
Dokoumetzidis, A., A. Iliadis, and P. Macheras (2001). Nonlinear dynamics and
chaos theory: concepts and applications relevant to pharmacodynamics. Pharm.
Res. 18(4), 415–426.
Domokos, G. and P. J. Holmes (1993). Euler’s problem and Euler’s method, or the
discrete charm of buckling. J. Nonlinear Sci. 3, 109–151.
Doob, J. L. (1942). The Brownian movement and stochastic equations. Ann. of
Math. 43(2), 351–369.
Dormayer, P. (1994). Examples for smooth bifurcation for x(t) = −alphaf (x(t −1)).
Appl. Anal. 55(12), 25–40.
Drossos, L. and T. Bountis (1992). On the convergence of series solutions of noninte
grable systems with algebraic singularities. In T. Bountis (Ed.), Chaotic dynamics
and practice. London: Plenum.
Drossos, L., O. Ragos, M. N. Vrahatis, and T. Bountis (1996). Method for computing
long periodic orbits of dynamical systems. Phys. Rev. E 53(1), 1206–2111.
Duﬃng, G. (1918). Erzwungene Schwingungen bei ver¨ anderlicher Eigenfrequenz.
Braunschweig: Vieweg & Sohn.
314 Chaotic Modelling and Simulation
Dulac, M. H. (1923). Sur les cycles limites. Bull. Soc. Math. Anal. 51, 45–188.
Eckmann, J.P. (1981). Roads to turbulence in dissipative dynamical systems. Rev.
Mod. Phys. 53, 643–654.
Eckmann, J.P., H. Koch, and P. Wittwer (1984). A computerassissted proof of
universality for areapreserving maps. Mem. Am. Math. Soc. 47, 1–122.
Eckmann, J.P. and D. Ruelle (1985). Ergodic theory of chaos and strange attractors.
Rev. Mod. Phys. 57, 617–656.
Eckmann, J.P. and P. Wittwer (1987). A complete proof of the Feigenbaum conjec
tures. J. Stat. Phys. 46, 455–477.
ElRifai, E. A. and E. Ahmed (1995). Knotted periodic orbits in R¨ ossler’s equations.
J. Math. Phys. 36(2), 773–777.
Elsner, J. B. and A. A. Tsonis (1992). Nonlinear predicting, chaos and noise. Bull.
Amer. Meteor. Soc. 73, 49–60.
Endler, A. and J. A. C. Gallas (2001). Period four stability and multistability domains
for the H´ enon map. Physica A 295(12), 285–290.
Epstein, H. (1986). Newproofs of the existence of the Feigenbaumfunctions. Comm.
Math. Phys. 106, 395–426.
Epstein, H. R. (1983). Oscillations and chaos in chemical systems. Physica D 7, 47.
Fang, H. P. (1994). Studying the Lorenz equations with onedimensional maps from
successive local maxima in z. Z. Phys. B 96(4), 547–552.
Farady, M. (1831). On a peculiar class of acoustical ﬁgures, and on certain forms
assumed by groups of particles upon vibrating elastic surfaces. Phil. Trans. Roy.
Soc. London 121, 299–340.
Fatou, P. (1919). Sur les ´ equations fonctionelles. Bull. Soc. Math. France 47, 161–
271.
Fatou, P. (1920). Sur les ´ equations fonctionelles. Bull. Soc. Math. France 48, 33–94,
208–314.
Fatou, P. (1926). Sur l‘it´ eration des fonctions transcendentes enti` eres. Acta Math. 47,
337–370.
Feichtinger, G. (1992a). Hopf bifurcation in an advertising diﬀusion model. J. Econ.
Behav. Organ. 17, 401–411.
Feichtinger, G. (1992b). Limit cycles in dynamic economic systems. Ann. Operation
Res. 37, 313–344.
Feichtinger, G., L. L. Ghezzi, and C. Piccardi (1995). Chaotic behavior in an adver
tising diﬀusion model. Int. J. of Bifurcation and Chaos 5(1), 255–263.
References 315
Feigenbaum, M. J. (1978). Quantitative universality for a class of nonlinear trans
formations. J. Stat. Phys. 19, 25–52.
Feigenbaum, M. J. (1979). The universal metric properties of nonlinear transforma
tions. J. Stat. Phys. 21, 669–706.
Feigenbaum, M. J. (1980a). The metric universal properties of period doubling bi
furcations and the spectrum for a route to turbulence., Volume 357 of Ann. N. Y.
Acad. Sci. New York: The New York Academy of Sciences.
Feigenbaum, M. J. (1980b). The onset spectrum of turbulence. Phys. Lett. A 74,
375–378.
Feigenbaum, M. J. (1980c). The transition to aperiodic behaviour in turbulent sys
tems. Commun. Math. Phys. 77, 65–86.
Feigenbaum, M. J. (1980d). Universal behaviour in nonlinear systems. Los Alamos
Science 1, 4–27.
Feigenbaum, M. J. (1983). Universal behavior in nonlinear systems. Physica D 7,
16–39.
Feigenbaum, M. J. (1987a). Scaling spectra and return times of dynamical systems.
J. Stat. Phys. 46, 925–932.
Feigenbaum, M. J. (1987b). Some characterizations of strange sets. J. Stat. Phys. 46,
919–924.
Feigenbaum, M. J. (1988). Presentation functions, ﬁxed points, and a theory of
scaling function dynamics. J. Stat. Phys. 52, 527–569.
Feigenbaum, M. J. (1990). Presentation functions and scaling function theory for
circle maps. In D. K. Campbell (Ed.), Chaos — Soviet American perspectives on
nonlinear science, pp. 3–35. New York: AIP.
Feller, W. (1956). An introduction to probability theory and its applications, Vol
ume 1. New York: Wiley.
Field, R. J., E. K¨ or¨ os, and R. M. Noyes (1972). Oscillations in chemical systems.
II. Thorough analysis of temporal oscillations in the bromateceriummalonic acid
system. J. Am. Chem. Soc. 94, 8649–8664.
Field, R. J. and R. M. Noyes (1974). Oscillations in chemical systems. V. quantita
tive explanation of band migration in the BelousovZhabotinskii reaction. J. Am.
Chem. Soc. 96, 2001–2006.
Fisher, R. A. (1937). The wave of advance of advantageous genes. Ann. Eugenics 7,
355–369.
FitzHugh, R. (1969). Mathematical models of excitation and propagation in nerve
and muscle. In H. P. Schwan (Ed.), Biological engineering. New York: McGraw
Hill.
316 Chaotic Modelling and Simulation
FitzHugh, R. A. (1961). Impulses and physiological states in theoretical models of
nerve membrane. Biophys. J. 1, 445–466.
Flytzanis, N., E. Yiachnakis, and J. Micheloyannis (1991). Analysis of EEG signals
and their spatial correlation over the scalp surface. In M. Markus and A. V. Holden
(Eds.), Nonlinear wave processes in excitable media. New York: Plenum.
Fokas, A. S. and T. Bountis (1996). Order and the ubiquitous occurrence of chaos.
Physica A 228(14), 236–244.
Fowler, A. C. and M. J. McGuinness (1982). Hysteresis in the Lorenz equations.
Phys. Lett. A 92, 103.
Fowler, A. C., M. J. McGuinness, and J. D. Gibbon (1982). The complex Lorenz
equations. Physica D 4, 139.
Frank, M., R. Gencay, and T. Stengos (1988). International chaos? Europ. Econ.
Rev. 32, 1569–1584.
Frank, M. and T. Stengos (1988a). The stability of Canadian macroeconomic data as
measured by the largest Lyapunov exponent. Econ. Lett. 27, 11–14.
Frank, M. Z. and T. Stengos (1988b). Chaotic dynamics in economic timeseries. J.
Economic Surveys 2, 103–133.
Frank, M. Z. and T. Stengos (1988c). Some evidence concerning macroeconomic
chaos. J. Monetary Economics 22, 423–438.
Frederickson, P., J. L. Kaplan, E. D. Yorke, and Y. J. A. (1983). The Lyapunov
dimension of strange attractors. J. Diﬀ. Eqns. 49, 185–207.
Frouzakis, C. E., R. A. Adomaitis, and I. G. Kevrekidis (1991). Resonance phe
nomena in an adaptivelycontrolled system. Int. J. of Bifurcation and Chaos 1(1),
83–106.
Frouzakis, C. E., L. Gardini, I. G. Kevrekidis, G. Millerioux, and C. Mira (1997).
On some properties of invariant sets of twodimensional noninvertible maps. Int.
J. of Bifurcation and Chaos 7(6), 1167–1194.
Fukuda, W. and S. Katsura (1986). Exactly solvable models showing chaotic behav
ior II. Physica A 136, 588.
Galias, Z. and P. Zgliczynski (1998). Computer assisted proof of chaos in the Lorenz
equations. Physica D 115, 165–188.
Gallas, J. A. C. (1993). Structure of the parameter space of the H´ enon map. Phys.
Rev. Lett. 70(18), 2714–2717.
Garbaczewski, P., M. Wolf, and A. Weron (Eds.) (1995). Chaos — The interplay
between stochastic and deterministic behaviour. Proc. of the 31st Winter School of
Theor Phys., Karpasz, Poland, 13.24. Feb. 1995. Berlin: SpringerVerlag.
References 317
Gause, G. F. and A. A. Witt (1934). On the periodic ﬂuctuations in the numbers
of animals. A mathematical theory of the relaxation interaction between predators
and prey and its application to a population of protozoa. Bull. Acad. Sci. U.R.S.S.,
1551–1559.
Georgiou, I. T. (1999). On the global geometric structure of the dynamics of the
elastic pendulum. Nonlinear Dynamics 18(1), 51–68.
Geysermans, P. and F. Baras (1997). Stochastic description of a period2 limit cycle.
Europhys. Lett. 40(1), 1–6.
Giovanis, A. and C. H. Skiadas (1995). Forecasting the electricity consumption by
applying stochastic modeling techniques. In J. Janssen, C. H. Skiadas, and C. Zo
pounidis (Eds.), Advances in stochastic modelling and data analysis. Dordrecht:
Cluwer Acad. Publishers.
Glass, L. (1975). Global analysis of nonlinear chemical kinetics. In B. Berne (Ed.),
Statistical mechanics, pp. 311–349. New York: Plenum.
Glass, L. (1988). Simple mathematical models for complex dynamics in physiologi
cal systems. In B.L. Hao (Ed.), Directions in chaos, Volume 2, pp. 90. Singapore:
World Scientiﬁc.
Glass, L. (1991). Cardiac arrhythmias and circle maps. Chaos 1, 13–19.
Glass, L., M. R. Guevara, J. Belair, and A. Shrier (1984). Global bifurcations of a
periodically forced biological oscillator. Phys. Rev. A 29, 1348–1357.
Glass, L. and P. Hunter (1990). There is A theory of heart. Physica D 43, 1–16.
Glass, L. and M. C. Mackey (1988). From clocks to chaos. Princeton NJ: Univ.
Press.
Glass, L. and J. S. Pasternack (1978). Stable oscillations in mathematical models of
biological control systems. J. Math. Biol. 6, 207–223.
Glass, L. and R. Young (1979). Structure and dynamics of neural network oscillators.
Brain Res. 179, 207–218.
Glass, L. and W. Zeng (1990). Complex bifurcations and chaos in simple theoretical
models of cardiac oscillations. In Mathematical Approaches to Cardiac Arrhyth
mias, Volume 591 of Ann. N.Y. Acad. Sci., pp. 316–327.
Gleick, J. (1987). Chaos — Making a new science. New York: Viking.
Goetz, A. (1998). Perturbations of 8attractors and births of satellite systems. Int. J.
of Bifurcation and Chaos 8(10), 1937–1956.
Gollub, J. P. and H. L. Swinney (1975). Onset of turbulence in a rotating ﬂuid. Phys.
Rev. Lett. 35, 927.
Gompertz, B. (1825). On the nature of the function expressing of the law of human
318 Chaotic Modelling and Simulation
mortality. Philosoph. Trans. Royal Soc. 36, 513–585.
Goriely, A. (1992). From weak to full Painlev´ e property via time singularities trans
formations. In T. Bountis (Ed.), Chaotic dynamics and practice. London: Plenum.
Grammaticos, B., B. Dorizzi, and A. Ramani (1983). Integrability of Hamiltonians
with third and fourth degree polynomial potentials. J. Math. Phys. 24, 2289.
Grammaticos, B., A. Ramani, and V. G. Papageorgiou (1991). Do integrable map
pings have the Painlev´ e property? Phys. Rev. Lett. 67, 1825.
Grassberger, P. (1981). On the Hausdorﬀ dimension of fractal attractors. J. Stat.
Phys. 26, 173.
Grassberger, P. (1983a). Generalized dimensions of strange attractors. Phys. Lett.
A 97, 227–230.
Grassberger, P. (1983b). On the fractal dimension of the H´ enon attractor. Phys. Lett.
A 97, 224–226.
Grassberger, P. (1985). Information ﬂow and maximum entropy measures for 1D
maps. Physica D 14, 365–373.
Grassberger, P. (1986). Toward a quantitative theory of selfgenerated complexity.
Int. J. Theor. Phys. 25(9), 907–938.
Grassberger, P. and I. Procaccia (1983a). Estimation of the Kolmogorov entropy
from a chaotic signal. Phys. Rev. A 28, 2591–2593.
Grassberger, P. and I. Procaccia (1983b). Measuring the strangeness of strange at
tractors. Physica D 9, 189–208.
Grassberger, P. and I. Procaccia (1983c). On the characterization of strange attrac
tors. Phys. Rev. Lett. 50, 346–349.
Grassberger, P. and M. Scheunert (1981). Some more universal scaling laws for
critical mappings. J. Stat. Phys. 26, 697.
Grassberger, P. and T. Schreiber (1999). Statistical mechanics — microscopic chaos
from Brownian motion? Nature 401(6756), 875–876.
Grau, M. and et al (1983). On the periodic orbits of the Contopoulos Hamiltonian.
In L. Garrido (Ed.), Dynamical systems and chaos, Volume 179 of Lecture notes
in physics, pp. 284–286. Berlin: SpringerVerlag.
Grebogi, C., E. Ott, S. Pelikan, and J. A. Yorke (1984). Strange attractors that are
not chaotic. Physica D 13, 261–268.
Grebogi, C., E. Ott, and J. A. Yorke (1982). Chaotic attractors in crisis. Phys. Rev.
Lett. 48, 1507–1510.
Grebogi, C., E. Ott, and J. A. Yorke (1987). Unstable periodic orbits and the dimen
sion of chaotic attractors. Phys. Rev. A 36, 3522–3524.
References 319
Grebogi, C., E. Ott, and J. A. Yorke (1988). Roundoﬀ induced periodicity and the
correlation dimension of chaotic attractors. Phys. Rev. A 38, 366.
Guckenheimer, J. (1976). A strange strange attractor. In J. E. Marsden and M. Mc
Cracken (Eds.), The Hopf bifurcation and applications, pp. 368–381. Berlin:
SpringerVerlag.
Guckenheimer, J. (1979a). The bifurcation of quadratic functions. Annals of N.Y.
Acad. Sc. 316, 78–85.
Guckenheimer, J. (1979b). Sensitive dependence to initial conditions for one
dimensional maps. Commun. Math. Phys. 70, 133.
Guckenheimer, J. (1980a). Dynamics of the van der Pol equation. IEEE Trans. Circ.
and Systems CAS 27, 983.
Guckenheimer, J. (1980b). Symbolic dynamics and relaxation oscillations. Physica
D 1, 227–235.
Guckenheimer, J. and P. J. Holmes (1983). Nonlinear oscillations, dynamical sys
tems, and bifurcations of vector ﬁelds. Berlin: SpringerVerlag.
Guckenheimer, J. and P. J. Holmes (1988). Structurally stable heteroclinic cycles.
Math. Proc. Cambridge Philos. Soc. 103, 189–192.
Guevara, M. R., L. Glass, M. C. Mackey, and A. Shrier (1983). Chaos in neurobiol
ogy. IEEE Trans. SMC 13, 790.
Hadamard, J. (1898). Les surfaces ` a courbures oppos´ ees et leurs lignes g´ eod´ esiques.
J. Math. Pure Appl. 4, 27.
Haken, H. (1978). Synergetics; An introduction. New York: SpringerVerlag.
Haken, H. and G. MayerKress (1981). ChapmanKolmogorov equation and path
integrals for discrete chaos in presence of noise. Z. Physik 43, 185–187.
Hao, B.L. (Ed.) (1984). Chaos. Singapore: World Scientiﬁc.
Hao, B.L. (Ed.) (1990). Chaos II. Singapore: World Scientiﬁc.
Hassard, B., S. P. Hastings, W. C. Troy, and J. Zhang (1994). A computer proof that
the Lorenz equations have “chaotic” solutions. Appl. Math. Lett., in press.
Hausdorﬀ, F. (1919). Dimension und ¨ außeres maß. Math. Annalen 79, 157.
Helleman, R. H. G. (1980a). Nonlinear dynamics, Volume 357 of Ann. N. Y. Acad.
Sci. New York: The New York Academy of Sciences.
Helleman, R. H. G. (1980b). Selfgenerated chaotic behavior in nonlinear mechan
ics. In E. G. D. Cohen (Ed.), Fundamental problems in statistical mechanics,
Volume 5, pp. 165. Amsterdam: NorthHolland.
Helleman, R. H. G. (1981). One mechanism for the onsets of largescale chaos in con
320 Chaotic Modelling and Simulation
servative and dissipative systems. In W. H. et al. (Ed.), Nonequilibrium problems
in statistical mechanics, Volume 2. New York: Wiley.
Helleman, R. H. G. (1984). Feigenbaum sequences in conservative and dissipative
systems. Springer Ser. in Synerg. 11.
H´ enon, M. (1969). Numerical study of quadratic areapreserving mappings. Quart.
Appl. Math. 27, 291–312.
H´ enon, M. (1976). A twodimensional mapping with a strange attractor. Commun.
Math. Phys. 50, 69–77.
H´ enon, M. (1982). On the numerical computation of Poincar´ e maps. Physica D 5,
412–414.
H´ enon, M. (1989). Chaotic scattering modelled by an inclined billiard. Physica
D 33, 132–156.
H´ enon, M. and C. Heiles (1964). The applicability of the third integral of motion:
Some numerical experiments. Astron. J. 69, 73–79.
H´ enon, M. and Y. Pomeau (1977). Two strange attractors with a simple structure. In
R. Temam (Ed.), Turbulence and NavierStokes equations, Volume 565 of Lecture
notes in mathematics, pp. 29. Berlin: SpringerVerlag.
Hilborn, R. C. (1994). Chaos and nonlinear dynamics: an introduction for scientists
and engineers. Oxford New York: Oxford Univ. Press.
Hirsch, M. W. and S. Smale (1974). Diﬀerential equations, dynamical systems, and
linear algebra. New York: Academic Press.
Hodgkin, A. L. and A. F. Huxley (1952a). Current carried by sodium and potassium
ions through the membrane of the giant axon of loligo. J. Physiology 116, 449–
472.
Hodgkin, A. L. and A. F. Huxley (1952b). A quantitative description of membrane
current and its application to conduction and excitation in nerve. J. Physiol. 116,
500–544.
Holmes, C. and P. J. Holmes (1981). Second order averaging and bifurcations to
subharmonics in Duﬃng’s equation. J. Sound Vib. 78, 161–174.
Holmes, P. J. (1977a). Behavior of an oscillator with even nonlinear damping. Int. J.
Nonlinear Mech. 12, 323–326.
Holmes, P. J. (1977b). Strange phenomena in dynamical systems and their physical
implications. Applied Mathematical Modelling 7(1), 362–366.
Holmes, P. J. (1979a). A nonlinear oscillator with a strange attractor. Phil. Trans.
Roy. Soc. London 292, 419–448.
Holmes, P. J. (1979b). Recurrent periodic and nonperiodic behavior in simple dy
References 321
namical systems. In Proc. Joint National Meeting, pp. 107.
Holmes, P. J. (1980). Periodic, nonperiodic and irregular motions in a Hamiltonian
system. Rocky Mountain J. Math 10(4), 679–693.
Holmes, P. J. (1982). Proof of nonintegrability for the H´ enonHeiles Hamiltonian
near an exceptional integrable case. Physica D 5, 35–347.
Holmes, P. J. (1984). Bifurcation sequences in horseshoe maps: inﬁnitely many
routes to chaos. Phys. Lett. 104A, 299–302.
Holmes, P. J. (1986). Spatial structure of timeperiodic solutions of the Ginzburg
Landau equation. Physica D 23, 4–90.
Holmes, P. J. (1990). Poincar´ e, celestial mechanics, dynamical systems theory and
“chaos”. Physics Reports 193(3), 137–163.
Holmes, P. J. and J. E. Marsden (1982). Horseshoes in perturbations of Hamiltonian
systems with two degrees of freedom. Comm. Math. Phys. 82, 523–544.
Holmes, P. J. and D. A. Rand (1976). The bifurcations of Duﬃng’s equation: An
application of catastrophy theory. J. Sound Vib. 44(2), 237–253.
Holmes, P. J. and D. A. Rand (1978). Bifurcation of the forced van der Pol oscillator.
Quart. Appl. Math. 35, 495–509.
Hopf, E. (1942). Abzweigung einer periodischen L¨ osung von einer station¨ aren
L¨ osung eines diﬀerentialsystems. Ber. Math.Phys. Akad. Wiss. Leipz. 94, 3–22.
Hopf, E. (1948). A mathematical example displaying features of turbulence (ex
cerpt). Commun. Pure Appl. Math. 1, 303.
Hopf, F. A., D. L. Kaplan, H. M. Gibbs, and R. L. Shoemaker (1982). Bifurcations
to chaos in optical bistability. Phys. Rev. A 25, 2172.
Hoppensteadt, F. C. (1993). Analysis and simulation of chaotic systems. New York:
SpringerVerlag.
Hudson, J. L., O. E. R¨ ossler, and H. C. Killory (1988). Chaos in a fourvariable
piecewiselinear system of diﬀerential equations. IEEE Trans. Circ. and Systems
CAS 35, 902–908.
Hunter, C. and A. Toomre (1969). Dynamics of the bending of the galaxy. Astrophys.
J. 155, 747.
Ide, K. and S. Wiggins (1989). The bifurcation to homoclinic tori in the quasiperi
odically forced Duﬃng oscillator. Physica D 34, 169–182.
Ikeda, K. (1979). Multiplevalued stationary state and its instability of the transmit
ted light by a ring cavity system. Opt. Commun. 30, 257.
Ikeda, K. and O. Akimoto (1982). Instability leading to periodic and chaotic self
pulsations in a bistable optical cavity. Phys. Rev. Lett. 48, 617.
322 Chaotic Modelling and Simulation
Ikeda, K., H. Daido, and O. Akimoto (1980). Optical turbulence: Chaotic behavior
of transmitted light from a ring cavity. Phys. Rev. Lett. 45, 709.
Ikeda, K. and K. Kondo (1982). Successive higherharmonic bifurcations in systems
with delayed feedback. Phys. Rev. Lett. 49, 1467.
Ikeda, K. and K. Matsumoto (1987). Highdimensional chaotic behavior in systems
with timedelayed feedback. Physica D 29, 223–235.
Ikeda, T. and S. Murakami (1999). Dynamic response and stability of a rotating
asymmetric shaft mounted on a ﬂexible base. Nonlinear Dynamics 20(1), 1–19.
Ikeda, Y. and S. Tokinaga (2000). Approximation of chaotic dynamics by using
smaller number of data based upon the genetic programming and its applications.
IEICE Trans. Fund. Elec. Com. Com. E83A(8), 1599–1607.
Iooss, G., R. Helleman, and R. Stora (Eds.) (1983). Chaotic behavior of deterministic
systems. Amsterdam: North Holland.
Ishida, Y., T. Ikeda, T. Yamamoto, and S. Murakami (1989). Nonstationary vibration
of a rotating shaft with nonlinear spring characteristics during acceleration through
a critical speed (A critical speed of a 1/2order subharmonic oscillation). JSME
Int. J. Ser. III 32(4), 575–584.
Ito, A. (1979). Successive subharmonic bifurcations and chaos in a nonlinear Math
ieu equation. Prog. Theor. Phys. 61, 815.
Ito, H. and L. Glass (1991). Spiral breakup in a new model of discrete excitable
media. Phys. Rev. Lett. 66, 671–674.
Itoh, H. (1985). Nonintegrability of H´ enonHeiles system and a theorem of ziglin.
Kodai Math. J. 8, 120.
Joy, M. P. (1995). On the bifurcation in a modulated logistic map: Comment. Phys.
Lett. A 202(23), 237–239.
Julia, G. (1918). M´ emoire sur l’it´ eration des fonctions rationelles. J. Math. Pure
Appl. 8, 47–245.
Kac, M. (1947). Random walk and the theory of Brownian motion. Am. Math.
Month. 54(7), 369–391.
Kadanoﬀ, L. P. (1981). Scaling for a critical KolmogorovArnoldMoser trajectory.
Phys. Rev. Lett. 47, 1641–1643.
Kadanoﬀ, L. P. (1993). From order to chaos: Essays: Critical, chaotic and other
wise. Singapore: World Scientiﬁc.
Kahlert, C. and O. E. R¨ ossler (1984). Chaos as a limit in a boundary value problem.
Z. Naturforsch. A 39, 1200–1203.
Kahlert, C. and O. E. R¨ ossler (1985). Analytical properties of Poincar´ e halfmaps in
References 323
a class of piecewiselinear dynamical systems. Z. Naturforsch. 40, 1011–1025.
Kahlert, C. and O. E. R¨ ossler (1986). The separating mechanisms in Poincar´ e half
maps. Z. Naturforsch. A 41, 1369–1380.
Kahlert, C. and O. E. R¨ ossler (1987). Analogues to a boundary away fromanalyticity.
Z. Naturforsch. 42, 324–328.
Kalman, R. E. (1956). Nonlinear aspects of sample data control systems. In Proc.
Symp. on Nonlin. Circ. Anal., Brooklyn, April 1956, pp. 273–313.
Kalman, R. E. (1960). A new approach to linear ﬁltering and prediction problems.
J. Basic Eng. 82, 35–45.
Kapitaniak, T. (1990). Analytical condition for chaotic behavior of the Duﬃng os
cillator. Phys. Lett. A 144, 322–324.
Kapitaniak, T. (1991). On strange nonchaotic attractors and their dimension. Chaos
Solitons and Fractals 1, 67–77.
Kaplan, H. (1983). New method for calculating stable and unstable periodic orbits
of onedimensional maps. Phys. Lett. A 97, 365.
Kaplan, J. L. and J. A. Yorke (1979a). Chaotic behavior of multidimensional dif
ference equations. In H. O. Walter and H.O. Peitgen (Eds.), Functional diﬀeren
tial equations and approximation of ﬁxed points, Volume 730 of Lecture notes in
mathematics, pp. 204–227. Berlin: SpringerVerlag.
Kaplan, J. L. and J. A. Yorke (1979b). The onset of chaos in a ﬂuid ﬂow model of
Lorenz, Volume 316 of Ann. N. Y. Acad. Sci. New York: The New York Academy
of Sciences.
Kappos, E. (1996). The Conley index and global bifurcations. II. Illustrative appli
cations. J. of Bifurcation and Chaos 6(12B), 2491–2505.
Karamanos, K. and G. Nicolis (1999). Symbolic dynamics and entropy analysis of
Feigenbaum limit sets. Chaos Solitons Fractals 10(7), 1135–1150.
K´ arolyi, G., I. Scheuring, and T. Cz´ ar´ an (2002). Metabolic network dynamics in
open chaotic ﬂow. Chaos 12(2), 460–469.
K´ arolyi, G. and T. T´ el (1997). Chaotic tracer scattering and fractal basin boundaries
in a blinking vortexsink system. Physics Reports 290, 125–147.
Katsamaki, A. and C. H. Skiadas (1995). Analytic solution and estimation of para
meters on a stochastic exponential model for a technological diﬀusion process.
Appl. Stoch. Mod. Dat. Analys. 11, 59–75.
Katz, L. (1996). Note on the Schwarzian derivative. Chaos Solitons Fractals 7(9),
1495–1496.
Keppenne, C. and C. Nicolis (1988). Toward a quantitative view of predictability of
324 Chaotic Modelling and Simulation
weather. Annal. Geophys., 207.
Keppenne, C. L. and C. Nicolis (1989). Global properties and local structure of the
weather attractor over western Europe. J. Atmos. Sci. 46, 2356.
Kerr, R. A. (1989). Does chaos permeate the solar system? Science 244, 144–45.
Kevrekidis, I. G., R. Aris, L. D. Schmidt, and S. Pelikan (1985). Numerical compu
tation of invariant circles of maps. Physica D 16, 243–251.
Kevrekidis, I. G. and M. S. Jolly (1987). On the use of interactive graphics in the nu
merical study of chemical dynamics. 1987 AIChE Annual Meeting, New York 22.
Kevrekidis, I. G., R. RicoMart´ınez, R. E. Ecke, R. M. Farber, and A. S. Lapedes
(1994). Global bifurcations in RaleighB´ enard convection. Experiments, empirical
maps and numerical bifurcation analysis. Physica D 71, 342–362.
Kilias, T. (1994). Generation of pseudochaotic sequences. Int. J. of Bifurcation and
Chaos 4(3), 709–713.
Kim, J. H. and J. Stringer (Eds.) (1997). Applied chaos. New York: Wiley.
Kirner, T. and O. E. R¨ ossler (1997). Frequency clustering in a chain of weakly
coupled oscillators. Z. Naturforsch. A 52(89), 578–580.
Kloeden, P. E. and J. Lorenz (1986). Stable attracting sets in dynamical systems and
in their onestep discretizations. SIAM J. Numer. Anal. 23, 986–995.
Kloeden, P. E. and E. Platen (1989). A survey of numerical methods for stochastic
diﬀerential equations. J. Stoch. Hydrol. Hydraul. 3, 155–178.
Kloeden, P. E. and E. Platen (1992). The numerical solution of stochastic diﬀerential
equations. Berlin: SpringerVerlag.
Kloeden, P. E., E. Platen, and H. Schurz (1991). The numerical solution of nonlinear
stochastic dynamical systems: A brief introduction. Int. J. of Bifurcation and
Chaos 1(2), 277–286.
Kocak, H. (1989). Diﬀerential and diﬀerence equations through computer experi
ments. Berlin: SpringerVerlag.
Kolmogorov, A. (1942). Interpolation and extrapolation of stationary series. Bulletin
de l’Acad´ emie des Sciences de l’U.R.S.S. Series Math´ ematiques 2, 3.
Kolmogorov, A., I. Petrovsky, and N. Piscounoﬀ (1937). Etude de l’´ equation de la
diﬀusion avec croissance de la quantit´ ede mati` ere et son application ` aun probl` eme
biologique, Volume 1. Bull. Univ. Moskou: Ser. internat.
Kolmogorov, A. N. (1936). Sulla teoria di Volterra della lotta per l’esistenza, (in
Italian). Giorn. Istituto Ital. d. Attuari 7, 74–80.
Kolmogorov, A. N. (1954). Preservation of conditionally periodic movements with
small change in the Hamiltonian function. Dokl. Akad. Nauk. SSSR 98, 527.
References 325
Kolmogorov, A. N. (1958). A new invariant of transitive dynamical systems. Dokl.
Akad. Nauk. SSSR 119, 861.
Kolmogorov, A. N. (1965). Three approaches to the quantitative deﬁnition of infor
mation. Inf. Trans. 1, 3–11.
Komineas, S., M. N. Vrahatis, and T. Bountis (1994). 2D universality of period
doubling bifurcations in 3D conservative reversible mappings. Physica A 211(2
3), 218–233.
Kopell, N. and L. N. Howard (1973a). Horizontal bands in the Belousov reaction.
Science 180, 1171–1173.
Kopell, N. and L. N. Howard (1973b). Plane wave solutions to reactiondiﬀusion
equations. Studies in Appl. Math. 52, 291–328.
Kosmatopoulos, E. B., M. M. Polycarpou, M. A. Christodoulou, and P. A. Ioannou
(1995). High order neural network structures for identiﬁcation of dynamical sys
tems. IEEE Trans. Neural Networks 6, 422–431.
Koumoutsos, N. and C. H. Skiadas (1995). Applied stochastic models and data
analysis for engineering education. In J. Janssen, C. H. Skiadas, and C. Zopounidis
(Eds.), Advances in stochastic modelling and data analysis. Dordrecht: Cluwer
Acad. Publishers.
Kounadis, A. N. (1999). Dynamic buckling of autonomous systems having potential
energy universal unfoldings of cuspoid catastrophe. Nonlinear Dynamics 18(3),
235–252.
Kowalevskaya, S. (1889a). Sur le probl` eme de la rotation d’un corps solide autour
d’un point ﬁxe. Acta Math. 12, 177.
Kowalevskaya, S. (1889b). Sur une propri´ et´ e du syst` eme d’´ equations diﬀ´ erentielles
qui d´ eﬁnit la rotation d’un corps solide autour d’un point ﬁxe. Acta Math. 13, 81.
Kremmydas, G. P., A. V. Holden, A. Bezerianos, and T. Bountis (1996). Repre
sentation of sinoatrial node dynamics by circle maps. Int. J. of Bifurcation and
Chaos 6(10), 1799–1805.
Kuczma, M. (1968). Functional equations in a single variable. Warszawa: PWN 
Polish Scientiﬁc Publishers, Warszawa.
Kugiumtzis, D. (1997). Correction of the correlation dimension for noisy time series.
Int. J. of Bifurcation and Chaos 7(6), 1283–1294.
Kuske, R. and G. Papanicolaou (1998). The invariant density of a chaotic dynamical
system with small noise. Physica D 120(34), 255–272.
Kutta, W. (1901). Beitrag zur n¨ aherungsweisen integration totaler diﬀerentialgle
ichungen. Z. Math. Phys. 46, 435–453.
Kuznetsov, S. P., A. S. Pikovsky, and U. Feudel (1995). Birth of strange nonchaotic
326 Chaotic Modelling and Simulation
attractor: a renormalization group analysis. Phys. Rev. E 51(3, pt. A), 1629–1632.
Kuznetsov, Y. A. (2004). Elements of applied bifurcation theory. New York:
SpringerVerlag.
Kuznetsov, Y. A., S. Muratori, and S. Rinaldi (1992). Bifurcations and chaos in a
periodic predatorprey model. Int. J. of Bifurcation and Chaos 2(1), 117–128.
Lamb, H. (1879). A Treatise on the Mathematical Theory of the Motion of Fluids.
University Press.
Landau, L. D. (1944). On the problem of turbulence, (Russ. original); English trans
lation in. Akad. Nauk. Doklady 44, 339.
Landberg, P. (1956). Vibrations caused by chip formation. Microtenic 10, 219–228.
Lanford, O. E. (1982a). A computer assisted proof of the Feigenbaum conjectures.
Bull. Am. Math. Soc. 6, 427–434.
Lanford, O. E. (1982b). The strange attractor theory of turbulence. Ann. Rev. of
Fluid Mech. 14, 347.
Lanford, O. E. (1983). Period doubling in one and several dimensions. Physica D 7,
124–125.
Lanford, O. E. (1984). Functional equations for circle homeomorphisms with golden
ratio rotation number. J. Stat. Phys. 34, 57–73.
Lasota, A. and M. C. Mackey (1985). Probabilistic properties of deterministic sys
tems. Cambridge: University Press.
Lasota, A. and M. C. Mackey (1994). Chaos, Fractals, and Noise (Applied Mathe
matical Sciences), Volume 97. Berlin: SpringerVerlag.
Lauterborn, W. (1976). Numerical investigation of nonlinear oscillations of gas bub
bles in liquids. J. Acoust. Soc. Am. 59, 283–293.
Leith, C. E. (1996). Stochastic models of chaotic systems. Physica D 98(24), 481–
491.
Leonov, N. N. (1959). Transformations d’une droite en ellem` eme. Radioﬁscia 2,
942.
Leonov, N. N. (1960a). Th´ eorie des transformations discontinues d’une droite en
ellem` eme. Radioﬁscia 3, 872.
Leonov, N. N. (1960b). Transformation ponctuelle d’une droite en ellem` eme dis
continue, lin´ eaire par morceaux. Radioﬁscia 3, 496.
Leontaritis, I. J. and S. A. Billings (1987). Experimental design and identiﬁcability
for nonlinear systems. Int. J. Syst. Sci. 18(1), 189–202.
Lewis, J. E. and L. Glass (1991). Steady states, limit cycles, and chaos in models of
References 327
complex biological networks. Int. J. of Bifurcation and Chaos 1(2), 477–483.
Li, K. T. and L. Z. Wang (2001). Global bifurcation and long time behavior of the
VolterraLotka ecological model. Int. J. Bifurcation Chaos 11(1), 133–142.
Li, T. Y. and J. A. Yorke (1975). Period three implies chaos. Am. Math. Monthly 82,
985–992.
Lorenz, E. (1993). The essence of chaos. Seattle: Univ. of Washington Press.
Lorenz, E. N. (1956). Empirical orthonogal functions and statistical weather predic
tion. Scientiﬁc Report No. 1, Statistical Forecasting Project, Cambridge, MA.
Lorenz, E. N. (1963). Deterministic nonperiodic ﬂow. J. Atmos. Sci. 20, 130–141.
Lorenz, E. N. (1964). The problem of deducing the climate from the governing
equations. Tellus 16, 1.
Lorenz, E. N. (1969). Atmospheric predictability as revealed by naturally occurring
analogies. J. Atmos. Sci. 26, 636.
Lorenz, E. N. (1979). Predictability: Does the ﬂap of a butterﬂy’s wings in Brazil set
oﬀ a tornado in Texas? Talk given at the annual meeting of the AAAS December
29, 1979 in Washington, American Association for the Advancement of Science.
Lorenz, E. N. (1980). Noisy periodicity and reverse bifurcation. Ann. N. Y. Acad.
Sci. 357, 282–291.
Lorenz, E. N. (1991). Dimension of weather and climate attractors. Nature 353,
241–244.
Lotka, A. (1910). Zur theorie der periodischen raktionen. Z. phys. Chemie 72, 508.
Lozi, R. and S. Ushiki (1988). Organized conﬁnors and anticonﬁnors and their bi
furcations in constrained. Ann. T` el` ecommun. 43, 187–208.
Lozi, R. and S. Ushiki (1991a). Coexisting chaotic attractors in Chua’s circuit. Int.
J. of Bifurcation and Chaos 1(4), 923–926.
Lozi, R. and S. Ushiki (1991b). Conﬁnors and boundedtime patterns in Chua’s
circuit and the double scroll family. Int. J. of Bifurcation and Chaos 1(1), 119–
138.
Lyapunov, M. A. (1893). in Russian, for an English translation see: Ann. Math.
Studies 17, Princeton (1949); cited after W. Hahn (1959) “Theorie und anwendung
der direkten methode von Lyapunov”. Mat. Sbornik 17, 252–333.
Lyapunov, M. A. (1947). Probl` eme g´ en´ eral de la stabilit´ e du mouvement, (1907
translation from the 1892 Russian original). Annal. Math. Stud. 17.
LyndenBell, D. (1969). Galactic nuclei as collapsed old quasars. Nature 223, 690.
MacKay, R. S. (1993). Nonareapreserving directions from areapreserving golden
328 Chaotic Modelling and Simulation
circle ﬁxed points. Nonlinearity. 6(5), 799.
MacKay, R. S. (2001). Complicated dynamics from simple topological hypotheses.
Philos. Trans. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci. 359(1784), 1479–1496.
Mackey, M. C. and L. Glass (1977). Oscillation and chaos in physiological control
systems. Science 197, 287–289.
Mahajan, V. and M. Schoeman (1977). Generalized model for the time pattern of the
diﬀusion process. IEEE Transactions on Engineering Management 24, 12–18.
Malliaris, A. G. and J. L. Stein (1999). Methodological issues in asset pricing: Ran
dom walk or chaotic dynamics. J. Banking and Finance 23(11), 1605–1635.
Malthus, T. (1798). An essay on the principle of population. London: Johnson.
Mandelbrot, B. (1963). The variation of certain speculative prices. J. Business 36,
394–413.
Mandelbrot, B. (1967). The variation of some other speculative prices. J. Busi
ness 40, 393–413.
Mandelbrot, B. (1972). Statistical methodology for nonperiodic cycles: From the
covariance to R/S analysis. Ann. Econ. Soc. Measurement 1.
Mandelbrot, B. (1977). Fractals — form, chance, and dimension. San Francisco:
Freeman.
Mandelbrot, B. (1980). Fractal aspects of the iteration of z → λ∗z(1−z) for complex
λ and z. Ann. N. Y. Acad. Sci. 357, 249–259.
Mandelbrot, B. (1982). The fractal geometry of nature. San Francisco: Freeman.
Mandelbrot, B. and J. Wallis (1969a). Computer experiments with fractional Gaus
sian noises. Part 1. Averages and variances. Water Resourc. Res. 5.
Mandelbrot, B. and J. Wallis (1969b). Computer experiments with fractional Gaus
sian noises. Part 2. Rescaled ranges and spectra. Water Resourc. Res. 5.
Mandelbrot, B. and J. Wallis (1969c). Computer experiments with fractional Gaus
sian noises. Part 3. Mathematical appendix. Water Resourc. Res. 5.
Mandelbrot, B. B. (1974). Intermittent turbulence in selfsimilar cascades: Diver
gence of high moments and dimension of the carrier. J. Fluid Mech. 62, 331.
Mandelbrot, B. B. (1990). Two meanings of multifractality, and the notion of nega
tive fractal dimension. In D. K. Campbell (Ed.), Chaos — SovietAmerican per
spectives on nonlinear science, pp. 79–90. New York: AIP.
Mandelbrot, B. B. and J. V. Ness (1968). Fractional Brownian motions, fractional
noises and applications. SIAM Review 10, 422–437.
Manneville, P. (1980). Intermittency, selfsimilarity and 1/ f spectrum in dissipative
References 329
dynamical systems. J. d. Phys. 41, 1235–1241.
Manneville, P. (1989). Dissipative structures and weak turbulence. New York: Aca
demic Press.
Manneville, P. and Y. Pomeau (1979). Intermittency and the Lorenz model. Phys.
Lett. A 75, 1.
Manneville, P. and Y. Pomeau (1980). Diﬀerent ways to turbulence in dissipative
dynamical systems. Physica D 1, 219–226.
Markstein, G. H. (1951). Experimental and theoretical studies of ﬂamefront stabil
ity. J. Aeronaut. Sci. 18, 199–209.
Markus, M. and S. K. (1994). Observation of chemical turbulence in the Belousov 
Zhabotinsky reaction. Int. J. of Bifurcation and Chaos 4(5), 1233–1243.
Marotto, F. R. (1979). Chaotic behavior in the H´ enon mapping. Commun. Math.
Phys. 68, 187.
Marsden, J. and M. McCracken (1975). The Hopf bifurcation. Berlin: Springer
Verlag.
Matsumoto, T. (1984). A chaotic attractor from Chua’s circuit. IEEE Trans. Circ.
and Systems CAS 31, 1055–1058.
May, R. M. (1972). Limit cycles in predatorprey communities. Science 17, 900–
902.
May, R. M. (1974). Biological populations with nonoverlapping generations: Stable
points, stable cycles, and chaos. Science 186, 645–647.
May, R. M. (1976). Simple mathematical models with very complicated dynamics.
Nature 261, 459–467.
May, R. M. (1987a). Chaos and the dynamics of biological populations. In M. V.
Berry, I. C. Percival, and N. O. Weiss (Eds.), Proc. Royal Soc. Lond. A, Volume
413, Princeton, NJ, pp. 27–44. University Press.
May, R. M. (1987b). When two and two make not four: nonlinear phenomena in
ecology. Proc. Roy. Soc. London A 413, 27–44.
May, R. M. and G. F. Oster (1980). Period doubling and the onset of turbulence. An
analytic estimate of the Feigenbaum ratio. Phys. Lett. A 78, 1.
MayerKress, G. and H. Haken (1981). The inﬂuence of noise on the logistic model.
Journ. Stat. Phys. 26, 149–171.
McKean, H. P. (1970). Nagumo’s equation. Adv. in Math. 4, 209–223.
Meakin, P. (1986). A new model for biological pattern formation. J. Theor. Biol. 118,
101.
330 Chaotic Modelling and Simulation
Meiss, J. D. (1999). Physics of chaos in Hamiltonian systems. Nature 398(6725),
303.
Metropolis, N., A. Rosenblut, M. Rosenbluth, A. Teller, and E. Teller (1953). Equa
tions of state calculations by fast computing machine. J. Chem. Phys. 21, 1097.
Metropolis, N., M. L. Stein, and P. R. Stein (1973). On ﬁnite limit sets for transfor
mations on the unit interval. J. of Combinatorial Theory 15(1), 25–44.
Michelson, D. (1986). Steady solutions of the KuramotoSivashinsky equation.
Physica D 19, 89–111.
Michielin, O. and P. E. Phillipson (1997). Map dynamics study of the Lorenz equa
tions. J. of Bifurcation and Chaos 7(2), 373–382.
Misiurewicz, M. and B. Szewc (1980). Existence of a homoclinic point for H´ enon
map. Commun. Math. Phys. 75, 285.
Mitsi, S., S. Natsiavas, and I. Tsiaﬁs (1998). Dynamics of nonlinear oscillators under
simultaneous internal and external resonances. Nonlinear Dynamics 16(1), 23–39.
Modis, T. (1997). Genetic reengineering of corporations. Technol. Forecast. Social
Change 56, 107–118.
Modis, T. and A. Debecker (1992). Chaoslike states can be expected before and
after logistic growth. Technol. Forecast. Social Change 41, 111–120.
Murakami, C., W. Murakami, and K. Hirose (2002, July). Sequence of global period
doubling bifurcation in the H´ enon map. Chaos, Solitons and Fractals 14(1), 1–17.
Muzzio, F. J. and J. M. Ottino (1990). Diﬀusion and reaction in a lamellar system:
Selfsimilarity with ﬁnite rates of reaction. Phys. Rev. A 42, 5873–5884.
Myrberg, P. J. (1958). Iteration der reellen Polynome zweiten grades. Ann. Acad.
Sci. Fenn. 256, 1–10.
Myrberg, P. J. (1962). Sur l’it´ eration des polynomes r´ eels quadratiques. J. de Math.
Pures et Appl. 9, 339.
Nagumo, J., S. Arimoto, and S. Yoshizawa (1962). An active pulse transmission line
simulating nerve axon. In Proc. IRE, Volume 50, pp. 2061–2070.
Namba, T. (1986). Bifurcation phenomena appearing in the LotkaVolterra compe
tition equation: A numerical study. Math. Biosci. 81, 191–212.
Naschie, M. S. E. (1989). Generalized bifurcation and shell buckling as special
statical chaos. ZAMM 69, 376–377.
Naschie, M. S. E. and S. A. Athel (1989). On the connection between statical and
dynamical chaos. Z. Naturforsch. 44, 645–650.
Naschie, M. S. E. and T. Kapitaniak (1990). Soliton chaos models for mechanical
and biological elastic chains. Phys. Lett. A 147, 275–281.
References 331
Natsiavas, S., S. Theodossiades, and I. Goudas (2000). Dynamic analysis of piece
wise linear oscillators with time periodic coeﬃcients. Int. J. Non Lin. Mech. 35(1),
53–68.
Natsiavas, S. and G. Verros (1999). Dynamics of oscillators with strongly nonlinear
asymmetric damping. Nonlinear Dynamics 20(3), 221–246.
Navarro, J. F., C. S. Frenk, and D. M. White (1995). Simulations of Xray clusters.
Mon. Not. R. Astron. Soc. 275, 720.
Neimark, Y. I. and P. S. Landa (1992). Stochastic and chaotic oscillations. Dordrecht,
Netherlands: Kluwer Academic.
Nicolis, C. and G. Nicolis (1984). Is there a climatic attractor? Nature 311, 529–532.
Nicolis, C. and G. Nicolis (1987). Evidence for climatic attractors. Nature 326, 523.
Nicolis, C. and G. Nicolis (1991). Dynamics for error growth in unstable systems.
Phys. Rev. A 43, 5720–5723.
Nicolis, C. and G. Nicolis (1993). Finite time behavior of small errors in deter
ministic chaos and Lyapunov exponents. Int. J. of Bifurcation and Chaos 3(5),
1339–1342.
Nicolis, C. and G. Nicolis (1998). Closing the hierarchy of moment equations in
nonlinear dynamical systems. Phys. Rev. E 58(4), 4391–4400.
Nicolis, G. (1971). Stability and dissipative structures in open systems far from
equilibrium. Adv. Chem. Phys. 19, 209–324.
Nicolis, G. (1996). Natural laws and the physics of complex systems. In P. Wein
gartner and G. Schurz (Eds.), Law and prediction in the light of chaos research,
Volume LNP 473, pp. 36–39. Berlin: SpringerVerlag.
Nicolis, G., T. Erneux, and M. HerschkowitzKaufman (1978). Pattern formation in
reacting and diﬀusing systems. Adv. Chem. Phys. 38, 263–314.
Nicolis, G. and J. Portnow (1973). Chemical oscillations. Chem. Rev. 73, 365–384.
Nicolis, G. and I. Prigogine (1977). Selforganization nonequilibrium systems. New
York: WileyInterscience.
Nicolis, G. and I. Prigogine (1981). Symmetrybreaking and pattern selection in
farfromequilibrium systems. In Proc. Natl. Acad. Sci. USA, Volume 78, pp.
659–663.
Nicolis, J. S., G. MayerKress, and G. Haubs (1983). Nonuniform chaotic dynamics
with implications to information processing. Z. Naturforsch. 38a, 1157–1169.
Nikolaev, E. V., V. N. Biktashev, and A. V. Holden (1999). On bifurcations of spiral
waves in the plane. Int. J. of Bifurcation and Chaos 9(8), 1501–1516.
Nusse, H. E. and J. A. Yorke (1996). Basins of attraction. Science 271(5254), 1376–
332 Chaotic Modelling and Simulation
1380.
Oono, Y. (1978). Period 2
n
implies chaos. Prog. Theor. Phys. 59, 1028–1030.
Ornstein, D. S. (1995). In what sense can a deterministic system be random? Chaos
Solitons Fractals 5(2), 139–141.
Ostriker, J. P. and E. P. J. Peebles (1973). A numerical study of the stability of
ﬂattened galaxies: Or, can cold galaxies survive? Astrophys. J. 186, 467–480.
Ott, E. (1981). Strange attractors and chaotic motions of dynamical systems. Rev.
Mod. Phys. 53(4), 655–671.
Ott, E. (1993). Chaos in dynamical systems. Cambridge: University Press.
Ott, E., C. Grebogi, and J. A. Yorke (1989). Theory of ﬁrst order phase transitions
for chaotic attractors of nonlinear dynamical systems. Phys. Lett. A 135, 343–348.
Ott, E., C. Grebogi, and J. A. Yorke (1990). Controlling chaos. Phys. Rev. Lett. 64,
1196–1199.
Ott, E., T. Sauer, and J. A. Yorke (1994a). Coping with chaos. New York: Wiley.
Ott, E., T. Sauer, and J. A. Yorke (Eds.) (1994b). Coping with chaos; Analysis of
chaotic data and the exploitation of chaotic systems. New York: Wiley & Sons
Inc.
Ott, E., E. D. Yorke, and J. A. Yorke (1985). A scaling law. How an attractor’s
volume depends on noise level. Physica D 16, 62–78.
Ottino, J. M. (1989). The kinematics of mixing: Stretching, chaos, and transport.
Cambridge: University Press.
Ottino, J. M. (1990). Mixing, chaotic advection, and turbulence. Ann. Rev. Fluid
Mech. 22, 207–253.
Ozaki, T. (1982). The statistical analysis of perturbed limit cycle processes using
nonlinear time series models. J. Time Series Anal. 3, 29.
Paidoussis, M. P., G. X. Li, and F. C. Moon (1989). Chaotic oscillations of the
autonomous system of a constrained pipe conveying ﬂuid. J. Sound Vib. 135, 1–
19.
Palis, J. (2002). Chaotic and complex systems. Curr. Sci. 82(4), 403–406.
Panas, E. and V. Ninni (2000). Are oil markets chaotic? A nonlinear dynamic
analysis. Energy Econ. 22(5), 549–568.
Papadimitriou, S., T. Bountis, S. Mavroudi, and A. Bezerianos (2001). A probabilis
tic symmetric encryption scheme for very fast secure communication based on
chaotic systems of diﬀerence equations. Int. J. Bifurcation Chaos 11(12), 3107–
3115.
References 333
Papaioannou, G. and A. Karytinos (1995). Nonlinear time series analysis of the stock
exchange: The case of an emerging market. Int. J. of Bifurcation and Chaos 5(6),
1557–1584.
Papoulias, F. A. (1991). Bifurcation analysis of line of sight vehicle guidance using
sliding modes. Int. J. of Bifurcation and Chaos 1(4), 849–865.
Papoulias, F. A. and M. M. Bernitsas (1988). Autonomous oscillations, bifurcations
and chaotic response of moored vessels. J. Ship Res. 32(3), 220–228.
Pasta, J. R., S. M. Ulam, and E. Fermi (1965). Studies on nonlinear problems. In
Collected works of Enrico Fermi, Volume 2, pp. 978. Chicago: Univ. of Chicago
Press.
Patsis, P. A., C. Efthymiopoulos, G. Contopoulos, and N. Voglis (1997). Dynamical
spectra of barred galaxies. Astron. Astrophys. 326, 493–500.
Pearl, R. and L. J. Reed (1920). On the rate of growth of the population of the
United States since 1790 and its mathematical representation. Proceedings of the
National Academy of Sciences of the United States of America 6(6), 275–288.
Pecora, L. M. (Ed.) (1993). Chaos in communications, Volume 2038 of SPIE pro
ceedings. Bellingham, Washington: SPIEThe International Society for Optical
Engineering.
Pecora, L. M. and T. L. Carroll (1990). Synchronization in chaotic systems. Phys.
Rev. Lett. 64, 821–824.
Pecora, L. M. and T. L. Carroll (1991). Driving systems with chaotic signals. Phys.
Rev. A 44, 2374–2383.
Pecora, L. M. and T. L. Carroll (1993). Using chaos to eliminate multiple period
behavior. Phys. Rev. A 48, 2426.
Pedersen, P. O. (1935). Subharmonics in forced oscillations in dissipative systems.
part I. J. Acoust. Soc. Am. 6, 227–238.
Perez, R. and L. Glass (1982). Bistability, period doubling bifurcations and chaos in
a periodically forced oscillator. Phys. Lett. A 90, 441.
Petit, J. M. and M. H´ enon (1986). Satellite encounter. Icarus 60, 536.
Phatak, S. C. and S. S. Rao (1995). Logistic map: A possible randomnumber gen
erator. Phys. Rev. E 51(4), 3670–3678.
Poincar´ e, H. (1890a). M´ emoire sur les courbes d´ eﬁnies par les ´ equations
diﬀ´ erentielles IVI, oeuvre I. In Oeuvre I. Paris: GauthierVillars.
Poincar´ e, H. (1890b). Sur les ´ equations de la dynamique et le probl` eme de trois
corps. Acta Math. 13, 1–270.
Poincar´ e, H. (1892). Les m´ ethodes nouvelles de la m´ ecanique c´ eleste. Paris:
334 Chaotic Modelling and Simulation
GauthierVillars.
Poincar´ e, H. (1899). Les m´ ethodes nouvelles de la m´ ecanique c´ eleste, Volume 3.
Paris: GauthierVillars.
Poincar´ e, H. (1905). Le¸ cons de m´ ecanique. Paris: GauthierVillars.
Poincar´ e, H. (1908). Science et methode. Biblioteque Scientiﬁque.
Pomeau, Y. and P. Manneville (1980). Intermittent transition to turbulence in dissi
pative dynamical systems. Commun. Math. Phys. 74, 189–197.
Poznanski, K. Z. (1983). International diﬀusion of steel technologies. Timelag and
the speed of diﬀusion. Technol. Forecast. Social Change 23, 305–323.
Prigogine, I. (1995). Why irreversibility? The formulation of classical and quantum
mechanics for nonintegrable systems. Int. J. of Bifurcation and Chaos 5(1), 3–16.
Prigogine, I. (1996). Time, chaos and the laws of nature. In P. Weingartner and
G. Schurz (Eds.), Law and prediction in the light of chaos research, Volume LNP
473, pp. 3–9. Berlin: SpringerVerlag.
Prigogine, I. (1997). Nonlinear science and the laws of nature. Int. J. of Bifurcation
and Chaos 7(9), 1917–1926.
Prigogine, I. and R. Lefever (1968). Symmetry breaking instabilities in dissipative
systems. II. J. Chem. Phys. 48, 1695–1700.
Prigogine, I., R. Lefever, A. Goldbeter, and M. HerschkowitzKaufman (1969). Sym
metry breaking instabilities in biological systems. Nature 223, 913–916.
Procaccia, I. (1987). Exploring deterministic chaos via unstable periodic orbits,
Volume 2. Nucl. Phys.: proc. suppl.
Rabinovich, S., G. Berkolaiko, S. Buldyrev, A. Shehter, and S. Havlin (1997). An
alytical solution of the logistic equation. Int. J. of Bifurcation and Chaos 7(4),
837–838.
Raha, N., J. A. Sellwood, R. A. James, and F. D. Kahn (1991). A dynamic instability
of bars in disk galaxies. Nature 352, 411–412.
Rayleigh, L. (1883a). On maintained vibrations. Phil. Mag. 15, 229–235.
Rayleigh, L. (1883b). On the crispations of ﬂuid resting upon a vibrating support.
Phil. Mag. 16, 50–58.
Rayleigh, L. (1887). On the maintenance of vibrations by forces of double frequency,
and on the propagation of waves through a medium endowed with a period struc
ture. Phil. Mag. 24, 145–159.
Rayleigh, L. (1891). On the problem of random vibrations, and of random ﬂights in
one, two, or three dimensions. Phil. Mag. 37, 321.
References 335
R´ enyi, A. (1959). On the dimension and entropy of probability distributions. Acta
Math. Acad. Sci. Hung. 10, 193–215.
Rice, S. O. (1945). Mathematical analysis of random noise. Bell System Tech. J. 23
and 24, 1–162.
Riess, P. (1859). Das anblasen oﬀener rohre durch eine ﬂamme. Ann. Phys.
Chem. 108, 653–656.
Rijke, P. L. (1859). Notiz ¨ uber eine neue art, die in einer an beiden enden oﬀenen
R¨ ohre enthaltene luft in schwingungen zu versetzen. Ann. Phys. Chem. 107, 339–
343.
Ritt, F. C. (1923). Permutable rational functions. Trans. Am. Math. Soc. 25, 399–448.
R¨ ossler, O. E. (1976a). Chaotic behavior in simple reaction systems. Z. Natur
forsch. 31, 259–264.
R¨ ossler, O. E. (1976b). Chemical turbulence: Chaos in a simple reactiondiﬀusion
system. Z. Naturforsch. A 31, 1168–1172.
R¨ ossler, O. E. (1976c). Diﬀerent types of chaos in two simple diﬀerential equations.
Z. Naturforsch. 31, 1661.
R¨ ossler, O. E. (1976d). An equation for continuous chaos. Phys. Lett. A 57, 397–398.
R¨ ossler, O. E. (1979a). Continuous chaos — four prototype equations, Volume 316
of Ann. N. Y. Acad. Sci. New York: The New York Academy of Sciences.
R¨ ossler, O. E. (1979b). An equation for hyperchaos. Phys. Lett. A 71(2/3), 155–159.
R¨ ossler, O. E. (1983a). The chaotic hierarchy. Z. Naturforsch. 389, 788–801.
R¨ ossler, O. E. (1983b). Macroscopic behavior in a simple chaotic Hamiltonian sys
tem. In Dynamical systems and chaos, Volume 179 of Lecture notes in physics,
pp. 67–78.
R¨ ossler, O. E. and G. C. Hartmann (1995). Attractors with ﬂares. Fractals 3(2),
285–296.
R¨ ossler, O. E., J. L. Helson, M. Klein, and C. Mira (1990). Self similar basin in
continuous system. In W. Schiehlen (Ed.), Nonlinear Dynamics in Engineering
Systems, pp. 265–273. Berlin: SpringerVerlag.
R¨ ossler, O. E. and J. L. Hudson (1989). Selfsimilarity in hyperchaotic data. In
E. Basar and T. H. Bullock (Eds.), Brain Dynamics, Volume 2 of Springer Series
in Brain Dynamics, pp. 113–121. Berlin: SpringerVerlag.
R¨ ossler, O. E., J. L. Hudson, and M. Klein (1989). Chaotic forcing generates wrin
kled boundary. J. Phys. Chem. 93, 2858–2860.
R¨ ossler, O. E., J. L. Hudson, M. Klein, and R. Wais (1988). Selfsimilar basin
boundary in an invertible system (foldedtowel map). In J. A. S. Kelso, A. J.
336 Chaotic Modelling and Simulation
Mandell, and M. F. Schlesinger (Eds.), Dynamic patterns in complex systems, pp.
209–218. Singapore: World Scientiﬁc.
R¨ ossler, O. E. and C. Kahlert (1979). Winfree meandering in a twodimensional
twovariable excitable medium. Z. Naturforsch. 34, 565–570.
Roux, J.C. (1983). Experimental studies of bifurcations leading to chaos in the
BelousovZhabotinsky reaction. Physica D 7, 57.
Rowlands, G. (1983). An approximate analytic solution of the Lorenz equations. J.
Phys. 16, 585.
Rowlands, G. and J. C. Sprott (1992). Extraction of dynamical equations from
chaotic data. Physica D 58, 251–259.
Ruelle, D. (1973). Some comments on chemical oscillations. Trans. N. Y. Acad. Sci.
( Ser. II) 35(1), 66–71.
Ruelle, D. (1978). Sensitive dependence on initial condition and turbulence behavior
of dynamical systems. Ann. N. Y. Acad. Sci. 316, 408–416.
Ruelle, D. (1979a). Ergodic theory of diﬀerentiable dynamical systems. Publ. Phys.
Math. IHES 50, 275–306.
Ruelle, D. (1979b). Microscopic ﬂuctuations and turbulence. Phys. Lett. A 72(2),
81–82.
Ruelle, D. (1981). Diﬀerentiable dynamical systems and the problem of turbulence.
Bull. Amer. Math. Soc. 5(2942).
Ruelle, D. (1983). Five turbulent problems. Physica D 7, 40.
Ruelle, D. (1985). Rotation numbers for diﬀeomorphisms and ﬂows. Ann. Inst. H.
Poincar´ e 42, 109–115.
Ruelle, D. (1986a). Locating resonances for axiom A dynamical systems. J. Stat.
Phys. 44, 281.
Ruelle, D. (1986b). Resonances of chaotic dynamical systems. Phys. Rev. Lett. 56,
405–407.
Ruelle, D. (1987). Diagnosis of dynamical systems with ﬂuctuating parameters. In
Proc. Roy. Soc. Ser. A, Volume 413, Princeton, NJ, pp. 5–8.
Ruelle, D. (1989a). Chaotic evolution and strange attractors: The statistical analysis
of time series for deterministic nonlinear systems. Cambridge: University Press.
Ruelle, D. (1989b). Elements of the diﬀerentiable dynamics and bifurcation theory.
Boston: Academic Press.
Ruelle, D. (Ed.) (1995). Turbulence, strange attractors, and chaos, Volume 16 of
World Scientiﬁc Series on Nonlinear Science, Ser. A. Singapore: World Scientiﬁc.
References 337
Ruelle, D. and F. Takens (1971). On the nature of turbulence. Comm. Math. Phys. 20,
167–192.
Runge, C. (1895).
¨
Uber die numerische auﬂ¨ osung von diﬀerentialgleichungen. Math.
Ann. 46, 167–178.
Sakaguchi, H. and K. Tomita (1987). Bifurcations of the coupled logistic map. Prog.
Theor. Phys. 78(2), 305–315.
Sarkovskii, A. N. (1964). Coexistence of cycles of a continuous map of a line into
itself. Ukr. Mat. Z. 16, 61–71.
Sauer, T. and J. A. Yorke (1991). Rigorous veriﬁcation of trajectories for the com
puter simulation of dynamical systems. Nonlinearity 4, 961–979.
Schmelcher, P. and F. K. Diakonos (1997). Detecting unstable periodic orbits of
chaotic dynamical systems. Phys. Rev. Lett. 78(25), 4733–4736.
Schuster and H. G. (Eds.) (1998). Handbook of chaos control. WileyVCh.
Schuster, H. G. (1988). Deterministic chaos: An introduction. Weinheim: VCH.
Schwartz, I. B. and I. T. Georgiou (1998). Instant chaos and hysteresis in coupled
linearnonlinear oscillators. Phys. Lett. A 242(6), 307–312.
Scovel, C., I. G. Kevrekidis, and B. Nicolaenco (1988). Scaling laws and the pre
diction of bifurcations in systems modeling pattern formation. Phys. Lett. A 130,
73–80.
Sellwood, J. A. (1983). Quiet starts for galaxy simulations. J. Comp. Phy. 50(3),
337–359.
Sellwood, J. A. (Ed.) (1989). Dynamics of astrophysical discs. Cambridge, UK:
Cambridge University Press.
Sellwood, J. A. and A. Wilkinson (1993). Dynamics of barred galaxies. Reports on
Progress in Physics 56(2).
Shannon, C. E. and W. Weaver (1949). The mathematical theory of information.
Urbana, IL: University Press.
Sharif, M. N. and C. Kabir (1976). Ageneralized model for forecasting technological
substitution. Technol. Forecast. Social Change 8, 353–364.
Sharkovsky, A. N. (1994). Ideal turbulence in an idealized timedelayed Chua’s
circuit. Int. J. of Bifurcation and Chaos 4(2), 303–309.
Sharkovsky, A. N. (1995). Coexistence of cycles of a contiuous map of the line into
itself. Int. J. of Bifurcation and Chaos 5(5), 1263–1273.
Sharkovsky, A. N., P. Deregel, and L. O. Chua (1995). Dry turbulence and period
adding phenomena from a 1D map with time delay. Int. J. of Bifurcation and
Chaos 5(5), 1283–1302.
338 Chaotic Modelling and Simulation
Shil’nikov, A. (1991). Bifurcation and chaos in the moriokashimizu system. Selecta
Mathematica Sovietica 10, 105–117.
Shil’nikov, A. (1993). On bifurcations of the Lorenz attractor in the shimizumorioka
model. Physica D 62, 338–346.
Shilnikov, A. L. and L. P. Shilnikov (1991). On the nonsymmetrical Lorenz model.
Int. J. of Bifurcation and Chaos 1(4), 773–776.
Shilnikov, A. L., L. P. Shilnikov, and D. V. Turaev (1993). Normal forms and Lorenz
attractors. Int. J. of Bifurcation and Chaos 3(3), 1123–1139.
Shilnikov, L. P. (1965). A case of the existence of a countable number of periodic
motions. Sov. Math. Dokl. 6, 163–166.
Shilnikov, L. P. (1967). On the Poincar´ eBirkhoﬀ problem. Matem. Sbornik 3, 353–
371.
Shilnikov, L. P. (1984). Bifurcation theory and turbulence, Volume 2. Harward
Academic Publishers.
Sirovich, L. and P. K. Newton (1986). Periodic solutions of the GinzburgLandau
equation. Physica D 21, 115.
Skiadas, C. H. (1985). Two generalized rational models for forecasting innovation
diﬀusion. Technol. Forecast. Social Change 27, 39–61.
Skiadas, C. H. (1986). Innovation diﬀusion models expressing asymmetry and/or
positively or negatively inﬂuencing forces. Technol. Forecast. Social Change 30,
313–330.
Skiadas, C. H. (1987). Two simple models for early and middle stage prediction of
innovation diﬀusion. IEEE Trans. Eng. Manag. 34, 79–84.
Skiadas, C. H. (1994). Methods of growth functions formulation. In R. Guti´ errez and
M. Valderrama (Eds.), Selected topics on stochastic modelling. Singapore: World
Scientiﬁc.
Skiadas, C. H. (1995). A Lagrangian approach for the selection of growth functions
in forecasting. In J. Janssen, C. H. Skiadas, and C. Zopounidis (Eds.), Advances
in stochastic modelling and data analysis. Dordrecht: Kluwer Acad. Publishers.
Skiadas, C. H. (2007). Recent advances in stochastic modeling and data analysis,
Chapter Exploring and simulating chaotic advection: A diﬀerence equations ap
proach. Singapore: World Scientiﬁc.
Skiadas, C. H. and A. N. Giovanis (1997). A stochastic Bass innovation diﬀusion
model for studying the electricity consumption in Greece. Appl. Stoch. Mod. Dat.
Analys. 13, 85–101.
Skiadas, C. H., A. N. Giovanis, and J. Dimoticalis (1993). A sigmoid stochastic
growth model derived from the revised exponential. In J. Janssen and C. H. Ski
References 339
adas (Eds.), Applied stochastic models and data analysis, Volume II. Singapore:
World Scientiﬁc.
Skiadas, C. H., A. N. Giovanis, and J. Dimoticalis (1994). Investigation of stochastic
diﬀerential models: the Gompertzian case. In R. Guti´ errez and M. Valderrama
(Eds.), Selected topics on stochastic modelling. Singapore: World Scientiﬁc.
Skiadas, C. H., C. Zopounidis, and J. Dimoticalis (1993). Exploring the ﬁtting and
forecasting performance of sigmoid forecasting models: A comparative study in
the Greek economic data series. In J. Janssen and C. H. Skiadas (Eds.), Applied
stochastic models and data analysis, Volume II. Singapore: World Scientiﬁc.
Smale, S. (1967). Diﬀerentiable dynamical systems. Bull. Am. Math. Soc. 73, 747–
817.
Smale, S. (1976). On the diﬀerential equations of species in competition. J. Mathem.
Biology 3, 5–7.
Smale, S. (1977). Dynamical systems and turbulence. In Chorin, Marsden, and
Smale (Eds.), Turbulence seminar, Volume 615, pp. 71–82. Berlin: Springer
Verlag.
Smale, S. (1980). The mathematics of time; Essays on dynamical systems, economic
processes, and related topics. New York: SpringerVerlag.
Smale, S. and R. F. Williams (1976). The qualitative analysis of a diﬀerence equation
of population growth. J. Mathem. Biology 3, 1–4.
Sondhauss, C. (1850).
¨
Uber die schallschwingungen der luft in erhitzten glasr¨ ohren
und in gedeckten pfeifen von ungleicher weite. A. Phys. Chem. 79, 1–34.
Sonis, M. (1996). Once more on H´ enon map: Analysis of bifurcations. Chaos
Solitons Fractals 7(12), 2215–2234.
Sophianopoulos, D. S., A. N. Kounadis, and A. F. Vakakis (2002). Complex dynam
ics of perfect discrete systems under partial follower forces. Int. J. NonLinear
Mech. 37(7), 1121–1138.
Spano, M. L. and W. L. Ditto (1994). Controlling chaos. AIP Conf. Proc. 296,
137–156.
Sparke and J. S. Gallagher (2000). Galaxies in the universe. Cambridge, UK: Cam
bridge University Press.
Sprott, J. C. (1994a). Predicting the dimension of strange attractors. Phys. Lett.
A 192(56), 355–360.
Sprott, J. C. (1994b). Some simple chaotic ﬂows. Phys. Rev. E 50(2), 647–650.
Sprott, J. C. (1997). Simplest dissipative chaotic ﬂow. Phys. Lett. A 228, 271–274.
Stewart, I. (1989). Does God play dice? The mathematics of chaos. Oxford: Black
340 Chaotic Modelling and Simulation
well.
Strogatz, S. H. (1994). Nonlinear dynamics and chaos. Redwood City: Addison
Wesley.
Swinney, H. L. (1978). Hydrodynamic instabilities and the transition to turbulence.
Theor. Phys. Supp. 64, 164.
Swinney, H. L. and J. P. Gollub (1978). Transition to turbulence. Phys. Today 31,
41.
Tata, F. and J. C. Vassilicos (1991). Is there chaos in economic time series? A
study of the stock and foreign exchange markets. discussion paper 120, Financial
markets group, London School of Economics.
Taylor, M. A. and I. G. Kevrekidis (1991). Some common dynamic features of
coupled reacting systems. Physica D 51, 274–292.
Thompson, J. and H. Stewart (1987). Nonlinear dynamics and chaos. New York:
Wiley.
Thompson, J. M. T. and S. R. Bishop (Eds.) (1994). Nonlinearity and chaos in
engineering dynamics. Chichester: Wiley.
Tjahjadi, M. and J. M. Ottino (1991). Stretching and breakup of droplets in chaotic
ﬂows. J. Fluid Mech. 232, 191–219.
Tomita, K. and H. Daido (1980). Possibiliy of chaotic behaviour and multibasins in
forced glycolytic oscillations. Phys. Lett. A 79, 133.
Tomita, K. and D. K. (1979). Thermal ﬂuctuation of a selfoscillating reaction system
entrained by a periodic external force. Prog. Theor. Phys. 61, 825.
Toomre, A. (1963). On the distribution of matter within high ﬂattened galaxies.
Astrophys. J. 138, 385–392.
Toomre, A. (1964). On the gravitational stability of a disk of stars. Astrophys. J. 139,
1217–1238.
Tsiganis, K. and H. Varvoglis (2000). Chaotic evolution of (719) Albert, the recently
recovered minor planet. Astron. Astrophys. 361(2), 766–769.
Tsonis, A. A. (1988). On the dimension of the weather attractor. (Special issue).
Ann. Geophys., 208.
Tsonis, A. A. (1992). Chaos: From theory to applications. New York: Plenum.
Tsonis, A. A. (1996). Dynamical systems as models for physical processes. Com
plexity 1(5), 23–33.
Tsonis, A. A. (2001a). The impact of nonlinear dynamics in the atmospheric sci
ences. Int. J. Bifurcation Chaos 11(4), 881–902.
References 341
Tsonis, A. A. (2001b). Probing the linearity and nonlinearity in the transitions of the
atmospheric circulation. Nonlinear Process Geophys. 8(6), 341–345.
Tsonis, A. A. and J. B. Elsner (1988). The weather attractor on very short time
scales. Nature 333, 545–547.
Tsonis, A. A. and J. B. Elsner (1989). Chaos, strange attractors, and weather. Bull.
Amer. Meteor. Soc. 70, 14.
Tsonis, A. A. and J. B. Elsner (1990). Multiple attractors, fractal basins and longterm
climate dynamics. Contrib. Atmos. Phys. 63, 171.
Tsonis, A. A. and J. B. Elsner (1992a). Estimating the dimension of weather and
climate attractors: Important issues about the procedure and interpretation. J.
Atmos. Sci. 50, 2549–2555.
Tsonis, A. A. and J. B. Elsner (1992b). Nonlinear prediction as a way of distinguish
ing chaos from random fractal sequences. Nature 359, 217–220.
Tucker, W. (1999). The Lorenz attractor exists. C. R. Acad. Sci. Ser. I Math. 328(12),
1197–1202.
Turing, A. M. (1952). The chemical basis of morphogenesis. Phil. Trans. Roy.
Soc. 237, 5.
Tyson, J. J. (1984). Relaxation oscillations in the revised oregonator. Journ. Chem.
Phys. 80, 6079–6082.
Tzafestas, S. and E. Tzafestas (2001). Computational intelligence techniques for
shortterm electric load forecasting. J. Intell. Robot. Syst. 31(13), 7–68.
Ueda, Y. (1979). Randomly transitional phenomena in the system governed by Duﬀ
ing’s equation. J. Stat. Phys. 20, 181–190.
Ueda, Y. (1980). Explosion of strange attractors exhibited by Duﬃng‘s equation,
Volume 357 of Ann. N. Y. Acad. Sci. New York: The New York Academy of
Sciences.
Ueda, Y. (1985). Random phenomena resulting from nonlinearity in the system
described by Duﬃng’s equation. Int. J. Nonlinear Mech. 20(56), 481–491.
Ueda, Y. (1992a). The road to chaos. Santa Cruz, CA: Aerial Press.
Ueda, Y. (1992b). Strange attractors and the origin of chaos. Nonlinear Science
Today 2, 1–16.
Ueda, Y., C. Hayashi, and N. Akamatsu (1973). Computer simulation of nonlin
ear ordinary diﬀerential equations and nonperiodic oscillations. Electronics and
Commun. in Japan 56, 27.
Ueda, Y., H. Nakajima, T. Hikihara, and H. B. Stewart (1988). Forced twowell
potential Duﬃng’s oscillator. In F. M. A. Salam and M. L. Levi (Eds.), Dynamical
342 Chaotic Modelling and Simulation
systems approaches to nonlinear problems in systems and circuits, pp. 128–137.
Philadelphia, PA.: Soc. for Industrial and Appl. Math. (SIAM).
Ueda, Y., H. Ohta, and H. B. Stewart (1994). Bifurcation in a system described by a
nonlinear diﬀerential equation with delay. Chaos 4(1), 75–83.
Uhlenbeck, G. E. and L. S. Ornstein (1930). On the theory of the Brownian motion.
Phys. Rev. 36(3), 823–841.
Ulam, S. M. and J. von Neumann (1947). On combinations of stochastic and deter
ministic processes. Bull. Am. Math. Soc. 53, 1120.
Ushiki, S. (1982). Central diﬀerence scheme and chaos. Physica D 4, 407.
Ushiki, S. and R. Lozi (1987). Conﬁnor and anticonﬁnor in constrained “Lorenz”
system. Japan. J. Appl. Math. 4, 433–454.
Utida, S. (1957). Population ﬂuctuation, an experiment and theoretical approach.
Cold Spring Harb. Symp. Quant. Biol. 22, 139.
van der Pol, B. (1927). Forced oscillations in a circuit with nonlinear resistance.
Phil. Mag. Ser. 7 3, 65–80.
van der Pol, B. and J. van der Mark (1927). Frequency demultiplication. Nature 120,
363–364.
Vayenas, D. V. and S. Pavlou (1999). Chaotic dynamics of a food web in a chemostat.
Math. Biosci. 162(12), 69–84.
Vayenas, D. V. and S. Pavlou (2001). Chaotic dynamics of a microbial system of
coupled food chains. Ecol. Model. 136(23), 285–295.
Verhulst, F. (1990). Nonlinear diﬀerential equations and dynamical systems. New
York: SpringerVerlag.
Verhulst, P. F. (1845). Recherches math´ ematiques sur la loi d’acroissement de la
population. Mem. Acad. Roy. Belg. 18.
Voglis, N. and G. J. Contopoulos (1994). Invariant spectra of orbits in dynamical
systems. J. Phys. A 27, 4899–4909.
Volterra, V. (1959). Theory of functionals and of integral and integrodiﬀerential
equations. New York: Dover.
Voyatzis, G. and S. Ichtiaroglou (1999). Degenrate bifurcations of resonant tori in
Hamiltonian systems. Int. J. of Bifurcation and Chaos 9(5), 849–863.
Vozikis, C. L. (2001). The transition phase of the deviation vector of nearby orbits.
J. Phys. AMath. Gen. 34(7), 1513–1527.
Vrahatis, M. N. (1995). An eﬃcient method for locating and computing periodic
orbits of nonlinear mappings. J. Comput. Phys. 119(1), 105–119.
References 343
Vrahatis, M. N., T. C. Bountis, and M. Kollmann (1996). Periodic orbits and in
variant surfaces of 4D nonlinear mappings. Int. J. of Bifurcation and Chaos 6(8),
1425–1437.
Walker, G. (1931). On periodicity in series of related terms. In Proc. Roy. Soc.
London A, Volume 131, pp. 518–532.
Walker, G. and E. W. Bliss (1937). World weather IV. Mem. Roy. Met. Soc. 4,
119–139.
Wang, M. C. and G. E. Uhlenbeck (1945). On the theory of the Brownian motion II.
Rev. of Mod. Phys. 17(2and3), 323–342.
Wegmann, K. and O. E. R¨ ossler (1978). Diﬀerent kinds of chaotic oscillations in the
BelousovZhabotinskii reaction. Z. Naturforsch. 33, 1179.
Weierstrass, F. (1872).
¨
Uber kontinuierliche funktionen eines reellen arguments, die
f¨ ur keinen wert des letzteren einen bestimmten diﬀerential quotienten besitzen. In
Mathematische Werke II, pp. 71–74.
West, R. W. (1924). The action between bromine and malonic acid in aqueous solu
tion. J. Chem. Soc. 125, 1277–1282.
White, R. B., S. Benkadda, S. Kassibrakis, and G. M. Zaslavsky (1998). Near thresh
old anomalous transport in the standard map. Chaos 8, 757–767.
Wiener, N. (1930). Generalized harmonic analysis. Acta Math. 55, 117–258.
Wiener, N. (1938). The homogeneous chaos. Amer. J. Math. 60, 897.
Wiener, N. (1949). The extrapolation, interpolation and smoothing of stationary
time series. New York: Wiley.
Wiener, N. (1958). Nonlinear problems in random theory. New York: Wiley.
Williams, R. F. (1979a). The bifurcation space of the Lorenz attractor, Volume 316
of Ann. N. Y. Acad. Sci. New York: The New York Academy of Sciences.
Williams, R. F. (1979b). The structure of Lorenz attractors. Publ. Math. IHES 50,
321.
Winfree, A. T. (1972). Spiral waves of chemical activity. Science 175, 634–636.
Winfree, A. T. (1973). Scrollshaped waves of chemical activity in three dimensions.
Science 181, 937–939.
Winfree, A. T. (1974a). Rotating chemical reaction. Sci. Amer. 230, 82–95.
Winfree, A. T. (1974b). Rotating solutions to reaction/diﬀusion equations in simply
connected media. In D. S. Cohen (Ed.), Mathematical aspects of chemical and bio
chemical problems and quantum chemistry, Volume 8, pp. 13–31. SIAM  AMS.
Winfree, A. T. (1978). Stable rotating patterns of reaction and diﬀusion. Theoret.
344 Chaotic Modelling and Simulation
Chem. 4, 1–51.
Wisdom, J. (1987a). Chaotic behavior in the solar system. In M. V. Berry, I. C.
Percival, and N. O. Weiss (Eds.), Dynamical Chaos, Proc. of a Roy. Soc. Disc.
Meeting, Feb. 1987, pp. 109–129. Princeton, NJ: University Press.
Wisdom, J. (1987b). Chaotic dynamics in the solar system. Icarus 72, 241.
Yamada, T. and Y. Kuramoto (1976). A reduced model showing chemical turbulence.
Prog. Theor. Phys. 56, 681.
Yamaguti, M. and S. Ushiki (1981). Chaos in numerical analyses of ordinary diﬀer
ential equations. Physica D 3, 618–626.
Yannacopoulos, A. N., J. Brindley, J. H. Merkin, and M. J. Pilling (1996). Approxi
mation of attractors using Chebyshev polynomials. Physica D 99(23), 162–174.
Yannacopoulos, A. N., I. Mezic, G. Rowlands, and G. P. King (1998). Eulerian
diagnosis for Lagrangian chaos in threedimensional NavierStokes ﬂows. Phys.
Rev. E 57(1), 482–490.
Yorke, J. A., E. D. Yorke, and J. MalletParet (1987). Lorenzlike chaos in a partial
diﬀerential equation for a heated ﬂuid loop. Physica D 24, 279.
Yoshida, H., A. Ramani, and B. Grammaticos (1987). Painlev´ eresonances vs.
Kowalevski exponents: exact results on singularity structure and integrability of
dynamical systems. Acta Appl. Math. 8, 75.
Yuan, G. C. and J. A. Yorke (2000). Collapsing of chaos in one dimensional maps.
Physica D 136(12), 18–30.
Zaslavsky, G. M. (1978). The simplest case of a strange attractor. Phys. Lett. A 69,
145–147.
Zaslavsky, G. M. (1993). Selfsimilar transport in complete chaos. Phys. Rev. E 48,
1683.
Zaslavsky, G. M. (1999). Chaotic dynamics and the origin of statistical laws. Physics
Today 52(8), 39–45.
Zaslavsky, G. M. and B. V. Chirikov (1972). Stochastic instabillities in nonlinear
oscillations. Sov. Phys. Usp. 14, 549.
Zhabotinskii, A. M. (1964). Periodic course of oxidation of malonic acid in solution
(investigation of the kinetics of the reaction of Belousov). Biophysics 9, 329–335.
Zhabotinskii, A. M., A. N. Zaikin, M. D. Korzukhin, and G. P. Kreitser (1971). Math
ematical model of a selfoscillating chemical reaction (oxidation of bromomalonic
acid with bromate catalyzed by cerium ions). Kinetics and Catalysis 12, 516–521.
Chaotic Modelling and Simulation
Analysis of Chaotic Models, Attractors and Forms
Christos H. Skiadas Charilaos Skiadas
Chapman & Hall/CRC Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 334872742 © 2009 by Taylor & Francis Group, LLC Chapman & Hall/CRC is an imprint of Taylor & Francis Group, an Informa business No claim to original U.S. Government works Printed in the United States of America on acidfree paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number13: 9781420079005 (Hardcover) This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot assume responsibility for the validity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint. Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 9787508400. CCC is a notforprofit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress CataloginginPublication Data Skiadas, Christos H. Chaotic modelling and simulation: analysis of chaotic models, attractors and forms / Christos H. Skiadas, Charilaos Skiadas. p. cm. Includes bibliographical references and index. ISBN 9781420079005 (hardcover : alk. paper) 1. Chaotic behavior in systems. 2. Mathematical models. I. Skiadas, Charilaos. II. Title. Q172.5.C45S53 2009 003’.857dc22 Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com and the CRC Press Web site at http://www.crcpress.com 2008038183
Preface
“Is there a chaotic world? To what extent? Mathematics, geometry or simply chaos?”
The above questions arise time and time again when dealing with chaotic phenomena and the related attractors and chaotic forms. During the lengthy preparation of this manuscript, computers were used in the same way that geometers of old used geometric tools. Simple or complicated ideas are designed, estimated and tested using this new tool. The result is more than 500 graphs and illustrations that ﬁll the pages in front of you. Verbal explanations are kept to a minimum, and a lot of eﬀort is devoted to ensuring that the presentation of ideas progresses from the most elementary to the most advanced, in a clear and comprehensible way, while requiring little previous knowledge of mathematics. In this way, we hope that a more general audience will beneﬁt from the material of this book, especially from the large variety of chaotic attractors included. A lot of new material is presented alongside the classical forms and attractors appearing on the literature on chaos. However, most of the illustrations are based on new simulation methods and techniques. A rather lengthy introduction provides the reader with an overview of the subsequent chapters, and the interconnection between them. Our inspirations came from various disciplines, including geometry from the ancient Greek and Alexandrian period (a great deal of which is taught in the Greek school system), mathematics from the developments of the last centuries, astronomy and astrophysics from recent developments (the Conference on Chaos in Astronomy organized by George Contopoulos in 2002 was very stimulating) and the amazing chaotic illustrations of Hubble and Chandra and the vortex movements from ﬂuid ﬂow theory. Chaotic theory is developing in a new way that inﬂuences the world around us, and consequently also inﬂuences our ways of approaching, analyzing and solving problems. It is not surprising that one of the central models in the chaos literature, the H´ nonHeiles model, is presented in a paper with the title “The applicability of e the third integral of motion: Some numerical experiments.” Numerical experiments in 1964 were the basis for many signiﬁcant changes in astronomy in the decades that followed. In 1963 Edwin Lorenz, in his pioneering work on “Deterministic Nonperiodic Flow,” proposed a more prominent title for chaotic modelling, by including the term “deterministic.” His work spearheaded numerous studies on chaotic phe
nomena. Fifteen years later, Feigenbaum explored the rich chaotic properties of the simple discrete Logistic model, known since 1845 following the work of Verhulst on the continuous version of this model, and introduced scientists to the chaotic ﬁeld. Our intention in this work was to present the main models developed by the pioneers of chaos theory, along with new extensions and variations to these models. The plethora of new models that can arise from the existing ones through analysis and simulation was surprising. This book is suitable for a wide range of readers, including systems analysts, mathematicians, astronomers, engineers and people in any ﬁeld of science and technology whose work involves modelling of systems. It is our hope that this book will prompt more people to become involved in the rapidly advancing ﬁeld of chaotic models and will inspire new developments in the ﬁeld. Christos H. Skiadas Charilaos Skiadas
List of Figures
1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 1.10 1.11 1.12 1.13 1.14 1.15 1.16 1.17 1.18 1.19 1.20 1.21 1.22 1.23 1.24 1.25 1.26 1.27
The (x, y) diagram . . . . . . . . . . . . . . . . . . . . Continuous and discrete R¨ ssler systems . . . . . . . . o Bifurcations of the logistic and the Gaussian . . . . . . Delay models . . . . . . . . . . . . . . . . . . . . . . The H´ non and Holmes attractors . . . . . . . . . . . . e The Lorenz and autocatalytic attractors . . . . . . . . . A chemical reaction attractor . . . . . . . . . . . . . . Nonchaotic portraits . . . . . . . . . . . . . . . . . . A rotation chaotic map . . . . . . . . . . . . . . . . . . The Ikeda attractor . . . . . . . . . . . . . . . . . . . . Ikeda attractors . . . . . . . . . . . . . . . . . . . . . . Translationreﬂection attractors . . . . . . . . . . . . . Symmetric forms . . . . . . . . . . . . . . . . . . . . Separation of chaotic forms . . . . . . . . . . . . . . . Flowers . . . . . . . . . . . . . . . . . . . . . . . . . . Orbits in twobody motion . . . . . . . . . . . . . . . . Motion in the plane . . . . . . . . . . . . . . . . . . . Motion in space . . . . . . . . . . . . . . . . . . . . . Ring systems . . . . . . . . . . . . . . . . . . . . . . . Spiral galaxies . . . . . . . . . . . . . . . . . . . . . . Spiral galaxies . . . . . . . . . . . . . . . . . . . . . . Reﬂection forms . . . . . . . . . . . . . . . . . . . . . Spiral Patterns . . . . . . . . . . . . . . . . . . . . . . H´ nonHeiles systems . . . . . . . . . . . . . . . . . . e A twoarmed spiral galaxy . . . . . . . . . . . . . . . . Chaotic paths in the H´ nonHeiles system (E = 1/6) . . e The Contopoulos system: orbits and rotation forms at ratio 4/1 (left) and 2/3 (right) . . . . . . . . . . . . . . The solution to the logistic equation . The two Gompertz models . . . . . . Chaotic forms . . . . . . . . . . . . A snail’s pattern . . . . . . . . . . . Chaos as randomness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3 4 5 6 7 8 9 9 10 11 12 13 14 15 16 17 17 18 19 20 21 22 23 24 25 26 27 35 37 41 42 43 48
2.1 2.2 2.3 2.4 2.5 3.1
Tracing iterations . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 3.20 3.21 3.22 3.23 3.24 3.25 3.26 3.27 3.28 3.29 3.30 3.31 3.32 3.33 3.34 3.35
Second order stationary points . . . . . . . . . . . . . . . . . . . . The logistic and the inverse logistic map for b = 3.5 . . . . . . . . The third order cycle of the logistic and the inverse logistic map for √ b=1+ 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chaotic region for the logistic and the inverse logistic map at b = 3.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The logistic map . . . . . . . . . . . . . . . . . . . . . . . . . . . The logistic map, b ≤ 3 . . . . . . . . . . . . . . . . . . . . . . . The logistic map, b > 3 . . . . . . . . . . . . . . . . . . . . . . . Order4 bifurcation of the logistic map; real time and phase space diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Order8 bifurcation of the logistic map; real time and phase space diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Order16 bifurcation and real time and phase space diagrams of the logistic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Order32 bifurcation and real time and phase space diagrams of the logistic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The order5 bifurcation and real time and phase space diagrams . . The order6 bifurcation and real time and phase space diagrams . . The bifurcation diagram of the logistic . . . . . . . . . . . . . . . Order3 bifurcations . . . . . . . . . . . . . . . . . . . . . . . . . The order3 bifurcation . . . . . . . . . . . . . . . . . . . . . . . The May model: an order4 bifurcation (left) and the bifurcation diagram (right) . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bifurcation diagrams of the Gaussian model . . . . . . . . . . . . Bifurcation diagram of the Gaussian model (a = 17) . . . . . . . . Bifurcation diagram of the GL model . . . . . . . . . . . . . . . Cycles in the GL model . . . . . . . . . . . . . . . . . . . . . . . Patent applications in the United States . . . . . . . . . . . . . . . The GRM1 model . . . . . . . . . . . . . . . . . . . . . . . . . . Spain, oxygen steel process (1968–1980) . . . . . . . . . . . . . . Italy oxygen steel process (1970–1980) . . . . . . . . . . . . . . . Luxemburg oxygen steel process (1962–1980) . . . . . . . . . . . Bulgaria oxygen steel process (1968–1978) . . . . . . . . . . . . . The modiﬁed logistic model (ML) . . . . . . . . . . . . . . . . . The modiﬁed logistic model (ML) for b = 1.3 and e = 0.076 . . . . A chaotic stage of the modiﬁed logistic model (ML) for b = 1.3 and e = 0.073 . . . . . . . . . . . . . . . . . . . . . . . . . . . . A third order cycle of the GRM1 model for b = 1.3 and σ = 0.025 A third order cycle of the GRM1 model for b = 1.2 and σ = 0.01175 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A third order cycle of the modiﬁed logistic model for b = 1.3 and e = 0.0719 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The ﬁrst bifurcation cycle of the GRM1 model for b = 1.3 and σ = 0.032 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
49 50 50 51 53 54 54 55 56 56 56 57 57 59 60 61 62 63 63 63 64 65 67 69 70 71 72 73 73 73 74 75 75 75
Chaotic attractors of a delay logistic model . . . . . .5 4. . . . . .4.11 5. . 121 . A complicated delay model . . . . . . . The delay logistic model . . . . . The sine delay model for large b . . . . . . . . e Orbits in the cosine H´ non model . . . . . . . . . . . . . . . . .20 6. . . . . . . . . . . Irregular patterns .9 4. . . . . . . . . . . . . . . . . . . .18 5. . . . . . . . Comparing the Glass (A) and the logistic (B) delay models (m = 40) The delay exponential model . .17 5. . . . . . . . . . . . . . . . . . . . .4 4. . . . . . . . . . . . . . . . . . .2 5. . . . . . . . . . . . . .3 A simple delay oscillation scheme (a = −1) . . . . . . . . . . . . A secondorder delay model . . . . . . . . Carpetlike forms . . .15 5. . . . . . . . .765. The order2 delay model . . . . . . . . . . . . . . . . . . .7 5. . . . . . e Chaotic patterns of the delay cosine model . . . . . . . . . . . . . . A simple delay model (n = 15) . . . . . . . . . . . . . . . 80 80 81 82 83 85 85 87 88 89 90 91 92 93 94 94 95 96 96 97 100 102 103 104 105 106 106 106 107 108 108 109 110 110 111 112 112 113 114 114 The Arneodo model . . . . .10 5. . . . . . . . . . . . . Triangular islands . . . . . . .20 5. 120 A modiﬁed Arneodo model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The sine delay model (a = 0. b = −0. 120 An autocatalytic attractor . . . . . . . . .1) . . . . . . . . . . . . . .1 4. . . . . . . . . . . . . . . . . The Holmes model (a = 2. . . . .19 4. . . .16 4. . . . . . . . . . . . . . .5 5. . . .10 4. . . . . . Oscillations and limit cycles in delay logistic models . . . . . . . . Folds in the delay logistic model . .8 4. . . . e A variation of the H´ non model . . . . .13 4. . . Delay model (b = 2). . . . . . . . e An exponential variant of the H´ non model . . . .8 5. . . . . . . . . . . Delay logistic model .11 4. . . . . .19 5. . . . . . . . . . . . .12 5. . . .3) . . .3) . . .3 5. . . .3 4. . . . . . . . . . .17 4. . . . . . . . . . . . . .1 5. . .6 4. . . . Chaotic attractors of the logistic delay model (m = 40) . . . a square pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 6. . . . . . . . . Delay logistic and delay cosine H´ non variants . . . . . . .2) . . .2 6. . . . . . . . . .2 4. . . . . . . . . . Variation of the simple delay model . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 5. . . . . . . . .14 5. . . . . . . .9 5. . . . . . . . e A thirdorder delay model . . . . . . . . . The Holmes model (a = 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Carpetlike forms . . . . .7 4. . . A threedimensional model . . The sine delay model . . .14 4. . . . . . . . .4. . . . . The H´ non map . . . . . . . . e Bifurcation forms in the cosine H´ non model .18 4. . . . . . .15 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . .16 5. . . . . . . . . . . . . The sine delay model for b = 1 . . . . . . . A H´ non model with a logistic delay term . .12 4. . . . . . . . . . . . . . . . . . . . . . . e Higherorder bifurcations in the cosine H´ non model . .4 5. . . . . . . b = 0. . . . . . . Squarelike forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . e A 2diﬀerence equations system . . . Islands in the chaotic sea . . . . . . . . . . . . . .13 5. . . . . . . . . . . . The sine delay map (b = 0.
. . . . . . . . . . . . . . . . A double eggshaped form with an envelope . . . . .6. . . . . . . . .2 8. . . . . . . . . . . . . . . . .1 8. . .14 6.13 6. . . Rotations . . . . . . y). . . .6 7. . o Plane views of the total spiral chaos in the variant of the R¨ ssler o model (a = 0. . . . . .3 7. . . . . . . . . . The R¨ ssler attractor . . The modiﬁed Lorenz model — left to right: (x. . . . . . o Another view of the modiﬁed R¨ ssler model . . . . . . . . . .8 6. Rotating particles . . . . . . . . . . . . . . . . . . . . . . . . . (y. . The chaotic bulge . . . . . . . . . . . . . . . . The chaotic bulge . . . . .1 9. . . . . . . . . . . . . . . . . o Spiral chaos in the variant of the R¨ ssler model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . .1 7. . . . . 181 . . .6 8. . .2 Threedimensional views of the autocatalytic model . . . . The chaotic bulge . . . . . .7 6. . . . . . . . . . . . . . . . . . More complex forms . . . . . . . . . . . . . . . . .13 8. . . . . . . Symmetry/reﬂection along a line . . . . . . . . . . .4 7. . . . . . .5 8. . Development of rotating particles over time (a = 0. . . . . . . z) and (x. o The Lorenz attractor . z) views .14 8. . . . . . . . .10 8. . . . . . . . . . . . . . . . . . and the chaotic bulge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rotationtranslation. Equilibrium points . . . . .10 6. . . . . Symmetric forms . . . . . . . Modiﬁed R¨ ssler models . . . . . . . . . . . The simplest conservative system . . . . . . . . . . . . . . . . . . . . . . .9 6. . . . . A simple nonsymmetric form . . . . . . . . 180 Planar isometries are determined by their eﬀect on three points . . . The elapsed time is t = 10. . . . . . . . . . . . . . . .8 8. . . . . . . . . . . . . . . . .11 8. . . . . . . . . . . . . . . . . . . . . .4 8. . . . . . . . . . . . . . .12 8. . . . .8) . . . . . . . . . . . .7 8.5 6. . . . . Comparison of the continuous and the discrete models . o b = 0. . . . . . Higherorder forms . . . . . . . .7 7. The chaotic bulge . . . . . . . Qualitative classiﬁcation of equilibrium points Eggshaped forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9 8. . . . . . .6 6. . . . . 122 123 124 125 126 127 128 128 129 131 132 132 137 143 146 147 149 149 150 152 153 159 161 165 166 167 167 168 169 170 172 172 173 174 175 177 The limiting trajectory . . . . . . . . . . . . . . . . . . . . . . . . . .11 6. A symmetric threefold form . . . . .4 6. o Limit cycles in a variant of the R¨ ssler model . . . The characteristic trajectory . . . . . . . . . . .5 7. . .3 8. Development of rotating particles according to the magnitude of the translation parameter. . .8 7. . . . . . . . . . . . . . . . . .12 6. . . . . .60) . . . . . . .15 7. . . . . . . . . . . . . .15 9. . . .2 7. . . . . . . .3 and c = 4.9 8. . . . . . . . . . . . . . . Rotating paths . . . . . . . . . . . . . . . . . A fourdimensional autocatalytic model . . . . . . . . . . . . . Elliptic forms and the chaotic bulge . . The modiﬁed Lorenz chaotic attractor . . . . . . . . . . . . .6). . . . . . . View of a simple cycle of the variant of the R¨ ssler model (a = 0. . . . . . . . . .
. . . . . . . . . . . . The reﬂection Ikeda attractor . . . . . . . . . . . 182 183 185 187 188 189 189 190 190 191 192 193 193 194 195 195 196 198 199 200 201 206 208 210 211 212 212 213 214 215 216 217 217 218 220 221 226 226 227 228 231 232 . . . . . . . . . Two other reﬂection attractors from a quadratic rotation angle . . The composition of two reﬂections . . . . . .5 10. . . . . . . .8 10. . . . . . . . . . . . . . . .3 11. . . . . Spiralling towards the sink .10 9. . . . . .11 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . .17 9. . . . . The Ikeda attractor . . . . . . . . . . . .6 Glide reﬂection seen in two diﬀerent ways . . . . . . . . . . A symmetric graph . . . . . . Ring systems . . . . . The eﬀect of a complicated rotation angle on chaotic attractors . .2 11. . A simple area preserving rotationtranslation model . . . .6 9. . Comparing rotationreﬂection case . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16 9. . . . . . . . . . . . . . . . . . . . . .11 10. . . . A sinusoidal rotation angle chaotic attractor (c = 0. Chaotic images of rotationtranslation maps . . . . . . . . . . . . . . . . . . . The eﬀect of space contraction on chaotic images . . . . . . . . Chaotic forms without space contraction . . . . . . . . . . .15 11. . . Chaotic images with area contraction rotation angles . Chaotic attractor in the twosink problem . . . . Sinusoidal rotation angle chaotic attractors . . . . . . Bifurcation diagrams . . . . . .15 9. . . . . . . Attracting bodies in space and relative movement . . . . . . . . .5 11. . .9. . . .4. . . . . . . .21 9. . . . . . . . . . Geometric analogue of rotationtranslation (θ = 2) . .22 9. . . . . . . . . . . . . . . . Twopiece chaotic attractors . . . . . . . Motions in space . . . . . . . .1 11. . . . . . . . .12 10. Spiral forms . . . . . d = 6) . . Chaotic forms of a simple rotationtranslation model Chaotic forms of a simple rotationtranslation model . .9 10. . . . Higher order vortices . . .10 10. . Circular chaotic forms . . . . . . . . . . . . . . . . . . . . . . .12 9. . . . . . . . . . . . . . . . . .3 9. . . . . . . . . . . . . Chaotic attractors in a noncentral sink . . . . . Elliptic orbits . . . . .4 11. . . . . . . . . . . . . . . . . . .13 10. . . . . . . . . . . Comparing rotationreﬂection cases . . . Two distinct vortex forms (d = 1) . . . . . . . . . . . . . . . The two symmetric sinks model . . . . . . . . . . . . . . . . . . . Reﬂection: a twopiece chaotic attractor . . . . . . . . . . . . . . . The eﬀect of a complicated rotation angle on chaotic attractors . . . . . . .5 9. . . . . A reﬂection chaotic attractor resulting from a quadratic rotation angle . . . . . . . . . . . . . . . . . .9 9. . . . . . . . . . . . . . Geometric analogues of rotationtranslation .14 10. . . . . . The eﬀect of the reﬂection parameter on chaotic images . . . . . . . . . .19 9. . Sinusoidal rotation angle chaotic attractors . . . The chaotic attractor with strong vortex strength parameter d = 2π Vortex forms . . . .4 10. .14 9. . . . . . . . . . . . .7 9. . . . . . .20 9. Space contraction . . . . . . . .13 9. . . . . . .7 10. . .3 10. . . . . . . . . . . . . .4 9. . . . . . . . . . . . . . . . . . . . . . . . . .2 10. . . . . . .6 10. . . .18 9. . . . . . . .23 10. . . . . . . . . . . . . . . . . .1 10. . . . . . . . .8 9. . . .
.15 11. . . .3 12. Two attracting masses and the disk of rotating particles at time t = 0 Rotating disks with two attracting masses . . . . . . . . . . . . . . . . Comparing chaotic forms: rotationreﬂection . . . . . . . . . e Chaotic paths in the H´ nonHeiles system . .18 11. .13 11. . . . . . . . . . . . . . Rotating particles under the inﬂuence of an attracting mass . . . . . . . The eﬀect of two perpendicularly attracting masses . . . .11.7 Chaotic forms of a simple rotationtranslation model .26 11. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rotating disks . . . . . .1 12. . .35 11. . . . . . . .27 11. . . . . Rotating disks with two attracting masses over a long time . . . . . radius r . . . . . . . . . . . . . . . . . . .4 12. . . .25 11. . . . . . .12 11. . .28 11. .32 11. . . . . Twoarmed galaxies .21 11. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ˙ e A discrete analogue to the H´ nonHeiles system (E = 1/12) e The discrete H´ nonHeiles model . .7 11. . . . .2 12. . y) diagrams for the H´ nonHeiles system . . . . . . . Relativistic reﬂectiontranslation . . A rotation image following a ContopoulosBozis paper (approaching galaxies) . . . .38 12. . . . . . .10 11. Relativistic rotationtranslation forms for high relativistic speeds . . e e The H´ nonHeiles system . . . . . . A rotationtranslation chaotic image . Rotation speed vs. . . . . . . . Rotating disc at early times . . A galacticlike cluster . . . . . . . . . . Other relativistic chaotic images . . . . . . . . . .31 11. . . . . . . . . . . .8 11. . . . . . . . . . . . . . . . . . . .34 11. . . . . . . . . . . . An asymptotically stable chaotic attractor .22 11. . . . . . . . . . . . . . . . .9 11. . .29 11. . A relativistic chaotic image when b = 0. . . . . . . . . . . . . . . . . . . . . The original H´ nonHeiles (y. . . . . . Relativistic rotationtranslation forms for medium and high relativistic speeds . . . . . . . . . . . . . . . . . . . . . . . .36 11.17 11. . . . . . . . . . . . . . . . . . . . .14 11. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11 11. . .23 11. . Chaotic forms for low relativistic speeds . . . . Attractors and attracting points in rotationtranslation model . . . . . . . . . . 233 234 235 236 237 239 240 241 242 243 243 244 245 246 246 247 247 249 250 252 252 255 256 256 257 259 259 260 261 262 263 264 268 269 270 271 271 272 273 . . . . A relativistic chaotic image when b = 0. . . Other relativistic chaotic images . . .24 11. . . . . . y) diagram . Two galaxylike chaotic images . . . . . . . . . . . . . . . . . . . . . . . . . Nonrelativistic chaotic images in the early period of a rotationtranslation process with b = 1 . . . . Chaotic attractors in the case of two unequal symmetric attracting masses . . . . . . Various relativistic chaotic forms . . . . .9 . . . . . . . . .33 11. . . . . . . . . . . . . .20 11. e ˙ (y. . . Twoarmed galaxies . . . . . . . . . .19 11. . . Multiarm images . . .6 12. . .5 12.16 11. . . . . . . . . . . . . . . . . . . . . . . . . . . e Discrete analogues to the H´ nonHeiles system . . . . . . . . . . . Boxlike paths of a twoarmed spiral galaxy . . . . . . . Relativistic and initial chaotic images of a rotationtranslation process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37 11. . . . .9 . . . . . . .30 11. .
12. . . The dolphin attractor . . . Orbits and rotation forms at time t = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The eﬀect of noise on threedimensional models . . . . . . . . . .6 13. x) (heavy line) and the (t. . . .11 12. . . . . . y) (light line) graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 13.8 13. . . . The pendulum and the (t. . . . . Logarithmic potential: orbits and rotation forms at time t = 1 . . .9 12. . . . . . . . 274 276 278 280 281 282 286 287 289 290 291 292 293 294 294 13. The LotkaVolterra system . . . . . . . The Contopoulos system: orbits at resonance ratio 4/1 (left) and 2/3 (right) . . . .10 12. . . . . . . . . . . The (y. . . . x) oscillations . . . . . . . . . .13 Another simple Hamiltonian system . . . . . . 13. . . . . . A simple Hamiltonian: orbits and rotation forms at various times . . . . . The ship and signature attractors . . . . . . . .12 Boxlike orbits . . . . . y) curves of a special second order diﬀerential equation sys˙ tem . . . . . . . . . .4 13. . . . . . . . . . . . . . . . The Ushiki and Tinkerbell attractors . . . . . . . . . . . . . . .2 13. . . . . . The LotkaVolterra system: limit cycle (upper image) and (t. . . . . . . . . . .8 12. . . . . . . .5 13. .3 Adding a sinusoidal forcing term to a twodimensional system . . 12. .1 13. . . . .9 . . . . . . . . . .
. . . . .3 Galactictype potentials and the H´ nonHeiles system e 1.5 Odds and Ends. . . . .4 Shape and form . .3 Rotations . . . .2 Delay models . . . . . . . . . . . 2. . . . .2 Conservative systems . 1. . . . . . . . . 1. . . . . . . . . . . e 1. Models and Modelling 2. . . . . . . . . . . . . .4. . . . . . . . . . . 2. . . . . .3. . . . . 2. . .3. 1.3 More Complex Structures . . . . . . 1. . . . . . . . . . . . . . . . . . . . . .3. .4 The Contopoulos system .5 Delay growth functions . . . . 3 . . . . .3 The H´ non model . . . . . 2. . .4 Chaos and the Universe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . .1. . . 3. . . . . . . . . . .3. . . . .2. . . . 1. .2 Empirical model formulation . . 1. . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . .3.1 The Logistic Map .1 Chaos in Diﬀerential Equations Systems . .1 Threedimensional and higherdimensional models . . . . . . . . .1 Series approximation . . . . . . . . . .3 The calculus of variations approach . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . 1.2 Model Construction . . . 3. 1 1 4 5 6 6 8 8 8 10 12 14 14 19 23 25 27 29 29 30 30 32 33 36 38 39 40 41 42 47 47 47 51 58 61 2 .Contents 1 Introduction 1. . . . . . . . . . . . . . . . .1. 1. . Stochastic and Chaotic Models The Logistic Model 3. . . . . . . . . . . . . .4. . . .3 Other Models with Similar Behaviour . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . and Milestones . . . . . . . 1. . . .5 Deterministic. . . 1. 2. . . . . . . . . .2. . . . . . . . . . . . . 1. . . . . . . . . . . . . .1 The logistic map . . . . . . . . . 2. 1. . . . . . . .2 Algebraic analysis of the logistic . . . . . . .1 Geometric analysis of the logistic 3. 3. .4 The probabilisticstochastic approach 2. .4 Chaotic Analysis and Simulation . . .4. . . . . . . . . . . .2 Chaos in galaxies . . . . . . . . . .2 The Bifurcation Diagram . . . .3. . . 2. . . . . . .3 Modelling Techniques . . . . .4. . . . . . . . . . . . . .1 Chaos in the solar system . . .1 Growth/decay models . . . 2. . . . . . . . . .3. . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Introduction . . . .2 Chaos in Diﬀerence Equation Systems . . . . . . . . . . . . . . . . . . .
e 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 The CosineH´ non Model . . . . . . 4. . . . . . . . . 5. . . . . . . . .7. . . . . . . . . .1. . . . . 5. . . . . 4. . . . .4 An Autocatalytic Attractor . . . . 3. . . . . . . . . . .1 Simple delay oscillation scheme 4. . . . . . .2 Secondorder delay models . 4. . . . . .5. . . . . . . 4. .6 The R¨ ssler Model . . . .5. . . . . . . .1 GRM1 and innovation diﬀusion modelling . . . . . . . . . o 6. . . . . . . 5 The H´ non Model e 5. . . . e 5. 4. . . . .2 The delay logistic model . . . . . . . . . . . . .5. . 6. .1 The Holmes model . . . . . . . . . . .6. . . . . . .5. . . . .1 Equilibrium Points and Characteristic Matrices 6. . . . . . . . . . .1 Introduction . . . . . . . . . . . . . . . . . .2 The sine delay model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Global Period Doubling Bifurcations in the H´ non Map e 5. 6. . . . . . .1 A variant of the R¨ ssler model . . .5. . . . . . . . . . . . . . . . . . . . . . . . e 5. . . . . . . . . . . . . . . . . . . .4 A More Complicated Delay Model . . . . . . . . . . . . . . .3 Firstorder delay variants . 6 . . . . . . . . . . . . . . . . .2 The LotkaVolterra Model . . 6. . . 62 64 64 66 68 69 71 79 79 79 79 81 82 84 86 86 89 91 99 99 100 101 101 103 103 104 104 105 107 108 109 110 110 111 117 117 118 119 121 122 123 124 The Delay Logistic Model 4. 5. . . . . . . . . .5 3. . .7 Model Behaviour for Large Delays . . . . . . . . . . . . . . . . .6 4 Models with Diﬀerent Chaotic Behaviour . . . .7. . . . . . . . . . .3 An Example of Bifurcation and Period Doubling . ThreeDimensional and HigherDimensional Models 6. o . . 3. . . . .7 The Holmes and Sine Delay Models . . . . . . . . . . 3. . . . .6 Variants of the H´ non System Equations . . . 4.1 Period doubling bifurcations when b = −1 . .5. . . . . . . . . Further Discussion . . 4. . . 5.5 Variants of the H´ non Delay Diﬀerence Equation . . . . . . . . .2. . . . . . . . . . . . . . . . 3. . .3 Time Delay Diﬀerential Equations . . . . . . . . . . . 5. . .2 The generalized rational model . . . . . . . . . . . . . . . . 5. . . . . . .3 Parameter estimation for the GRM1 model . . . . . . .4 Exponential variants . . . 5. . . . . The GRM1 Chaotic Model . . . . 6. . . . . . . . . . .2. . . .2 Delay Diﬀerence Models . . . . . . . . . . . . . . . . . . . .5 A Delay Diﬀerential Logistic Analogue 4. . . .4 A Diﬀerential Equation Analogue . . . . . . . . . . . . . . . . . 5. . . . . . . . . . . 5. . 5. . . . .3. . . . . . . .4 3. . . . . . . . .8 Another Delay Logistic Model . . . . . . . . . . .5. . . . . . .1 The thirdorder delay model . . . . . . . . . . . . .2 Period doubling bifurcations when b = 1 . . . .6 Other Delay Logistic Models . . . .4 Illustrations . . . . . . . . . . . . . . .5 A FourDimensional Autocatalytic Attractor . . .3 The Arneodo Model . . . . . .5. .1. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .4 A General RotationTranslation Model . . . . . . .2 The eﬀect of space contraction and change of reﬂection angle on translationreﬂection . . . . .7. . . . . .1 Transformations on linear systems . . . . . . . . . 9. . . . . . 7. . .1. . . . . . . . . . . . . . .6 Rotations Following an Inverse Square Law . . . . . . . . . . . . . . . . . .2. . . . .1 The eﬀect of space contraction on rotationtranslation . .2.3 EggShaped Forms . . . . . . . 7. . . . . . . . . . . .4 Symmetric Forms . . . . . . . . . . . . . . . . . . .1 A simple eggshaped form . . . . . . 9. . . .2 A Simple RotationTranslation System of Diﬀerential Equations 8. 8. . 9. . . . .2 Qualitative behaviour at equilibrium points 7. . . .1 Introduction . . . . . 9. . . . . . . . . . . . . . . . . . . .7 6. . . 8 Rotations 8. . .6. . . . . . . . . . . .5 Chaotic Attractors and RotationReﬂection . .5 More Complex Forms . . . . . . . . . . . . . . . . . . . . . . .3. . . .8 Further Analysis . . . . . . . . 8. . . . . . . . . . 7. . . . . . . . . . . 9. . . . . . . . 7. . . . . . . . . . . . . . .1. . . .6. . . . .3 A double eggshaped form with an envelope 7. . . . .1 The modiﬁed Lorenz model . . .1 Space contraction . . . . . . . . . 7. . 7. . . . .2 Linear Systems . . . . . . .1 Conservative Systems . . . . . . . . .1 Twodimensional rotation .3 Complicated rotation angle forms . 9. . . 127 o The Lorenz Model . . . . 9 Shape and Form 9.1 Symmetry and plane isometries . . . . . . . . .3 A Discrete RotationTranslation Model . . . . . .6. . . . . . . . . 9. 7. . . . . . . 9. . . . . . . .7 Chaotic Circular Forms .6.6. . . . . . . . . .6 Experimenting with Rotation and Reﬂection . . . . . . . . 8. . . . . . . . .2. . . . . . . . . . 9. . . . . . . 131 135 135 137 138 139 140 142 144 144 145 146 148 150 153 157 157 158 163 169 171 173 179 179 180 184 184 186 186 187 188 191 191 192 193 194 196 196 200 7 NonChaotic Systems 7. . . . . . . . . . . . . . . . . . . . . . . . 9. . . . . . . . . . . . . . . . . .2 Isometries in Modelling . .6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Reﬂection and Translation . . .4 Application in the Ikeda Attractor . . . 9. . .4 Comparing rotationreﬂection . . . . .1 The simplest conservative system . . . . . . . . . . . . . . . .3.6 HigherOrder Forms . . .5 A simple rotationtranslation model . 8. .2 Introducing rotation into the R¨ ssler model . 9. . . . . 9. . . . . . . . . .5 Rotating Particles inside the EggShaped Form . . 129 6.1. . . . . . . . . . . . . . . . . . . .2 A double eggshaped form . . . . . . . . . . . . . . . . . . . . . . . . .6. .2 Equilibrium points in Hamiltonian systems 7. . . . . . . . . . . . . . . . . . 9. . . . . . . . 9. . . . . . . . . . . . . . . 7. . . . . . . . . . . . 7. . . . . . . . . . . . . .1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . .3.3. . . . . . .
. . .6 Other Relativistic Forms . . . .12 Two Attracting Equal Nonsymmetric Masses . . . . . . . . . . . . . .1 A circular rotating disk . . . . . .1. 10. . . . . . . .11 Two Equal Attracting Masses in Opposite Directions . 10.2 The contraction process . . . .2 The rotating ellipsoid . . . . 12.1 One attracting mass . .2 The area of the chaotic region in galaxy simulations 11. .1 The space contraction rotationtranslation case . . 11. . . . .1 Central sink . . . . . .10. . . . . .5 Other Chaotic Forms .1 Elliptic rotationtranslation . . . . . .2 NonCentral Sink . . .2 The H´ nonHeiles System . . . . . .3 Discrete Analogues to the H´ nonHeiles System . . . . . . . . .1 A special rotationtranslation image . . 11. . . . . . . . . . . .5 Relativity in RotationTranslation Systems . .11. . . . . . . . . 11. . . . . . . .4 Chaotic Forms without Space Contraction . . . . . . . . . . . . 11. . 10. . . . . . . . . .8. . . . . . . 202 205 205 205 207 207 208 208 213 213 216 219 219 219 220 223 223 225 228 234 235 236 241 247 248 248 248 250 253 253 255 257 258 260 263 267 267 268 270 272 273 275 10 Chaotic Advection 10.10. . . . . . . 10. . . . . . . 11. .1 Introduction . . . . . . . . . . . .3 Two Symmetric Sinks . . .4 RotationReﬂection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . e 12. . 10. . . . . . . . . . . . . . . . . . . . . . . . .2 Chaos in the Solar System . . . . 11. . . . . 11. . . . . . . . . . 11. . . . . . . . . . . . . . . . . . . . . . . . 10. . . . . . . . . 10. . . e 12. . . . 10. .8. . . . . . . .3. .6 Complex Sinusoidal Rotation Angle . . . . . . . . . . . . . .2 Rotationtranslation with special rotation angle . . . . . 11 Chaos in Galaxies and Related Simulations 11. . . . . . .1 Introduction . . . . . . e 12. . .9. . . . . . . . . . . 11. . . . . . . . . . . . . . . .3 Galaxy Models and Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9. . . . . . . . .8. . . . . . . . 10. . . . . . . . . . . . . . . . . . 11.1 Aref’s blinking vortex system . . . . . . . 11. . . . . . . . . 11. . .9. . 10. . . . . . . . . . 12. . . . . . . . . . . . .3 The speed of particles . . . . . . . . . . . . . . . . . . . 11. . .5 Other Forms for the Hamiltonian .9 Rotating Disks of Particles . . . . . . . .1 Symmetric unequal attracting masses . . . . . . .8 Relativistic ReﬂectionTranslation . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . 12 GalacticType Potentials and the H´ nonHeiles System e 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10. . . . . . . . . . . . . . . . . . . . . 10. . . .10. . . . . . .8 Other RotationTranslation Models . . . 11. .10 Rotating Particles under Distant Attracting Masses . . . . . . . .7 Galactic Clusters . . . . . . . . . . .4 Paths of Particles in the H´ nonHeiles System . . . . . . . . . . . . .6 The Simplest Form for the Hamiltonian . . . 11. . . . .3. . . . . . . . . . . . . 11. . . . . . . . . . . . . . . . . . . . . . . . .7 A Special RotationTranslation Model . . .1 The Sink Problem . . 11. . . .
. . . . . . . . 281 285 . . . . 12. . . . . . . . . . . . . . . . . 13. . . . 12. . . .10 Another Simple Hamiltonian System . . . . . . . . . 13. . . . 275 . . . . 13 Odds and Ends 13. . . . . . . . . . . . . . . . . . . . . . . . . . .2 The Eﬀect of Noise in ThreeDimensional Models . . . . . . . . . . 279 . . . 13. . . . . . . . . .4 The Pendulum . .3 The LotkaVolterra Theory for the Growth of Two Conﬂicting Populations . .8 A Logarithmic Potential . . . . . . . 13. . . 14 Milestones References . .7 Gravitational Attraction . . . . . . . .5 A Special SecondOrder Diﬀerential Equation . . . . . . .9 Hamiltonians with a Galactic Type Potential: The Contopoulos System . . . . . . .12. . . . . . . . . . . . . . . . . . . 285 . . . 12. 13. .1 Forced Nonlinear Oscillators . . . . . . . . . . . . . . . . . . . . . . . . 285 . . .6 Other Patterns and Chaotic Forms . . 279 . . . . . . . . . . . 288 290 292 292 297 303 . . . . . . . . . . . .
Although this method was still quite laborious at those early days.Chapter 1 Introduction 1. The scientists of the last few centuries followed this theory in order to ﬁnd rational routes in the construction of models. for which the earlier method by Euler was not very accurate. It is no surprise that this change inﬂuenced even the French school. the stars of Ursa Major are found every year in predetermined places. the moon follows a fourweek period.1 Chaos in Diﬀerential Equations Systems Four centuries ago. It would be diﬃcult to say that determinism. Henri Poincar´ . especially for the study of highly nonlinear problems. How then does one introduce a theory that goes against the perfect world of Newtonian dynamics? Our solar system has been rotating for millions of years. The theory of determinism also had theoretical support from philosophy and other scientiﬁc ﬁelds. It was essential to simplify our description of the world ﬁrst. and the experience of the last centuries was in favor of determinism. Theory and practice could now equally beneﬁt each other. advances in applied mathematics made this transition easier. tried to resolve the 1 . How and why could this perfect deterministic world lose this certainty and lead a chaotic existence? It was not easy to convince the scientiﬁc community to introduce uncertainty into this deterministic world and accept chaotic theories. The movements of stellar bodies were assumed to follow predetermined paths according to laws that were known or had to be discovered by the scientiﬁc community. the RungeKutta method for numerical solutions to diﬀerential equations dramatically improved our ability to get more accurate solutions for numerous models. Following the introduction of the RungeKutta method. instead of entering into obscure paths of uncertainty and later on Chaos. As a result. At the end of the nineteenth century. and only then to solve fundamental problems. e one of the most prominent scientists in 19th century France. especially in mathematics and physics. The sun continues to rise every day. However. it gave new directions in the way scientists addressed various problems. traditionally the most theoretical of all. It was a real milestone in applied mathematics. as a general idea. future events in nature could be explained using knowledge of the past. Newton and other scientists introduced the idea of determinism in the mathematical representation of the real world. can solve all problems in nature. the paths traversed by a stellar body could now be estimated quite eﬃciently.
The system (1. the twodimensional paths do not exhibit a chaotic behaviour. Poincar´ tried to determine solutions to the restricted threebody problem. He understood that. y) = 0 and g(x.1) where x and y are the ﬁrst time derivatives. or c) approach a limit cycle.3) illustrates all three cases: x = −y(y − a)(y − b) ˙ (1. the Poincar´ Bendixson theorem says that a phase path will either: e a) terminate at an equilibrium point. Assuming that we have determined the equilibrium points at which both f (x. The system of diﬀerential equations (1. consider the twodimensional autonomous system (1. Nowadays we know that to solve this problem we need a set of more than three coupled diﬀerential equations.2) and we can ﬁnd phase paths in the plane describing the motion. he noticed that the orbits of three bodies moving under a central force due to gravity are quite complicated and can change drastically under small changes of the initial conditions. y) ˙ y = g(x. the motion had still a very complicated form. In stark contrast. directly from 1. He tried to explain the phenomenon more generally and to establish a theory describing the chaotic paths of the solutions to a system of diﬀerential equations: The threedimensional solutions to a system of nonlinear diﬀerential equations could be chaotic. and the paths of the solutions have a chaotic character. b) return to the original point. y is: x(x − a)(x − b) dy =− dx y(y − a)(y − b) . y) ˙ (1. namely the movement of a small mass around two larger masses revolving in two dimensions. there are no chaotic solutions. even in this simple case.3) y = x(x − a)(x − b) ˙ The resulting diﬀerential equation for x. To elaborate a bit.2. resulting in a closed path.1) can take the form of a ˙ ˙ twodimensional equation for x and y: dy g(x. Hence. when e one mass is very small relative to the other two. e 1890b.2 Chaotic Modelling and Simulation famous threebody problem in a contest proposed by the King of Norway (Poincar´ . 1892). as follows from the famous Poincar´ Bendixson theorem: Chaos cannot occur in a twodimensional aue tonomous system. Although he succeeded in solving a special case of this problem. y) (1. y) = 0. y) = dx f (x.1): x = f (x.
the simplest threedimensional nonlinear system of diﬀerential equations. y) diagram On the other hand.Introduction This equation can be easily integrated. The R¨ ssler system. .1. are further discussed in Chapter 7.3. However. 1976d) o o x = −y − z ˙ y = x + ey ˙ z = f + xz − mz ˙ easily produces chaotic paths for appropriate values of the parameters.2(a). There are nine equilibrium points. Such systems. The paths. in the last equation.1: The (x. with parameters equal to a = 0. are illustrated in Figure 1. the R¨ ssler model (R¨ ssler. chaotic paths appear with the form presented in Figure 1. and one in the middle of the ﬁgure.7.8 and b = 0. (1. is examined in greater detail in Chapter 6. not chaotic in themselves but nonetheless not without interest. o when the parameters have values e = f = 0. along with other interesting o threedimensional systems.4) The R¨ ssler system has only one nonlinear term. resulting in the solution: x4 + y4 x3 + y3 x2 + y2 − (a + b) + ab =h 4 3 2 3 where h is the constant of integration.2 and m = 5. four at centres four corners. The more complex paths pass from the four unstable equilibrium points at the sides. FIGURE 1. four in the sides.
4. suppose we replace the R¨ ssler system (1.5) Then.4) o by a corresponding system of diﬀerence equations: 1 xt+1 = xt − (yt + zt ) 6 1 yt+1 = yt + (xt+1 + eyt ) 6 zt+1 = zt + 0. for parameter value e = 0. It is therefore not surprising that the logistic model. period doubling bifurcations occur. was “invented” only in the last three decades.2.6) This very simple model already exhibits the key elements of chaotic behaviour. For example. On the other hand.1 + ext+1 zt (1. For centuries the mathematical development of methods and tools of analysis concentrated mainly on diﬀerential equations. y) (b) A discrete analogue FIGURE 1. The logistic model is described by a very simple equation: xt+1 = bxt (1 − xt ) (1. giving rise to stable orbits of ever increasing size. the most popular chaotic diﬀerence equation model today. culminating in completely chaotic behaviour. the diﬀerential equation analogue of this model was very popular: x = bx(1 − x) ˙ (1. an interesting chaotic form like a seashell results (Figure 1.2(b)).4 Chaotic Modelling and Simulation (a) The continuous case (x. and iterating under the condition xt+1 = xt − 0. or by a single diﬀerence equation with nonlinear terms. As the parameter b increases.2: Continuous and discrete R¨ ssler systems o 1. and not on diﬀerence equations.2 Chaos in Diﬀerence Equation Systems Radical changes in the behaviour appear when a system of diﬀerential equations is replaced by a similar system of diﬀerence equations.7) .
3: Bifurcations of the logistic and the Gaussian . are: the sinusoidal model. gives simple ﬁnite solutions without complications and chaotic behaviour. Also. the Gaussiantype models. and exhibit similar behaviour. the stable orbits become more and more complex as b increases. there are clear diﬀerences between the two classes of models. Other models that are examined. called the logistic model. along with chaotic analysis and simulation techniques. In the case of the logistic model.1 The logistic map Chapter 3 is devoted to a detailed analysis and simulation of the chaotic behaviour of nonlinear diﬀerence equations. as expected. As can be seen clearly from the bifurcation diagrams of the logistic (Figure 1. The bifurcation diagram exhibits the stable orbit of the x values for each value of the chaotic parameter b. 1.3(b)). (a) Logistic (b) Gaussian FIGURE 1. Chapter 2. stochastic and chaotic models are introduced. culminating in the complete chaos observed for b close to 4. was applied in demography by Verhulst already in 1838. starting with the logistic model. The logistic map and its bifurcation diagram are presented.7). The subsequent chapters deal with particular classes of models. a class of models with behaviour diﬀerent than that of the quadratic maps are studied. Deterministic. as well as some similarities in the chaotic windows. and their interesting behaviour. the Gompertz model and a model proposed by May.3(a)) and the Gaussian model (Figure 1. further discusses models and the modelling process in general. This continuous model.Introduction 5 Equation (1.2.
For instance.2865.4(b) presents the map of the delay logistic. Delay models can be found in several real life cases when the future development of a phenomenon is inﬂuenced from the past.3 The H´non model e In some cases a system of two diﬀerence equations is replaced by a single difference delay equation with the same properties.9 and the delay m = 30. a simple delay equation is the delay logistic: xt+1 = bxt (1 − xt−m ) which is analysed further in Chapter 4.4: Delay models 1. (a) The Glass model (b) The delay logistic FIGURE 1.2. c = 0. along with some variations and more complicated models. the Glass model gives a very complicated graph in the phase space and chaotic time oscillations of a very complex character (Figure 1. when the delay is m = 10.2.2 Delay models Delay models are models where future values depend on values far in the past. This is the case of the H´ non e . Figure 1. The chaotic parameter is b = 1. as well as the related oscillations.4(a)). Such chaotic behaviour is expressed by a model proposed by Glass xt+1 = cxt + a xt−m 10 1 + xt−m When the parameters are a = 0. The complexity of the delay system is higher when the delay is large and the equations used are highly nonlinear.6 Chaotic Modelling and Simulation 1.2.
5: The H´ non and Holmes attractors e Another model.10) Fig 1.5(b) presents the Holmes model when the parameters take the values a = 3. which is e now considered one of the standard models in the chaos literature. The model proposed by H´ non. has the form: xt+1 = 1 − axt2 + yt yt+1 = bxt The equivalent delay diﬀerence equation of the H´ non system is: e xt+1 = bxt−1 + 1 − axt2 (1. with very interesting properties.5(a)). 1976). was proposed by Holmes (Holmes. b = 0. extensions and variations.8) For parameter values a = 1. are presented in Chapter 5. which can be thought of either as a system of two diﬀerence e equations.8.9) (1.3 the famous H´ non attractor is formed e (Figure 1. (a) The H´ non attractor e (b) The Holmes attractor FIGURE 1. .Introduction 7 model (H´ non. or as a single delay equation. e such as the Holmes model.4 and b = 0. along with other related models. The H´ non model. 1979a): xt+1 = bxt−1 + axt − xt3 (1.
we consider conservative systems.6(b) respectively. .8 Chaotic Modelling and Simulation 1. the autocatalytic and Aro neodo (Arneodo et al. Figure 1.. 1980. appears in Figure 1.2 Conservative systems In Chapter 7. 1. A twodimensional view of this chaotic attractor.3. which gives rise to a rich variety of chaotic patterns. along with the time development of onedimensional oscillations.6: The Lorenz and autocatalytic attractors We also examine a fourparameter model of chemical reactions.6(a) and 1.8 shows an example of two of the forms analysed in this chapter. 1981b) models. 3dimensional views of the Lorenz and autocatalytic attractors are illustrated in Figures 1. systems that have a nontrivial ﬁrst integral of motion.1 More Complex Structures Threedimensional and higherdimensional models A number of interesting threedimensional models are studied in Chapter 6.7. The resulting patterns exhibit a varying degree of symmetry.3 1. (a) The Lorenz attractor (3d) (b) The autocatalytic attractor FIGURE 1. depending on the equations of the system.3. that is. and their extensions and variations. The existence of a ﬁrst integral of motion provides enough information to draw the phase portrait of the system in cartesian coordinates. and in particular Hamiltonian systems. 1963) and R¨ ssler models. These include the Lorenz (Lorenz.
6 0.2 −0.2 −0.2 0.2 0.3 −0.Introduction 9 FIGURE 1.0 0.8 0.0 −0.1 0.8: Nonchaotic portraits .4 0.3 FIGURE 1.1 0.4 −0.7: A chemical reaction attractor 0.
reﬂections and translations.3 Rotations Chapter 8 is devoted to rotations.11): xt+1 = xt cos(θt ) − yt sin(θt ) (1. followed by a translation. as well as nonlinear terms. A simple rotation in the plane. θ = 1. For appropriate values of the parameters.13)). The most popular of these is probably the Ikeda model (Ikeda.3. does not in itself lead to chaotic forms.11) yt+1 = xt sin(θt ) + yt cos(θt ) However. First of all. The model is based on a system of two rotation equations with a translation parameter a added to the ﬁrst .11) gives chaotic solutions. the system gives an interesting twodimensional map with chaotic regions and islands of stability inside these regions (Figure 1.10 Chaotic Modelling and Simulation 1. in its usual form.3.9).12) yt+1 = xt sin(θ) + yt cos(θ) − xt2 cos(θ) When the rotation angle is stable.14) and (1. 1979) (equations (1. is based on linear functions such as (1. let us consider the eﬀect of adding a quadratic term in both equations. and the chaotic forms and attractors that these give rise to. FIGURE 1. in the system equations. these models give rise to chaotic patterns and attractors. the introduction of nonlinear terms in system (1. as is the case of the famous Ikeda attractor. Rotation models express a large variety of systems in various disciplines. The rotation equation.9: A rotation chaotic map Another way to add nonlinearity is to assume that the rotation angle θ is a nonlinear function. The models in this category are characterized by the presence of rotations and translations. The system then becomes nonlinear: xt+1 = xt cos(θ) − yt sin(θ) + xt2 sin(θ) (1.
13) The rotationtranslation system in this case takes the following form: xt+1 = a + b xt cos(θt ) − yt sin(θt ) yt+1 = b xt sin(θt ) + yt cos(θt ) (1. In Figure 1.Introduction equation.14) This system was ﬁrst proposed by Ikeda (1979) to explain the propagation of light into a ring cavity. only the parameter c is changed (c = 2.11(a) and 1. The parameter values in that case where a = 1.11(b).22). some of which are illustrated in Chapter 8.11(b). The shape of the Ikeda attractor in the previous standard form depends on the selection of the values of the parameters. and they give rise to a very interesting attractor illustrated in Figure 1. A wide variety of these forms are presented and analysed in Chapter 8. When b < 1 several chaotic and nonchaotic patterns can be seen in the various simulations.11(a) an attractor pair appears. The properties of the chaotic maps in this category are so rich and often unexpected. b = 0.10.4 and d = 6. By varying these parameters. FIGURE 1.9. a space parameter b.83. very interesting new forms appear. .1 and d = 6. and a nonlinear rotation angle of the form: θt = c − d 1 + xt2 + y2 t 11 (1. c = 3.10: The Ikeda attractor This map has Jacobian J = b2 . Now three images of the attractors are present. The parameters are a = 6. b = 0. Two of these attractors are illustrated in Figures 1. In Figure 1. c = 0.
The main tools for shape formation are rotation and translation. and to pass from a qualitative to a quantitative point of view. based on the following set of equations: xt+1 = a + b xt cos(2θt ) + yt sin(2θt ) (1. a = 6 and the other parameters remain unchanged. The parameters in the ﬁrst case are a = 2. . c = 1. We try to ﬁnd mechanisms to design simple or complicated and chaotic forms. In the second case.3.4 Shape and form In Chapter 9 we study how chaotic analysis can be used as a tool to design forms that appear in physical systems. The analysis of these systems has moved from the classical analysis concerning simple and regular forms to chaotic analysis. but also reﬂection with translation. The equations used are 3dimensional and the original o inspiration comes from the R¨ ssler model.9.5.15) yt+1 = b xt sin(2θt ) − yt cos(2θt ) If the same function as in the Ikeda case is used for the angle θt θt = c − d 1 + xt2 + y2 t then we obtain an attractor (Figure 1. Even the most strange chaotic attractors and forms can have real signiﬁcance. Such an analysis is useful in the scientiﬁc ﬁeld related to shape and form.12(b)).12(a)) which is divided in two new separate forms (Figure 1. especially in biological systems but also in ﬂuid ﬂow or in crystal formation. The seashell formation illustrated earlier in this chapter is one case of an attractor with a special shape. as in the Ikeda case above.12 Chaotic Modelling and Simulation (a) A double Ikeda attractor (b) A triple Ikeda attractor FIGURE 1. The aim here is to go beyond the classical stability analysis.6 and d = 9. b = 0.11: Ikeda attractors 1.
but now with b = 1. One of these points is also located on the x = a axis.13(a). For instance.13(b) illustrates this case with parameter values a = 1. located at (x. The reﬂection attractor is on the left side of Figure 1. The ﬁve equilibrium points in the periphery are of the ﬁfth order: (xt . The parameter a is quite higher than in the previous case. These cases simulate the split of a vortex in two separated forms.12(b). 0) and (x. y) = a + ab cos(θ). ﬁve ﬁfthorder and eight eightorder equilibrium points. Interesting forms arise assuming that the rotation angle is a function of the distance rt = xt2 + y2 . yt ). There is a symmetry axis at x = a 2 and one equilibrium point in the central part of the ﬁgure. yt+1 ) = a (xt+5 . x = 2 . The parameters are a = 3. Furthermore. c = 1 and d = 4. In the ﬁgure there are two pairs of secondorder equilibrium points. whereas the other four 2 are located in symmetric places about this axis. the Ikeda model with b = 1 provides interesting forms with the same a axis of symmetry. ab sin(θ) 13 d where θ = c − 1+a2 .Introduction The centres of the two attractors are located at: (x. c = 5 and d = 5. .14). when t d rt2 θt = c + we obtain the form illustrated in Figure 1. applied to the Ikeda case. yt . yt+5 ) = 2 . Figure 1.12: Translationreﬂection attractors The rotation equation (1. y) = (a. (a) Reﬂection with 1 image (b) Reﬂection with 2 images FIGURE 1. is a spacepreserving transformation.
the graph in k 12 Figure 1. Objects in the solar system mainly follow circular or elliptic orbits around a centre of mass.85 and d = 16.” By using the rotation formula with rotation angle θt = c + d rtδ and with parameter values b = 0. 35. c = 2.15(a) appears. 45 and 55 respectively.14(d)). The attracting force is. k = 17 and δ = 2. and after the separation of the chaotic form into two symmetric forms (Figure 1.13: Symmetric forms In Chapter 9. When a = 2π . 3k k 1. of course. The rotation iterative procedure is used. gravity. and .14 Chaotic Modelling and Simulation (a) Fifthorder symmetry (b) Multipleorder symmetries FIGURE 1.1 Chaos and the Universe Chaos in the solar system Complicated and even chaotic orbits and paths in the solar system are studied and simulated in Chapter 11. The parameters are b = 1.15(b) results. several symmetric and nonsymmetric forms are presented and analysed. The following ﬁgures illustrate a symmetric chaotic form before the separation step (Figures 1. Another category of patterns includes forms which we call “ﬂowers. To simplify the problem the centre of mass is assumed to be located in the centre of the large mass. In this case the simulation can follow a purely geometric approach.14(b)).14(a) and 1.4 1.4. in the intermediate stage (Figure 1. c = d = 2π and δ = 3. with d rotation angle θt = c + rt .9.14(c)). a = c = d = 2π . Figure 1. and the parameter a takes the values a = 18.
Introduction 15 (a) One symmetric form (b) Starting separation (c) First separation stage (d) Final separation stage FIGURE 1.14: Separation of chaotic forms .
In the same ﬁgure. and the circular orbit is obtained when ν = 1. The motions of two attracting point masses are relatively simple. then the ˙ ˙ ˙ ˙ . y1 ) and (x2 . y2 ).16(a).15: Flowers is presented by an inverse square law of the distance. where the initial √ velocity is 1 < ν = 1. If we give the two masses initial positions (x1 . A hyperbolic orbit is also presented. x2 and y2 . The gravity constant is G = 1 and the total mass of the rotating system is M = m1 + m2 . y = 0. y1 .16) 3/2 x 2 + y2 The initial conditions are x = 1. are. The equations of motion for the point mass m in Cartesian coordinates are: x1 = − ¨ y1 = − ¨ (x1 − x2 )m2 3/2 (x1 − x2 )2 + (y1 − y2 )2 (y1 − y2 )m2 (x1 − x2 )2 + (y1 − y2 )2 (1.16(b) illustrates a large number of elliptic paths for the same value of the initial velocity as before.45 > 2 and the particle ﬂies oﬀ to inﬁnity. and initial velocities x1 .2 < 2 . The nondimensional equations of motion for a body with small mass m rotating around a large body with mass M. The circular orbit is illustrated in Figure 1.16 Chaotic Modelling and Simulation (a) δ = 2 (b) δ = 3 FIGURE 1. but deviate from the original position. The initial √ velocity is higher than the escape limit ν = 1. an elliptic orbit is presented. which we consider to stay in equilibrium position. The paths remain in the same plane.17) 3/2 Analogous is the case for the point mass m2 . The escape velocity is ˙ ˙ √ ν = 2. forming the shape presented in the ﬁgure. Figure 1. in Cartesian coordinates: x=− ¨ y=− ¨ x x2 + y2 y 3/2 (1. x = 0 and y = ν.
15. in special cases. Elliptic and Hyperbolic orbits (b) Many elliptic orbits FIGURE 1.Introduction 17 (a) Circular. through space. as a pair.17(b).6 and initial velocities x1 = −0. Their centre of mass drifts at a constant velocity. Each pointmass follows an elliptic path. The same motion. The points have masses m1 = 0. m2 = 0.17(a). Relative to an observer moving with the centre of mass.16: Orbits in twobody motion masses revolve around each other while drifting together. x2 = 0.45 and y2 = 0.17: Motion in the plane .4. The heavy line represents the movement of the centre of mass. their orbits are ellipses and. The masses revolve around the centre and drift together. The motion of two attracting points in space is presented in Figure 1. is illustrated in Figure 1.25. circles. (a) Attracting bodies in space (b) Relative movement FIGURE 1. viewed relative ˙ ˙ ˙ ˙ to the centre of mass. y1 = −0.
The satellite rotates in the same plane.18(a). m1 . y1 ) and (x2 . The total mass of the planets is m1 + m2 = 1.35) respectively. The equations of motion for the satellite then are: x=− ¨ y=− ¨ m1 (x − x1 ) 3/2 r1 − − m1 (y − y1 ) 3/2 r1 3/2 r2 m2 (y − y2 ) 3/2 r2 m2 (x − x2 ) (1.1. Figure 1. y) is the position of the satellite. It shows that the satellite can move between the two revolving planets. still gives very complicated and chaotic orbits. from space. To simplify the problem we assume that the third body. the planets. m2 = 0.9. 0) and (0. m2 are the masses of the two planets.18(b) illustrates the orbit of the satellite viewed in a coordinate system that rotates with the revolution of the planets (rotating frame). reduced to the plane. is of negligible mass relative to the other two bodies. In the following example the planets have masses m1 = 0. y2 ) are the positions of the planets at the same time t.18: Motion in space . . r1 . 0. with the Xaxis always passing through the planets according to the rotation formulas: X = x cos t + y sin t Y = −x sin t + y cos t (1.18) where (x. the movement is viewed from the outside of the system. r2 are the distances of the satellite from the two planets. The other two.18 Chaotic Modelling and Simulation The threebody problem.033. and (x1 . The complicated and chaotic paths are illustrated in the following ﬁgures. travel around the centre of the combined mass in circular orbits in the same plane. In Figure 1.19) (a) Motion in space (b) Motion in a rotating frame FIGURE 1. and the position and initial velocity of the satellite are (−1. a satellite.
4.68 FIGURE 1.6 (b) b = 3. Sellwood and Wilkinson (1993). Two realizations appear in the next ﬁgures.19(b).6 for Figure 1. it is possible to simulate chaotic patterns like the rings of Saturn and other planets.19(a) and b = 3.19: Ring systems 1. The simplest approach is to use an equation like the logistic.Introduction 19 By using diﬀerence equations to model solar systems. and to assume that the system rotates following elliptic orbits. which follow a steady rotational motion. The majority of the models we examined are based on numerical integration of the equations of motion for N mutually interacting particles with equal masses. The parameter of the logistic model is b = 3. The approach to the Nbody problem follows some simpliﬁcations.2 Chaos in galaxies The underlying literature on galactic simulations deals mainly with the Nbody problem and related simulations.20) yt = 2rt sin t where rt+1 = brt (1 − rt ). 1989).68 for Figure 1. The rotation formula is: xt = 2rt cos t (1.1 The acceleration of the ith particle is: N r j − ri j i 2 ri2j + rcut 3/2 1 See Sellwood (1983. (a) b = 3. .
r8 t = 200 FIGURE 1. Two characteristic examples of these forms are the twoarmed spirals illustrated in Figures 1. According to the theory of kinematic spirals.15. the oval orbits are nested to form a twoarmed spiral pattern of a galaxy.20: Spiral galaxies Another approach to producing a spiral pattern formation is based on the elliptical epicycle that a star follows around its guiding centre. but very large. familiar galaxy forms appear after some iterations. accelerations at close encounters. are based on the following equation in polar coordinates (r. The simulations start from an elliptic disk with diﬀerent density distribution. The parameters are a = 0. n = 1 for the onearmed spiral (Figure 1.20 Chaotic Modelling and Simulation The cutoﬀ radius rcut greatly simpliﬁes the numerical computation by eliminating the infrequent. .20(b).21(a). When N is large.20(a) and 1.08.2 The simulation results. In Figure 1. (a) Twoarmed spiral ρ = c2 .2 < Rg < 1. n = 2 for the twoarmed spiral and a = 0. The gravity constant and the mass of each particle are set to 1.20(b).21(b)). the high central density leads to the formulation of a barred spiral form. r4 t = 100 (b) Barred spiral ρ = c1 . illustrated in Figure 1. 2 See also Sparke and Gallagher (2000). θ): r= Rg 1 + a cos n(5 − 5Rg + φ) The values of the guiding centre are 0. The details are presented in Chapter 11. the main computational problem is the large number (∝ N(N − 1)) of operations required to determine the accelerations. When a small perturbation is applied to the rotating system.
22(b) the central bulge is already present. intergalactic forms.20) with parameters a = 1. c = 2. b = 1. only the central bulge remains.23(a) illustrates a simple spiral pattern with one companion. the galactic forms arise by using very simple models and simulation tools.6.22(b). black holes and other objects give inspiration for new simulations of the large variety of observed forms and shapes.6.22(c). planetary nebulae. The outer part has the form of an electromagnetic ﬁeld. The questions about black matter and how it inﬂuences the movement and shape of galaxies and other objects change our approach to model building and the related simulations of galactic forms. There is a centralbulge connected with the outer periphery by two symmetric routes.22(d) also illustrates a case where d = 800. whereas in Figure 1. Figure 1. but now a = 0. In both cases the parameter c is set to 1. When d = 800. and a is 0. d = 6.22(a) and 200 for Figure 1. Figure 1. Rotations. In both cases b = 0.9.21: Spiral galaxies Recent discoveries of new galaxies. In some instances a central chaotic bulge is created.23(b). whereas in Figure 1. Figure 1. while the parameter d is 35 for Figure 1.Introduction 21 (a) Twoarmed spiral (b) Onearmed spiral FIGURE 1. The rotation angle in both . In this case.23(a) and 5 for Figure 1. In many instances. the reﬂection image takes the form presented in Figure 1. The forms presented below arise from the reﬂection iterative formula (1. reﬂections and translations with appropriate selection of the equation of the rotation angle give rise to a plethora of galaxy formations.22(a) illustrates the beginning of the mirror image formation.3 for Figure 1. and rotation angle: d θt = c − 2 rt The reﬂection procedure leads to mirror image symmetric shapes. while the form in the outer region is just beginning to take shape.23(b) two spiral galaxies with two companions appear.
6 FIGURE 1.22: Reﬂection forms .22 Chaotic Modelling and Simulation (a) d = 35 (b) d = 200 (c) d = 800 (d) d = 800. a = 0.
3 (b) a = 5 FIGURE 1. Contopoulos explains that in 1958 he expected the orbits to be ergodic and ﬁll all the space inside the energy surface. The works of Contopoulos and H´ nonHeiles are examples of what we will call e computer experiments. Later he could prove (Contopoulos. The work of H´ non and Heiles (1964) on galactic motion is the ﬁrst complete work e that establishes the existence of chaos in galaxy formations. as in the case of deformed Lissajous ﬁgures. It provided considerable evidence of the existence of chaotic motions in galaxies.4. but instead ﬁlled curvilinear parallelograms. Their work started as a computer experiment in order to explore the existence and applicability of the third integral of galactic motion. and found that the orbits in the meridian plane of an axisymmetric galaxy are like Lissajous ﬁgures. 1965) explored the boxlike paths of a body in a galactic potential. Chapter 12 is devoted to this work of H´ nonHeiles and e Contopoulos. Sometimes the computer results were surprising and contradicted the existing theories of that time.3 Galactictype potentials and the H´nonHeiles system e Contopoulos (1958. .Introduction case is given by: θt = c + d rt 23 (a) a = 0.23: Spiral Patterns 1. This new type of experiments gave new directions in various scientiﬁc ﬁelds. In his book (Contopoulos. he found that the orbits did not ﬁll all the available space. 2002). 1960. Instead. 1960) that such orbits can be explained both qualitatively and quantitatively by a formal third integral of motion. and especially in astronomy.
24: H´ nonHeiles systems e We have reproduced here. c = 0. y′ ) diagram when x = 0.21) So the potential U(x.24(b) for the energy level E = 1/12.24 Chaotic Modelling and Simulation The Hamiltonian applied by H´ non and Heiles (1964) has the form e H= 2y3 1 2 1 2 x + y2 + 2x2 y − =h ( x + y2 ) + ˙ ˙ 2 2 3 (1.22) . H´ non and Heiles estimated the paths of a pare e ticle according to the Hamiltonian (1. θ) = 1 2 1 3 r + r sin(3θ) 2 3 1 2 2y3 x + y2 + 2x2 y − 2 3 (a) The original H´ nonHeiles (y.24(a) presented in the H´ nonHeiles paper. The discrete model is based on the following set of equations: xn+1 = cxn + un yn+1 = cyn + vn un+1 = −xn − 2xn+1 yn+1 2 vn+1 = −yn − xn+1 + y2 n+1 (1. y) = or in polar coordinates: V(r. the original ﬁgure Figure 1.1. y) that is used has the form: U(x. and then they provided the (y. A discrete alternative of the H´ nonHeiles model results in the image presented e in Figure 1. In this case.21) at the energy level E = h = 1/12. y) diae ˙ gram (b) A discrete alternative (E = 1/12) FIGURE 1. In this low energy level mainly regular orbits appear. via computer simulation. At a higher energy level chaotic regions are formed.
The resulting .25 is produced by using the following rotation model: 2 xn+1 = b(xn cos a − yn sin a) + xn sin a 2 yn+1 = b(xn sin a + yn cos a) − xn cos a (1. All the (x.Introduction The Jacobian of the system (1. The initial values are (x. y) plane. 0. The value of u.4 The Contopoulos system A Hamiltonian with a “galactictype” potential was ﬁrst introduced by Contopoulos (1958.2 and b = 0. J is 1.5668627. Since the Jacobian is J = b2 . The H´ nonHeiles system gives also interesting paths in the (x.9. which is the escape limit. 1.1 in Figure 1. The twoarmed spiral galaxy in Figure 1.26(b). is estimated so that the value of the Hamiltonian is h = 1/6. 1960) in his pioneering work on galaxies. FIGURE 1. very interesting ﬁgures arise (c = 0. a = 0. 0). The energy level in both cases is E = h = 1/6. The paths e are characterized by the level of potential. y) paths of the particle will be restricted inside the triangle.25: A twoarmed spiral galaxy Twoarmed spiral galaxy forms can be obtained from simulations of very simple models.26(a) and 1. when c takes values less than 1. and therefore the spacepreserving property is satisﬁed. The similarities with the H´ nonHeiles system (Fige ure 1. However. For this limit the equipotential triangle is drawn.22) is: 2 J = 1 + 2(c − 1)yn+1 + 4(c − 1) xn+1 + y2 n+1 25 It is obvious that for c = 1. y) = (0. along with higher order terms. b is an area contracting parameter. and v = 0.23) In this case.24(a)) are obvious.1. The potential is based on the addition of two harmonic oscillators.24(b)).4. Totally chaotic trajectories is illustrated in Figures 1. The form of the potential is triangular.
u = 0 and v = 0. y) = with corresponding Hamiltonian: H= 1 2 u + v2 + U(x.27. . x. and the initial conditions are x = 0.1 and k = w1 /w2 = 4/1. y = 0.6. y.26: Chaotic paths in the H´ nonHeiles system (E = 1/6) form is U(x.1.2. For the 4/1 case the parameters are e = 0. u = 0 and v = 0. y = 0. For the 2/3 case the parameters are e = 0.1 and k = w1 /w2 = 2/3. e have been rescaled.26 Chaotic Modelling and Simulation (a) A chaotic path over a long time period (b) A chaotic path over a short time period e FIGURE 1. characterized as 4/1 (left image) and 2/3 (right image). are illustrated in Figure 1. and k = w1 /w2 is the important resonance ratio. The equations of motion are given by: u=− ˙ ∂U = −k2 x − 2ey2 ∂x ∂U = y − 2exy v=− ˙ ∂y (1. v. and the initial conditions are x = 0.9977753. y) = h 2 1 2 2 w x + w2 y2 − exy2 2 2 1 Without loss of generality this Hamiltonian can be simpliﬁed to give the form: H= 1 2 1 1 u + v2 + k2 x2 + y2 − exy2 = 2 2 2 where u.24) Resonant orbits.
including the eﬀect of introducing noise into models. The interesting images of Section 13.5 Odds and Ends. A “milestones” table in Chapter 14 should provide a useful reference for future reading. We hope they will act as an inspiration for future researchers. and an extensive discussion of the very interesting LotkaVolterra system. which is a collection of interesting systems of or variations on systems. .27: The Contopoulos system: orbits and rotation forms at resonance ratio 4/1 (left) and 2/3 (right) 1. and Milestones The book ends with Chapter 13.Introduction 27 FIGURE 1. The history of chaotic modelling is quite complex.6 could certainly be considered as works of art.
.
” is based on a purely geometric formulation of a model describing planetary motion.” His explanation of Kepler’s second law of planetary motion. 1 See the introduction in Acheson (1997). also includes a section on the “Applications on the Geometry of CurveLines. The paper.1 Introduction For many years scientists from all ﬁelds have been trying to describe the real world by constructing logical forms of varying complexity. Calculus and analysis came many centuries later. All the models we will examine fall under two categories. Nevertheless. or more commonly philosophy. Since the beginning of time. mainly as a subset of geometry. During the modelling process. These verbal constructs became the ﬁrst models.Chapter 2 Models and Modelling 2. a process known as dialectics.” where his treatment of calculus is presented. 29 . equilibrium points and characteristic trajectories. or also model building. We pay particular attention to the methods of nonlinear analysis based on singular points. The development of language enabled man to form important verbal constructs. the standard methods of linear and nonlinear analysis are extensively used. Indeed Newton’s “Method of Fluxions. called models. the characteristic matrix of the system. expressing not just real objects.1 is reminiscent of Archimedes’s computation of the area of the circle by inscribed polygons. In this book we analyse models that are expressed in mathematical terms. their foundations can be found in the Pythagorean theory of fractions and in the Archimedean theory of polygonal approximations. and eigenvalues of the Jacobian. titled “De motu corporum in gyrum. but also abstract ideas and innovations. The process of selecting the main characteristics of a real life phenomenon and ﬁnding the basic mechanisms that describe the development of the phenomenon is called modelling. Ancient Greeks established the ﬁrst schools and academies. Mathematical model building was also studied in Ancient Greece. diﬀerential equation models and diﬀerence equation models. as described in his famous letter to Halley. in order to encourage learning and to enhance the dissemination of information and exchange of ideas on scientiﬁc model building. humans have been using models to understand the real world. also called maps.
they describe the real situations quite well.” In this section we will see numerous methods commonly used for constructing models. Growth functions can be described in a number of diﬀerent ways. The second step in the process is the estimation of these parameters so that the model approximates the real situation to the best of its potential a process typically known as “model ﬁtting. The modelling process thus proceeds in two steps.2. Practical applications that lack a theoretical background do not contribute much to the development of any scientiﬁc ﬁeld. becomes relevant only when supported by related applications (Skiadas. During the modelling process. and the expected behaviour of the system. and the practical application of these models in real situations. on the other hand. A purely theoretical approach. We discuss some of these ways in the sections that follow. in general. In such models. a general form for the model is provided. and they have. 1994).1 Diﬀerential equation models Models based on diﬀerential equations have been extensively used ever since the invention of calculus. These are functions that express the growth of a system or of some special characteristics of a system. The mathematical treatment in both cases is very similar. First. It is important here to emphasise that one needs to strike a balance between the theoretical constructions of models. 2. One then proceeds to estimate the parameters involved. This function will depend on some unknown parameters.2. where the rate change of the system is proportional to the system’s current state. and to study the eﬀect that these parameters have on the behaviour of the solution. initial conditions.30 Chaotic Modelling and Simulation 2. based on general considerations of the form of the solutions. typically in the form of a modelling function. and possibly other special characteristics. However. Similarly. Skiadas et al. but a number of simulation and approximation techniques are available. and would in general describe a decay process. and thus can describe a number of diﬀerent situations. Thus. one doesn’t always have a closed formula for the function. The most widely used growth/decay model is the exponential model. the exact form of a model is not always possible to determine. This can be .1. 1994. a positive ﬁrst time derivative. one usually agrees upon the general form of the system. functions with a negative ﬁrst derivative would express the decline of a system. the system under consideration is described by one or more functions that satisfy a system of diﬀerential equations.2 Model Construction Models are approximations of reality..1 Growth/decay models An important general class of modelling functions are the so called Growth Functions. In some cases. 2.
Positive values of b will give rise to growth functions. referred to as steps. for the time ˙ derivative of x: DFN ∂x x = ˙ ∂t In equation (2. equation (2. The time parameter t is here assumed to be taking only discrete. The time derivative is then replaced by a ﬁnite time diﬀerence: x= ˙ xt+1 − xt dx ∆x ≈ = = xt+1 − xt dt ∆t (t + 1) − t (2. corresponding to consecutive stages of the system. These diﬀerence equations are referred to as maps in the chaotic literature.2) The “rate of change” thus becomes simply the change in the system from the one stage to the next. A diﬀerence equation is then nothing more than an algebraic relation between xt and its successors.1) Here. In this particular case. . while negative values will describe decay functions. 2. as is typically shown in any calculus class..1). .. closed form solutions can sometimes be obtained by recursion and induction. or if you prefer of xt+1 and its predecessors. For equation (2. xt+2 . we will use Newton’s notation.4) . Sometimes systems such as (2. . 2. For instance. values: t = 0. and one has to rely on other quantitative and qualitative techniques for describing the solutions. 1.1) has the following discrete analog: xt+1 − xt = bxt or more simply: xt+1 = bxt (2.Models and Modelling described by the simplest ﬁrst order diﬀerential equation: x = bx ˙ 31 (2. as in the rest of this book.1.2. xt+1 . the solution is the familiar exponential function: x(t) = x(0)ebt Often such an explicit description is not possible. x.2 Diﬀerence equation models Diﬀerence equations are the discrete analog of diﬀerential equations. . this can be readily achieved and the solution is xt = x0 bt (2. .1) can be explicitly solved.3). . b is the model parameter that needs to be speciﬁed.3) In diﬀerence equation systems.
σ(x. as you will notice. the growth rate x = dx is a function only of the magnitude x of ˙ dt the system. without loss of generality.1. the Calculus of Variations approach is discussed. with its source in the Lagrangian reformulation of clas 2 In fact. t) dt 2. It is a technique that we will use many times throughout the book. The most common source of models is the close investigation of real situations. t) dwt (2. and we discuss this process in section 2.1). σ(x. 1930. Series approximation (section 2.3 Modelling Techniques We will now discuss several general techniques for model construction.2. A measure expressing the inﬁnitesimal ﬂuctuations. where the ﬂuctuations are assumed to be zero: dx(t) = µ(x.1) can help us reduce a complex system to a much simpler one that closely resembles it.3. Every stochastic diﬀerential equation has a deterministic analogue. t) dwt A combination of these two components provides a stochastic diﬀerential equation of the general form: dx(t) = µ(x.3.3.3) here corresponds to the 1+b of (2. be considered to be bounded (0 ≤ x(t) ≤ 1). t)dt In many applications. t).” xt = x0 (1 + b)t = x0 et ln(1+b) ≈ x0 ebt .3 Stochastic diﬀerential equations The growth or decline of a system expressed by a variable xt over time can often be formulated by two components: 1. and wt is the standard Wiener process (Wiener. t) dt + σ(x. if we keep in mind that the b in (2. so that x(t) measures the percentage of the current state of the system over its “maximum state.2. In section 2. 1958).2 2.4) as for small values of b. The growth part or inﬁnitesimal growth. 3 Since the system can be rescaled.5) where the growth function x = x(t) can. µ(x. greatly resembles the diﬀerential equation solution. 1949.3. we can rewrite (2. and not of time. Hence the deterministic analogue takes a much simpler form: dx(t) = µ(x)dt 2. t) are to be speciﬁed.3 the functions µ(x. 1938.32 Chaotic Modelling and Simulation which.
section 2. consider a diﬀerential system of the form: x = f (x) ˙ (2.6) by its linear approximation. Diﬀusion equations are discussed in section 2. If we are interested in values of x near 0. smooth and continuous.3.4. we obtain the exponential model: x = a1 x ˙ In the majority of cases. 2. if we replace f in (2. quite common in physical systems. Often even this quadratic form can be simpliﬁed further. it is not surprising that systems where f is a quadratic polynomial approximate very well other more complicated systems. and can lead directly to an expression of the model equation. can be simpliﬁed to: f (x) = a0 + a2 x2 + a4 x4 + · · · (2. Thus. As an example.3. f is symmetric around the origin.7). if necessary. the expansion (2. and possibly second (curvature). the higher terms of the Taylor expansion may also be used. For instance.5 discusses delay diﬀerential equations.e. and possibly after replacing x by x + c. using only the ﬁrst (direction).7) Here the coeﬃcients of the expansion can be readily interpreted as the derivatives of f at 0: f (n) (0) an = n! As an example of this approximation process.1 Series approximation When the function xt describing a system takes values close to a particular point.8) 4 i. This method often gives suﬃcient results when the original model is simple. then we can replace f with the ﬁrst couple of terms of its Taylor expansion: f (x) = a0 + a1 x + a2 x2 + a3 x3 + a4 x4 + · · · (2.4 we are led to the functional equation: f (x) = f (−x) Since the odd derivatives of an even function vanish at 0. often leads to important simpliﬁcations of the approximation. then one can approximate a complex model by expanding the functions in the system in a Taylor series around that point. Of course. Symmetry of the system.Models and Modelling 33 sical mechanics.3. Finally. . and the derivatives of the general equation of the model exist. and hence to the model. if the xspace is isotropic.6) where f (x) is a reasonably behaving function of x. that have the ability to express the delay in propagation of changes inherent in many natural and social systems. derivative terms provides an adequate approximation to f .
The logistic diﬀerential equation has the simple form fkin = a2 v2 = x = bx(1 − x) ˙ (2. . The other parameter.6) through the process of a Taylor series approximation. 1]. by rewriting it as 1 dx = bdt x(1 − x) 5 For instance in birth processes or innovation diﬀusion cases. over the maximum level that the population could reach in the future. In other words. consider that x is restricted in the interval [0. To see how equation (2. so we can reasonably assume that a = 0. then a good approximation for small v (close to 0) would be: f (v) = a0 + a2 v2 The two terms can be identiﬁed physically as the energy at rest f (0) = a0 and the kinetic energy 1 2 mv 2 On the other hand. Equation (2. x is the saturation level. We are thus led to equation (2. a simple approximation will involve linear and quadratic terms. In that case. is related to 1 the growth rate at x = 2 . social and biological phenomena do not have this isotropic property. The same lack of symmetry holds for the time space.9) where x is the population at the present time. and assume that the system will exhibit symmetry around x = 1 . 2 the Taylor approximation to the second order would be: f (x) = a0 + a2 x − 1 2 2 = a0 + 1 − a2 x − x2 = a + bx(1 − x) 4 Here a is the rate of growth of the population when the population is at x = 0 or x = 1. We can consider the value x = 1 where the population is at half its potential as the 2 centre. Later on.9). If f (v) represents the energy of the system. b.” ﬁrst proposed by Verhulst (1845) to model the population growth in France. This approximation gives rise to a diﬀerential equation known as “The Logistic.9) follows naturally from (2.9) is a separable diﬀerential equation. economic.5 Hence.34 Chaotic Modelling and Simulation Probably the most wellknown example of this type is the equation for the kinetic energy of a mass. and so can be easily solved explicitly. and the relevant Taylor approximations tend to include a linear term as well. The constant term in such a system can often be assumed to be zero. Pearl and Reed (1920) applied this model to express the population growth in the United States.
When b < 3. using the approximation of the ﬁrst derivative by a ﬁnite diﬀerence as in (2.1). when b > 3 chaotic oscillations appear. The behaviour of the solutions of the discrete logistic will be explored further in Chapter 3.1: The solution to the logistic equation A diﬀerence equation whose form closely parallels that of equation 2.11) The behaviour of the solutions to (2.10) exhibits a very characteristic smooth sigmoid form (Figure 2. .Models and Modelling The solution then would satisfy: − ln(1 − x) + ln(x) = bt + C This last equation can be rewritten in the ﬁnal form: x(t) = x0 ebt 1 + x0 ebt − 1 35 (2.10) The graph of (2.11) depends on the value of the parameter b. The oscillations become progressively more chaotic as b approaches 4. is: xt+1 = xt + bxt (1 − xt ) Various second.9. but whose behaviour is a lot more interesting. 1 x(t) = x0ebt 1 + x0(ebt − 1) x0 0 −6 −4 −2 0 t 2 4 6 FIGURE 2.11). third and higher order forms of diﬀerential and diﬀerence equations models can be constructed with similar methods as above. they follow a discrete sigmoid form. However. The correct diﬀerence equation analogue to (2. These equations may in some cases exhibit chaotic behaviour. also goes under the name “logistic”: xt+1 = bxt (1 − xt ) (2. These models will be examined more closely in subsequent chapters.2).
2. the construction of an appropriate general model follows by the empirical investigation of realistic situations.2 Empirical model formulation In many cases in Physics. it would be more realistic to assume a declining relative growth rate. is deﬁned as y= and can be approximated by ∆x x That is. whose descriptions follow. This ˙ leads us to the exponential growth model x = b discussed already. Forecasting.36 Chaotic Modelling and Simulation 2.3.12) or (ln x)′ = −b ln x which can ﬁnally be rewritten as: x = −bx ln x ˙ (2.14) (2. the relative magnitude of the system.3. Another example is based on the relative growth rate d x ˙ = ln x = (ln x)′ = y ˙ x dt where y. Examples of using the relative magnitude of the system as the standard measure are the Gompertz. which is assumed as the metric scale of the system. the standard measure of growth is the relative magnitude of the system. This assumption leads to a convenient model construction especially when systems measured in completely diﬀerent scales are to be compared. the relative magnitude y is a measure of all the relative changes of the system from its original formation until its current status. Gaussian and Gamma models. Assuming that this decline is proportional to the relative growth y of the system. For instance. as it is x reasonable to expect that a system has limits to its growth. y≈ 2. In this case. we are led to a simple diﬀerential equation in y: y = −by ˙ (2. Direct integration gives as solution the Gompertz function: x = eln(x0 )e −bt . Marketing. Economics. when the growth rate x is assumed to be propor˙ tional to the magnitude x of the system.1 The Gompertz model dx = ln x x The simplest assumption about the relative growth rate is that it is constant.13) This model was ﬁrst proposed in Gompertz (1825). Biology. etc. However. then we end up with the exponential model discussed on page 31.
2 The Gaussian and Gamma models The well familiar Gaussian and Gamma distributions arise naturally from the relativegrowthrate setting. A comparison of the two models is given in Figure 2.2.3. This model has skewness opposite to that of the Gompertz model. as it is more intimately related to mortality. For instance. So for x near 1. resulting in: x = −b(1 − x) ln(1 − x) ˙ This model arises by considering the relative decay of the system. which leads . instead of the relative growth.2: The two Gompertz models 2. after some rather reasonable assumptions. The second variant is favoured in the actuarial sciences. e Both models are referred to in the literature as “the Gompertz model. b e ⋅ x 0 1 e x 1−1 e 1 FIGURE 2.1. the Gompertz model can be approximated by the logistic model.” with diﬀerent disciplines preferring one model over the other. we can examine the Gompertz system near x = 1: The logarithm ln(x) can be approximated by: ln(x) ≈ x − 1 Using this approximation in (2.2.Models and Modelling 37 Using the Taylor approximation methods of section 2. if may assume that the relative growth rate is a linear function of time. Another interesting variant occurs if we replace x by 1 − x in the Gompertz diﬀerential equation. as now the maximum growth rate is achieved when x=1− 1 e instead of when x = 1 .3.14) would result exactly in the logistic model.
summarises the dynamics of the system in question. and V(x) expresses the limits of the ˙ system. x. suppose the Lagrangian consists of two basic components: L( x.16) involves a summation of the inﬁnitesimal parts of a path. The solution to the optimisation problem is then given by the EulerLagrange equation: ∂L d ∂L − =0 dt ∂ x ˙ ∂x (2.38 Chaotic Modelling and Simulation us to the following equation: y = (ln x)′ = a − bt ˙ where b > 0. . If. it is in many cases valid to approximate the functions G( x) and V(x) using Taylor series. However. t) = G( x) + V(x) ˙ ˙ where G( x) expresses the forces of growth. as the ˙ integral (2.17) As a speciﬁc example. called the Lagrangian. σ2 = b. The solution to (2. on the other hand. x and t.16) where the functional L.3. one assumes a slightly diﬀerent decreasing pattern for the relative growth rate: y = (ln x)′ = a(ln t)′ − λ ˙ then the solution is related to the Gamma function: x = c(λt)a e−λt 2.15) is a Gaussian function. Then the EulerLagrange equation becomes: d ∂G( x) ˙ ∂V(x) − =0 dt ∂x ˙ ∂x (2.15) where µ = a . and the integration constant c is determined by normalizab tion. especially ˙ when the growth system is continuous and smooth. in terms of x.3 The calculus of variations approach Calculus of variations is concerned with optimising a functional: I= L( x. x. namely: x = ce− (t−µ)2 2σ2 (2. The ﬁrst step in the modelling process in this ˙ case is the construction of an appropriate Lagrangian function.18) The form of the functions G( x) and V(x) is not always known. t) dt ˙ (2.
we are led to the exponential model: y = 0 ⇒ y = b ⇒ (ln x)′ = b ⇒ x = bx ¨ ˙ ˙ As we’ve seen before. the ˙ simplest expression for G( x) is x2 . A reactiondiﬀusion equation has the general form ∂u = f (u) + D∇2 u ∂t . when b < 0 this equation expresses a decay process. and this is suﬃcient in many systems. Equation (2. while when b < 0. which give rise to certain partial diﬀerential equations.19) The simplest interesting expression for V is V(x) = 1 bx2 . Hence.19) then 2 takes a very familiar form: x = bx ¨ When b > 0.4 The probabilisticstochastic approach A lot of models have their roots in probabilisticstochastic considerations.3. This as˙ ˙ sumption leads to a Lagrangian of the form L= 1 2 x + V(x) ˙ 2 Equation (2. 2.17) yields. V(y) = c. y = ln x.18): The ﬁrst order term of a Taylor series expansion of G( x) does not contribute to the equation. this model expresses a growth process.Models and Modelling 39 Note further an interesting property of equation (2. If we consider the Lagrangian as a function of the relative growth rate of a system. it expresses an oscillating process. then the exact same analysis will yield the equation: y= ¨ ∂V ∂y Assuming V is constant. These are usually diﬀusion equations. the following diﬀerential equation of growth: x= ¨ ∂V ∂x (2. in this case. The simple diﬀusion equation (D is the diﬀusion coeﬃcient) ∂2 u ∂u =D 2 ∂t ∂x and the FokkerPlanck equation ∂u ∂(au) ∂2 u =− +D 2 ∂t ∂x ∂x are the basis for many models.
for instance. but also of some earlier time (t − T ). many diﬀerent models arise.20) is not easy to handle. 1962). Simulation of models of this type gives rise to important chaotic cases of pattern formation. where T is the delay. 1961. the Laplacian ∇2 is given by: ∇2 ≡ ∂2 ∂2 ∂2 + 2+ 2 ∂x2 ∂y ∂z A classical model of this type is the socalled FitzHughNagumo model (FitzHugh. Nagumo et al. The general form for a continuous ﬁrstorder delay growth model is: x = f x. σ = 1 and (2.5 Delay growth functions Delay growth models assume that the rate of growth. we obtain the system: x= ˙ bxg(x) 1 + bT g(x) (2. These are examined in some detail in Chapter 4. t ˙ A very simple delay growth model is the following multiplicative process: x = bxt−T g(xt ) ˙ (2.. x = dx . give chaotic solutions for certain parameter values.22) where σ = 1 + bT .9). However. 2. t). When no delay is present. 1969. .21) in (2.22) reduces to the logistic diﬀerential equation (2.20) Equation (2. an approximation of the delay term is possible if we retain the ﬁrst two terms of a Taylor series expansion at t: xt−T ≈ xt − T x ˙ Substituting (2. as usual.40 Chaotic Modelling and Simulation where.20) and rearranging terms. is a function not only ˙ dt of (x.21) Depending on the the function g(x). g is set to be the decreasing function g(x) = 1 − x. then we obtain the GRM1 model proposed in Skiadas (1985) to express an innovation diﬀusion process: x= ˙ bx(1 − x) σ + (1 − σ)x (2. (t − T ). The discrete alternatives of delay growth models. expressed by diﬀerence equations. The onevariable form of this model is given by u = u(1 − u2 ) + K + D∇2 u ˙ where K is a parameter. If. These models explain the development of chaotic waves in ﬂuids and chemical kinetics.3.
which appears on the left side of Figure 2.3(a) and b = 1. (a) (b) FIGURE 2. The equations used are: ft+1 = b ft (1 − ft ) xt = (1 − ft ) cos(θt ) yt = (1 − ft ) sin(θt ) (2. Determining the number of equilibrium islands and their locations inside a chaotic sea is done ﬁrst by using chaotic simulations. when possible.Models and Modelling 41 2.23) The rotation angle θt changes by h = 0. and its use of various methods and tools not applied before. by using the logistic diﬀerence equation and a simple rotation scheme. If we exclude computers and simulations from chaotic analysis. an essential part of the chaotic phenomena.” The innovation of this new ﬁeld lies in its multidisciplinary approach to the problems studied. extensive studies in various ﬁelds led to the creation of a new ﬁeld of analysis. f ). and the chaotic para√ √ meter b is set to b = 1 + 8 in Figure 2.01 in each time step. rotates . will be lost. consider the following formation of two variants of a twodimensional sea urchin.3: Chaotic forms The shape of a snail is formed from the same logistic diﬀerence equation as above.4. that has become an integral part of Chaotic Analysis. called “Chaotic Analysis.3(b).4 Chaotic Analysis and Simulation In the past decades. and only later by exact calculations. 1993). but now the chaotic band (t. Most notable among these is the use of computers for the related analysis and simulation. As an example of the advantages of simulation.1 + 8 in Figure 2. “the essence of chaos” according to Lorenz (Lorenz.
To achieve that. and the time step is h = 0. but for some other values of the chaotic parameters they will give radically diﬀerent ﬁnal values for inﬁnitesimally small changes in the initial values. deterministic equations. due to the addition of an exponential decay factor. for the same initial conditions. and different ﬁnal results. which may give diﬀerent paths. As an illustration of this idea. The system of equations used is: ft+1 = b ft (1 − ft ) yt = (1 − ft ) sin(θt )e−at √ The chaotic parameter used here is 1. given the sensitivity of chaotic models to the initial conditions. From a practical point of view however. in that the ﬁnal values are determined completely from the precise initial values. one could attempt to use the logistic model as a random number generator. The chaotic models can act as deterministic models for some parameter values. which give the same results for the same initial conditions. consider the onedimensional stochastic . in order to form the twodimensional shape of a snail (right).42 Chaotic Modelling and Simulation and simultaneously sinks. xt = (1 − ft ) cos(θt )e−at (2.24) FIGURE 2.005. the unpredictability of the ﬁnal results and the nonuniqueness of the paths followed by the system in chaotic systems are in some ways similar to the unpredictability and nonuniqueness present in stochastic models.4: A snail’s pattern 2. Stochastic models on the other hand are given by stochastic equations.03 + 8. Stochastic and Chaotic Models Deterministic models are expressed by speciﬁc.5 Deterministic. Chaotic models are in a sense deterministic.
25) Here. reminiscent of Brownian motion. Several paths of this process are illustrated in Figure 2. e is a small parameter (e = 0. Another example.5(a).Models and Modelling process generated by the following set of diﬀerence equations: xt+1 = 1 − 2xt2 43 yt+1 = yt + ext (2.5: Chaos as randomness . (a) Onedimensional diﬀusion (b) Twodimensional diﬀusion FIGURE 2. This ﬁgure shows a twodimensional stochastic process based on a fourdimensional mapping.5(b). The resemblance of these paths to those of a Wiener process is striking.17). appears in Figure 2.
what simple model do you ﬁnd that approximates the model dx = cos(bx) for small x? dt 8. Consider various dt values for b. For each of the following diﬀerential equations. Compare this map with the solution to the diﬀerential equation. end up dt dt explaining very frequently reallife situations. including cases with positive as well as negative values for the parameters a. (a) (b) (c) dx dt dx dt dx dt = bx = bx2 = bx2 (1 − x) 6. Using the Taylor expansion techniques. and verify that the approximation is indeed good near x = 0.3 to explain 2 why the very simple models described there. Compare dt your results to the exponential (Malthus) model dx = bx. dt 3. Solve the more general Exponential diﬀerential equation: dx = bxa . Use the calculus of variations methods described in section 2.3. 9. Provide examples and reallife applications modelled by the above two diﬀerential equations. 2. Use a simulation to compare the two models. construct both the “correct” and the “naive” diﬀerence equation analogue. x) map. 4. b. Using the Taylor expansion techniques. 7.44 Chaotic Modelling and Simulation Questions and Exercises 1. as well as values less than 1 and greater than 1. but for negative values of b? What is the diﬀerence between the two cases? . The second order diﬀerential equation x′′ + bx = 0 provides oscillating solutions for positive values of the parameter b. Solve the diﬀerential equation dx = bx2 . Discuss the special cases resulting and draw graphs. Construct the (t. including both negative and positive values. What happens in the diﬀerence equation from the previous problem. Compare this graph to the graph of the solution to the diﬀerential equation dx = bx. show that the diﬀerential equation model dx = sin(bx) may be approximated by the exponential model when dt x is small. Find the diﬀerence equation alternative of this diﬀerential equation and check if the oscillating character is retained in the map form. x) graph for the map: xt+1 = bxt . dx = bx and d 2x + bx = 0. where the dt constant a is both positive and negative. What is the result when b = 1? 10. and graph the (t. and graph the solution. 5.
Models and Modelling 45 11. How much is the maximum e e growth rate? . Conﬁrm that the maximum growth rates for the two Gompertz models are indeed at the points 1 and 1 − 1 respectively.
.
A point of particular interest is the point where the graph of f meets the line y = x. at least for all quadratic maps (Feigenbaum. then b xt+1 = f (xt ) = xt .1(a)). 1979. by considering ﬁrst of all the graphs of the functions y = f (x) = bx(1 − x) and y = x.1 Geometric analysis of the logistic We will present in this section the basic properties of the logistic map based on the classical analysis proposed by Feigenbaum and others. The analysis of this relatively simple iterative equation revealed a rich variety of inherent chaotic principles and rules. and therefore the system is stationary. 47 . 3. we consider the set of points (xt . However.1) Equation (3. with coordinates x = y = 1 − 1 .d. 1983). 1978.1 The Logistic Map The logistic map is given by the logistic diﬀerence equation xt+1 = bxt (1 − xt ) (3.Chapter 3 The Logistic Model 3. Given a value xt . . we will approach the subject from a more geometric perspective. the study of the period doubling bifurcations that appear when the chaotic parameter b changes from one characteristic value to the next led to the formulation of a universal law in order to explain the phenomenon.1(b). .1) via this set of points allows us to examine the logistic system in a geometric way.1. 1980a. we can trace the evolution of the system. If xt is the coordinate of that point. where: yt = xt+1 = f (xt ) = bxt (1 − xt ) Thinking of the system (3. 2. We then use the line y = x to project this y coordinate back to the x axis (Figure 3. as in Figure 3. we locate xt+1 = yt = f (xt ) by locating the point on the graph of y = f (x) with x = xt . In particular.1) has been extensively analysed in the last decades and has become the basis for numerous papers and studies. yt )t = 1.1). . If we continue this process. In order to understand the evolution of (3.
48 Chaotic Modelling and Simulation 1 1 yt 0 0 xt xt+1 + 1 0 0 1 (a) Tracing an iteration (b) More iterations FIGURE 3.1: Tracing iterations Things get more interesting. The resulting point is f ( f (xt )). namely the origin and the stationary point we discussed already. Then we see that xt+2 = xt and yt+2 = yt . and useful. Algebraically. we can perform two iterations.3) has four roots. in other words it is xt+2 . Of particular importance is the inverse function to the logistic map. and then use the inverse logistic to locate x = f (yt ). Suppose now that (x.2) . with equation: x = f (y) = by(1 − y) Suppose we follow steps similar to what we did above. y) = (xt . If (xt . locate yt = f (xt ). except now we use this inverse logistic instead of the line y = x.3) Equation (3. we would have the two equations: y = f (x) = bx(1 − x) x = f (y) = by(1 − y) This system can be solved by observing that: f ( f (x)) = f (y) = x or in our case: x = b2 x(1 − x) 1 − bx(1 − x) (3. Let us consider for a moment the geometric signiﬁcance of the other two points. so the system enters a twopoint orbit. This way. corresponding to the four points where the curves of the logistic map and the inverse logistic map meet. Namely. yt ) is a point on the intersection of the graphs of f and its inverse. yt ) is one of these points. yt+1 ) will be the other (3. then (xt+1 . Two of these will be the points where these maps meet the line y = x. when we consider other curves in place of y = x. starting from a value xt .
e. systems of the form: y = f k (x) x = f m (y) (3. the polygonal segments are related to f and its inverse. The starting value is x0 = 0. tracing the ﬁrst map followed by the second map results in transitioning from xt to xt+k+m . yt+2 ) will be the same as (xt .5. is the function obtained by performing the function k times in succession. the system will alternate between these two states. 2 In this example. and (xt+2 . These are the ﬁrst bifurcation points of the logistic map (Figure 3. An illustrative example of the bifurcation of the logistic map is presented in Figure 3. denoted by f k . In other words. and in general f n+1 (x) = f ( f n (x)). 1 0 0 1 FIGURE 3. higher order bifurcation points can be found by considering the points of intersection of the iterates1 of f . Algebraically.2: Second order stationary points Similarly.2).1. f 2 (x) = f ( f (x)). For instance.2 1 The nth iterate of f .The Logistic Model 49 one. i. . hence each subsequent x point is the result of two applications of f . The ﬁnal equilibrium is at one of the points where the logistic curve joins the inverse logistic curve. yt ) and so on.4) Geometrically.3 when b = 3. we would have to solve a polynomial of degree 2(k + m). We will call these points stationary of second order.
3 Those two curves have 5 points of intersection. .50 Chaotic Modelling and Simulation FIGURE 3.4: The third order cycle of the√ logistic and the inverse logistic map for b = 1 + 8 √ The value b = 1 + 8 ≈ 3. when b = 3.828 is the smallest value of b for which this third order orbit is present.3 FIGURE 3.4 illustrates a third order bifurcation cycle resulting when b = 1 + 8.9.5 for instance presents a chaotic region for the logistic and the inverse logistic maps. but two of these points are the ﬁrst order stationary points.5 √ Figure 3. A large number of repetitions is drawn. The three points in the stationary orbit are the points of intersection of the quartic y = f 2 (x) with the inverse x = f (y).3: The logistic and the inverse logistic map for b = 3. Figure 3. Higher values of b quickly lead to chaotic behaviour.
if b < 1. For a given value of b. . However. 1 The map y = f (x) has two ﬁxed points. xt+1 is even closer to 0. 0) and x0 = 1 − 1 . b > 1 is exactly the range of values of b for which the second ﬁxed 1 point. since then zt+1 = xt − xt2 < xt . 1] × [0. enters the picture. The same is true when b = 1. the graph of the logistic is a parabola passing through the origin (0. we have f (x) = bx(1 − x) = bx − bx2 ≈ bx so if xt is close to 0.2 Algebraic analysis of the logistic We will now proceed to analyse the behaviour of the logistic map from a more algebraic/analytic point of view.1.4 When x is close to 0. 1]. this is not the case any more when b > 1. Keeping in mind that f (x) will be used as the x for the next iteration step.The Logistic Model 51 FIGURE 3. xt+1 is about equal to bxt . xt+1 would tend to move away from 0. (0. The maximum of the b 1 parabola is obtained when x = 2 . and consequently 0 will be an attracting point.9 3. x0 = 1 − b . if starting from any point suﬃciently close to S the map stays close to S .5: Chaotic region for the logistic and the inverse logistic map at b = 3. and y = f (x) = 4 . 1). we see that we can only hope to constrain the system when 0 ≤ b ≤ 4. since it is now a point in the square [0. 1 − b . 4 A set S of points is called attracting. 0) and the point (0. The derivab tives of f at the critical points are: f ′ (0) = b and f ′ (x0 ) = 2 − b We would like to know in which cases these points are attracting. Hence. as can be seen by calculating the ﬁrst derivative of the logistic function: f ′ (x) = b(1 − 2x) and equating it to zero. Incidentally.
i. we consider the diﬀerence from x0 : f (x) − x0 = ( f (x0 ) − x0 ) + f ′ (x0 )(x − x0 ) + f ′′ (x0 ) (x − x0 )2 2 (3. xt ) diagram. the second order diagram (xt .e. i.52 Chaotic Modelling and Simulation We will now carry a similar analysis at this point. the ﬁrst term in (3.6 (b = 2. the ﬁxed point is x0 = 1 − b .7(a). attracting orbit appears. In order to obtain this orbit geometrically. xt+1 ) diagram. the graph of f ( f (x)). the graph of x = f (y). and so: f (x) − x0 ≡ f ′ (x0 )(x − x0 ) Therefore. is: f ′ (x0 ) = 2 − b More generally. equation (3. It can be seen from the graph. . if  f ′ (x0 ) < 1. When b > 3. 1 In our case. f actually equals its second order Taylor expansion.5) becomes f ′′ (x0 ) f (x) − x0 = f ′ (x0 )(x − x0 ) + (x − x0 )2 (3. the graph of y = f (x). x0 is an attracting point for 1 ≤ b < 3.9. and the derivative.5) becomes f (x) − x0 = (x − x0 ) + f ′′ (x0 ) (x − x0 )2 2 and the sign of f ′′ (x0 ) determines whether x0 is attracting or not.6) 2 For x − x0 close to 0. the second order term will be negligible. f (x0 ) − x0 .e.e that x0 is a ﬁxed point.07). In that case. we obtain Figure 3. A Taylor expansion of f at x0 will provide us with some insight into the behaviour of the system near the point:5 f (x) = f (x0 ) + f ′ (x0 )(x − x0 ) + f ′′ (x0 ) (x − x0 )2 2 Since we want to determine if the system will move towards x0 or not. When 5 Note that.5). When f ′ (x0 ) = 1. as computed earlier. An example of the attraction process is illustrated in Figure 3. xt+2 ).5) We are interested in determining for which values of b. is used. a small x − x0 will result in an even smaller f (x) − x0 . in addition to the (xt . When x − x0 is small. then x0 is an attracting point. and the (xt+1 . then x0 is not attracting. With a parameter of b = 2. will dominate the other two. equation (3. that the only ﬁxed points of f ( f (x)) are the two ﬁxed points of x. since f is a polynomial of degree 2. second order. x0 = 0.7. i. the ﬁxed point is no longer an attracting point. if it is nonzero. i. and a twopoint. the Taylor expansion of f around x0 in this case is: f (x) − x0 = (2 − b)(x − x0 ) − b(x − x0 )2 Consequently.e. in order for x0 to be attracting. so. while if  f ′ (x0 ) > 1. it is necessary that f (x0 ) − x0 = 0.
4495. and remains attracting as long as b is such that the slope of the order2 graph at the attracting points does not exceed −1. This leads to the equation b3 − 4b2 + 8 = 0 b2 x(1 − x) (1 − bx(1 − x)) = x (3.4 = 2b It is clear now why these points appear only when b ≥ 3. Figure 3. Figures 3. Another point to consider is the value of x at which the inverse of the logistic curve cuts the logistic curve at its maximum. This leads to the equation: b2 − 2b − 5 = 0 √ which provides the upper value of b. with slope very close to −1.8(a) and 3. we can compute the second order orbit by solving the equation f ( f (x)) = x that is: Equation (3.7(b)). When b > 3. b = 1 + 6 ≈ 3. One can also clearly see that the ﬁxed point of y = f (x) is no longer an attracting point.43 and x0 = 0.The Logistic Model 53 FIGURE 3. The slopes of the 3 3 tangent lines to the curves at this point are −1 for y = f (x) and x = f (y). and 1 for y = f ( f (x)).7) has four roots: The two ﬁxed points of f (x1 = 0 and x2 = 1 − 1 ).8(b) illustrate this order2 bifurcation. The order2 bifurcation appears in the form of a square with corners located at the 4 characteristic points of the diagram. This twopoint orbit is attracting for b > 3. for which this twopoint orbit is attracting. b as well as two new roots.8(a) shows the case where b = 3. all the curves pass through the point 2 . given by: √ (b + 1) ± (b + 1)(b − 3) x3. The tangent line to the order2 graph at the larger of these points appears. 2 (Figure 3.6: The logistic map b = 3.7) .01.
7 (b) b = 3 FIGURE 3.54 Chaotic Modelling and Simulation (a) b = 2.7: The logistic map.8: The logistic map. b ≤ 3 (a) b = 3. b > 3 .43 (b) b = 1 + √ 5 FIGURE 3.
5440903 · · · < b < 3.9 is the real time diagram (xt . xt ) diagram.8(b) √ illustrates the case where b = 1 + 5.5696916 · · ·.6 Figure 3. The parameter b is con√ strained in the range 1 + 6 < b < 3. FIGURE 3. real time and phase space diagrams The right of Figure 3.5440903 · · ·.The Logistic Model 55 √ which has three real roots.5644073 · · · the period8 bifurcation appears. xt+4 ) diagrams. The range for b is 3. xt+4 ). When 3. The range covered by the parameter b is 3. The 4point orbit can be thought of either as the points of intersection of the (xt . b = 2 and b = 1 + 5. The period4 bifurcation appears in Figure 3. and in order to locate the characteristic points. For better presentation.9. The order8 logistic bifurcation is expressed by the (xt .5687594 · · ·.56969). oscillate between theses four limiting levels. Two of them are acceptable.11.25. or as the points of intersection of the xt+3 . The period16 bifurcation is illustrated in Figure 3. A real time diagram of the chaotic oscillations appears on the right of Figure 3.5687594 · · ·.10 illustrates this case.5644073 · · · < b < 3. and the inverse function is given by the (xt+7 . with x0 = 0.12 presents the period32 bifurcation (b = 3. Figure 3. . There are 4 characteristic values for x.9: Order4 bifurcation of the logistic map. after enough time has elapsed. For the application we set b = 3. t). then asking for f (y) = x is the same as asking for f ( f (x)) = x. we introduce the inverse (xt+3 . In the real time diagram the order4 oscillations appear.5687594 · · · < b < 3. xt+4 ) diagram (purple line) with the line y = x. The values of xt . xt ) and (xt . 6 Since if y = f (x). Figure 3. whereas the left is the phase space diagram (xt . xt diagram (blue line) with the curve y = f (x). The chaotic parameter is b = 3.5644.10. xt+8 ) diagram. These are the same points where the order2 curve meets the main diagonal.
10: Order8 bifurcation of the logistic map. real time and phase space diagrams FIGURE 3.56 Chaotic Modelling and Simulation FIGURE 3.12: Order32 bifurcation and real time and phase space diagrams of the logistic .11: Order16 bifurcation and real time and phase space diagrams of the logistic FIGURE 3.
14: The order6 bifurcation and real time and phase space diagrams . The chaotic oscillations appear in the real time diagram (right). xt+6 ) diagram. The chaotic oscillations appear in the real time diagram (right). The order6 logistic bifurcation is expressed by the (xt .13: The order5 bifurcation and real time and phase space diagrams The order6 bifurcation appears in Figure 3.14 (b = 3. xt ) diagram. The inverse function is given by the (xt+4 . The order5 logistic bifurcation is expressed by the (xt . FIGURE 3.The Logistic Model 57 The order5 bifurcation appears in Figure 3. xt ) diagram. The inverse function is here given with the (xt+5 .13. FIGURE 3.6267). xt+5 ) diagram.
is due to a quantitative convergence of the parameter b. At this point. These points can be calculated by solving the appropriate equation of the form f n (x) = x. The characteristic values of the parameter b at each bifurcation stage follow the Feigenbaum law.449 becomes unstable and splits up into an order4 orbit and so on.15(b). For 1 < b < 3. which tends to a universal parameter δ = 4. under some mild conditions on the map f . to achieve stability much quicker. though this becomes exceedingly diﬃcult as n increases.54409 and so on. and perform repeated applications of the iterative process xn+1 = f (xn ) until stability is achieved. which at b 3. This parameter is given by the expression bi−1 − bi δ = lim i→∞ bi − bi+1 The behaviour of the system can be understood better by considering a scaled version. Use the previously found stable x values as starting points for the iteration process. if we consider a system xn+1 = f (xn ) then we consider the function: g(x) = lim n→∞ 1 n n f 2 x f 2 (0) f 2n (0) Then. the ﬁxed point is x = 1 − 1 .6692016 · · ·.2 The Bifurcation Diagram The period doubling and other properties of the logistic map are presented in the socalled bifurcation diagram.4495. More generally. presented in the above ﬁgure by a bifurcation tree. The ﬁrst bifurcation point is located at b = 3. b the next four at b ≈ 3. The full bifurcation diagram for the logistic map appears in Figure 3. there is only one ﬁxed point. this splits up into the ordertwo orbit. 4]) into small intervals.5. 2. The greyed out lines show the values that these orbits would have had if they were still stable. 3. choose a b in that interval (say the left endpoint). According to this law the sequence of period doubling. Break the range of b ([0. xt ) formed as follows: For each value of the chaotic parameter b. at x = 0.8) .15(a) shows how the bifurcation diagram looks for b < 3. Figure 3. the values of xt that form the stable orbit are plotted. x = 1 − 1 . g satisﬁes the universal equation: g(αx) = −αg g (−x) (3. Move on to the next value of b. and the next two at b ≈ 3. according to the theory.58 Chaotic Modelling and Simulation 3. At b = 3. This is a twodimensional graph (b. For each interval. the desired values xn will be determined. The construction of the bifurcation diagram typically follows these steps: 1. For 0 < b < 1.
The Logistic Model 59 1 0 0 1 b 3 3.449 4 (a) The ﬁrst two bifurcations (b) The whole diagram FIGURE 3.15: The bifurcation diagram of the logistic .
and from that it is possible to estimate α = 2. related to the ratio of the width between the prongs in a bifurcation. the ﬁrst period doubling occurs at c = − 4 . (a) Order3 bifurcation (b) Order 2 × 3 bifurcation FIGURE 3. which corre5 sponds to b = 3.16: Order3 bifurcations A change of variables 1 b(1 − 2x) 2 results in the following expression for the logistic equation: z= zt+1 = z2 + c t where c= b b2 − 2 4 3 From this point of view.74 and b3 = 1 + 8.845). The period3 bifurcation can be seen in Figure 3. ﬁfth and third order bifurcations respectively. These values of b give sixth. 10 and 6 and higher order bifurcation forms. 1982a). while the period3 solutions appear at a saddlenode bifurcation at .60 Chaotic Modelling and Simulation Here α is the second Feigenbaum constant. b5 = 3.16(b) the order 2 × 3 = 6 bifurcation is illustrated (b = 3. The second period doubling occurs at c = − 4 . The function g can be approximated by a ﬁnite polynomial. and the related period doubling bifurcations lead to 12.16(a). which corresponds √ to b = 1 + 6. with the normalisation convention g(0) = 1 (Lanford.50290787 · · ·. Other interesting properties arising from the bifurcation diagram come from observing the three “windows” which appear in a range of √ characteristic values of the parameter b around b6 = 3. In Figure 3.627. relative to the width in the previous bifurcation.
FIGURE 3. 1976) to describe the population growth of a single species population.3 Other Models with Similar Behaviour Functions with varying equation forms show chaotic behaviour similar to the logistic. right). 1825): xt+1 = −bxt ln(xt ) where 0 < b ≤ e = 2. . One of them is the sinusoidal map. 1974. Another model is the Gompertz type model (Gompertz.18 (r = 2. with equation xt+1 = b sin(xt ) where b is restricted to the interval 0 < b ≤ π. A more convenient form is: xt+1 = b sin(πxt ) Now the interval for the parameter is 0 < b ≤ 1. which is regulated by an epidemic disease at high density: xt+1 = xt er(1−xt ) The chaotic behaviour of this model is similar to that of the logistic model.6).18.The Logistic Model 61 √ 7 c = − 4 . as illustrated by the bifurcation diagram (Figure 3. Another model was proposed by May (May. The order4 bifurcation is presented on the left of Figure 3. corresponding to b = 1 + 8.17: The order3 bifurcation 3.71829 · · ·. An illustration of the period3 bifurcation using the zform of the logistic equation appears in Figure 3.17. 1972.
a = 7. A rich bifurcation diagram appears when the Gaussian model is combined with the logistic to form the GL model: xt+1 = b − xt (1 − xt ) + e−axt 2 When a = 7. gives a full sequence of period doubling bifurcations and order3 saddlenode bifurcations (Figure 3. with period doubling bifurcations and order3 and order5 saddlenode bifurcations. A simple Gaussian equation is: xt+1 = b + e−axt 2 The bifurcation diagram for a = 4 and −1 < b < 1 is shown on the left of Figure 3. When the parameter a takes higher values. right).7 has the form illustrated in Figure 3. . There are two main bifurcation regions.19(a).5.11). the chaotic behaviour of the model changes radically. This is presented in Figure 3.5.20 (a = 17.22(a) (a = 7.18: The May model: an order4 bifurcation (left) and the bifurcation diagram (right) 3. Figure 3. −1. The order5 bifurcation is presented in Figure 3.1 < b < 0.19). The model has two period doubling bifurcations. the bifurcation diagram in the range −1.5. b = 0.21.5.19(b).2 < b < 1).4 Models with Diﬀerent Chaotic Behaviour Some Gaussian models show special chaotic behaviour. b = 0.22(b) illustrates a perfect order12 circle (a = 7. mainly due to the quadratic exponent in the equations of these models.62 Chaotic Modelling and Simulation FIGURE 3. A further increase in the parameter a of the quadratic term. and then it returns to the original state in a reverse order.
21: Bifurcation diagram of the GL model .19: Bifurcation diagrams of the Gaussian model FIGURE 3.20: Bifurcation diagram of the Gaussian model (a = 17) FIGURE 3.The Logistic Model 63 (a) a = 4 (b) a = 7.5 FIGURE 3.
1987. Time series data expressing the cumulative percentage of steel produced by the oxygen process in various countries are used. Here.22: Cycles in the GL model 3. 1977. and this explosion of innovation activity presents signiﬁcant challenges.S. the chaotic behaviour is expressed through the discrete alternative to the continuous GRM1 model. Several innovation diffusion models have been presented. Characteristic graphs of the chaotic behaviour are given and applications are presented. 000 in 2006 (Figure 3. The model shows symmetric and nonsymmetric behaviour expressed by a parameter σ. When the diﬀusion parameter b and the σ parameter are in the range b/σ ≥ 2. with the total number of patent applications exceeding 450.1 GRM1 and innovation diﬀusion modelling It has become commonplace to call this the information age. Mahajan and Schoeman. but an even more appropriate name might be the innovation age. Companies must have an appropriate way to describe the competitive dynamics in a market (Modis. analysed and applied in Skiadas (1985). .64 Chaotic Modelling and Simulation (a) Order5 bifurcation (b) Order12 cycle FIGURE 3. The deterministic continuous version of this model was proposed.5 The GRM1 Chaotic Model In this section we study the chaotic behaviour of the generalized rational (GRM1) innovation diﬀusion model. analysed and applied to real life data (Bass. A method is proposed for ﬁtting the model to the data.23). While not all patents translate to new products or new production methods. 1985. The number of patent applications that the U. 1976. 1969. this growth clearly demonstrates a tendency. then the chaotic aspects of the model appear.5. 1997). Sharif and Kabir. Patent and Trademark Oﬃce receives every year has been increasing exponentially since 1985. Skiadas. 3. 1986.
by using the logistic model it is not possible to express chaotic behaviour in real situations. data provided for various cases shows that . The GRM1 model has a point of inﬂection. lower than 1. A main direction of these applications focused on the nonsymmetric behaviour of the models. Another interesting property of σ is that it provides a measure of the asymmetry of the model. On the other hand. various applications of the logistic model in several disciplines showed that the parameter b of the logistic model lies in very low limits. b is the diﬀusion parameter. as the estimated values of b fail to reach the limit at which chaotic behaviour appears. 1986).The Logistic Model 65 G 400 G Patents (thousands) G G G G 300 G G G G G G 200 G GG GG G G G G G 100 G GG GG G G G G GGGGGGGGGGGG G 1970 1980 1990 2000 Year FIGURE 3. and σ is a dimensionless parameter. The GRM1 model is based on a family of generalised rational models. Thus.10) f˙ = b f 1 − F In the logistic model. F is the total number of potential adopters. bifurcation and further chaotic behaviour appear when 2 < b ≤ 3.9) where f is the number of adopters at time t. It was proposed as a relatively simple but very ﬂexible model to express asymmetry during the innovation diﬀusion process (Skiadas. This model is expressed by the following diﬀerential equation f˙ = b f (F − f ) F − (1 − σ) f (3. expressed by speciﬁc parameters. 1992).23: Patent applications in the United States Modis and Debecker. Perfect symmetry appears for σ = 1 when. 1985. equation (3. However. varying from 0 to F as σ decreases from ∞ to 0.9) reduces to the logistic: f (3.
oscillation and chaotic behaviour appear by gradually augmenting the b fraction σ .24(e). and 3. which follows. As we show here. which gives a measure of the asymmetry of the model. and especially when the diﬀusion process is close to the upper limit F. In all cases presented here the starting value is f0 = 1.5.11) 3.5. the resulting condition on the parameters b and σ is: b =2 σ b When σ > 2. when the logistic model is applied in the form xt+1 = bxt (1 − xt ) (3. The chaotic behaviour of the model is analysed and illustrated by using appropriate graphs.9) is given by: ft+1 = ft + b (3. In Figure 3. It is also possible to obtain an analytical form for the values of ft between the ﬁrst two bifurcation points.24(e).12) Some interesting properties of this model are illustrated in Figures 3. the proposed model takes the classical sigmoid form.24(f) a totally chaotic form appears. the GRM1 model exhibits chaotic behaviour for values of b that are quite low and are in accordance with the values estimated in real situations. b = 0. The presence of the ﬁrst oscillations and the onset of chaos. To achieve this we consider the equation ft+2 = ft . a bifurcation appears as a simple oscillation. f ) diagrams. a more complicated oscillation with four distinct oscillating levels appears. bifurcation appears for b > 2.66 Chaotic Modelling and Simulation where xt = ft /F. then bifurcation and chaotic behaviour appear at even higher values of b. In Figure 3. whereas in Figures 3. According to the theory of chaotic models. when 3 < b ≤ 4. The value selected for b is within the range 0.12). the upper limit F = 100. several forms of the model appear. oscillations and chaotic behaviour appear quite frequently.13) ft+1 = ft Using equation (3.3 and σ takes various values.24(a) to 3. whereas in Figure 3. which is the case for the logistic model. especially (t.24(b). which is valid in real situations.24(a). 3.1 to 0. When σ = 1.24(d). By varying the dimensionless parameter σ.24(c). bifurcation for the GRM1 model starts when: ′ ft+1 = −1 (3. Moreover. It is very important to consider the estimation of the values of the parameters b and σ for which bifurcation appears. This is accomplished with the help of the ﬂexible parameter σ. is a signiﬁcant factor when studying innovation diﬀusion systems.2 The generalized rational model ft (F − ft ) F − (1 − σ) ft The discrete version of the continuous model (3.
24: The GRM1 model .3 and σ = 0.3 and σ = 0.13 (c) b = 0.3 and σ = 0.3 and σ = 0.08 FIGURE 3.The Logistic Model 67 (a) b = 0.3 and σ = 2 (b) b = 0.12 (d) b = 0.3 and σ = 0.10 (e) b = 0.09 (f) b = 0.
68 Chaotic Modelling and Simulation The exact formula of the solutions is: (b + 2) ± b(b+2)(b−2σ) (b−2σ+2) ft = F 2(b + σ − 1) (3. until the sum of squared errors is minimised. This parameter is gradually changed during the iterative procedure.15). given a set of initial values for the parameters ai . or b > 2 in (3. When b is higher than these values.3 Parameter estimation for the GRM1 model The parameters of the discrete GRM1 model are estimated by an iterative nonlinear regression analysis algorithm by minimizing the sum of squared errors (S = S S E): n S = ǫt2 = t=1 (yt − ft )2 where ǫt is the error term of the stochastic equation: n yt = ft + i=1 ∂ ft ∆ai + ǫt ∂ai where yt denotes the provided data and ft is calculated for every t from the GRM1 equation. four distinct oscillating levels appear. n is very high and the system exhibits chaotic oscillations. The starting values for the partial derivatives need an estimation of the following forms given a set of initial values for the model parameters: f0 (F − f0 ) ∂ f1 = ∂b F − (1 − σ) f0 2 F 2 − 2F f0 + (1 − σ) f0 ∂ f1 =1+b ∂ f0 (F − (1 − σ) f0 )2 f0 ∂ f1 = bσ ∂F F 2 (3.5. 3. and later eight and ﬁnally 2n points.16) . the system oscillates at the values of ft given by the above formulas respectively.14). it was decided to use the nonlinear estimation method proposed by Nash for the discrete logistic model for only three parameters of the model.15) When b > 2σ in (3. For sufﬁcient speciﬁcally high values of b. and ﬁxing the dimensionless parameter σ.14) For the logistic model (σ = 1) this reduces to: √ (b + 2) ± (b + 2)(b − 2) ft = F 2b (3. For a better ﬁt. The estimation of parameters is highly sensitive to the presence of oscillations and chaotic oscillations in the provided data.
. Before the introduction of this method. was used instead.25: Spain. pure oxygen is introduced in order to burn the carbon inside melted steel. so as to improve the quality of the produced steel.The Logistic Model 69 FIGURE 3. the dotted line is the path of the logistic model and the solid line is the path of the GRM1 model. The actual data includes 18 years. Figure 3. as this part shows the characteristic oscillations that are of special interest to us. Parameter estimates and the sum of squared errors are summarised in Table 3.4 Illustrations Time series data expressing the cumulative percentage of steel produced by the oxygen process7 in various countries are used from Poznanski (1983). the iterative procedure continues the estimation by using the formulae: ft (F − ft ) ∂ ft+1 ∂ ft = (1 + bkt ) + ∂b ∂b F − (1 − σ) ft ∂ ft+1 ∂ ft = (1 + bkt ) ∂ f0 ∂ f0 bσ ft2 ∂ ft+1 ∂ ft = (1 + bkt ) + ∂F ∂F (F − (1 − σ) ft )2 kt = F 2 − 2F ft + (1 − σ) ft F − (1 − σ) ft 2 (3.25 illustrates the diﬀusion of oxygen steel technology in Spain from 1968 to 1980.1. The small circles indicate the actual data. air. but it is more appropriate to study the last part of the time series. The parameter b for the logistic model ﬁt is relatively high. but is still far from the 7 In this process. oxygen steel process (1968–1980) Following the estimation of the initial values of the partial derivatives.5. consisting of a mixture of oxygen and nitrogen. for a period of 13 years.17) where 3.
The ability of the GRM1 model to follow the oscillating behaviour of actual data is illustrated in the Figure 3. as the fraction b/σ = 3. the GRM1 model has a value for b lower than that of the logistic model. The process ends in an oscillating form. as it covers both the fast growth at the ﬁrst stages of the diﬀusion process.609 accounts for the chaotic region of . as the fraction b/σ = 2. as well as the sudden turn to the high platform of 100%.70 Chaotic Modelling and Simulation TABLE 3. from 1. whereas the discrete GRM1 shows a considerable ﬂexibility in approximating the data.474 — 72.27). but the extra parameter σ accounts for the presence of oscillating and chaotic behaviour. the small ﬂuctuations at the end of the process are simulated quite well. The model fails to express the oscillating behaviour of the actual case studied. Data for the diﬀusion of the oxygen steel process in Luxemburg are of considerable interest. and is also expressed by the strong improvement of the sum of squared errors (SSE). Also. The estimated values for l and F are very close for both models.5292 for the GRM1 model accounts for the chaotic behaviour.838 GRM1 0. The fraction b/σ = 3.6309 24. The form of the logistic path presented in the Figure 3. The discrete logistic fails to express these oscillations.2.775) 41.736 0. The GRM1 model showed a good ﬂexibility.2331 25. Figure 3.748 FIGURE 3.5% during 1962 to 100% in 1980 (Figure 3. The sum of the squared errors is very low for the GRM1 model.084 (2. compared to that of the logistic. as is shown in Table 3.25 has a smooth form.26 illustrates the diﬀusion of oxygen steel technology in Italy from 1970 to 1980.26: Italy oxygen steel process (1970–1980) value needed for the start of bifurcation (b = 2). On the other hand.1: Parameter estimates and sum of squared errors (SSE) for logistic and GRM1 models in Spain from 1968 to 1980 Model b l F σ(b/σ) SSE Logistic 0.779 51.25. as they cover a wide scale.373 51.775 > 2.
5292) 7. as well as chaotic oscillations.0535.957 44. the analysis of the logistic map shows that bifurcation.04046. are demonstrated in the following case of the diﬀusion of oxygen steel technology in Bulgaria from 1968 to 1978 (Figure 3. But. namely 1 = 100%.6 Further Discussion Several problems arise when applying the logistic map to real world data.The Logistic Model 71 TABLE 3.28). there are many cases where nonchaotic or even nonoscillating behaviour is present.669 and σ = 0. l = 49.3718 indicates that the model is in the region of chaotic behaviour.2425.2: Parameter estimates and sum of squared errors (SSE) for logistic and GRM1 models in Italy from 1970 to 1980 Model b l F σ(b/σ) SSE Logistic 0. l = 7. 3.0402 44.08823 36. Firstly.27 illustrates the case of Luxemburg for the following estimated values for the parameters: b = 0.330 GRM1 0. F = 58.431 and the fraction b/σ = 3. and thus oscillat .431 FIGURE 3.1931. The ﬂexibility and ability of the GRM1 model to simulate growth processes that show oscillations.012.968. at the end of the process. The sum of squared errors is S S E = 21. F = 99.5447 35.473 — 15.4614 0. The mean squared error is MS E = 20. Figure 3. even when the sigmoid logistic shape approaches the highest values.872.025 (3.27: Luxemburg oxygen steel process (1962–1980) the model.412 and σ = 0. The estimates for the parameters are b = 0.
Especially in diﬀusion and innovation diﬀusion processes. The second problem when applying the logistic map to timeseries data is that. for b = 1. Another approach is to select a transformed logisticlike model. this is not the case. for e = 0.10.30 A chaotic process at b = 1. where the term b/(1 − (1 − σ)x) is replacing the parameter b.073 for the modiﬁed logistic model is presented in Figure 3. and. The next time period of the process must be 0. The model approaches the equilibrium limit 0. The modiﬁed logistic model (ML). is illustrated in Figure 3. There are very few growth processes that can cover the gap between 10% and 72.725.3 and e = 0.27. then. the intermediate stages of the process are far apart to give reasonable results during simulation. This is the method employed in the GRM1 model. when a high value for the parameter b is introduced. but eventually the process reaches very high values.29.59 and 0. For e = 1.31 . the development is slower. However.3 and e = 0. An oscillating form of the modiﬁed logistic at b = 1. If we follow the logistic model.09. all the processes exceeding this level must show oscillating or chaotic behaviour when approaching a higher level. oscillating or even chaotic behaviour appears. consider a logistic process with b = 3 starting at x = 0.72 Chaotic Modelling and Simulation FIGURE 3. it reduces to the exponential model. this model reduces to the logistic model. As a simple example.66% level). or x = 2/3 (66. followed by 0.28: Bulgaria oxygen steel process (1968–1978) ing behaviour.3 and e = 0. whereas.076 is presented in Figure 3. This is the minimum level where the ﬁrst bifurcation starts. One approach is to use a logistic model with a varying function for the parameter b. starts at b = 3. The ﬁxed point is at x = 1 − 1/b(1/e) . Many logisticlike timeseries data approach high levels without showing oscillating or chaotic behaviour. which corresponds to x = 1 − 1/b.5% of the total process in only three time periods. This is easily achieved by selecting a modiﬁed logistic model of the form: xn+1 = bxn (1 − xn )e where 0 < e < 1.9458.
The Logistic Model 73 FIGURE 3.31: A chaotic stage of the modiﬁed logistic model (ML) for b = 1.29: The modiﬁed logistic model (ML) FIGURE 3.076 FIGURE 3.073 .3 and e = 0.30: The modiﬁed logistic model (ML) for b = 1.3 and e = 0.
Accordingly. During the ﬁrst time periods the process follows a growth path.33. the amplitude of the oscillations varies according to the values of the parameters selected. Later on. giving third order chaotic oscillations. the GRM1 model is more ﬂexible. The maximum value is 0.1494 and 0. The interval of this third order cycle is included inside the chaotic region as illustrated in the bifurcation diagram.74 Chaotic Modelling and Simulation FIGURE 3. In Figure 3.3 and σ = 0.3. σ = 0. .32).01175. The third order cycle for the modiﬁed logistic model is illustrated in Figure 3.2 and σ = 0.32: A third order cycle of the GRM1 model for b = 1. During the third order cycle. The (x.8069 and the maximum value is xmax = 0. for lower values of the parameter σ the amplitude of the oscillations during the third order cycle is smaller. valuable information for the chaotic region can be obtained. the GRM1 and the modiﬁed logistic models (Figure 3. the higher value of the third order cycle is xmax = 0.32. A comparative example is given by observing the third order cycle for the logistic.7047. By observing this third order cycle.025. The resulting third order oscillations cover 20% of the total amplitude. For this value of σ the GRM1 model shows a third order cycle for b = 1.9768 and the lower value is xmin = 0. that is.34. chaotic oscillations appear and the process is stabilised.7846.0719 and b = 1. in the cases of the GRM1 and the modiﬁed logistic. The oscillations during the third order cycle can vary according to the values selected for the parameter σ.025 Another serious problem when applying the logistic model is that the oscillations following the bifurcation process are quite large for most real life situations. The oscillations therefore cover only 27% of the total amplitude. On the other hand. the oscillations cover about 80% of the total amplitude. The minimum value is xmin = 0. The parameters selected are e = 0. t) oscillation diagram is presented in Figure 3.9987. When the parameters for the GRM1 model are b = 1. the logistic model oscillates between 0.9594.3. and the minimum value is 0.9693. The third order oscillations cover 19% of the total accepted amplitude of the process. On the contrary.
33: A third order cycle of the GRM1 model for b = 1.0719 FIGURE 3.35: The ﬁrst bifurcation cycle of the GRM1 model for b = 1.032 .34: A third order cycle of the modiﬁed logistic model for b = 1.2 and σ = 0.3 and σ = 0.3 and e = 0.The Logistic Model 75 FIGURE 3.01175 FIGURE 3.
76 Chaotic Modelling and Simulation Figure 3. bx(1 − x) 1 − (1 − σ)x (3. The parameters are b = 1.4 = 2b It is clear that the equation expressing every higher bifurcation level includes the solutions of the equations expressing the lower bifurcation stages. What is diﬀerent is the stability of solutions. this fourth order equation has the two roots from (3. giving the values for the ﬁrst bifurcation levels.18) The resulting equation for The roots of this equation are x1 = 0 and x2 = xn+2 = xn = x can then be transformed to x x− b−1 × b−1+σ × b(b + 1 − σ)x2 − (b + 1 − σ)(b + 1)x + (b + 1) = 0 (3.19) Clearly.18) as well as two other roots: (b + 1) ± (b + 1)2 − 4b(b+1) b+1−σ x3.3 and σ = 0. The fourth order equation for x has. among others. The process ends at a simple oscillation scheme. . The maximum and minimum levels of the oscillation are illustrated by the lines with two and one dots respectively.35 illustrates the GRM1 model at the ﬁrst bifurcation stage. The resulting fourth order equation can be solved explicitly. the roots that satisfy the relation xn+1 = xn = x or. These levels are obtained when xn+2 = xn = x. after substituting from the GRM1 map x= or after some simpliﬁcation: x x− b−1 =0 b−1+σ b−1 b−1+σ .032.
draw the bifurcation diagram and ﬁnd the ﬁrst and the second bifurcation points. as given by the random number generator of your computer. 3 4 . 6. start from an arbitrary value of x. The logistic map is deﬁned on the interval: 0 < xt ≤ 1. 5. 000 repetitions.7). For this latter case.The Logistic Model 77 Questions and Exercises 1 1. Show that this is indeed the case. Prove that the Gompertz map is deﬁned into the interval 0 < xt ≤ 1 . that. starting from an x in the interval [0. 8. 9. i. and ﬁnd the interval where every map is deﬁned. 3. Consider the correct diﬀerence equation analogue to the logistic model. and perform 100. and compare the results with the previous case. Compare the distribution of the 100. Find the related interval for the map xt+1 = 1 + axt2 . how do you explain these diﬀerences? 4. Draw the bifurcation diagram for the map xt+1 = b + e−axt . e 7. where f (x) = bx(1 − x) at the 2point orbit given in equation (3. Transform the Gompertz and the Mirror Gompertz model to the related discrete maps. Draw the bifurcation diagram for both cases and ﬁnd the ﬁrst bifurcation point. the process will remain in the interval. Compute the derivative of the transformation y = f ( f (x)). 000 values to that of the uniform distribution. 1]. Show that the ﬁxed point of the map y = bx(1 − x) is y = x = 1 − b .e. 2. If you found diﬀerences. and determine from this the range of values of b for which this orbit is attracting. Use the logistic map as a random number generator: For various high values of b. as described in page 31. Find the interval where the map is deﬁned and ﬁnd the formula for the ﬁrst and second order bifurcation points. Draw the bifurcation diagram for the map xt+1 = b + e−axt .
.
Glass and Mackey. . 1988. and for initial values x1 = 0.1 Delay Diﬀerence Models Simple delay oscillation scheme Simple oscillation schemes can arise very easily when using delay diﬀerence equations. 1997. Sharkovsky. Special attention is given to the development and presentation of the most frequently used delay models..3) When a = −1. 1978. Of major importance are the applications of delay models in physiological systems (Glass and Pasternack. 1991. . x2 = 0. Cabrera and de la Rubia. and then proceed to study more complicated models. 1988. 1999). 1987.1. The hexagon 79 .1 Introduction Delay models are applied in a large variety of cases in many scientiﬁc ﬁelds. Ueda et al. .2 4. xt ) where m is the delay. 1996. consider the simple ﬁrstdegreepolynomial delay form: xt+2 = xt+1 + axt (4.2. As an example. (4. Glass. Delay models can be divided in two broad categories: The delay models expressed by a diﬀerence equation xt+m = f (xt+m−1 . 1994. xt+m−2 . xt−T ) ˙ where T is the parameter expressing the delay. Bunner. . the resulting twodimensional map has the hexagonal form illustrated at the top left of Figure 4.Chapter 4 The Delay Logistic Model 4. it is diﬃcult to classify the particular models used and to study them systematically.3. In this section we start with the study of simple delay models. Celka. and the delay models expressed by a diﬀerential equation x = f (xt .2. Glass and Zeng. as well as to related extensions and new delay models of particular interest. de Olivera and Malta.1) 4. Consequently. 1990.2) (4. 1994.
1): xt−1 xt+1 = xt + a √ xt + xt−1 Figure 4. FIGURE 4. 1].1 can be obtained by introducing a Euclidean distance term in equation (4. uniformly chosen from the interval [0.1). This iterative scheme generates a discrete oscillation over time (bottom of Figure 4.2 shows the corresponding twodimensional map for a negative value of a. related to the size of the graph. FIGURE 4.80 Chaotic Modelling and Simulation at the top right is formed by considering random initials values (x1 .2: Variation of the simple delay model . x2 ).1: A simple delay oscillation scheme (a = −1) A variation of the hexagon in Figure 4.
The distance between two successive oscillations depends on the delay parameter: The higher the delay. during which the system takes zero values. (The parameters are m = 1. for m = 0 equation (4.4(a).3(b). (a) m = 1 (b) m = 10 FIGURE 4. These are the maximum values for b for which the model makes sense.44413).1 summarises the values of b used in our simulations.The Delay Logistic Model 81 4. The m = 5 case is even more interesting.3: The delay logistic model The special case with delay m = 1 has an interesting property. which can be seen when examining the left corner (corresponding to the origin) of the (xt−m .3(b)).4) are of a particular form.3(b). xt−m ) diagram can be seen at the top of Figure 4.2 The delay logistic model The delay logistic model is given by the equation: xt+1 = bxt (1 − xt−m ) (4. the higher the distance. The oscillations of the map (4. The resulting (xt . The simplest case is illustrated in Figures 4. xt ) diagram closer. Clearly.4) It has several interesting properties. There is also a marked time period. mainly. xt ) diagram at the bottom. Each oscillation is followed by a linear term with value of zero. Table 4.4(b). The graph exhibits a small “fold. b = 1.3(a) and 4. b = 2. .2865 respectively).2. b = 1.4) reduces to the logistic equation. The main diﬀerence when the delay is present is the formation of a limit cycle (Figure 4. with three folds (Figure 4. on the values of the delay parameter m.271 and m = 10.” as illustrated in Figure 4. depending of course on the values of b. with the (t. but.
5) where T > 0 is the time delay. The delay increases along with a.8395 1. and 2 for a > π . we can perform a simulation of the above simple time delay equation. 2 Using a linear spline approximation method.52195 1.6) The transformation T = at turns (4.4: Folds in the delay logistic model TABLE 4.3 Time Delay Diﬀerential Equations The general form of a time delay diﬀerential equation is given by x = f (xt . whereas for a < π solutions converge to zero.6393 1. The simplest time delay diﬀerential equation form is the linear equation x = −axt−1 ˙ (4. xt−T ) ˙ (4. A very interesting observation here is that the parameter a is directly related to the delay.1: Limiting b values for the delay logistic m 1 2 3 4 5 ··· 10 b 2. solutions diverge to inﬁnity (Hoppensteadt.82 Chaotic Modelling and Simulation (a) Delay logistic (m = 1) (b) Delay logistic (m = 5) FIGURE 4.44413 · · · 1.271 1. 1993). The key step in this method is to take the delay . Theory suggests that for a = π the delay 2 equation has periodic solutions.6) into y = −yT −a ˙ where yT = xT/a .2865 4.
a delay equation with delay m is equivalent to an mdimensional system.The Delay Logistic Model 83 interval T and divide it into n equal parts. Figure 4. . yi = −n(yi − yi−1 ) ˙ . xt+1 ).5 illustrates the twodimensional Poincar´ e map. the accuracy of the method is good.5: A simple delay model (n = 15) The behaviour of the simple oscillating scheme simulated above shows that oscil1 A Poincar´ map is a twodimensional representation of a multidimensional ﬂow of a nonlinear e system. Geometrically. (4. FIGURE 4. 1895. . In our case.7) yn = −n(yn − yn−1 ) ˙ This system was solved by using a fourth order RungeKutta method of numerical integration.1 and the periodic solutions of the system when a approaches π . The selected time increment h must be suﬃciently small. (xt+m−1 . Several iterations were tried out for various values of n. . (xt . The method is widely used and. for instance (xt . xt+m ). showing us diﬀerent aspects of the ﬂow. gives good results. and several Poincar´ e maps may be selected. this map results by cutting the curves of a multidimensional ﬂow by a plane. . . xt+2 ) and so on. xt+m ). When n = 15 and h = 0. (xt . 1901). The iterative process is given by the following scheme: y0 = −ayn ˙ y1 = −n(y1 − y0 ) ˙ . and the number of intervals n large enough. Kutta. proposed at the end of the nineteenth century by the German mathematicians Carl Runge and Wilhelm Kutta (Runge. when properly applied. small changes of the parameter a from the value π lead to diverging 2 or converging oscillations. The oscillations 2 are not stable.001. Each of these maps leads to diﬀerent graphs.
The general delay equation for this model is ˙ x = −axt−1 1 − xtb (4. 1]. Prigogine et al.2. 1969.7. delay means oscillations. i. the linear spline approximation method was used with n = 30.7. Chemical oscillations and the BelousovZabontiski reactions modelled and studied extensively by Prigogine and his coworkers (Prigogine and Lefever. economic.e. Biological cycles.8) are quite stable when a > π . The resulting twodimensional map is an almost perfect square. Delays in human decisions on social. The rectangular oscillations appear to be quite perfect and may be useful in several applications. Short. This rectangular form shows for any a > π . simulation results in the very important form shown in Figure 4. h = 0. 1996. 4. 1997) are also examples of the eﬀect of delays in the formation of the intermediate substrates until the formation of the ﬁnal product.6. Prigogine. innovation or technological cycles may be expected when time delays are present. The parameter b is an integer. As a consequence. a = 3. b = 1..84 Chaotic Modelling and Simulation lations appear very naturally in a system with memory.9) Using the same method presented above. 1981. which has a more stable oscillating behaviour.or longterm economic cycles.8) . In the case studied in Figure 4. irrespective of the starting point. we simulate a more complicated delay model. 1968. technological or political systems may lead to oscillations. social or political changes. This is because the model expressing the behaviour of the system integrates into the diﬀerentialdiﬀerence equations the past as one or more delay variables. the model (4. Nicolis and Prigogine.001 and initial value x = 0. Experience conﬁrms the appearance of oscillations in very many systems where delays are inherent. When b = 1. When b = 2 and a = 14.4 A More Complicated Delay Model The oscillations of (4. and they exhibit a very inter2 esting behaviour for some values of a. depending on the parity of b.8) reduces to a continuous delay logistic model. The behaviour in this case is similar to that of the discrete delay logistic model (Figure 4. if the starting values are in [0. Of special importance is the case where b = 2. n = 15). with the sharpness of the corners depending 2 on the value of a. as long as this point is inside the square. The delay equation in this case is: x = −axt−1 (1 − xt ) ˙ (4. 1995.6. and it gave very good results. biological clocks and other oscillating mechanisms of humans and other living organisms are some of the numerous cases where delays are present in nature. and two distinct cases are considered. The oscillations are quite diﬀerent when b is odd.
7: Delay logistic model .6: Delay model (b = 2).The Delay Logistic Model 85 FIGURE 4. a square pattern FIGURE 4.
x1 = x1 + bx1 (1 − xm )d m x2 = x2 − (x2 − x1 )d s .001 and b = 2. .6. however.5 is: A Delay Diﬀerential Logistic Analogue The appropriate diﬀerential equation analogue to the delay diﬀerence equation (4.86 Chaotic Modelling and Simulation 4.01. As can be seen from the ﬁgure.12) is a continuous equation very appropriate for expressing a square. . The delay term is included only in the ﬁrst equation. 4. s = 3 and m = 14.8(b) is based on another delay equation: xm = bxm (1 − x2 ) ˙ (4. equation (4. x1 ) diagrams for the case where b = 1. This is due to the choice of the parameters and of the integration step. b was selected close to its upper limit. or replacing it by another. m xm = xm − (xm − xm−1 )d s (4.10). Close inspection. xm+1 ) and (t. d = 0.11) Figure 4. By changing this delay term.13) . d = 0. s = 6.10) In order to perform a simulation of (4. many interesting cases arise. However. The oscillation turns out to exhibit a step form. There is a clear similarity with the previous case of the diﬀerence equation above.22. xm+1 ) graph is an almost perfect square. The simulation producing Figure 4. In both cases presented above.9) xm = bx(1 − xm ) ˙ (4. Nevertheless.6 Other Delay Logistic Models Another form of a delay logistic model is expressed by the following diﬀerence equation: xt+1 = axt + bxt−m (1 − xt−m ) (4. will show that the corners are not very sharp.8(b) are m = 15. one can have several intermediate cases of particular interest. The intermediate step is s and the iteration step is d.8(a) illustrates the (x1 . It is the last term to the right of the ﬁrst equation. the method of linear spline approximation presented in page 83 will be used.12) The parameters in Figure 4. the (x1 .
74.706.22. the timedelay eﬀect is more apparent. and the bxt−m (1 − xt−m ) term represents the new adopters. 0.8: Oscillations and limit cycles in delay logistic models This is a delay model that integrates the inﬂuence of a logistic like interaction to the future. 1. The values of b are 0. The axt term represents the current adopters who continue. social or political systems. 1. even in the simplest cases of one or two units of delay. 1. 0. 1.965. bifurcation and then chaos appear as b increases. The more interesting forms appear in Figure 4. whereas b is 0.2. and leads to a chaotic picture with a period doubling.9(c)).252. The six realisations. consider the impact of a “wordofmouth” communication between x adopters and (1 − x) potential adopters of an innovation or an idea. a = 0. (Figure 4. A large variety of oscillations and chaotic oscillations appear. Again. As a simple example. 1.259 and 1. limit cycles.9 and x0 = 0.795 respectively.758 and 0.751. Figure 4.1. 0.762 respec . the behaviour of the system is quite complicated. a was set to 0. bifurcation and chaotic behaviour appear. and inﬂuencing people at the present time.991. limit cycles. ﬁgures A through F correspond to b values of 1. limit cycles are followed by chaotic behaviour.717. 0.994 and 1. High delay levels ensure chaotic behaviour.9(a). 1.754.77 and 1. In this model. economic. The bifurcation is more complicated.9 and x0 = 0.13) is when m = 1. depending on the level of the delay. illustrate the behaviour of the system when b is 1.95. The impact of this timedelayed inﬂuence to the evolution of x in the future will be studied by varying the width of the delay interval m and the values of the parameters a and b. A through F.016 respectively.743. As b increases.99.758.747. 0. The parameter a and the initial value x0 are as in the preceding example.The Delay Logistic Model 87 (a) Delay Logistic (m = 14) (b) An almost perfect square FIGURE 4. Throughout. But. 0.269 respectively (a = 0. 1. 0. When m = 5.241. 0.9(b) examines the case where m = 2. This happens very frequently in marketing.9(d)). communicated m time units in the past. 1. The simplest form of (4.1). The m = 3 time delay has an interesting eﬀect on the twodimensional graph (Figure 4.96.9 and x0 was set to 0.
88 Chaotic Modelling and Simulation (a) m = 1 (b) m = 2 (c) m = 3 (d) m = 5 FIGURE 4.9: Chaotic attractors of a delay logistic model .
0.10 shows great similarities to that arising from a discrete delay Glass model of the form xt+1 = axt + b xt−m 10 1 + xt−m (4. and the form of the attractors is inﬂuenced by the delay mechanisms. 4.3818 and 0.14) . the form of the graphs tends to be stable for changes of the delay parameter and keeps the form of the chaotic attractors. Graph A represents a classical limit cycle.10 illustrates the twodimensional graphs for the logistic delay model when m = 40. Higher order bifurcations appear in graph E.38.37. and chaotic behaviour is present in graph F.41 respectively.95. This can be seen by comparing the logistic delay model to the Glass model (Mackey and Glass.7 Model Behaviour for Large Delays When the time delay m is large or extremely large.3813. but the chaotic attractors and the resulting chaotic oscillations are very similar. 1977). 0. bifurcation and chaotic behaviour are again present. Limit cycles. and graphs C and D show second and third order bifurcations respectively. The model is expressed in a high dimensional space.9 and x0 = 0. a = 0. The two models look completely diﬀerent in terms of their analytic equations.10: Chaotic attractors of the logistic delay model (m = 40) The chaotic behaviour presented in graph F of Figure 4.36. Figure 4. not merely by the form of the delay equation. The form of the graphs is similar to the m = 3 case. FIGURE 4. The parameter b here takes the values 0. 0.The Delay Logistic Model 89 tively. whereas graph B corresponds to a ﬁrst order bifurcation. 0.
Such general models are of the form xt+1 = axt + F(xt−m ) Examples of speciﬁc models that exhibited behaviour similar to that of the Glass and the delay logistic models in the chaotic region are: LG2 LG3 GRM1 Gauss Gompertz 2 xt+1 = axt + bxt−m (1 − xt−m ) 3 xt+1 = axt + bxt−m (1 − xt−m ) xt−m (1 − xt−m ) xt+1 = axt + b 1 − (1 − c)xt−m 2 (4. Chaotic behaviour is expressed for the following values of the parameters: In all these cases. delay GRM1.11: Comparing the Glass (A) and the logistic (B) delay models (m = 40) Various general models were also tested. A diﬀerent chaotic behaviour is exhibited by the following delay exponential model: xt+1 = xt + bxt−m e−cxt−m (4.15) xt+1 = axt + be−c(xt−m −k) xt+1 = axt − bxt−m ln(xt−m ) We refer to the ﬁve models as delay LG2. The delay parameter is m = 40 and a = 0.18 for the Glass model and 0.9 for both models.11.11.4 for the logistic model. delay LG3. The similarities are striking.90 Chaotic Modelling and Simulation The Glass model (graph A) and the logistic model (graph B) are compared in Figure 4. the attractors have the same shape as that illustrated in Figure 4. and b is set to 0. FIGURE 4.16) . with similar results. both in the twodimensional attractors as well as in the chaotic oscillations. delay Gauss and delay Gompertz respectively. presented in the lower part of the ﬁgure.
Figure 4. A shape like that of a pigeon drawn by a modern painter appears.243 0. k) 0.3 0.9 b (c.8 Another Delay Logistic Model A somewhat diﬀerent delay logistic model is given by the following diﬀerence equation: xt+1 = bxt−m + axt (1 − xt ) (4.18) The analysis of system (4.8.The Delay Logistic Model 91 TABLE 4.12 illustrates this limit cycle and the related oscillations when the parameters of the model take the values a = 0. this system reduces to a system equivalent to that of the H´ non e model: xt+1 = bxt−1 + axt (1 − xt ) (4. b = 3. FIGURE 4.2: Parameter values on which various delay models exhibit chaotic behaviour Model LG2 LG3 GRM1 Gauss Gompertz a 0.17 (4.95 0.9 0.12: The delay exponential model 4. 0.9.9 0.7) 0.2 and c = 0.18) when b = −1 shows that it has a stable ﬁxed point .9 0.84) 0.17) When m = 1.28 Model (4.16) gives a relatively stable limit cycle for various values of the parameters.1 (0.
18) when a = 3. The small attractor produces the large attractor by rescaling and translation.3.19) However. The large attractor is the (xt .20) Equation (4.19) implies that yt = xt−1 . xt+1 ) image produced by equation (4. note that the second equation of (4.13(b).3. the two attractors look as if they were produced by reﬂection and translation of one to the other. (a) Delay cosineH´ non e (b) Delay logisticH´ non e FIGURE 4.4 and b = 0.20) when a = 1. as can be easily observed in Figure 4. The cross near this attractor is e . We can therefore substitute this into the ﬁrst equation.13: Delay logistic and delay cosine H´ non variants e In Figure 4. The attractor on the top right of Figure 4.13(a).13(a) is the H´ non attractor provided by the system (4.17 and b = 0.92 when Chaotic Modelling and Simulation 2 a x=1− which turns into the period2 orbit with: x= a+2± √ (a − 6)(a + 2) 2a The H´ non model is usually presented in another form.18).19). The small attractor is the one produced by equation (4. and then the following delay equation results: xt+1 = bxt−1 + 1 − axt2 (4. as a twodimensional e model expressed by a system of diﬀerence equations: xt+1 = byt + 1 − axt2 yt+1 = xt (4.20) is equivalent to (4.
4.13(a) is formed when b = 0. whereas the outer dimension of these squares is also equal to π. The attractor in Figure 4.5.3 and a = 3. The inﬂuence of the cosine function in the delay equation is clear: The various forms are repeated many times by following 45◦ and 135◦ degrees of symmetry.16(a).5. The islands have now become very ordered geometric structures. With the exception of the central square.22) (4. 0). An inﬁnite number of islands are included inside the chaotic sea.15(a) shows the bifurcation diagram (a. This attractor results from the system xt+1 = byt + a cos(xt ) yt+1 = xt and its corresponding delay equation xt+1 = bxt−1 + a cos(xt ) (4.15(b)). as its Jacobian is J = b = −1. algebraically. The map (4. −π).The Delay Logistic Model 93 located at the origin (0. Another carpetlike attractor appears in Figure 4. The center of the squarelike forms is located at distances proportional to π. x) for 0 < a < π + 0. presented in Figure 4. .21) is an areapreserving map.22) retains some of the properties of the H´ non map. Here b = −1 and a = π − 0.14 (a = −0. b = −1).21) This system looks very diﬀerent from (4.19).22. the islands cover more space 2 2 inside the chaotic sea. yet the attractors are very similar. e Equation (4. FIGURE 4. The squares are also related to each other in terms of their creation over time.16(b). Here a = π − 1. As a is less than the characteristic value π .1 and 2 b = −1. 2 There is a systematic way of drawing these graphs.14: Chaotic patterns of the delay cosine model Figure 4. The chaotic sea almost disappears in Figure 4. The parameter value a = π gives rise to an interesting carpetlike form 2 (Figure 4. and cover almost the entire plane. all the other squares are generated in groups of four. whereas the cross close to the attractor on the left of the same ﬁgure is located at (−π. but it also contains other interesting properties.
x) (b) A carpetlike map FIGURE 4.16: Carpetlike forms .15: Carpetlike forms (a) A carpetlike pattern (b) Groups of squareforms FIGURE 4.94 Chaotic Modelling and Simulation (a) The bifurcation diagram (a.
In Figure 4. 2π .19(a). Period doubling and totally chaotic regions are clearly present. the chaotic 2 sea covers the entire plane.17(a) illustrates the same map as in Figure 4. On the outer part of each pattern we can see 6 small islands. The six islands from the preceding case are now separated from the original form. 2 2 2 2 The bifurcation diagram (b. π ) and at ( π . On the boundary one can now see 4 islands. 2 When a = 2π the ﬁxed point bifurcates into a period2 orbit. . Figure 4. When the control parameter is very small.001). with their centers located at (− π . Beyond it. Interesting windows and special forms appear when a takes the special values π . a triangular form appears (Figure 4. along with the ﬁrst group of four surrounding squares 2 (starting with an initial value a bit more than π ).19(d) the bifurcation has been completed.18(a). but in the latter case only the centers and the corresponding squares are present. The control parameter at this point is a = π. − π ).01.19(b) shows the case where a = π . a = 0. This is illustrated in 3 Figure 4. The control parameter here is a = π . The oscillations seem to cover the entire plane.18(b). 2 (a) Squarelike forms (b) The generating squares FIGURE 4. The chaotic sea almost disappears and the nonsymmetric islands become geometric objects. When the control parameter is increased. almost perfect squares appear. The original central square is drawn (starting with an initial value a bit less than π ). and two distinct chaotic objects appear. x) appears in Figure 4.The Delay Logistic Model 95 Figure 4. This is illustrated in Figure 4. and are inside the chaotic sea. where a was set to 0.16(b).20(a). 3 2 3 2 The value 3π is the lower limit for the control parameter a.17(b).19(c). At the end of the diagram the chaotic region has a stochastic character. The building blocks for the construction of the above carpetlike form are two 3 interrelated objects close to the origin (marked by a cross in Figure 4.17: Squarelike forms The case of b = 1 is illustrated in Figure 4. The system is at the starting point of the bifurcation. π . with no islands present. The two forms have the shape of almost perfect squares. π and 3π .
19: Islands in the chaotic sea .18: Irregular patterns (a) Bifurcation diagram (b. a = π 2 (c) Islands in the chaotic sea. a = 2π 3 (d) The order2 bifurcation FIGURE 4. x) (b) Islands in the chaotic sea.96 Chaotic Modelling and Simulation (a) Irregular patterns (b) The generating squares FIGURE 4.
.The Delay Logistic Model 97 a = 6π −0. 5 (a) A triangular island (b) An order3 bifurcation FIGURE 4.20: Triangular islands The H´ non model was introduced here in order to compare it to the other delay e models. a = 6π − 0. A more detailed analysis of the H´ non model will be the subject of the next e chapter. the triangular form splits into three objects 5 5 around the central one (Figure 4. However.008).20(b). the H´ non model is a very interesting model. When a gets closer to 6π . deserving its own e study.02).
What happens to the number of “folds” in Figure 4.1. 4.4(b) as the delay parameter m increases? 9. b and m.8) is allowed to take noninteger values. 5. .98 Chaotic Modelling and Simulation Questions and Exercises 1. and generalise your result to a map of the form xt+1 = −xt−1 + a f (xt ) where f (xt ) is a continuous smooth function. xt+1 ) graphs for the Gompertz delay model and for various positive values of the parameters a. a = 0. 8.28. xt+1 ) and (t. 2. Find the Jacobian of the map in the previous problem. Can you draw carpet like forms by using the map resulting from the approximation of the cosine? Justify your answer. Examine what happens when the parameter b in (4. Find similarities and diﬀerences with the previous case. Determine the sides of the large hexagon appearing on the right of Figure 4. but now for the delay map x2 xt+1 = −xt−1 + a 1 − t c where c is a positive parameter.9 and b = 0. Draw the graph of the same model for m = 40. 3. Replace the cos(xt ) term into the delay cosine map (Equation (4. 6. Repeat the analysis and graphs of the previous problem. and compare the resulting ﬁgure with that of the Glass and the logistic models.22)) by Taylor approximation to the cosine cos(xt ) = 1 − xt2 2 and draw the bifurcation diagram for b = −1 and compare the resulting ﬁgure with that provided for the delay cosine model. 7. Draw the (xt−m . Find the characteristic points and compare them to those of the delay cosine.
1. The system becomes stable for 0 < b < 1. The ﬁxed points (xt+1 = e xt . and the resulting maps take the form of chaotic attractors. (2002).2) = −b When b = 1. The analysis of the H´ non map is based on (5.3. yt = xt+1 . 5. and a very detailed analysis can be found in Murakami et al. the resulting map is area preserving.1) By introducing a new variable.2). but the system is unstable. A variety of maps appear when a and b vary over a range of values.2) is given by: det J = ∂xt+1 ∂xt+1 ∂xt ∂yt ∂yt+1 ∂yt+1 ∂xt ∂yt (5.4 and b = 0.1) is transformed into a system of two ordinary diﬀerence equations: xt+1 = yt + 1 − axt2 yt+1 = bxt The Jacobian determinant of (5.1) and (5. (5.1 Global Period Doubling Bifurcations in the H´non e Map Global period doubling bifurcations in the H´ non map have been studied by many e researchers. This model can also be expressed by a delay diﬀerence equation: xt+1 = bxt−1 + 1 − axt2 (5. yt+1 = yt ) of the map are the solutions of the secondorder polynomial: ax2 − (b − 1)x − 1 = 0 namely xt = (b − 1) ± (b − 1)2 + 4a . and is illustrated in e Figure 5.Chapter 5 The H´non Model e The H´ non model is a twodimensional model introduced and analysed by H´ non e e (1976). yt = xt 2a 99 . The most known chaotic attractor of the H´ non model appears when a = 1. We will see a variety of these attractors in this chapter.
yt = xt a for −1 < a < 3. This last value of a is the root of: √ 4a(2 a − a) + 1 = 0 .4) The third period doubling bifurcation takes place when a = 4. This orbit remains stable for 3 < a < 4. giving rise to a period8 orbit. y4 = x1 1/2 (5.100 Chaotic Modelling and Simulation FIGURE 5. y2 = a a at the value a = 3.1. the map has the stable ﬁxed point √ −1 + 1 + a xt = . y2 = x1 x3 = x1 .3) √ √ 1− a−3 1+ a−3 x2 = . y1 = √ 1 − √ a a a x2 = −y1 . This turns into the period2 orbit √ √ 1− a−3 1+ a−3 . y3 = −y1 x4 = y1 . giving rise to the period4 orbit with 2 1 1 x1 = √ .1 Period doubling bifurcations when b = −1 When b = −1. The resulting fourthorder equation for x has the two extra solutions xt = −(b − 1) ± −3(1 − b)2 + 4a . The second period doubling takes place at a = 4. y1 = x1 = a a (5. y t = xt 2a 5.12045.1: The H´ non map e The period2 orbit can be found from the relations xt+2 = xt and yt+2 = yt .
2(a).2) are at (5. and we have a bifurcation into a period4 √ orbit. The period8 orbit bifurcates into the period16 orbit at a = 1. Also present in this case are 7 islands on the boundary. based on trigonometric functions: e xt+1 = byt + 2a (cos (xt ) − 1) + 1 yt+1 = xt (5. 5. e but it also has other distinct properties.2 The CosineH´non Model e During the simulation process related to the global period doubling bifurcation.The H´ non Model e 101 5.6) when the cosine term e is replaced by its secondorder Taylor approximation: cos(xt ) = 1 − xt2 2 Therefore.13397 and leads to a period8 orbit. Figure 5. Figure 5.y=x a When b = 1.23.2(c) illustrates the period4 orbit of the cosine map with b = −1 and a = 4. We leave the details to the reader. y = −x a This orbit becomes unstable when a = 1. The relation between the two models e is that the H´ non model (5. therefore. more convenient to use an analogue of the H´ non system. just before the completion of the next global period doubling.2 Period doubling bifurcations when b = 1 1 x=±√ . y) plane. the values of the control parameter related to the period doubling bifurcations of the H´ non map are close to the corresponding e values of the control parameter for the cosine model (b = −1). The period4 orbit undergoes a period doubling bifurcation at a = 8− 4 12 = 1. This model accepts a wider range of initial values in the (x. .5) These are both unstable when a > 0. as well as of the local or global instabilities related to the period doubling.2(b) illustrates the period2 orbit at a = 3. The period2 orbit consists of the points 1 x = ± √ . The system is in subcritical state.6) This model is called the cosine H´ non model.1. this new model retains many of the properties of the H´ non model. the ﬁxed points of (5. The ﬁxed point at the critical value a = 3 is illustrated in Figure 5.11. It is. diﬃculties arise on the estimation of the starting parameters. We can see 11 islands on the boundary.2) is the model obtained from (5.15135. Furthermore.
2: Orbits in the cosine H´ non model e .102 Chaotic Modelling and Simulation (a) The ﬁxed point (a = 3) (b) The period2 orbit (c) The period4 orbit FIGURE 5.
3(b) illustrate some interesting cases regarding the bifurcation and period doubling when the parameter b of the cosine H´ non model takes the e value b = 1. the central island has turned into three separate forms.4(b) show the map when the control parameter is a = 0.8 (b) a = 0. we see that.3 respectively.3(a) and 5. The system will undergo period doubling when a = 1.02 and 1. 1. in each of the original structures.3(b). the period4 islands are completely separated.3(a). 5.3(a).98 FIGURE 5. (a) a = 0. Figures 5.8. In Figure 5.The H´ non Model e 103 5.3 An Example of Bifurcation and Period Doubling Figures 5.4(a) and 5.7) .98.4 A Diﬀerential Equation Analogue A diﬀerential equation analogue to the H´ non model is given by the following e system of two coupled diﬀerential equations x = by + 1 − ax2 − x ˙ y= x−y ˙ (5. a double structure appears. 5. 0.4(b). After the period doubling.3: Bifurcation forms in the cosine H´ non model e 5. In Figure 5.
5.4: Higherorder bifurcations in the cosine H´ non model e This system has ﬁxed points at x=y= b−1± (b − 1)2 + 4a 2a √ 1+a a When b = −1. the provided value is x = y = ± √a which is also the same value as that provided from the diﬀerence equation case.02 (b) a = 1.5. or simply changes in the way the ﬁrstorder terms are introduced (sections 5. 5.2).5. 5.5 Variants of the H´non Delay Diﬀerence Equation e A number of variants of the H´ non model give rise to some interesting attractors.8) .1. involving either higherorder delays (sections 5.4). 1 Similarly.5.104 Chaotic Modelling and Simulation (a) a = 1.3. 5.2). these ﬁxed points become: x=y= −1 ± which is precisely the same as the ﬁxed points for (5.1 The thirdorder delay model An extension to higher delay dimensions is given by xt+1 = bxt−3 + axt (1 − xt ) (5. e We discuss a couple of these variants.5. 5. when b = 1.3 FIGURE 5.
The H´ non Model e 105 Similar to (5.11) This is a twoparameter delay model with both ﬁrst.8.10) When a = 0.2 and b = 0. The model (5.9) The attractors of the model are however very interesting. x) are illustrated in Figure 5.5: A thirdorder delay model 5.75. A more complicated model is given by: 2 xt+1 = bxt−1 − axt + xt xt−2 (5. with a secondorder delay term.6 for instance shows the attractor when b = 1. FIGURE 5.8) exhibits three timedelay periods.5). Figure 5.2 and b = 0.5. Another interesting variant with a secondorder delay is the following: 3 xt+1 = xt + a(xt−2 − xt−2 ) (5. and is characterised by a very interesting attractor when the parameters are a = 3.8) can be turned into a system of four ordinary diﬀerence equations. The number of time delays in the delay diﬀerence equation is equal to the number of new equations of the ordinary diﬀerence equation system. the diﬀerence delay equation (5.2 (Figure 5.52. the resulting map and the related oscillations (t. is extremely simple. There is a characteristic limit cycle obtained when the parameters have values a = 1.2 Secondorder delay models The following variant.37. .7. having only one parameter b: xt+1 = −bxt + xt xt−2 (5.and secondorder delay terms. The repeated oscillations and the limit cycle are presented in Figure 5.2).
7: The order2 delay model FIGURE 5.106 Chaotic Modelling and Simulation FIGURE 5.6: A secondorder delay model FIGURE 5.8: A complicated delay model .
13) For a = 1.12) Figure 5. FIGURE 5. (5.The H´ non Model e 107 5.14) is det J = −bxt .13) into a system of two diﬀerence equations: xt+1 = xt yt + 1 − axt2 yt+1 = bxt The Jacobian determinant of (5. For example. when a = 2.1) is altered.10.6. This indicates that the model (5. the resulting model has the form: xt+1 = bxt−1 + axt (1 − xt−1 ) (5. it is more convenient. if we replace the quadratic term with a delay logistic term.14) has properties diﬀerent from those of the H´ non e model.72 and b = 0.9 illustrates the twodimensional map.14) .3 Firstorder delay variants Interesting ﬁrstorder delay variants of the H´ non model arise when one of the two e terms on the righthand side of (5. to transform (5.35 and b = 0.2. As in the case of the original H´ non model.9: A H´ non model with a logistic delay term e Another variant arises by altering the ﬁrst term: xt+1 = bxt xt−1 + 1 − axt2 (5. and the related oscillations. the resulting map and related oscillations are illustrated in Figure 5. for further e analysis.5.
36).11. b = 0.10: A variation of the H´ non model e 5.108 Chaotic Modelling and Simulation FIGURE 5.8.15) FIGURE 5.5.11: An exponential variant of the H´ non model e . a = 0.20) gives rise to a variant with a very complex attractor (Figure 5. The equation is xt+1 = bxt−1 + 1 − axt2 e−axt−1 (5. which we will call the Henia attractor.4 Exponential variants The introduction of an exponential term into (4.
628.16) For appropriate values of the parameters. The parameters a and c are ﬁxed at 0.12 (a = 0. An example is the following: xt+1 = bxt + cyt − 2xt yt yt+1 = a(yt − xt ) (5.The H´ non Model e 109 5. 0.6337. 0.17) Depending on the values of the three parameters.4 respectively. Figure 5.48.12: A 2diﬀerence equations system Another set of extensions involves the introduction of a third variable: xt+1 = bxt − yt − xt zt yt+1 = −czt + xt zt zt+1 = a(zt − xt ) (5.6 Variants of the H´non System Equations e Other interesting variants arise from altering the system (5. interesting stoneshaped attractors appear (Figure 5. bifurcation and ﬁnally chaotic behaviour.8. The graphs represent the (x.1)).6348. FIGURE 5. 0. while b takes the values 0. 0. .637 respectively.6355 and 0. c = 1. y) map.13 illustrates these cases. b = 0.62. this model exhibits limit cycles.2).95 and 0.
when b = −0. When b > 0. b = 0. provides a characteristic sigmoid map and quite complicated chaotic oscillations (Figure 5.14: The Holmes model (a = 2.2 and a = 2.765. e 1979a). b = −0. the FIGURE 5.2) Holmes model gives a rich sigmoid form (Figure 5.765.110 Chaotic Modelling and Simulation FIGURE 5.13: A threedimensional model 5.3 and a = 2.1 The Holmes and Sine Delay Models The Holmes model The Holmes model can be thought of as an extension of the H´ non model (Holmes. The general form of the Holmes model is: xt+1 = bxt−1 + axt − xt3 The Holmes model. .14).4).7.7 5.15.
there is a close relation between the maxima and minima of this map.3 and a = 6. the map changes into a series of sigmoid forms.17(a).0772 for the case B. the situation changes radically.7.2 The sine delay model The model proposed by Holmes is an approximation1 of a sinusoidal one.17(b) shows the case where b = 0.3) 5. of the form: xt+1 = bxt−1 + a sin(xt ) (5. say b = 0. b = 0. However.5(1 − b)π. approximation can be seen by replacing sin(x) with its third degree Taylor approximation. Figure 5. for relatively small values of a. Note further that the range of x values for the map ends exactly at the last solution to f (x) (the solutions to f (x) = 0 are exactly the ﬁxed points of the sine delay map).18) This model. It is clear.18(a)). In the example presented in Figure 5. When b takes values closer to 1.16. When b is small. x − .15: The Holmes model (a = 2. There is now 1 The 1 3 6x .3 in both cases. that by doubling the value of a. the number of maxima or minima is also doubled.4.88. The relation can be seen in Figure 5. and the parameter a is set to a = 3. the delay parameter is ﬁxed at b = 0.The H´ non Model e 111 FIGURE 5. and the maxima and minima of the function y = f (x) = (1 − b)x − a sin(x). when a takes large values. by observing the resulting maps. as the attractor is now separated in two very similar parts (Figure 5. The number of maxima or minima of the resulting curves can be estimated according to the number of solutions of the equation (1 − b)x − a sin(x) = 0. called the sine delay model. shows similar behaviour to that of the Holmes model.0772 for the case A and a = 2 × 3.
3) (a) Solutions to the ﬁxed point equation for the sine delay model (b) Maxima and minima of the sine delay model and the corresponding ﬁxed point equation FIGURE 5.16: The sine delay map (b = 0.112 Chaotic Modelling and Simulation FIGURE 5.17: The sine delay model .
a = 0. The resulting simulation is shown in Figure 5. The basic forms of the two generating shapes can be seen more clearly in Figure 5. The process follows a random sequence that gradually occupies the entire (xt−1 .1). Both shapes become closer to a square with rounded corners as the parameter a takes very small positive values (Figure 5.19(a) for a = 1. xt ) plane.20. (a) b = 0.The H´ non Model e 113 little relation with the maxima and minima of the ﬁxed point equation. The values a = 3 and b = 0.19(b).88 (b) a = 3. Both shapes are identical but they are a ninetydegrees rotation of each other. where a = 0.18: The sine delay model for large b Things change quite a bit when b = 1.2.8.18(b). In this case. the map takes the form xt+1 = xt−1 + a sin(xt ). .8 produce the very interesting attractor of Figure 5. b = 0.8 FIGURE 5.
19: The sine delay model for b = 1 FIGURE 5.1) . a = 0.2 (b) Close look at the two shapes.114 Chaotic Modelling and Simulation (a) a = 1.8 FIGURE 5.20: The sine delay model (a = 0.
1. Explain why for small b there is a relationship between the maxima and minima of the sine delay model (xt .9). (b) Rewrite them as systems of 3 diﬀerence equations without delay. (5.18) when b = 1. 5.11): (a) Determine the ﬁxed points and secondorder ﬁxed orbits. For each of the maps (5. Examine the behaviour of the H´ non map for b = 3. Continuing from the previous question. e 3. . 4. Determine the ﬁxed points of the sine delay model given by equation (5.The H´ non Model e 115 Questions and Exercises 1. run simulations with diﬀerent parameter values. Verify that the ﬁxed points and perioddoubling bifurcations of the H´ non map e for b = 1 and b = −1 are as described in section 5. and how they spread across the plane.19(a)). and their stability. and estimate the n beyond which f (c) > 0 at the minima. and calculate the Jacobian determinants of these systems. 6. Use this information to explain the carpetlike forms arising when b = 1 (Figure 5. 7. and their stability. Estimate the number of maxima and minima of the ﬁxed point equation for the sine delay model: f (x) = (1 − b)x − a sin(x) Hint: Find an equation for those maxima and minima from f ′ (x) = 0. xt+1 ) map and the maxima and minima of the equation for the ﬁxed points of the map. 2.10) and (5. and examine whether the carpetlike formations are present.
.
0). at a point x = x0 + h suﬃciently close to x0 . z) ˙ ˙ x = z(x. a general discussion of the behaviour of such systems near their equilibrium points is necessary. using a ﬁrst order ˙ Taylor approximation. 117 .1) Then. z0 ) is an equilibrium point for it. 0. y. z) ˙ ˙ and suppose x0 = (x0 .e. a point where ˙ x0 = ( x0 . though. i. say λ1 . that x is given by: ˙ ˙ x = x0 + A · h = A · h Suppose now that the matrix A has a real eigenvalue. y0 .1 Equilibrium Points and Characteristic Matrices Let us consider a general autonomous system of the form x = x(x. 6. we can assume. the orbit of a particle placed at x0 + δh1 will move away from x0 . Then. y0 . in the direction of the vector h1 .Chapter 6 ThreeDimensional and HigherDimensional Models In this chapter. z0 ) = (0. y. since its position after a short time t would be: ˙ x(t) ≈ (x0 + δh1 ) + x · t = x0 + (δ + λ1 t)h1 . y. ˙ ˙ ˙ Consider the matrix: ∂x ˙ ∂x ∂˙ y A = ∂x ∂˙ z ∂x ˙ ∂y ∂˙ y ∂y ∂˙ z ∂x ∂y ∂x ˙ ∂z ∂˙ y ∂z ∂˙ z ∂z (6. z) ˙ ˙ y = y(x. we present a number of higherdimensional models that exhibit chaotic behaviour. Before we delve into particular examples. with eigenvector h1 .
the system can be replaced by the simpler system x= x ˙ y = −y ˙ . Using linearization. the characteristic equation is λ2 − 1 = 0 and the system has two real eigenvalues λ1. The system will cycle around this point. y) = (1. 6. y) = (0. and where there are no prey.2 = ±1. λ1. at (x.2) In the terminology of Chapter 7. the LotkaVolterra model. y) = (1. the characteristic equation is λ2 + 1 = 0 and the system has two imaginary eigenvalues. this is a conservative system with ﬁrst integral of motion: f (x. Since one of these eigenvalues is positive. In its simpliﬁed form. The characteristic polynomial is given by −y + 1 − λ −x = λ2 − (x − y)λ + (x + y − 1) y x−1−λ At (x. 0).2 = ±i. we see that near (0. the LotkaVolterra model has the form: x = x − xy ˙ y = −y + xy ˙ (6. To determine these eigenvalues. 0).118 Chaotic Modelling and Simulation So any equilibrium point with a positive eigenvalue will be unstable. we need to compute the roots of the characteristic polynomial ∂x ˙ ∂x ∂˙ y ∂x ∂˙ z ∂x A − λI = −λ ∂˙ y ∂y ∂x ˙ ∂y ∂˙ z ∂y −λ ∂˙ z ∂z ∂x ˙ ∂z ∂˙ y ∂z −λ The matrix A − λI is called the characteristic matrix of the system. At (x. 1). y) = (0. 1). y) = xye−x−y It has two equilibrium points. also known as the predatorprey model.2 The LotkaVolterra Model As an illustration of the above theory in a simpler case. let us consider a classical twodimensional system. the system is unstable. 0) and at (x. The eigenvectors correspond to the two directions where there are no predators.
An example of such a system is the Arneodo model: x=y ˙ y=z ˙ z = − z + sy − mx + x ˙ 2 (6. 0) m.8 and m = 7. 6. in their work on systems that exhibit multiple periodicity and chaos. This model was proposed by Arneodo et al. if we use coordinates (ξ. 0). (1979. 0) (6.1.5 and y0 = z0 = 1). 0. y). The bottom part of this ﬁgure shows the chaotic oscillations in the (y. 1. Near (1. replacing the last equation in (6. z) respectively. t) space. 1981a).3 The Arneodo Model The simplest threedimensional chaotic models are based on a system of three linear diﬀerential equations. 1980. the system can be replaced by the simpler system ˙ ξ = −η η= ξ ˙ whose orbits are counterclockwise circles ξ2 + η2 = (x − 1)2 + (y − 1)2 = c. 0) The Jacobian of the modiﬁed system is J = −m + 3x2 . the initial values are (x0 = 0.2 shows the projections of the threedimensional solution path of this system to the three planes (x. (x. η) = (x − 1. 0. and (m.ThreeDimensional and HigherDimensional Models 119 whose orbits are the hyperbolas xy = c.3). 0). the (x. y) diagram of the system is shown (s = 3. except that m = 5.3) where s and m are parameters.4) (− m. There are three stationary points: ( ( √ √ 0. The parameters are as above. the origin (0. 0.5.001. A slight variation. 0. on which we add a nonlinear term. 1). The resulting paths of this modiﬁed system have a somewhat more complicated form than those of (6. There are two stationary (equilibrium) points. . y − 1) centred at 1. Figure 6. In Figure 6. An integration method was used with integration step d = 0. The Jacobian determinant of the model is J = −m + 2x. 0. z) and (y.3) with z = − z + sy − mx + x3 ˙ leads to a model with a more stable behaviour.
120 Chaotic Modelling and Simulation FIGURE 6.1: The Arneodo model FIGURE 6.2: A modiﬁed Arneodo model .
. may demonstrate chaotic behaviour. 1969. y) phase portrait. representing an autocatalytic reaction with three stages.4 An Autocatalytic Attractor Autocatalytic reactions often give rise to chaotic phenomena.5) take the values k = 0.e. FIGURE 6. Nicolis et al. Figure 6. Nonlinear modelling theory suggests that at least one nonlinear part must be present in each reaction cycle in order for chaotic phenomena to appear. The number of these stages inﬂuences the complexity of the system. consequently. for appropriate parameter values. and. gives rise to a process with the chaotic behaviour. 1971.3: An autocatalytic attractor . 1977. 1968. 1973.013. while on the left side we see the temporal variation of chaotic oscillations in the y direction (i. 1983).3 and e = 0. A typical such model is given by the following system: x = k + λz − xy2 − x ˙ y= ˙ xy2 + x − y e z=y−z ˙ (6. The representation of chemical reaction models by mathematical nonlinear models is possible and has been studied extensively (Prigogine and Lefever. Prigogine et al. A threedimensional chaotic model is perhaps the simplest case. Nicolis and Portnow.7567.. Nicolis. y) diagram). 1981. λ = 0.3 arises. the (t.ThreeDimensional and HigherDimensional Models 121 6. It is more diﬃcult to ﬁnd a good representation of the numerous intermediate stages of a chemical reaction. Such a model. On the right side we see the twodimensional (x. giving rise to a threedimensional chaotic attractor.5) When the parameters of system (6. Nicolis and Prigogine.
y) coordinate system (b) In the (y. z. z.017 and e = 0. y) of the phase portrait of the autocatalytic reaction is presented in Figure 6.5.4: Threedimensional views of the autocatalytic model 6. and to the couplings x1 x2 in the ﬁrst and the .4(a). The view from the (y. x) coordinate system FIGURE 6. In this case the system (6.24x1 − 100x4 ˙ x3 = 0. which appears in the ﬁrst and fourth equations.24x1 + 200x4 + a ˙ x2 = x1 x2 − x2 − 10x2 x3 ˙ 2 x4 = 0. with the ﬁrst equation replaced by: x= ˙ 1 + m (k + z) − xy2 − x 1+k (a) In the (x.0001. m = 0.5 A FourDimensional Autocatalytic Attractor The following fourdimensional model expresses a sequence of four chemical autocatalytic reactions: 2 x1 = x1 − x1 x2 − 0.013.122 Chaotic Modelling and Simulation A threedimensional view (x. z. x) coordinate system is presented in Figure 6.5) has been modiﬁed slightly. The parameters in both cases are k = 2. z. and the integration step is d = 0.4(b).01 + 10x2 x3 − 20x3 ˙ (6.6) is due to the second order term in x1 .6) The nonlinear character of (6.
. yet more complicated than in the previous cases. f = 0.003. o o 1976d): x = −y − z ˙ y = x − ez ˙ z = f + xz − mz ˙ (6. whereas the time oscillations are of a purely chaotic nature.ThreeDimensional and HigherDimensional Models 123 second equations and x2 x3 in the second and third equations. t). the (x1 . Figure 6. This is a very good example of chemical oscillations and the related chaotic waves.6 The R¨ssler Model o One of the simplest threedimensional models is the R¨ ssler model (see R¨ ssler.5(b) shows the threedimensional (x1 . has value −0. (x2 . The control parameter. which enters in the third equation. Figure 6. xz. x2 . x2 . a.2. t). from top to bottom.5: A fourdimensional autocatalytic model 6.7) The model contains a single nonlinear term. x3 ) diagram FIGURE 6. The chaotic behaviour is similar. from left to right. (a) Temporal chaotic oscillations (b) The (x1 . t) and (x4 .12. and three parameters. and the integration step is d = 0. t) diagrams. e.6(a) is a threedimensional view of the R¨ ssler model when the parao meters are: e = 0.2 and m = 5. (x3 .5(a) illustrates.7. f and m. with several peaks and intermediate chaotic stages. Figure 6. x3 ) diagram representing the chaotic attractor of (6.6).6(b) shows. Figure 6.
35.26.5. This chaotic form is usually referred to as spiral chaos.6: The R¨ ssler attractor o 6. and the ﬁrst order limit cycle of system (6. y). after the characteristic spiral paths of the resulting chaotic image.42).6. z) and (y.2.8) z = bx + xz − cz ˙ where a.7(a). y and z coordinates respectively. along with the temporal variation in the x. Figure 6. The order two limit cycle is presented in Figure 6.7(c) illustrates the ﬁrst order cycle when a = 0. (x. b and c are the parameters of the model. (a) A threedimensional view of the R¨ ssler o attractor (b) Chaotic oscillations and twodimensional views FIGURE 6. In all the cases that follow. .7(b).124 Chaotic Modelling and Simulation the (x. When a = 0.8) appears.1 A variant of the R¨ssler model o An interesting variant of the R¨ ssler model is given by the following system of o diﬀerential equations: x = −y − z ˙ y = x + ay ˙ (6.9) 1 c=1+ a In Figure 6. where a = 0. This is illustrated in Figure 6. total chaos ensues.7(d) we see the second order cycle (a = 0. z) views. the parameters b and c are speciﬁed in terms of a as follows: b=a (6.8(a). whereas in Figure 6. a is set to 0.
ThreeDimensional and HigherDimensional Models 125 (a) First order cycle.26 (c) First order cycle. a = 0.7: Limit cycles in a variant of the R¨ ssler model o . (a = 0.2) (b) Second order cycle.35 (d) Second order cycle. a = 0.42 FIGURE 6. a = 0.
8: Spiral chaos in the variant of the R¨ ssler model o The corresponding characteristic polynomial is A − λI = −λ −1 −1 1 a−λ 0 b+z 0 x−c−λ The characteristic matrix of the R¨ ssler model is o −λ −1 −1 0 A − λI = 1 a − λ b+z 0 x−c−λ = −λ3 + (x + a − c)λ2 − (ax − ac + z + b + 1)λ + (x + az + ab − c) This polynomial determines the stability of the system at its equilibrium points. z) and (x. y). 0. M1 = (0.8(b). The other equilibrium point is located . A threedimensional view of the total spiral chaos is illustrated in Figure 6. the equilibrium points are the solutions of the system: y+z=0 x + ay = 0 bx − cz + xz = 0 (6. In our case.126 Chaotic Modelling and Simulation A more complicated case of spiral chaos appears when a = 0. 0). the (x. Now the trajectories pass close to the origin. from left to right.10) The ﬁrst equilibrium point is at the origin. Figure 6.9 shows.50 (b) Total spiral chaos. (a) a = 0.60 FIGURE 6. The most interesting paths are in the neighbourhood of this point. z) views for the same value of a.6. a = 0. (y.
we used h = 0. where the parameters are a = 0.11) If a.ThreeDimensional and HigherDimensional Models 127 FIGURE 6.10. b.60) at: M2 = (x. then equation (6.27 x2 + y2 .6.8.12) c − ab c − ab . a a (6. c are such that c − ab > 0. − The characteristic equation at the origin is λ3 + (c − a)λ2 + (1 + b − ac)λ + (c − ab) = 0 The characteristic equation at M2 is λ3 − a(1 − b)λ2 + 1 − a2 b + c λ − (c − ab) = 0 a (6. The resulting equations have the form: x = −z sin(h) − y cos(h) ˙ y = x cos(h) + ey sin(h) ˙ z = f + xz − mz ˙ (6. The (x.9: Plane views of the total spiral chaos in the variant of the R¨ ssler o model (a = 0. . the characteristic polynomial is illustrated.13) The rotation angle h is a function of the distance r from the origin.2 Introducing rotation into the R¨ssler model o A simple way to introduce rotation into the R¨ ssler model is to insert a simple o rotation in two of the equations of the model. The variant of the R¨ ssler model is illustrated in Figure 6.4.3 and c = 4. y. In the same ﬁgure. b = 0. This in turn makes M2 into an unstable point. z) = c − ab. For the following simulations. It is a thirddegree polynomial with only one real and two complex roots (eigenvalues).12) has at least one positive root. y) graph is o presented. 6.
11: Modiﬁed R¨ ssler models o .3 and c = 4.4.10: View of a simple cycle of the variant of the R¨ ssler model o (a = 0.128 Chaotic Modelling and Simulation FIGURE 6. b = 0.8) (a) (b) FIGURE 6.
The movement in the z direction now tends to higher limits.7. gives the paths presented in Figure 6.12. e = 0. while cool ﬂuid above sinks. and ﬁnally the trajectories continue in the (x. There is a complicated movement.7 The Lorenz Model In 1963.ThreeDimensional and HigherDimensional Models 129 The inﬂuence of the rotation appears in Figure 6. The width of the ﬂow rolls is proportional to b. A better threedimensional illustration appears in Figure 6. the Rayleigh number r and the Reynolds number b are parameters of the system. x is proportional to the circulatory ﬂuid ﬂow velocity. whereas the paths around the origin follow concentric elliptical orbits. The equations .12. A fourth order RungeKutta procedure was used.3. y) plane and then in the positive z direction and back.12: Another view of the modiﬁed R¨ ssler model o 6.6. FIGURE 6. and z is proportional to the distortion of the vertical temperature proﬁle from its equilibrium.005.11(b).4 and m = 5. 1963) proposed an idealized model of a ﬂuid like air or water. y) plane. A diﬀerent set of parameters for the model.2 and m = 4. The parameters in this case were e = 0. y is proportional to the temperature diﬀerence between ascending and descending ﬂuid elements. If x > 0 the ﬂuid rotates clockwise.2. with an integration step d = 0. The Prandtl number σ. where we increased f to 0. f = 0. f = 0.11(a). setting up a clockwise or counterclockwise current. Edward Lorenz (Lorenz. the warm ﬂuid below rises. ﬁrst in the (x.
The vertical axis is z and the horizontal axis is y. 1 See Strogatz (1994). Two other equilibrium points exist when r ≥ 1: c+ = b(r − 1). r − 1 (6. Perhaps the famous “butterﬂy eﬀect” associated with the presence of chaos owes its name to this ﬁgure. Two small circles correspond to the unstable equilibrium points c± . a twodimensional (x. it is stable. z) view of the Lorenz attractor is presented. and almost all of them are neither periodic solutions nor convergent to periodic solutions or equilibrium points. For 0 < r < 1. is an equilibrium point for all r. it is unstable.15) c− = − b(r − 1). that the system behaves “chaotically” whenever the Rayleigh number r exceeds the critical value rh : All solutions are sensitive to the initial conditions. r − 1 These represent convective circulation (clockwise and counterclockwise ﬂow). 0.130 Chaotic Modelling and Simulation of this threedimensional model are: x = −σx + σy ˙ y = −xz + rx − y ˙ z = xy − bz ˙ (6.16).16) σ = 10 8 b= 3 the Hopf bifurcation point1 is rh = 24 14 ≈ 24. In Figure 6. the Lorenz model has two purely imaginary eigenvalues: λ = ±i when r = rh = Using the standard settings 2σ(σ + 1) σ−b−1 σ(σ + b + 3) . . Kuznetsov (2004).13(a).73684 19 For rh < r. numerically. − b(r − 1).13(b). A threedimensional view of the Lorenz attractor appears in Figure 6. The parameters take the standard values (6. (0. Lorenz found.14) The origin. This is the most common illustration of this famous attractor. all three equilibrium points of the Lorenz model are unstable. 0). which corresponds to a ﬂuid at rest. for r ≥ 1. b(r − 1). At the equilibrium points c± . while. σ−b−1 (6.
z) = (0. The equilibrium points are the solutions of the following system of equations: −σx + σy = 0 rx − xz = 0 xy − bz = 0 (6.13: The Lorenz attractor 6. . the resulting model is simpler. 0).18) Two equilibrium points appear. r √ √ M − = − br. r The characteristic matrix of the modiﬁed Lorenz model is −σ − λ σ 0 A − λI = r − z −λ −x y x −b − λ (6.19) The third equilibrium point.17) The only way (6. located at: √ √ M+ = br. Thus.1 The modiﬁed Lorenz model A modiﬁed Lorenz model is given by the following system of equations: x = −σx + σy ˙ z = xy − bz ˙ y = rx − xz ˙ (6.17) diﬀers from the original Lorenz model is that the term −y is excluded from the second equation. M = (x. − br. 0.7. y. br. is easily seen to be unstable.ThreeDimensional and HigherDimensional Models 131 (a) A threedimensional view (b) The butterﬂy attractor FIGURE 6.
15(a) illustrates the (x. z) views of the attractor. from left to right. The two stationary points are visible.14. y) view of the modiﬁed Lorenz attractor. (y. they are the (x.14: The modiﬁed Lorenz model — left to right: (x. Three twodimensional chaotic graphs appear. This graph gives an illustrative example of the bifurcation.15: The modiﬁed Lorenz chaotic attractor Figure 6. z) and (x. The image on the right presents the Poincar´ (x. . (y. (x. y). z) views 8 The chaotic paths of the modiﬁed Lorenz model. y) view. z) and (x. z. y). y) diagram of the e same attractor resulting when the plane Z = r cuts the chaotic paths. z. There appear to be many similarities with the Lorenz chaotic paths presented earlier in this chapter. x) view FIGURE 6. for s = 10.15(b) illustrates a threedimensional (y. are shown in Figure 6. (a) The modiﬁed Lorenz model: left. b = 3 and r = 27. The image on the left of Figure 6.132 Chaotic Modelling and Simulation FIGURE 6. right. x) view of the modiﬁed Lorenz chaotic attractor. cut of the chaotic attractor by the plane Z = r (b) A 3D (y.
Explain why this is so. while the system x=y ˙ ˙ y=x is not. Linearise the system x = µx − y − x x2 + y2 ˙ y = x + µy − y x2 + y2 ˙ at (0. Show that the linearisation scheme described in section 6. 4. 0).1 is not applicable to the system: x= y ˙ y = −x3 ˙ Show.ThreeDimensional and HigherDimensional Models 133 Questions and Exercises 1. and determine the stability of the system at the origin. The system y= x ˙ is stable at the origin. 3. (a) Find the ﬁxed points and stability of the system: x=y ˙ y = − sin(x) ˙ x = −y ˙ . 6. and discuss the form of these curves for various values of the integration parameter. (b) Solve the system and draw various level curves. furthermore. 5. and check the stability at the ﬁxed point. 2. that the system is stable at the origin. (b) Draw the level curves. Consider the system: x=y ˙ y = −x2 ˙ (a) Characterise the stability of the system at the origin. Consider the system: x = −2xy ˙ y = 2xy − y ˙ (a) Solve the system.
and examine the resulting systems. given by equation (6. Consider the system: y= ˙ x = a − (b + 1)x + x2 y ˙ bx − x2 y (a) Find the equilibrium points of the system. 8. .12) has at least one positive root.134 Chaotic Modelling and Simulation (b) Compare the behaviour of the system to that of the system: x=y ˙ y = − sin(x) ˙ (c) In both cases. M − for the Lorenz model. 12. are stable. (d) What is the diﬀerences when we use the thirdorder Taylor approximation for sin(x)? 7. (c) Analyse the special cases when b = 1 and b = −1. Let b > 0. 10. replace sin(x) with its ﬁrstorder Taylor approximation. 9. Compute the eigenvalues of the Lorenz model. 11.19). Determine if the equilibrium points M + . equation (6. Show that when c − ab > 0. (b) Determine the system’s stability at these points. Explain the butterﬂy eﬀect. Find the ﬁxed points of the system x=y ˙ y = −bx(x − 1) ˙ and determine the stability of the system at the points.
that is. t) ˙ y = G(x. We consider a particular level curve of f .2). that is independent of time. we look at a number of interesting conservative systems. systems. as follows.2 we give a brief account of linear systems of diﬀerential equations. by direct integration of the diﬀerential equation: dy G = dx F (7. The ﬁrst integral derives its name from the usual method of computing it. As a consequence: ∂f ∂f d f (x. every trajectory is part of some level curve of f .1). y) = x ˙ +y ˙ =0 (7. more complex. and consequently we think of f as an implicit equation for y as a function 135 .2) dt ∂x ∂y There is a direct connection between the level curves of the ﬁrst integral and the trajectories of the system: If f is a ﬁrst integral of the system (7. y). that will be helpful in analysing other. Thus. systems that have a nontrivial ﬁrst integral of motion. y) = c. 7. Hence. of x and y.Chapter 7 NonChaotic Systems In this chapter we discuss various aspects of nonchaotic systems. For a system x = F(x.1) A ﬁrst integral of motion is a function f (x. y. f (x. while in section 7. each level curve of f is a union of trajectories. determined by the value of f at the initial conditions (x0 . y0 ).1 we deﬁne conservative and Hamiltonian systems. t) ˙ (7.1 Conservative Systems In this section we consider twodimensional conservative systems. y. These are models for the local behaviour of more complicated systems near their equilibrium points. In section 7. and the qualitative behaviour of their solutions near the equilibrium point. In the remaining sections. then f is constant on every trajectory of the system.3) This follows from (7.
Before we proceed. However. This means that there is a family of conservative systems sharing the same ﬁrst integral of motion. we will call f the Hamiltonian of the system. notice that the trajectories are determined completely by the direction of their tangent vector at any point. The value of the integration parameter is directly related to the level of the curve. y) = x ˙ +y ˙ = −x ˙ +y ˙ = ˙ dt ∂x ∂y dx ∂y ∂y ∂y dx (7. As we saw in (7. G are both multiplied by the same function of x. in the case where F.4) Equating the last term to zero. G. and this direction is that of the vector ( x. G in general depend on t. let us elaborate a bit more on the equations. this integral may not be independent of t.5) Note that this equation will remain true if F.5) can be rewritten as: G F = − ∂f ∂f ∂y ∂x Rescaling F. All the systems we will consider in this section are Hamiltonian. and using equation (7. Therefore.3) will result in a ﬁrst integral of motion for the system. are completely determined from f by: F= ∂f ∂f and G = − ∂y ∂x (7. From all these systems.1). To see this geometrically. multiplying both F. G with the same ˙ ˙ factor has no eﬀect on this direction. y. y). G). and we will call the system a Hamiltonian system. we have Chaotic Modelling and Simulation ∂ f ∂ f dy + =0 ∂x ∂y dx which gives us: d dy ∂f ∂f dy ∂ f ∂f ∂f −x + y ˙ f (x. the ﬁrst integral of motion.136 of x. Since F. G amounts to selecting a value for this ratio. G are independent of the time t. are determined by the equation: F ∂f ∂f +G =0 ∂x ∂y (7.2). since they imply that: ∂F ∂G ∂2 H =− = ∂x ∂y ∂x∂y . f (x. and hence the solutions of the system.3).6) provide a necessary condition for the Hamiltonian to exist. and consequently the implicit equations for the trajectories of the system. Note that (7. there is one that stands out. In that case. we end up with equation (7. integration of (7. so a natural choice would be to set it to 1. y) = (F. in which case F. Equations (7. We will examine a number of particular such systems in this chapter.6) In this case.
from equations (7. y) = 1 2 x + y2 2 (7.1: The simplest conservative system .6).8) (7. Consequently.7) The level curves are therefore concentric circles. 7.1). The equations of motion of Hamiltonian systems can be completely described using their Hamiltonian.NonChaotic Systems 137 In the case where F. FIGURE 7. this is also suﬃcient. Hamiltonian systems play a central role in Hamiltonian Mechanics. Its ﬁrst integral of motion is: H(x. and the Hamiltonian can be computed by simply solving the system (7.1 The simplest conservative system The simplest conservative system is given by x= y ˙ y = −x ˙ The corresponding diﬀerential equation for the integral of motion is x dy =− dx y The solutions to (7. the system is conservative.5) and (7. whose radius is determined by the initial conditions (x0 .6). y0 ).8) are concentric circles x2 + y2 = h where h is a positive constant (Figure 7.1. G are independent of t.
y0 ) + h xx (x − x0 )2 + h xy (x − x0 )(y − y0 ) + hyy (y − y0 )2 + · · · (7. y) = 2 y2 + c(x). 0). y0 ∂y2 are the secondorder partial derivatives of H at (x0 . y) = H(x0 .138 Chaotic Modelling and Simulation Note that this is indeed a Hamiltonian system.7 the origin is the only equilibrium point. y0 . then the Taylor expansion of H around (x0 . this ˙ ˙ corresponds to ∂H ∂H = =0 ∂x ∂y So the equilibrium points of a Hamiltonian system are precisely the critical points of the Hamiltonian H.2 Equilibrium points in Hamiltonian systems Before we proceed to more complicated examples. as: ∂H ∂H = y = x and − ˙ = −x = y ˙ ∂y ∂x We could have come to the same conclusions by solving the system: ∂H = x=y ˙ ∂x ∂H = −˙ = x y ∂y 1 Integrating the ﬁrst equation would yield H(x. then the equation (7. y) = h0 + h xx x2 + h xy xy + hyy y2 + · · · 2 2 . 0).1. y0 ). and the terms ignored are of higher order in x − x0 . y − y0 . y0 ) = H(0.9) 2 2 where h xx = ∂2 H x0 . in the system 7. y0 ) = (0. for simplicity. For instance. ∂x∂y hyy = ∂2 H x0 . Equilibrium points are found by setting x and y to 0. it is worthwhile discussing the behaviour of a system in a neighbourhood of an equilibrium point. and we further denote h0 = H(x0 . y0 ) will be: 1 1 H(x. ∂x2 h xy = ∂2 H x0 . y0 . If (x0 . y0 ) is an equilibrium point. In the case of a Hamiltonian system. If. we change coordinates so that (x0 . and substituting into 1 2 the second equation would yield c(x) = 2 x + c.9) becomes: 1 1 H(x. 7.
NonChaotic Systems 139 So. in the case that the critical point is nondegenerate. in the case where the characteristic polynomial is not identically zero.1) can be approximated by a linear system: x = ax + by ˙ ˙ y = cx + dy (7. then the trajectories near the equilibrium point are elliptical.10) Depending on the number of real roots of the characteristic polynomial2 k(X) = h xx + 2h xy X + hyy X 2 . Hence. 0). real functions of t of a special form.11).4 7. system (7. of course. in a more general analysis. locally. the system will exhibit elliptical orbits near (0. hyperbolic. or parabolic. c and d are real numbers or. of the level curves of H. the orbits will remain elliptical even away from 0. 0).3 If k(X) has no roots. y) = (0. i. For instance. and the point is a stable equilibrium point. 0). We will 1 This number depends of course on the discriminant of the polynomial. This system has.7) has an equilibrium point at (0. 3 That is.2 Linear Systems Near an equilibrium point (we will assume this point is 0. whereupon H is written. If k(X) has two distinct roots. then the trajectories are hyperbolas. 4 In this case. near the equilibrium point. these trajectories may look elliptic.e. b. the system (7. 0.11) where a. an equilibrium point at (x. 0). and their asymptotes are the lines with slopes the two roots (this will be elaborated further in the next section). as: 1 2 1 2 x + y 2 2 In this case. which is the determinant of the Hessian matrix. i. We will now discuss in greater detail the classical theory of system (7. 2 This quadratic form is the quadratic form associated to the Hessian matrix: ∂2 H ∂2 H ∂x2 ∂x∂y 2 ∂ H ∂2 H ∂y∂x ∂y2 .e. which has no real roots. the behaviour of the trajectories of the system. the polynomial in question is 1 + X 2 . as already mentioned. is determined by the quadratic form:1 1 h xx x2 + 2h xy xy + hyy y2 2 (7.
as usual. we can write the equation for the eigenvalues as: λ2 − tr(A)λ + det (A) = 0 The solutions to this are tr(A) ± λ1.13) Before we proceed to a classiﬁcation of equilibrium points according to the qualitative behaviour of the solutions near these points. and they are given by the equation A − λI = or.1 Transformations on linear systems We start by considering a simple coordinate rescaling: u = kx v = k′ y (7.2. in simpler terms: λ2 − (a + d)λ + (ad − bc) = 0 If we note.11). is: A′ = 1 rc a rb d . 7.140 Chaotic Modelling and Simulation denote by A the matrix of coeﬃcients: A= ab cd The eigenvalues of this matrix are called the eigenvalues of system (7. au + rbv 1 cu + dv r So the matrix of the new system.14) The transformed system (7. ∆ denotes the discriminant: √ ∆ (7.2 = 2 where. that a + d and ad − bc are respectively the trace and determinant of A.11) then takes the form u= ˙ v= ˙ where r = k k′ .12) a−λ b =0 c d−λ ∆ = tr(A)2 − 4 det(A) (7. we will discuss the eﬀects of various transformations on the shape of the matrix A. A′ .
(7. and hence it amounts to a parallel translation of the eigenvalues by −k. Another simple transformation involves time rescaling:5 u(t) = x(kt) v(t) = y(kt) System (7.16) (7.17) A′ = A − kI tr(A′ ) = tr(A) − 2k .NonChaotic Systems The trace and determinant thus remain invariant: tr(A′ ) = tr(A) det(A′ ) = det(A) A more general change of coordinates amounts to the transformation uv =C xy 141 (7. that this transformation does keep the discriminant ∆ = tr(A)2 − 4 det(A) invariant. this again keeps the trace and determinant invariant. This transformation changes radically the behaviour of solutions near (0. the eigenvalues are simply rescaled by k.13). The corresponding matrix A′ is then a matrix similar to A: A′ = CAC′ Hence. provided k > 0. and we will utilise it in the next section. but in a very controlled way. 0). the transformation u(t) = e−kt x(t) v(t) = e−kt y(t) results in a system with matrix with the corresponding changes this implies to the trace and determinant: det(A′ ) = k2 − k tr(A) + det(A) Note however.15) where C is an invertible matrix of constants. Finally.12) and (7. 5 This does not alter in any essential way the behaviour of solutions near the point (0.11) then becomes: u = kau + kbv ˙ ˙ v = kcu + kdv Thus the matrix of the system becomes A′ = kA and hence tr(A′ ) = k tr(A) det(A′ ) = k2 det(A) According to (7. 0).
we may assume to be ±i. i. After a change of coordinates. and we will do so for the next couple of paragraphs.2(c) and 7. the sign of tr A is of paramount importance. we may assume f (x. Then. after performing transformation (7. that the linear system is: x=0 ˙ y=x ˙ The solutions are then: x(t) = A y(t) = At + B This case is illustrated in Figures 7. y) = cx2 − 2axy − by2 We have to distinguish a couple of cases here.11) to be conservative.2.e.17). i.e. which. i. Then there are two distinct real eigenvalues. there are two complex eigenvalues. i. we can always reduce to the case where tr(A) = 0. y) = x2 + y2 . y) = xy. Using transformation (7. ∆ > 0. 1. 2.142 Chaotic Modelling and Simulation 7.2 Qualitative behaviour at equilibrium points We will now proceed to analyse the qualitative behaviour of solutions near equilibrium points. eigenvalue. the ﬁrst integral f is degenerate. and there is only one. In that case. which. y) = x2 . we may assume to be ±1.16).2.e.16). we may assume that f (x.e.2(d). after performing tranformation (7.1. namely 0.2(a) and 7. Then. we may assume f (x. where d = −a.e. the system has the ﬁrst integral: f (x. double. that the linear system is x = −y ˙ y= x ˙ The solutions are then: x(t) = A cos(t) + B sin(t) y(t) = B cos(t) − A sin(t) This case is illustrated in Figures 7.2(b). that the system is: x=x ˙ y = −y ˙ The solutions are then: x(t) = Aet y(t) = Be−t . The condition tr(A) = 0 is exactly the condition for system (7. using the transformations from section 7. First oﬀ. ∆ < 0. 3. up to a change of coordinates. i. ∆ = 0. Up to a change of coordinates.
tr(A) > 0 FIGURE 7. tr(A) > 0 (e) Saddle point. ∆ > 0. tr(A) = 0 (b) ∆ = 0. ∆ < 0 . tr(A) = 0 (d) Spiral source.NonChaotic Systems 143 (a) ∆ = 0 . tr(A) > 0 (c) Center point.2: Qualitative classiﬁcation of equilibrium points . ∆ < 0. tr(A) = 0 (f) ∆ > 0 .
2. The equilibrium point is then called a saddle point. and the point is classiﬁed as unstable.3. In this case. The behaviour in this case depends a lot on their sign. Then λ1. If they are both negative. The critical point is then called a center. the real part is positive.2 are both real. and the point is then called an unstable point or a spiral source. indicated by the negative and positive eigenvalues respectively. then the solutions spiral toward the equilibrium point. If the real part is negative. If they are both positive. ∆ < 0. but with opposite signs. This is the degenerate case. 7.2 are both real. the behaviour of solutions near an equilibrium point falls under one of the following cases: 1. when the real part is zero). To summarise.18) . or also a stable node. then the solutions spiral away from the critical point.2(f). y) = − x2 y2 y3 −b + +h 2 2 3 (7. ∆ > 0.2 have nonzero real and imaginary parts. which in this case is called a stable point. then the solutions tend away from the equilibrium point.e. Then λ1. A special case is when λ1. Otherwise. on the other hand. The ﬁrst such system is: x = y(y − b) ˙ y=x ˙ This is a Hamiltonian system.144 Chaotic Modelling and Simulation This case is illustrated in Figures 7.2(e) and 7. λ1. 3.1 EggShaped Forms A simple eggshaped form We now proceed to examine certain Hamiltonian systems that give rise to very interesting eggshaped forms.3 7. since ∂ ∂ y(y − b) = 0 = − (x) ∂x ∂y The Hamiltonian is easily seen to be H(x. which is called in this case a stable spiral or a spiral sink. then all solutions tend toward the equilibrium point.2 have only an imaginary part (i. ∆ = 0. there is a stable direction and an unstable direction. If.
− has two imaginary roots: b The second point.3(a) (b = 0. since its characteristic polynomial.2 A double eggshaped form A double eggshaped form with interconnected sharp corners is expressed by the following system: x = −y(y − a)(y − b) ˙ (7. . whereas the bottom of the form is at the point 0. y) = (0. For convenience. y) = 0 This will have a double root at y = b. namely x = ± respectively. that H is 0 at this. √ √ −2 + 2bX 2 = 2( bX − 1)( bX + 1) i λ1.6). as the quadratic form.2 = ± √ 1 2 x + by2 2 145 1 λ=±√ b These are exactly the slopes of the two tangent lines at (0. − 2 . marked with a heavy line. b3 3. It has equation: H(x. is unstable. y) = − x2 y2 y3 b3 −b + + 2 2 3 6 The sharp (top) corner of the eggshaped form is located at the equilibrium point b (0. we will choose the integration constant in H to be such. at (x. The speed at these points is y = ± ˙ b3 3 7. b) The ﬁrst of these is a stable point. equilibrium point.6 At this point. namely: h= b3 6 has two real roots: The level curve that passes through (0. and one more root.3. second. b).NonChaotic Systems This system has two equilibrium points. the speed of the particle is given by x = 3 b2 . y) = (0. 0) and (x. The boundaries of the eggshaped form in the x ˙ 4 direction are when y = 0.19) y=x ˙ 6 This is found by solving the cubic polynomial H(0. however. in Figure 7. b) is shown. b).
Inside this form the 6 trajectories follow closed loops. so that the Hamiltonian is zero at the unstable point. When the initial values are x0 = 0 and y0 = −b.3.0 0. This is a characteristic level curve marked by a heavy line in Figure 7. The ﬁrst eggshaped form at the top of the ﬁgure has a stable equilibrium point at (a. whereas the other eggshaped form has a stable equilibrium point located at (0.3 −0.2 −0. 0).0 −0.1 0.4 0.2 −0. y) = − y3 y2 x2 b3 y4 + (a + b) − ab − + (b − 2a) 4 3 2 2 12 Figure 7.2 −0. An unstable point is located at (0.2 0.3(b).2 0.3 A double eggshaped form with an envelope A double eggshaped form enclosed in an external envelope is provided by the following system: x = y(y2 − b2 )(y + a) ˙ (7.1 0. . b).146 Chaotic Modelling and Simulation 0.2 0. a ﬁgure eight shape appears. The Hamiltonian in this case is7 H(x.3(b) shows the trajectories when a = 1 and b = 0.3: Eggshaped forms where we will assume that 0 < b < a.1 0. 7. The integration 3 b4 constant has the value h = − 12 + ab for this particular case.6 0.3 −0.0 0.0 0.5 0.3 (a) An eggshaped form (b) A ﬁgure eight form FIGURE 7. and the same holds outside this characteristic level.20) y=x ˙ 7 The constant was again chosen.6.0 0. where the sharp corners of the two eggshaped forms coincide.8 1. 0).4 −0.3 −0.1 0.
the trajectories diverge to inﬁnity.4: A double eggshaped form with an envelope . b). outside of it. where the envelope is formed.0 −0. y) = y5 y4 y3 y2 x2 + a − b2 − ab2 − 5 4 3 2 2 where we’ve set the integration constant to zero for simplicity. whereas.5 0.21) Inside the envelope.4 FIGURE 7.6. −b). For this particular case. where the sharp corners of the two eggshaped forms coincide. the trajectories follow closed loops.4 −0. 0). There are two unstable points. one at (0. and one at (0. whereas the second eggshaped form has a stable equilibrium point at (0. The ﬁrst eggshaped form at the top of the ﬁgure has a stable equilibrium point at (0.0 0. Figure 7.0 −0.2 0. −a).2 0. The level curves through those two unstable points are marked with a heavy line in the graph.4 illustrates this case where a = 1 and b = 0. 1. corresponding values of the Hamiltonian on these curves are 2 5 b − 15 2 5 b − 15 1 4 ab 4 1 4 ab 4 hb− = − hb+ = (7.0 0.NonChaotic Systems The Hamiltonian of this system is given by: 147 H(x.5 −1.
marked with a thick line. y) = y3 + x3 y2 + x2 −b 3 2 3 (7. and is illustrated in Fig6 ure 7. A simple symmetric Hamiltonian form is based on the system: x = y(y − b) ˙ y = x(x − b) ˙ The corresponding Hamiltonian is H(x. b) respectively. 0). 0) and (0. There are two stable equilibrium points. y) = y3 + x3 by2 + ax2 − 3 2 (7. The three points of the level curve located on the line x = y are: the point b .24) There are two characteristic level curves.5(a). The stable equilibrium points are located at (0. There are two stable points. b . at (0. These level curves are shown in Figure 7. 0) and (a. The equations are: x = −y(y − b) ˙ y = x(x − b) ˙ The Hamiltonian is H(x. 0) and (b. y) = y2 − x2 y3 − x3 −b 3 2 (7.22) The level curves at level 0 are marked by a thick line in Figure 7.23) The level curve for the outer periphery has level − b . whereas the two unstable points are located at (0. An interesting threefold shape is given by the system x = y(y2 − b2 ) ˙ y = x(x − b) ˙ (7. at (0. b). b).6. 0) 1 1 and (0.148 Chaotic Modelling and Simulation 7. at (b. and the other two solutions of 2 2 the equation 4x3 − 6bx2 + b3 = 0.25) . b). b) and (b. and with corresponding levels ha = − 6 a3 and hb = − 6 b3 respectively. 0) and (b. and two unstable points.5(b). A slight change to (7.22) gives rise to another doubleshaped form. passing through the unstable points (a. with symmetry with respect to the y = x axis. b). A simple nonsymmetric form is expressed by the following two equations: x = −y(y − b) ˙ y = x(x − a) ˙ The Hamiltonian in this case is H(x.4 Symmetric Forms We consider now certain forms that exhibit symmetry.
8 (a) (b) FIGURE 7.4 −0.5 0.6 0.2 0.4 0.8 0.2 0.0 0.0 0.4 0.2 0.5 1.0 −0.8 0.4 0.4 −0.6 0.0 −0.NonChaotic Systems 149 0.2 −0.8 0.4 0.0 0.2 −0.0 −0.4 −0.4 −0.5 −0.6 0.2 0.0 FIGURE 7.0 0.2 0.2 0.6 0.5 0.5: Symmetric forms 1.6: A simple nonsymmetric form .
0).5 1.0 1.0 0. corresponds to the equilibrium point (0.0 1.2 −0. and the 3 2 corresponding graph is shown in Figure 7. There are three stable equilib4 3 rium points.0 0. we get a ﬁgure 3 similar to Figure 7.0 0.150 Chaotic Modelling and Simulation The corresponding level curves of the Hamiltonian y4 y2 x3 x2 − b2 − +b 4 2 3 2 are shown in Figure 7.26) y = x(x2 − b2 ) ˙ .85). and gives the central formation.5 −0. 3 4 The second characteristic level is h = b6 − b .0 0.6 0.5 0.4 0.5 −0.6 −0.6 (c) b = 0.7: A symmetric threefold form There are two characteristic levels for H. which provide the level curves marked by a heavy line.5 0. for which the level curve passes 4 through the two unstable equilibrium points (b. the two curves are not so interrelated (Figure 7.0 −0.5 1. When b > 2 . The shape varies a lot depending on the parameter b.4 0. h = 0. and one at (b.7(b).2 0.5 More Complex Forms More complex forms appear by adding new terms in the righthand side of the Hamiltonian equations of the twodimensional system.0 0. The fourtype symmetric form illustrated in Figure 7.7(c)). ±b).4 −1.2 0. the two characteristic curves coincide.2 0.8(a) corresponds to the system x = −y(y2 − b2 ) ˙ (7.85 > 2 3 (b) b = 0. 7.7(a) (b = 0.0 0.7(a). 0) 2 where x takes its maximum value in the bounded part of this curve. where H = b6 . ±b).6 −0. two at (0. 1. 0). The ﬁrst value.5 0.5 −0. When b < 3 . When b = 2 .3 < 2 3 FIGURE 7.0 −0. This curve also passes through the point ( 3 b.0 (a) b = 0. where H = − b4 .0 −1.5 0.
4 a heavy line marks this critical level curve. b) and (−b.7. for values H = − 1 a4 and H = − 1 b4 respectively. Introducing a new parameter a into (7. There are also four unstable equilibrium points located at (0. (0. Two distinct level curves appear.3. 2 . b = 0. with parameters equal to a = 0. 0). (b.8(c). 0).26). then an interesting system arises: x = −y(y − a)(y − b) ˙ (7. (a. then hb = 0.NonChaotic Systems with Hamiltonian: x4 + y4 x2 + y2 − b2 4 2 151 There are ﬁve stable equilibrium points. and the other four located at the symmetric points (b.8(a). a). one at the origin (0. −b).8(b). 0). y) = x4 + y4 x3 + y3 x2 + y2 − (a + b) + ab 4 3 2 The model. −b). and the system leads to a perfect symmetry. also marked with a heavy line. The 4 level curve passing through those equilibrium points has level − b . a). −b). illustrated in Figure 7. 1 There is an outer characteristic level curve. (b. The four points located on the x axis (y = 0) are given by: x = ± a2 ± √ a4 − b4 If the parameter a is introduced in both equations. with level hb = − 12 b4 + 1 ab3 .5).27) y = x(x2 − a2 ) ˙ Its Hamiltonian is x4 + y4 a2 x2 + b2 y2 − 4 2 This system has a richer structure.8 and b = a = 0. 0) and (−b. In Figure 7. (a. b) and (b.8(b) (a = 0. and four unstable points at (b. as illustrated in Figure 7. (0.4. (b. is illustrated in Figure 7. 4 4 The ﬁrst curve is associated with two 8shape forms of Figure 7. with level ha = − 12 a + 6 a b. b). 1 When b = 2 a. The second curve is associated with the other more complicated form.8 and b = 0. 0).8(d) when the parameters are a = 0. b). a). There are ﬁve stable points located at (0. the following system arises: x = −y(y2 − b2 ) ˙ (7. (−b. b). and 6 1 4 1 3 an inner characteristic level curve. 0) and (0.28) y = x(x − a)(x − b) ˙ Its Hamiltonian is: H(x. b). (a.
2 0.2 −0.152 Chaotic Modelling and Simulation 0.0 −1.4 0.2 0.0 0.8 0.8: More complex forms .5 0.5 0.2 −0.0 0.8 0.0 0.0 −0.0 0.2 0.5 −0.0 (c) A complicated form (d) A perfect symmetry FIGURE 7.0 0.0 (a) A 4type symmetric form (b) Two 8shape forms with an envelope 1.5 0.0 −0.4 0.0 0.0 1.6 0.2 0.6 0.4 0.6 0.8 1.2 0.8 1.5 1.0 −0.6 0.5 −1.4 0.0 −0.2 0.0 −0.5 −1.5 0.0 0.
5 and f = 2.9(b).6 HigherOrder Forms We conclude this section with a couple of much more complicated examples. c = 0. e = 1.9: Higherorder forms appropriate values for the parameters lead to symmetric forms. One such example.5. Equilibrium regions and regions with high movement (ﬂows) appear. There are 13 stable equilibrium points.1. is the system: x = −(y − a)(y − b)(y − c)(y − e)(y − f ) ˙ (7.9. where the parameters are a = 0. and 12 unstable equilibrium points. (a) Symmetry only along y = x axis (b) Multiple symmetries FIGURE 7.29) y = (x − a)(x − b)(x − c)(x − e)(x − f ) ˙ The Hamiltonian can be easily computed from the above equations. e = 1. arising from systems whose equations involve higherorder polynomials. The parameters were set to a = 0. illustrated in Figure 7.NonChaotic Systems 153 7. b = 0. as in Figure 7. .9(a). As in the simpler cases. b = 0. c = 1.6.5 and f = 2. but we will omit the tedious task here.
and ﬁnd the type of stability at the equilibrium points. (a) y = ax ˙ x = y(y − b) ˙ (b) x = y2 − b2 ˙ y=x ˙ (c) y = x 2 − a2 ˙ x = y2 − b2 ˙ 4. and (d) plot. including the curves passing through the ﬁxed points. y 2. and classify them. y) plane. (e) Draw various trajectories of the system in the (x.154 Chaotic Modelling and Simulation Questions and Exercises 1. (b) Find the eigenvalues and eigenvectors at the equilibrium points. Consider the system y = x(1 − x) ˙ (a) Find its equilibrium points. 3. (g) Find the maximum speed of the ﬂow at the y direction (˙ ). Consider the system y = x + ay ˙ x=1− y ˙ x = −y ˙ (a) Find the equilibrium points for the system. . (a) show that it is Hamiltonian. the level curves of the Hamiltonian. Show that the system x = 2xy ˙ y = x2 − y2 ˙ is conservative. (d) Find a ﬁrst integral of motion for the system. (c) Characterise the equilibrium points for various values of a. (b) Find the characteristic equation at the equilibrium points. For each of the following systems. (c) determine the ﬁxed points. (f) Draw the trajectories that pass through equilibrium points. (b) compute the Hamiltonian function. (c) Linearise the system. for various values of the parameters. (d) Examine in particular the case a = 2. Find a system equivalent to it that is Hamiltonian. but not Hamiltonian. and characterise the points according to their stability.
Show that each of the transformations described in section 7.2. Show that the system 1 x 1 y=− ˙ y x= ˙ 155 is conservative.NonChaotic Systems 5. but not Hamiltonian. Find a system equivalent to it that is Hamiltonian. .1 have the indicated eﬀect on the matrix A and its eigenvalues. 6.
.
then the tranformation (8.4) yn+1 = xn sin(∆θ) + yn cos(∆θ) When the rotation is followed by a translation equal to a and parallel to the x axis. A wealth of very interesting models arise from this seemingly simple change. the eﬀect of rotation can be written directly using the following.Chapter 8 Rotations In this chapter we examine the eﬀects of introducing aﬃne transformations to models.4) is replaced by: xn+1 = a + xn cos(∆θ) − yn sin(∆θ) (8.1) where r = x2 + y2 and θn is the rotation angle.5) yn+1 = xn sin(∆θ) + yn cos(∆θ) 157 . yn ) of a system is given in parametric form (rn . 8.3) (8.1 Introduction xn = rn cos(θn ) yn = rn sin(θn ) The location (xn . in particular the eﬀect of introducing rotation. θn ) by: (8. If we assume that transitioning from the nth state to the n + 1st state amounts to a rotation by an angle ∆θ.2) Therefore. which may be familiar to you already: xn+1 = xn cos(∆θ) − yn sin(∆θ) (8. then the new coordinates would be given by: xn+1 = rn cos(θn + ∆θ) yn+1 = rn sin(θn + ∆θ) or after using the familiar trigonometric identities for the sum of two angles: xn+1 = rn cos(θn ) sin(∆θ) − sin(θn ) sin(∆θ) yn+1 = rn cos(θn ) sin(∆θ) + sin(θn ) cos(∆θ) (8.
we obtain the diﬀerential equations: ˙ x = a + x cos(∆θ) − 1 − y sin(∆θ) ˙ x sin(∆θ) (8. taking into account that x= ˙ dx ∆x ≈ = xn − xn−1 dt ∆t and similarly for y. Let us work out an analogous system of diﬀerential equations. whose ﬁrst integral of motion can be computed from the diﬀerential equation: ∂y x∆θ = ∂x a − y∆θ ady = (∆θ)rdr (8. Then. depends only on the distance r from the origin. can be simpliﬁed to: x = a − y∆θ ˙ (8. using ﬁrst order Taylor approximations for cos and sin. based on the transverse component of the acceleration. .e. and not of the direction. i. that: ∆θ ≈ h1 r2 Another important case is that of a mass M rotating in a circular orbit under a central force MG f (r) = 2 r 1 We have used here that r2 = x2 + y2 . A reasonable assumption would be that ∆θ is “radial.5) as a simple system of diﬀerence equations.158 Chaotic Modelling and Simulation We can consider the system (8.6) y= ˙ + y cos(∆θ) − 1 8. the system (8.8) The solution of equation (8. A standard choice in mechanics.” i.2 A Simple RotationTranslation System of Diﬀerential Equations If the rotation angle is small.8) depends on the form of the function ∆θ. The time elapsed between two successive iterations is ∆t and is usually set to 1.6).7) y= ˙ x∆θ This equation leads to the form:1 This is a Hamiltonian system. and consequently rdr = xdx + ydy. is that the rotational change is inversely proportional to the square distance. ∆θ ≪ 1.e.
1: The limiting trajectory . The equations of motion then result in ˙ ∆θ ≈ θ = In this case. equation (8. y) = 0. When h < MG the rotating mass moves inside the space bounded by a the above trajectory.9) The ﬁxed point of the system is at (x.Rotations where G is the gravity constant.8) becomes: ady = Its solution is r= MG dr r MG r3 159 (ay + h)2 4MG where h is an integration constant. at which point the value of h is: h= MG a MG a2 The path of the trajectory in the (x. When h > MG . the trajectories diverge and the rotating object heads a oﬀ to inﬁnity. This can also be rewritten as x2 = (ay + h)2 −y 4MG (ay + h)2 +y 4MG (8. FIGURE 8. y) plane for this value of h divides the plane in two segments.
7) we get: dy y ˙ x∆θ = = = dx x a − y∆θ ˙ h 4a MG a2 . then x= so from (8. z = 5 − 2. and the solution to the equation: a2 (ay + h)2 = −4ahy The largest solution of this equation is: √ MG MG y= 2 2−3 ≈ −0. The second is the one we MG √ ( 5 − 2) a2 MG h h2 h 8a2 = . z = − 5 − 2 The ﬁrst of these corresponds to the ﬁxed point 0. which is not stable. This can be found by setting x = 0 in equation (8. that is: ˙ a MG 3 ∂x x 1 = = ay + −y =0 (8.1. r= = and ∆θ = 2 4a 4MG 4a h 4a · 8a h =2 a 2 . The trajectory in the limit of escape has an egg shape.9).160 Chaotic Modelling and Simulation Some very interesting properties of equation (8.11) becomes: (z + 1)3 = 8z This can be easily seen to have the three roots: √ √ z = 1. and the corresponding y value is: y= When y = 0. similar to but diﬀerent from those discussed in Chapter 7.8) are illustrated in Figure 8. are after. The minimum value of y is that where the maximum rotation speed is achieved. This gives us two solutions. The sharp corner is at the maximum value of y = MG . This is the ﬁxed point for the system.10) xmax This is computed by equating to zero the ﬁrst derivative of x with respect to y.1716 2 2 a a The maximum value of x is given by: MG = 2 a √ MG 5 5 − 11 ≈ 0. a2 where x = 0. the ﬁxed point at y = MG as already discussed.11) ∂y y x 8(MG)2 ˙ a a Setting z = h y.3 2 2 a (8. equation (8.
Rotations 161 This tangent appears in Figure 8.2(a) illustrates the path of a moving particle. z).1.8) in a coordinate system centered at this point. The paths outside the eggshaped limits a2 . z) = (0. our orbit behaves exactly like the function: 1 1 0 = −x2 + z2 = √ z − x 2 2 So the two tangent lines are given by the equations: y= h √ ± 2x a 1 √ z+x 2 (a) Rotating paths (b) Paths in the plane FIGURE 8. a large number of paths are drawn.2: Rotating paths Figure 8. In Figure 8. in terms of (x.2(b). let: z=y− h MG =y− a a2 Then. For this reason. The tangent at the top sharp corner of the eggshaped path can be seen by writing equation (8. The particle follows an anticlockwise direction. equation (8.8) can be written as: 0 = −x2 + a 2 2h a 1 z · + 2z + z2 = −x2 + z2 + · · · 4h a 4h 2 where the omitted terms are of higher order in z. it escapes to inﬁnity following the tangent on the top of the eggshaped path. 0). This means that near (x. It starts from a point near zero and rotates away from the centre until the highest permitted place. Then. Every path has the same value for h = MG .
05. The iterative formula used for the above simulations is: √ 3 xn+1 = xn + a − yn MGrn d √ 3 yn+1 = yn + xn MGrn d (8.13) The escape speed for a mass rotating around a large body is known from mechanics to be: 2MG vesc = y Thus. MG = 100 and d = 0. equation (8. at the top and at the bottom of the eggshaped ˙ ˙ form only the x component of the velocity is present. . for the above ﬁgures.12) rn = 2 xn + y2 n where. the value of y at the top. from this last equation the value of y is: y= MG √ ( 2 − 1)2 a2 2MG y This is. up to a sign. Thus. This speed must be y equal to a. x = 0. or the transverse component of the velocity.10). y. Thus. the same result obtained in equation (8. MG =a y and MG a2 To obtain the equation for the total speed at the lower point of the curve.162 Chaotic Modelling and Simulation have direction from left to right and they do not transverse the eggshaped region of the plane. a = 2. This egglike form is a set of vortex curves with relative stability and strength. The rotational speed. is estimated from the equality v − a = 0. we must take into account that this speed must be equal to the escape speed: y= v + a = vesc (8. is v = MG . it is clear that the parameter a is a constant speed with direction parallel to the x axis. The rotational speed has two components: x. However. It is obvious that the system in these particular cases obeys the laws of classical mechanics. without using the equations for the trajectories. since x = 0. From the equations of motion. Other interesting properties of the aforementioned eggshaped forms arise by using elements from mechanics.13) becomes: MG +a= y Finally.
In the equations (8. and the translation parameter is a. No approximation is needed.Rotations 163 8. The points inside the disk remain inside it under the map. A very interesting property of the map (8. and the ﬁnal equations used are the following xn+1 = a + b (xn cos ∆θ − yn sin ∆θ) yn+1 = b (xn sin ∆θ + yn cos ∆θ) where the same approximation ∆θn ≈ MG 3 rn (8. The iterative scheme based on the diﬀerence equations for rotationtranslation is more diﬃcult to handle analytically. the trajectories inside the eggshaped boundary are perfectly closed loops. A convenient formulation of the last iterative map expressing rotation and translation is achieved by using the following diﬀerence equation fn+1 = a + b fn ei∆θ (8. we introduce an area contracting parameter b (0 < b < 1). the chaotic nature of the rotationtranslation procedure near the centre becomes apparent.2 To locate the ﬁxed points of this transformation.4). Moreover. on the other hand. the eggshaped scheme and the trajectories are retained. whereas.15) where fn = xn + iyn . The Jacobian determinant of this map is J = b2 = λ1 λ2 . The eigenvalues of the Jacobian are λ1 and λ2 . y) plane. Additionally. it is closer to a real life situation. .3 A Discrete RotationTranslation Model The analysis above is signiﬁcant in helping us understand the dynamics underlying the rotation analogue in the discrete case.14) is the existence of a disk F of radius ab 1−b centered at (a. let us change coordinates by setting: a zn = fn − 1 − b2 Then the transformation can be rewritten as: zn+1 − zn = bei∆θ − 1 zn + ab ei∆θ − b 1 − b2 2 This fact is left as an exercise for the reader. 0) in the (x.14) is used. This is an area contracting map if the Jacobian J is less than 1. but.
16) back to an equation for f = fn .164 Chaotic Modelling and Simulation So for a ﬁxed point we would have: z= It can easily be veriﬁed that ei∆θ − b =1 1 − bei∆θ and consequently all ﬁxed points lie on the circle centered at ab . particularly when chaotic behaviour is present. As a possible explanation for this symmetry. where z is as usual the complex conjugate of a complex number. the eggshaped form is shifted to the right of the original positions of the system of coordinates. In the diﬀerence equation case. This is diﬀerent from the symmetry axis of the diﬀerential equation analogue where the symmetry axis is the line x = 0. if we transform equation (8.16) and with radius + y2 = a2 b2 (1 − b2 )2 Further. then we see that: gn+1 = a − f¯ = a − a + f¯ e−i∆θ = (−a + gn )e−i∆θ n+1 n 3 The derivation of this is also left as an exercise to the interested reader. the ﬁxed points f further satisfy:3 cos(∆θ) = a2 1 1 + b2 − 2 2b r (8. let us consider the map in the form: fn+1 = a + fn ei∆θ For any complex number z. in other words they satisfy: 1−b2 x− a 1 − b2 2 a . . Be begin our analysis with the case where no area contraction takes place. we get: f = ei∆θ − b a a 1+b = 1 − b2 1 − bei∆θ 1 − bei∆θ Therefore. 2 If we denote the reﬂection of fn by gn .17) This equation can easily be solved provided that the angle ∆θ is a function of r. it can be easily seen that its reﬂection with respect to the ¯ ¯ axis x = a is a − z. An interesting property of the system (8. In this case the disk F is replaced by the entire plane.5) in this case is that the resulting map often exhibits a symmetry a along the line x = 2 .0 1−b2 ei∆θ − b ab · 1 − b2 1 − bei∆θ (8. so that b = 1.
equilibrium points will tend to exhibit symmetry around this axis.Rotations 165 Multiplying through with ei∆θ . so the formulas don’t quite match up. However. The system may easily escape to inﬁnity when y is higher than MG . we see that the reﬂections satisfy a very similar recursive relation. We proceed now to discuss the equilibrium points of the system. and the evolution of the system from that point on will be very close to the reﬂection of the history of the system up to that point. However. it is not a 2 a2 stable equilibrium point. as y does not show stable magnitude since it is a function of the angle ∆θ. except that it goes “backwards”:4 gn = a + gn+1 ei∆θ Assuming that at some time the map passes very close to the x = a line. a2 (a) Rotationtranslation (b) The central chaotic bulge FIGURE 8. It is clear that there is no symmetry axis in the x direction. This would cause the map to seem symmetric. y = MG is therefore an equilibrium point. an approximation of y is achieved when the angle ∆θ is small. .3: Rotationtranslation. and the chaotic bulge 4 An important detail here is that ∆θ depends on n. unless ∆θ n+1 is close to ∆θn . This leads to x = 2 and a ∆θ y = 2 cot 2 . We are therefore a searching for the points where xn+1 = xn and yn+1 = yn . especially when it exhibits chaotic behaviour. the corre2 sponding point will be very close to its reﬂection. Similarly. Then y≈ MG a = 2 ∆θ a The point x = a .
3(b) illustrate a rotationtranslation case with parameters a = 2 and MG = 100.2. The parameters are a = 0.2. .4(b).3(b). indicated by a small cross and a bigger cross respectively. MG = 100 and the rotation angle is ∆θ = MG . as illustrated in Figure 8. In this case. A chaotic attractor that resembles a bulge appears around the centre of coordinates. r3 In Figure 8. two equilibrium cases appear. third and fourth order equilibrium points appear. and xn+4 = xn and yn+4 = yn respectively.8 for Figure 8.3(a). when a = 1. All the paths are analogous to those obtained by the diﬀerential equation analogue studied above. The relations for the new equilibrium point are xn+5 = xn and yn+5 = yn . In Figure 8.4(b) present an illustration of the chaotic bulge of the rotationtranslation model. A ﬁfth order equilibrium point is added to those presented before. but they are moved to the right at a distance a x = 2 . The convergence is quite good provided that the starting values are in the vicinity of the equilibrium points. an algorithm is introduced for the estimation of the equilibrium points. An enlargement of this attractor is illustrated in Figure 8. In both cases.4(a) and a = 1 for Figure 8. a number of paths inside the eggshaped formation are drawn. Figure 8. in Figure 8. Only a ﬁrst order equilibrium point is present.166 Chaotic Modelling and Simulation Figures 8.4(a). 2 (a) a = 0.3(a) and 8. Accordingly.5(b).4(a) and 8.5(a) illustrates the case where a = 0. higher order equilibrium points appear. In both cases.4: The chaotic bulge Figures 8. This is a symmetric chaotic formation with the line x = a as an axis of symmetry.8 (b) a = 1 FIGURE 8.4(b). As the parameter a takes higher values. one of second order (indicated by small cross) where xn+2 = xn and yn+2 = yn and yn+2 = yn and one of the third order where xn+3 = xn and yn+3 = yn . the following relations hold: xn+3 = xn and yn+3 = yn .
9. the bifurcation is so high that the chaotic region covers all the space bounded by the eggshaped form.6: The chaotic bulge It is diﬃcult to ﬁnd the coordinates of equilibrium points that are located in the centre of the islands that are in the middle of the chaotic sea of the attractor. At this value. a tenth order bifurcation form appears in the periphery of the chaotic attractor.2 (b) a = 1.6(a) and 8.6(a).5: The chaotic bulge Higher order bifurcation is present in Figures 8.5.2 FIGURE 8.Rotations 167 (a) a = 0. The magnitude of the translation parameter is a = 1.6(b).6(b). the translation parameter is a = 1. (a) a = 1. In Figure 8.9) FIGURE 8. In Figure 8. The point masses following chaotic trajectories eventually escape to inﬁnity.5 (b) Complete chaos (a = 1. though .
. A ﬁrst order equilibrium point is characterised by the simple relations xn+1 = xn and yn+1 = yn . The ﬁrst point located on the circle with the larger radius (k = 1) is indicated by a cross. on the periphery of which the equilibrium points must be located: Rk ≈ 3 MG . . Subsequent points follow in decreasing order. . . when the coordinates are xn+2 = xn and yn+2 = yn . A simple case is that of the estimation of the equilibrium points of the second order. The equilibrium points of higher order follow by using diﬀerent relations.168 Chaotic Modelling and Simulation a as discussed earlier those are going to be symmetric with respect to the axis x = 2 .5 FIGURE 8. This assumption leads to the following equation: (x2 + y2 ) cos2 (∆θ) + 2x2 cos(∆θ) + x2 − y2 = 0 This equation is fulﬁlled if the rotation angle obeys the relation ∆θ = (2k + 1)π.7: Equilibrium points The circles Rk introduce a set of decreasing order with regards to the magnitude of the radius. . as illustrated in Figure 8. A few isoclines indicating the limits of the island are drawn around this point.1). (a) a = 0. that is. They are located in the middle of the isoclines. which lead to the following relation for the radius of the circle on the periphery of which the equilibrium point must be located Rk ≈ 3 MG . 2. (2kπ)2 k = 1.7(a) (a = 0. .1 (b) a = 0. this result leads to the following relation for the radius of the circle. 2. Then. . ((2k + 1)π)2 k = 1.
but in opposite directions: one to the positive and one to the negative part 2 of the semiplane for y. The translation parameter is a = 0.5.4 A General RotationTranslation Model The solution of the diﬀerential equation of the model (8.04) (b) Order3 bifurcation (a = 0. (a) Equilibrium pairs (a = 0. The translation parameter is a = 0. The order three equilibrium points are in the a periphery of a circle centred at ( 2 . The ﬁrst two equilibrium points are distributed in the outer circle. Three pairs of order two equilibrium points are illustrated in Figure 8.7) leads to the form ay + h = (∆θ)rdr . the central chaotic bulge has a small radius.8: The chaotic bulge The order three equilibrium points are presented in Figure 8. As this parameter has a very low value. They are located in the middle of the three islands that are in the outer part of the chaotic bulge. They are indicated by a cross and are located on the axis of symmetry x = a .8(a). The parameter is a = 0. The concentric circles are distributed in decreasing order according to the magnitude of the radius Rk . the three equilibrium islands of the ﬁrst order are presented.Rotations 169 The simulation results are illustrated in Figure 8.7(b). Also.7.7) FIGURE 8. The radius of this circle is approximated by 5 R≈ 3 MG a − 4 (2π/3)2 2 8. a ).8(b). The equilibrium pairs are indicated by a cross.04.
Figure 8.9(a)).18) The special case β = 2 leads to the following equation form √ ay + h = r MG or: (x2 + y2 )GM = (ay + h)2 This equation expresses a family of ellipses (Figure 8.170 Chaotic Modelling and Simulation A simple approximation of the quantity ∆θ is: ∆θ ≈ Then. (a) A family of ellipses (b) The chaotic bulge FIGURE 8. the solution has the form ay + h = r2− 2 √ 2− β 2 β MG rβ MG (8. With the exception of β = 2. The parameter a is 6.9: Elliptic forms and the chaotic bulge When β = 4. The axis of symmetry is at x = a . and the constant of integration h takes several values.9(b) illustrates a chaotic central bulge. . y) is: √ MG ln(r) = ay + h The rotating forms are eggshaped and similar to those provided in earlier rotation forms. the cases with β > 2 provide eggshaped rotation forms. where the rotation paths are elliptic. the resulting equation for (x. The outer part of this chaotic 2 attractor is approximated by an ellipse with h = MG + a2 .
10(b). 0).6. in times t = 2. In the limit when the parameter a approaches zero the ellipses turn to concentric circles. However. the translation parameter is high enough so that all of the eggshaped space is contained in the chaotic region. In this region. and a disk of rotating particles of equal mass is centered at (0. After a while. In the case presented in Figure 8. and the particles are trapped inside the eggshaped form and remain there. Figures 8.Rotations 171 Another very important property of the elliptic case (β = 2) is that the rotating particles remain in the elliptic paths even if high values for the parameter h are selected. y) = (a. the translation parameter a is quite small. following the trajectories proposed by the theory discussed so far. The cross in Figure 8.11(a) is at (x. In the second case. In the ﬁrst case. but a part of the system remains around the location (x. the outer part of the rotating object extends to almost the entire space of the shape.12.5 Rotating Particles inside the EggShaped Form Two cases are of particular importance. On the contrary. These coordinates are calculated by using as starting values in a repeated procedure x = a and y = 0. The parameters are a = 0. 0). t = 3 and t = 4 respectively. The translation parameter is a = 0. the majority of the particles will leave the region. they escape by following the escape trajectories and move away from the eggshaped region. The rotation angle is given by ∆θ = GM0 r3 The resulting form that the disk of rotating particles takes after time t = 10 appears in Figure 8. Figures 8. The original cyclic form has now changed. The rotating system of particles is in an intermediate stage. It escapes from the eggshaped form.25 and GM0 = 0. the outer limits of the eggshaped form are drawn.10(a). The case when t = 10 appears in Figure 8. The system is unstable and the rotating particles are not retained inside the eggshaped region. a smaller region inside the trapping region exhibits chaotic behaviour. y) = (a. the particles follow chaotic paths that form the attractors presented above. The diameter of the disk is r chosen as to be exactly within the limits of the eggshaped form. but by following very speciﬁc characteristic paths. 0) inside the eggshaped form.11(a). When .10(d) illustrate the resulting picture after time t = 20 and t = 100 respectively.10(d). A cross indicates the characteristic centres of the chaotic forms. 8. In Figure 8.11(c) illustrate three instances.10(c) and 8. The particles are distributed by follow1 ing the inverse law for the density ρ given by ρ = c3 . providing an outer form of rotation and an inner chaotic attractorlike object.45. 8.11(b) and 8. The iterative procedure is based on the diﬀerence equations for x and y. The distinct inner attractor is more clearly formed.
6). .10: Rotating particles (a) t = 2 (b) t = 3 (c) t = 4 FIGURE 8.172 Chaotic Modelling and Simulation (a) t = 0 (b) t = 10 (c) t = 20 (d) t = 100 FIGURE 8.11: Development of rotating particles over time (a = 0.
However. The leaflike structure starts to disappear inside the eggshaped form when close to (x. The speed of these jets depends on the magnitude of the parameter a and on the original rotation speed.6 Rotations Following an Inverse Square Law We return to the original system of diﬀerential equations x = a − y∆θ ˙ y= ˙ x∆θ (8.12(a)). and a part of this disk lies outside the eggshaped formation.12(c) illustrates the case when a = 0.25) to an original large cyclic cloud has a critical impact to the cloud formation.13(b).8 FIGURE 8.19) . all the particles escape after a short interval (Figure 8.12(b). in Figure 8. y) = (a. all the rotating material escapes outside the eggshaped pattern. (a) a = 0. the cloud is separated into two formations.7 (c) a = 0. the ﬁrst inside the eggshaped form with the chaotic bulgeform in the middle. and the second outside. all the material follows the escape route as a compact formation. On the other hand.6 (b) a = 0. The elapsed time is t = 10. Only the last part of the jet remains connected to the source of the chaotic sea inside the eggshaped pattern. with parameter a = 0. 8. When the translation parameter is a = 1 (t = 10). In Figure 8.12: Development of rotating particles according to the magnitude of the translation parameter. the original cyclic disk is quite large. 0).7 and time t = 10. These jets eject material through the trajectories of the model.8. After t = 100. An intermediate stage appears in the Figure 8. the inﬂuence of a small translation parameter (a = 0. Figure 8.Rotations 173 the magnitude of the translation parameter takes higher values. The speed and the direction of the jets of material coincides to a after a large enough amount of time t.13(a).
This last equation can be transformed in the following form: c1 ln(x2 + y2 ) = ay + h 2 (8. y: x∆θ dy = dx a − y∆θ This equation leads to the following form ady = (∆θ)rdr The solution of this diﬀerential equation depends on the form of the function ∆θ. The function is expressed by the formula θ = c2 . r where c1 is a constant. t = 100 FIGURE 8. t = 10 (b) a = 0. In this section. Provided that ∆t = 1. that deals with the rotation angle.174 Chaotic Modelling and Simulation (a) a = 1. we proceed to the solution of the equation: ady = The solution is c1 ln(r) = ay + h where h is an integration constant.13: Rotations and the resulting diﬀerential equation for x.20) c1 dr r . we use a function known from mechanics.21) (8. the following approximation for ∆θ is obtained c1 ∆θ ≈ 2 r Finally.25. This function arises from the law related to the transverse component of the 1 ˙ acceleration in a circular movement.
The resulting equation for y is c1 ln(y) = ay + h The maximum value for y is achieved when y = ca1 . These values of x. This trajectory is the outer limit of the vortex region of the rotation. this point is not stable. . y) plane.14. FIGURE 8. gives an equation for the path of a trajectory in the (x.278 ca1 .21). When h < hcrit .14: The characteristic trajectory Some very interesting properties of the twodimensional function are illustrated in Figure 8. In this case the maximum or minimum values of y are obtained. The trajectory in the limit of escape is shaped as an egg. y set the derivatives x. When h > hcrit . The resulting equation for y is: ay ay ln +1=0 + c1 c1 A numerical solution gives y = 0. the trajectories diverge and the rotating object heads oﬀ to inﬁnity. we set x = 0 when y = r. which divides the plane in two segments. This maximum value ˙ ˙ is obtained by solving the following equation for y: ln ay ay +1=0 − c1 c1 The minimum value of y is obtained when the rotation speed is maximum. the value of the parameter h at this limit is c1 hcrit = c1 ln − 1 a This value of hcrit . The sharper corner is at the maximum value of y = ca1 where x = 0. the object rotates inside the limits set by the above trajectory. Then. y equal to zero. However.Rotations 175 Exploring the properties of this last equation. applied to equation (8.
15(b).7 and c1 = 100. The tangent at this point is dx = e. z) = (0.176 Chaotic Modelling and Simulation The maximum xmax is estimated by equating to zero the ﬁrst derivative of the following equation for x. the discrete analogue of the model appears. a numerical solution of the resulting equation gives xmax = 0. after simpliﬁcation. using Taylor approximation of the curves in a neighbourhood of the point 0. important properties of the model are presented. the particle follows an anticlockwise direction.15(a).15(b) illustrates the central bulge in the case of a = 2. Figure 8.15(b). As is illustrated in Figure 8.22) becomes: Finally.203 ca1 .15(a). The value of the translation parameter is quite high. we see that equation (8. The tangent at the top sharp corner of the eggshaped form can be computed as before. equation (8. then x = ae . in Figure 8. ca1 . Setting z = y − ca1 . It starts from a point near zero. and it rotates away from the centre until the highest .402 ca1 . This tangent appears in Figure 8. This causes the appearance of an enormous chaotic attractor in the middle of the eggshaped area.21) becomes: 2 c1 2 x + z2 + 2 c1 z + c1 = az + c1 + h ln a a a2 ln (a + x) = ln(a) + where a = c2 1 a2 (8.22) Using the expansion x 1 x2 − + ··· a 2 a2 and x = x2 + z2 + 2 ca1 z.15(a) and 8. the two tangent lines at the point (x. strong bifurcation is present. In Figure 8. the continuous model presented above is simulated. this becomes: 0= c1 c2 c1 a2 2 c1 c1 a4 2 c1 2 2 az + c1 + h = ln 1 + x + z2 + 2 z − x + z2 + 2 z + · · · 2 4 2 2 c1 a 2 2c1 a a c1 2 2 2 a c1 x − 2z2 + · · · 2 In other words. On the other hand. 0) are given by the equation: x2 = 2z2 In Figures 8.14. dy c1 When y = 0. As in the case discussed previously. This is achieved at y = 0. y: c1 c1 ln(x2 + y2 ) = ay + c1 ln − c1 2 a After appropriate diﬀerentiation.
The iterative formula for the above simulations is the following: xn+1 = xn + a − yn yn+1 = yn + xn rn = c1 d r2 c1 d r2 (8.7. c1 = 100 and d = 0. . it escapes to inﬁnity following the tangent on the top of the eggshaped path.23) 2 xn + y2 n The parameters in both ﬁgures are a = 2. Then.Rotations 177 (a) The continuous case (b) The central chaotic bulge FIGURE 8.0001.15: Comparison of the continuous and the discrete models permitted place.
draw escape paths of particles through an eggshaped form. and draw a chaotic bulge and escape paths for both small and large values of the parameter a. expressed via the simpler form (8. y) of equation (8. Construct the level curves in the x.7) go to inﬁnity. Use the equations (8. then fn+1 is also within that circle.x=0 a 2.7) where ∆θ = MG a3 to determine the rotation speed in the x and y directions for the level curves examined in the previous question. Draw trajectories for the discrete case of Figure 8.9) when h = for various values of a.17) for the ﬁxed points of the system (8. Show that the maximum speed in the rotating mass of system (8.15). Show that if z is a complex number. 5.7) is obtained exactly when MG y=− 2 . Derive equation (8. 0) and with radius 1−b .15(a).14). Following the methodology of section 8. Explain why the continuous case of Figure 8. What happens with point outside this circle? 4.14). 9. Consider the system (8. y plane of equation (8. . 13. 12. y plane of equation (8.15(a) does not show any chaotic behaviour. 11. and varying values of a. 2MG a . 8. 10. that if fn is within the circle ab centred at (a. Show. its reﬂection along the line x = ¯ a − z. by fn −a considering the resulting system for zn = ab/(1−b) . 3. Construct the levels curves in the x. a 2 will be 6.9) for h = 0. What is the form of the curves when a = 0? 7. Draw the level curves (x.178 Chaotic Modelling and Simulation Questions and Exercises 1.6. and especially for β = 1. Construct a rotating model providing an eggshaped form. Show that for h > MG a the orbits of system (8.18) for various values of β.
Interested readers may ﬁnd related topics in books on hyperbolic geometry. became the basis of what is now known as Euclidean geometry. galactic object formation. Translation. Reﬂection is generally associated with light refraction. and still plays an essential part in a bevy of topics. The development of science throughout the centuries brought new theories and improvements in all scientiﬁc ﬁelds.Chapter 9 Shape and Form 9. the Euclidean metric remains unchanged. Rotation and translation are present in stellar systems. This chapter includes a brief introduction to the elementary rules of analytical geometry that are used throughout this book. 179 . via simulations based on relatively simple aﬃne transformations. carries his name (the “Euclidean metric”). namely the notions of translation.1 Introduction About 2300 years ago. rotation and reﬂection are not only tools essential in forming geometric objects. The usual notion of distance in space. tornados and vortex movements. along with several applications on the shape and form of chaotic and nonchaotic objects. This theoretical treatment is combined with applications to special relativity and Penrose’s tiling theory. as well as vortex formation. galaxies. chaotic or not. The linear movements of physical objects are easily modelled geometrically by using translation. Even nonchaotic objects are similarly simulated by following the same geometric rules. a large number of chaotic objects that appear in the literature may be classiﬁed as geometric objects produced by following simple geometric rules. rotation and reﬂection. are deﬁned and explored in this chapter. along with applications related to the development of chaotic forms and shapes. This book. The essential components of this theory. Nevertheless. Euclid of Alexandria wrote his famous book on geometry called The Elements. The theory of the simulation of chaotic shapes included in this book is based on an analytic approach that for the most part uses the fundamental notions of traditional Euclidean geometry. but can also be used as a basis of understanding how a process. can generate shapes and forms. As one can readily verify. and its famous ﬁve postulates. essential when we design geometric forms in two or three dimensions.
180 Chaotic Modelling and Simulation 9. and the metric aspect based on the notion of distance. snowﬂakes. Figures with a high degree of symmetry are most interesting. X ℓ O FIGURE 9. The fact that isometries respect collinearity is left as an exercise. and look at their images. i. A standard theorem in Euclidean geometry is that every planar isometry is completely determined by its eﬀect on three noncollinear points.2).1.1: X′ Symmetry/reﬂection along a line An isometry is a transformation which preserves distances. The simplest kind of symmetry in a plane ﬁgure is that of symmetry with respect to a line.1 Symmetry and plane isometries Any subset of the plane is called a ﬁgure. C respectively. it is a transformation T : R2 → R2 with the property that the distance from T (A) to T (B) is the same as the distance from A to B. if the line l bisects and is perpendicular to the segment XX ′ .1). then T D must be the point of intersection of these three circles (Figure 9. we mean simply a transformation that preserves the ﬁgure. There are four basic kinds of planar isometries: . and with the same radius. Spiral galaxies. By symmetry for a ﬁgure A. Isometries respect the two fundamental aspects of geometry: the incidence aspect based on the notion of collinearity. Two points X and X ′ are symmetrical with respect to a line l.e. We will also say that X ′ is the reﬂection of X along ℓ (Figure 9. and with centres A. not only because of interesting forms arising in simulations. B. The proof is fairly straightforward and instructive: Since distances are preserved. the circle centred at A and with a given radius is mapped to the circle centred at T A. It can be easily observed that some ﬁgures are more interesting than others. If we consider the three circles through D. molecules and crystals are examples of objects with symmetric cross sections. but also because they often occur in nature.
3(a) and 9.2: Planar isometries are determined by their eﬀect on three points Translations These are fully described by a single vector v. . the transformation is given simply by the matrix: cos θ − sin θ rot θ = sin θ cos θ in other words: T w = rot(θ) x x cos θ − y sin θ = y x sin θ + y cos θ Reﬂections A reﬂection across a line ℓ passing through the origin and forming angle θ with the x axis is given by the matrix ref θ = and therefore it has the form: T w = ref(θ) x x cos 2θ + y sin 2θ = y x sin 2θ − y cos 2θ cos 2θ sin 2θ sin 2θ − cos 2θ Glide Reﬂections A glide reﬂection is a translation followed by a rotation along a line parallel to the translation direction. You can see that a glide reﬂection can be thought of also as the composition of a rotation followed by a reﬂection (or equivalently. The translation and rotation can be performed in either order. Assuming the rotation is around the origin.3(b). An example of a glide reﬂection is given in ﬁgures 9.Shape and Form 181 TB B D A C TA TD TC FIGURE 9. Such a transformation is given simply by:1 Tw = w + v Rotations These correspond to a rotation with a given angle and around a speciﬁc point. three reﬂections along lines that do not all have a point in common).
y) with the vector w starting at the origin and ending at (x. B to the points O. A.3) (9. along the line ℓ1 . followed by a reﬂection FIGURE 9. takes the points O. 2 Since a reﬂection ﬁxes all points on the reﬂection line. The ﬁrst reﬂection. A. which we summarise here: rotθrotφ = rot(θ + φ) rot(0) = I refθrefφ = rot(2(θ − φ)) refθrotφ = ref(θ − φ/2) rotφrefθ = ref(θ + φ/2) refθrefφrefy = ref(θ − φ + y) (9. as well as in the rest. B in Figure 9. along the line ℓ2 . takes those 1 Here. followed by a reﬂection O (b) Glide reﬂection as a translation.3: Glide reﬂection seen in two diﬀerent ways The matrices rot θ.182 Chaotic Modelling and Simulation TO ℓ′ TB ℓ TA B A TA B A TB TO O (a) Glide reﬂection as a rotation.5) (9.1) (9. y). B′ .6) We leave the veriﬁcation of these identities to the reader. It is worth pointing out that the composition of two reﬂections with the same centre is a rotation with angle twice the angle between the two reﬂection lines: ref θ ref φ = cos 2(θ − φ) − sin 2(θ − φ) = rot 2(θ − φ) sin 2(θ − φ) cos 2(θ − φ) For a geometric proof of this fact. A. . consider the eﬀect of the combined transformation on the points O. we will identify a point P(x.2 and the second. ref θ have a series of useful properties.4) (9.2) (9.4(a).
ℓ2 B′ A B ℓ1 C ′′ = C ′ B′′ A′′ = A′ ℓ2 ℓ1 C B A B′ A ′ O (a) The composition of two reﬂections along intersecting lines is a rotation around the intersection (b) The composition of two reﬂections along parallel lines is a translation FIGURE 9. The relation between isometry groups and symmetries is given by the following theorem. though it might move the points of S around).4: The composition of two reﬂections A very striking result is that these four types. B′ . A′ . which leaves everything ﬁxed. already known to Leonardo da Vinci: All ﬁnite isometry groups can occur as symmetry groups of ﬁgures. The same eﬀect would be obtained by the aforementioned rotation. A shape/set S is said to be left invariant by a transformation T . The result is a translation with a vector twice the distance between the two lines (Figure 9. We omit the identity. you might be wondering. . 2. while a reﬂection ﬁxes every point on the line of reﬂection. if the sets T (S ) and S coincide (i. what happens when we take the composition of two reﬂections along parallel lines. Every plane isometry is a composition of reﬂections. are all the plane isometries. At this point. and cannot contain translations or glide reﬂections. Every composition of reﬂections in the plane can be performed with at most three reﬂections.e. Points Translations and glide reﬂections have no ﬁxed points. along with the identity. We ﬁnish this section with a brief discussion of the points and shapes left invariant by the various isometries.Shape and Form 183 points to the points O.4(b)). namely: 1. A rotation ﬁxes its centre. if the transformation preserves S as a whole.
1 Twodimensional rotation Rotation in a plane is the most frequently arising case in chaotic modelling. Finally. 9.7) 2 Orbits of the above map stay in a circle of radius r = x0 + y2 . not directly containing the rotation angle. Also. as is easily veriﬁed by the relation: 2 2 rn+1 = rn + a2 + 2a(xn cos θn − yn sin θn ) It is clear that rn+1 is also a function of the rotation angle θn . as well as any line parallel to it. after an appropriate change of coordinates. A glide reﬂection leaves its line of reﬂection invariant. invariant.2. another relation. Without loss of generality. we will consider the translation as a single parameter a added in the x direction.8) By the addition of a translation parameter. in which case it leaves any line passing through the centre of rotation invariant. A map arising from rotation in the plane is expressed by: x cos θn − yn sin θn x xn+1 = rot θn n = n xn sin θn + yn cos θn yn yn+1 (9. This can be easily veriﬁed from the relation rn+1 = rn . and is independent of the form of the rotation angle θn . holds: 2 2 rn+1 − 2axn+1 + a2 = rn . unless it is a rotation by 180◦ . things change radically. y0 ) is 0 the set of initial values. in many cases. y) plane around an axis of rotation perpendicular to this plane. a reﬂection leaves invariant the line of reﬂection.2 Isometries in Modelling We now focus our attention on maps that make heavy use of isometries in their construction. where (x0 .184 Chaotic Modelling and Simulation Lines A translation along a line ℓ leaves that line. When a translation is added to rotation. the radius rn+1 is not generally equal to its previous value rn . is reduced to rotation in the (x. 9. The resulting map is given by: a x a + x cos θn − y sin θn xn+1 = = rot θn n + 0 yn x sin θn + y cos θn yn+1 (9. Even the three dimensional case of rotation. as well as any line perpendicular to it. Every system of this type will continue with circular movements without any change in the circular path. A rotation has no ﬁxed lines.
Shape and Form This relation. Selecting nonlinear functions for the rotation angle is essential for producing forms of high complexity. The ﬁgures in the subsequent sections are chaotic and nonchaotic cases of rotationtranslation iterative schemes. after summation. . as well as chaotic forms.5). we 2ax = 2 2 (rn − r0 ) + a2 n In the limit (n → ∞). the following simple relation arises: a 2 x= This is suggesting the possibility of symmetry around this axis in a rotationtranslation iterative procedure (see for example Figure 9. . 2. . yields n 1 2 (ri+1 − ri2 ) + na2 = 2a n 185 xi 1 where i = 1. One can easily obtain various symmetric forms by choosing a function for the rotation angle. This subject was discussed more extensively in Chapter 8. FIGURE 9.5: A symmetric graph . . Setting x = ¯ obtain: 1 n n 1 xi . assuming the radius rn is bounded. n. and after the appropriate cancellations.
due to rotation and other balancing mechanisms.9) provides symmetric iterative forms with a symmetry axis perpendicular to the x axis at a distance equal to a/2 from the origin. but also in weather modelling and wave propagation. Contraction is counterbalanced by the presence of either rotation or translation. the “attractor. The simplest model uses a parameter b by which the distance equations are multiplied.11) yt+1 = yt 1− v2 = byt c2 . Space expansion is present for example in supernova explosions. However. translation and reﬂection. In the case of contraction.3 Reﬂection and Translation It is easily veriﬁed that the reﬂectiontranslation map (9.186 Chaotic Modelling and Simulation 9. to a movement inside an attracting space.10) has a Jacobian determinant of J = b2 . Movement that follows reﬂection with translation is modelled as in the previous rotationtranslation case.10).9) = = ref θn n + 0 yn x sin 2θn − y cos 2θn yn+1 9.3. reﬂection and translation are based on a stable distance and space metric. in every iteration we would have: rn+1 = brn In the case of rotationtranslation. the iterative spiraling sink towards a point of attraction is replaced. Many chaotic objects are formed by applying space contraction rules along with rotation. space contraction or expansion could be a case of speciﬁc interest in relativistic problems in cosmology and elsewhere. On the other hand.” It is interesting that the relativistic space contraction may lead to the same iterative scheme as (9. the iterative formula takes the form x a xn+1 a + bxn cos θn − byn sin θn = b rot θn n + = 0 yn yn+1 bxn sin θn + byn cos θn (9. that is v2 xt+1 = xt 1 − 2 = bxt c (9. The map that expresses translationreﬂection has the general form: a x a + x cos 2θn + y sin 2θn xn+1 (9. This is easily demonstrated by observing the relativistic space equations.1 Space contraction The iterative procedures presented above using rotation.10) The system (9. Some illustrative examples follow. At a ﬁrst glance. several other very interesting cases appear. which means a space contraction or expansion proportional to b2 . space contraction is more interesting from the point of view of stability and form generation. In other words. Space contraction or expansion appears in vortex formation in chaotic advection or in chemical reactions.
as demonstrated in Chapter 12. that relativistic space contraction can be observed for relatively high speeds. However.4 Application in the Ikeda Attractor A very interesting approach regarding rotation and reﬂection along with translation and space contraction appears in the ﬁgures below. the system rotates with an angle θ ≈ c. b = 0.12) where c and d are rotation parameters. FIGURE 9.6: The Ikeda attractor . 1980). The iterative scheme for rotation is: xn+1 x a = b rot θn n + yn+1 yn 0 Ikeda proposed the following function for the rotation angle: θn = c − d 2 1 + xn + y2 n (9. close to the origin. whereas. 9.83.4 and d = 6 (Figure 9. the rotation angle is θ ≈ c − d. c = 0.Shape and Form where b= 1− v2 c2 187 It is evident in the previous formulae. The socalled Ikeda model and the famous associated attractor express the rotationtranslationspace contraction example (Ikeda.. 1979. Ikeda et al. The classical Ikeda attractor is obtained when a = 1.6). At large distances. changes can appear in chaotic images even in lower speeds. The rotation angle θ is a function of the square of the distance from the origin.
and d. Two important questions are: a) Are there parameter values that generate chaotic attractors in both cases? b) If there are. J = −b2 . then what are the interrelations between rotation and reﬂection attractors? . The parameters for this simulation were a = 2. The reﬂection attractor.5 Chaotic Attractors and RotationReﬂection Both rotation and reﬂection generate chaotic attractors for appropriate values of the parameters a. FIGURE 9. b = 0.2. presented in Figure 9.188 Chaotic Modelling and Simulation The reﬂectiontranslation map with a similar function to (9.5 and d = 2. c = 1.7. has two separate images. c.7: The reﬂection Ikeda attractor 9.15. The ﬁgure on the right looks like a mirror image of the Ikeda attractor.12) for the reﬂection angle is expressed by the iterative scheme: a x xn+1 = b ref θn n + yn 0 yn+1 where the reﬂection angle is given by: 2θn = c + d 2 1 + xn + y 2 n The Jacobian determinant in this case is negative. b.9.
8. E(α. θ). b = 0. G(−α. −θ). just as in the preceding section.9 illustrates the ﬁrst two graphs to the left of Figure 9. that is .Shape and Form 189 FIGURE 9. F(−α.8. FIGURE 9. θ). C(α.9: Comparing rotationreﬂection case Another reﬂectiontranslation chaotic attractor is formed by selecting a simple quadratic function for the reﬂection parameter θ. Figure 9. n In reﬂection. we see four combinations for rotation. B(−α. θ) and H(−α. The parameters were a = 1.8: Comparing rotationreﬂection cases In Figure 9. A(α. −θ). but with a rotationreﬂection angle given by the more simple formula θn = c 2 xn + y2 . the parameter 2θn was set by the same formula. θ). θ).68 and c = 10/3. −θ) and D(−α. compared to the four cases for reﬂection. A rotationreﬂection iterative scheme was used.
11.190 Chaotic Modelling and Simulation 2θn = c 2 xn + y2 . n The parameters for the simulation in Figure 9.68 and c = 10/3.10 were a = 1. b = 0. FIGURE 9.11: Two other reﬂection attractors from a quadratic rotation angle . Both ﬁgures are obtained for a = 1 and b = 0. FIGURE 9.10: A reﬂection chaotic attractor resulting from a quadratic rotation angle The same reﬂection model provides two other attractors. whereas the other parameter is c = 1 for the image on the left and c = 2 for the image on the right. as illustrated in Figure 9.68.
6. At time t = 1.64.13) d + y2 n The Jacobian determinant in this case is J = b2 = 0. However. the space contraction parameter is b = 0. the particles form an ellipsoid form.6 9. 0). The translation parameter is a = 6. The particles are attracted into the two chaotic attractors (Figure 9. whereas at time t = 2.12: Space contraction .12(a). The outer circle represents the original particles at time t = 0. consider the following rotationtranslation scheme: xn+1 = a + b(xn cos θn − yn sin θn ) yn+1 = b(xn sin θn + yn cos θn ) where the rotation angle is θn = c + 2 xn (9. (a) The space contraction eﬀect (b) Space contraction and chaotic images FIGURE 9.8 < 1). The time development of the process is illustrated in Figure 9.1 Experimenting with Rotation and Reﬂection The eﬀect of space contraction on rotationtranslation In order to explore the eﬀect of space contraction due to the introduction of a parameter b (b = 0.8 and the rotation parameters are c = a/2 and d = a. the outer part of the ﬁrst chaotic attractor appears.12(b)) inside the circle. According to this computer experiment the approach to the chaotic state is quite rapid. y) = (a/2.Shape and Form 191 9. The original particles are located in a circle of radius 6 centred at (x. the parameters of the model used are high and the model expresses a strong chaotic process.
13. 16.13: The eﬀect of the reﬂection parameter on chaotic images The inﬂuence that a change in the space parameter b has on a chaotic system is examined in the following reﬂectiontranslation scheme.94.14. the original object is divided into two distinct objects. 0. 8. 0. Once again. as the parameter d decreases. c = 3. and the other images follow from left to right in the top row and from left to right in the second row. The original form appears on the top left of Figure 9. 6. Six chaotic forms of the reﬂectiontranslation scheme are presented here.93 and 0. The parameter d varies. The original form appears on the top left of Figure 9.6. We see that. taking the values 0.975. The set of equations expressing this model are xn+1 = a + b(xn cos θn + yn sin θn ) yn+1 = b(xn sin θn − yn cos θn ) where the rotation angle is: θn = c + d 2 1 + xn + y2 n (9.2. .192 Chaotic Modelling and Simulation 9.3 and b = 1.2 The eﬀect of space contraction and change of reﬂection angle on translationreﬂection A translationreﬂection model is introduced and applied in the following computer simulation. 0. and the other images follow from left to right in the top row and from left to right in the second row.2.78 and 4. The original object results when a = 2. FIGURE 9. but in this case following a process diﬀerent from that illustrated in Figure 9. whereas for the other ﬁve objects the spacecontracting parameter decreases.9.14) The original set of parameters is a = 2. as b decreases. the original object is gradually divided into two distinct objects. taking the following values: 32. b = 0. c = 3. 4.99.30 respectively.9 respectively. d = 4.13.
Shape and Form 193 FIGURE 9. c = 0. The function for the rotation angle is θn = − 1 c2 + x2 + y2 − m c2 + x2 + dy2 The parameters for the simulation were set to a = 0.6.6 and m = 3.1. The resulting chaotic attractor is illustrated in Figure 9.74. In this case.14: The eﬀect of space contraction on chaotic images 9.5.15. The area contraction parameter is set to b = 0. the rotation angle is a function of the inverse of the radius r = x2 + y2 plus a distance parameter c and another rotation angle parameter m.93.3 Complicated rotation angle forms More complicated chaotic forms arise by selecting a rotationtranslation scheme with rotation angle that follows a more complicated equation. d = 1. This is a more complicated chaotic attractor that could result in chaotic ﬂows and in galactic formations. FIGURE 9.15: The eﬀect of a complicated rotation angle on chaotic attractors .
also. It also has a slightly diﬀerent form for the rotationtranslation xn+1 = a + b(xn cos θn − yn sin θn ) yn+1 = −a + b(xn sin θn + yn cos θn ) where the rotation angle is θn = c + d x−a 2 b y+a 2 b (9. The parameters in Figure 9. in other rotationtranslation systems. The twopiece chaotic attractor is presented in Figure 9. 9.4 Comparing rotationreﬂection Figure 9.15) + The other parameters were set to a = 9.16: The eﬀect of a complicated rotation angle on chaotic attractors A twopiece chaotic attractor is formed by using a rotationtranslation formula with a special rotation angle.17(b) illustrates a twopiece chaotic attractor. . resulting from a set of equations expressing rotation and translation without space contraction (b = 1).17(a).16 were a = 1.9). that is θn = c + d 2 xn + y2 n . FIGURE 9. The resulting chaotic system provides an attractor expressing ﬂows appearing in the mixing of ﬂuids but. d = 1 and m = 3. The model is a space contracting one (b = 0.6.87.1. The rotation angle is a function of r. c = a and d = 6.194 Chaotic Modelling and Simulation A simpler form is produced by assuming that in the function for the rotation angle the parameter c is 0. b = 0. The rotation angle has a constant term denoted by c and a varying term related to a transformed distance.
18.5. as illustrated in Figure 9. c = 2. The reﬂectiontranslation equations without space contraction (b = 1) are used. the reﬂectiontranslation model provides two objects with mirror symmetry. The reﬂection angle (θre f ) is as in the previous rotationtranslation case but it is related to the rotation angle (θrot ) by the relation θre f = 2θrot . When the parameters take the values a = 80.18: Reﬂection: a twopiece chaotic attractor The images presented in Figure 9. . c = 5 and d = −20.18 are the reﬂection analogue of this system. FIGURE 9. The parameters were set to a = 37.Shape and Form 195 (a) With space contraction (b) Without space contraction FIGURE 9.85 and d = 15.17: Twopiece chaotic attractors The process shows how an original rotating disk of particles can form two distinct symmetric chaotic forms.
yn+1 = yn ). and four symmetric islands (in two pairs) with equilibrium points that follow the rule xn+4 = xn and yn+4 = yn . The simulations arising out of these cases start by forming concentric circles with diﬀerent particle density and can lead to more complicated forms.16) . There is an island with equilibrium points achieved when (xn+1 = xn .19: A simple area preserving rotationtranslation model A simple space preserving (b = 1) rotationtranslation model with a rotation angle. centred on the symmetry axis (x = a/2). The parameters were set to a = 1 and c = 0. also gives an interesting rotationtranslation image.196 Chaotic Modelling and Simulation 9.19.7 Chaotic Circular Forms The study of a rotation process in which the rotating particles move at circular or other paths from the centre following a speciﬁc formula is of particular interest. 9.5 A simple rotationtranslation model FIGURE 9.65. The iterations follow the simple scheme: xn = rn cos θn yn = rn sin θn where θn increases in a constant way: θn+1 = θn + θ (9. illustrated in Figure 9. depending on the selected set of nonlinear functions. 2 varying according to the square of the distance r from the origin (θn = c(xn + y2 ) = n 2 cr ).6.
2 < a < 6. Figure 9. three values for r appear. the bifurcation has already started to divide the outer ring in two rings (Figure 9.20(a).21(c)). A small change in b leads to the onset of chaos.4.88 · · ·. r) bifurcation diagram for this model is shown in Figure 9. The bifurcation process leads to chaos when higher values for b are selected. is partly explained by using Figure 9.4 and the parameter a varies in 0. r) bifurcation diagram of the model: rn+1 = b − a 4 rn The parameter b is set to 2. the particles are distributed in a particular chaotic way.21(f) presents the bifurcation diagram (b. while the parameter a varies in 1.21(e).2.2 < a < 60. By giving to the chaotic parameter b the value 1. The model gives two values for r in the region of 0 < b < 0. At 0. consider the iterative formula: rn+1 = b − a rn  The (a. As a simple example.9.21(a).21(d). Figure 9.17) Chaotic behaviour appears when the parameter b varies. The parameter b is set to 2. The radius r follows an inverse square law: rn+1 = 1 (rn − b)2 (9.195.4.17).21(b).20(c) illustrates the (a.Shape and Form 197 Depending on the iterative scheme for the radius rn . There are cyclical regions with higher or lower densities. There is only one parameter b = 1. At the end of this region. then four values. For b = 1.204. The parameter b varies between 0 < b < 2.629 · · ·. r) for the model (9. leading to a chaotic region. making up a new chaotic region. while the parameter a varies in 0. r) of the model expressed by the formula: a rn+1 = b − 2 rn The parameter b is set to 2. and so on until a totally chaotic region at 1. fourth and higher order bifurcations appear. interesting patterns arise. or with varying rotation angle related to the distance r = x2 + y2 . Figure 9.629 · · ·. as b increases. third. there appear the distinct circles. Second. along with chaotic behaviour. In Figure 9. . a second bifurcation starts.20(b) illustrates the bifurcation diagram (a. where OC = r.255. The development of the theory of rotation with varying rotation distance from the origin. A very interesting rotation scheme is presented in Figure 9.2 < a < 60. as presented in Figure 9.
198 Chaotic Modelling and Simulation (a) rn+1 = b − a rn  (b) rn+1 = b − a 2 rn (c) rn+1 = b − a 4 rn FIGURE 9.20: Bifurcation diagrams .
17) FIGURE 9.21: Circular chaotic forms .Shape and Form 199 (a) The geometry of circular chaotic forms (b) A circular form (c) The ﬁrst bifurcation stage (b = 1.21 (e) b = 1.204) (d) b = 1.255 (f) Bifurcation diagram for the model (9.
y). yn ) = (x. x= whereas for y y= a 2 a θ cot 2 2 FIGURE 9. when (xn+1 . In other words.8 Further Analysis xn+1 = a + xn cos θn − yn sin θn xn sin θn + yn cos θn Consider a rotationtranslation case expressed by the relations (9. ﬁnally.22: Geometric analogue of rotationtranslation (θ = 2) A geometric analogue of this rotationtranslation case is illustrated in Figure 9. yn+1 ) = (xn . solutions of the form zn+1 = zn = z.e. the rotation angle is θ = 2 and the initial .200 Chaotic Modelling and Simulation 9. This map is symmetric with the line x = a/2 as the axis of symmetry.18) yn+1 = or in the equivalent form zn+1 = a + zn eiθn where zn = xn + iyn . i. Now the following relation holds: z − a = z or (x − a)2 + y2 = x2 + y2 and.22 where the translation parameter is a = 2. This is easily demonstrated if we consider the ﬁxed points.
In this case the point (x. y) = (a/2. y) = (0. 0) indicated by two concentric small circles is the equilibrium point.23(b) where the value of the rotation angle is θ = π/2. the Jacobian for the rotationtranslation case is J = 1 when: xn θy = yn θ x This is the areapreserving case resulting when the rotation angle is expressed by any function of the form 2 2 θn = f xn + y2 = f rn n .23: Geometric analogues of rotationtranslation The Jacobian determinant for the rotationtranslation case above is J = 1 + xn where θx = ∂θn ∂x ∂θn θy = ∂y ∂θn ∂θn − yn = 1 + xn θy − yn θ x ∂y ∂x Analogously. y) = (a/2.23(a) illustrates the same case. Figure 9. An equilibrium point located at (x. 0. but now the rotation angle is θ = π.Shape and Form 201 values are (x. a/2) is presented in Figure 9. the Jacobian determinant for the reﬂection case is of the form J = −1 + xθy − yθ x Clearly.2).9. (a) θ = π (b) θ = π/2 FIGURE 9.
y) will give the relation or. yn+1 ) and (xn = a/2. yn ) By applying the rotationtranslation equation above and the last relation yields: xn+1 + xn = a Clearly. the last equation provides the values of θ for which zn+1 = zn . yn+2 ) = (xn . when xn = a/2 then from the last relation xn+1 = a/2.8. n Another issue is to explore relations regarding the argument zn+2 = zn or: (xn+2 . The Jacobian determinant of the space contraction rotationtranslation case is: J = b2 (1 + xn θy − yn θ x ) The same limit argument (xn+1 . yn ) imply that yn+1 = yn . .1 The space contraction rotationtranslation case zn+1 = a + bzn eiθn This case is expressed by the wellknown function where 0 < b < 1 is the space contraction parameter.0 1−b2 . yn ) = (x. y) = a . after some algebra: (x − a)2 + y2 = b2 (x2 + y2 ) a 1 − b2 2 x− + y2 = ab 1 − b2 ab 1−b2 2 This is the equation of a circle with radius R = The resulting relations for x and y are: x= centred at (x.202 Chaotic Modelling and Simulation 2 or by functions of the radius rn = xn + y2 . Both points coincide as the relations (xn+1 = a/2. yn+1 ) = (xn . The more general case zn+m = zn yields a relation of the form: a xn+m−1 + xn+m−2 + · · · + xn = m 2 9. The equation for cos θ is: a(1 − b cos θ) 1 + b2 − 2b cos θ ab sin θ y= 1 + b2 − 2b cos θ cos θ = a2 1 1 + b2 − 2 2b r When the angle θ is a function of r.
Then use the fact that the distances between T A. B. 6. 8. Show that plane isometries preserve collinearity: Let A. In the system from the previous question. while retaining the translation parameter a in the ﬁrst equation. B. Repeat the same analysis as in the two previous questions. but now using the relations rn+1 = b − a 2 rn 203 . Change the system (9. while retaining the translation parameter a in the ﬁrst equation. What circles are left invariant by each of the diﬀerent isometries? 4. 5. Find the axis of symmetry for this new system. 7. use relation (9. TC must be the same as those between A. T B. 2. You will likely have to use several values for the parameters. and in fact appear in the same order (possibly reversed).12) to ﬁnd a chaotic attractor of the Ikeda type. C respectively to show that T A.8) by adding a translation parameter a1 to the second equation.6).8) by adding a translation parameter −a to the second equation. Draw chaotic simulations for the circular system described by: rn+1 = 1 (rn − b)2 10. TC must also be collinear.Questions and Exercises 1. 3. Find the axis of symmetry for this new system. Repeat the previous question. but now using reﬂection equations. C be three distinct collinear points. Find the ﬁxed points of the relation: rn+1 = 1 (rn − b)2 What is the smallest value of b for which there are three ﬁxed points? How does the system behave when there is only one ﬁxed point? 9. T B. Verify the properties of rotations and reﬂections given in (9. Change the system (9.
. and explore if there is a relation between a and b related to the chaotic behaviour of the system. Try several values for the parameter a.204 and Chaotic Modelling and Simulation rn+1 = b − a 4 rn respectively.
mostly directed towards the solution of a NavierStokes boundary value problem. 1821 and George Stokes. Geometrically the problem is that of rotation with contraction by a parameter b < 1. 205 . the formulation of a theory that partially explains the vortex problem and gives results that coincide with the real life situations has only been achieved in recent years. in order to study chaotic advection problems. The rotationtranslation model (9. the formulation and analysis of vortex ﬂow problems by means of diﬀerence equations can be very useful for several cases if a systematic study is applied. vortex rings. along with a space contraction parameter. 1 The ﬁrst three sections of this chapter have previously appeared in Skiadas (2007).1 10.1 Central sink Consider a circular bath with a sink in the centre at (x. ﬁrst edited in 1879 (Lamb.Chapter 10 Chaotic Advection 10. Few interesting cases are based on diﬀerence equations analogues. A coloured ﬂuid is injected in the periphery of the bath. y) = (0. The water inside the bath is rotating counterclockwise. thanks to progress in computer simulations. the formulation and solution of the problem followed the diﬀerential equations approach. In most cases. in order to deﬁne and analyse speciﬁc vortex ﬂow cases. The question we propose to address is to determine the shape of the ﬂuid ﬁlaments if the sink is open. 0). 1845).1. vortex pairs and vortex systems can be found in the classical book by Horace Lamb. 1879). especially the introduction of the NavierStokes equations (Claude Navier. However. However. The introduction of terms like chaotic advection and the blinking vortex system appeared only in last decades. The vortex ﬂow case and the related forms including vortexlines and ﬁlaments.1 The Sink Problem Questions related to chaotic advection go back to the nineteenth century and the development of hydrodynamics. In this chapter we follow the diﬀerence equations methodology by introducing rotationtranslation diﬀerence equations and a nonlinear rotating angle. The interconnections between the diﬀerence and the diﬀerential equations case is also studied in speciﬁc cases.10) is applied.
2.2) xt2 + y2 = brt t The rotation angle is assumed to follow a function of the form φt = c + d .206 Chaotic Modelling and Simulation with the translation parameter a being zero. c = 0 and d = 0.85. . Figure 10. rt2 (10. When b < 1.1(a). . 1. y0 ) = (1. The equations of ﬂow would then be: xt+1 = b(xt cos φt − yt sin φt ) yt+1 = b(xt sin φt + yt cos φt ) The contraction in the radial direction (r = rt+1 = b (10.3) The space contraction is given by estimating the Jacobian of the ﬂow J = b2 . (a) Spiral particle paths (b) Spiral forms directed to the sink (c) Spiral formation toward a central sink FIGURE 10.1(c) illustrates an advection case when b = 0.1: Spiral forms . The spiral forms start from the periphery and are directed towards the central sink. These circles have a smaller diameter. or even disappear completely. The parameters are b = 0. c = 0. d = 0. ..1) x2 + y2 ) is found from system (10.1(b) results.4.1): (10. a particle is moving from the periphery of the bath to the sink in the centre of coordinates following spirals. The parameter b also inﬂuences the spiral. 0). It is also interesting that full concentric circles appear while the spiraling ﬂow continues. When the same case is simulated for particles entering from the periphery of the rotating system at time t = 0.01. Figure 10. as illustrated in Figure 10. when the rotation parameter d gets smaller.95. and the initial point is (x0 .
as the equation for the rotation angle has been given earlier. The equations of ﬂow are: xt+1 = b (xt − a) cos φt − yt sin φt (10. 0). The ﬁxed points for this map are given by the equation t x2 + y2 = b2 (x − a)2 + y2 or.2). When the movement covers a full circle.1) and (10.4) yt+1 = b (xt − a) sin φt + yt cos φt d rt2 The rotation angle is assumed to follow an equation of the form φt = c + where rt = (xt − a)2 + y2 .1. The sink is located at (x.2 The contraction process From the rotationcontraction equations (10. The paths form spirals towards the centre.Chaotic Advection 207 10. a circular bath with a noncentral sink is examined. it follows that the radial contraction is: ∆rt = rt+1 − rt = brt − rt = −(1 − b)rt The diﬀerential equation for the contraction process is found by observing that: rt+1 − rt dr ∆r ≈ = = −(1 − b)r dt ∆t (t + 1) − t The resulting diﬀerential equation expressing the radial speed r = −(1 − b)r ˙ is solved to give r = r0 e−(1−b)t where r0 is the initial radius. y) = (a. The movement is then totally determined. after transformation: x+ ab2 1 − b2 2 + y2 = ab 1 − b2 2 . the new radius will be: r = r0 e−(1−b) φ 2π 10.2 NonCentral Sink In the following case.
2(a).15. The attractor is also larger than in the previous case. The coloured particles are introduced in a square region (0. The parameters are a = 0. y) = (1.85. 0).3.208 Chaotic Modelling and Simulation 2 ab ab This is the equation of a circle with radius R = 1−b2 centred at (x. 1984. The coloured ﬂuid starting from the outer periphery of the bath approaches the sink in few time periods. the chaotic attractor appears at the 6th time step of the process. as presented in Figure 10.85. Aref’s system models the outﬂow from a large bath tub with two sinks that are opened in . The ﬂow is not symmetric. The attractor appears even if the coloured particles are introduced into a small region of the bath. As the vortex parameter d is higher than in the previous case. (a) (b) FIGURE 10. the circular form of the original coloured line is gradually transformed into a chaotic attractor located at the sink’s centre (x.1) at the right end of the bath at (x. 0).2: Chaotic attractors in a noncentral sink 10. y) = 1−b2 . 1986). invented by Aref (see Aref. b = 0. Aref and Balachandar.1 × 0. The attractor is quite stable in terms of form and location. y) = (a.8.1. 0 .1 Two Symmetric Sinks Aref ’s blinking vortex system Chaotic mixing in open ﬂows is usually modeled by using the blinking vortexsink system.2(b). it is assumed that the coloured ﬂuid is introduced simultaneously in the periphery of the bath. as illustrated in Figure 10.3 10. Then. 1983.15. c = 0 and d = 0. The parameters are a = 0. b = 0. c = 0 and d = 0. To simplify the process.
However. The theoretical studies also include simulations using large grids (1000 × 1000). If z = x + iy is the complex coordinate in the plane of ﬂow. when particles are injected into the ﬂow in few time periods. they are attracted to a speciﬁc region (the attractor) of the ﬂow system.5) φ = φ0 − K rt ln Q r0 (10. a2 ξ= K Q (10. These studies suggest that the attractors are time periodic according to the time periodicity of the ﬂow. as the simulations presented above show. 2πQ is the sink strength and 2πK the vortex strength. given by: η= QT . 2002). As indicated in the literature (K´ rolyi and T´ l.Chaotic Advection 209 an alternating manner. The symmetric sinks are located at (x. The imaginary part of w(z) is the stream function: Ψ = −K ln r − Qφ The streamlines are logarithmic spirals deﬁned by the function: φ = −(K/Q) ln r + const The diﬀerential equations of motion in polar coordinates are r=− ˙ ˙ rφ = and their solutions are: r= 2 r0 − 2Qt Q r K r (10..6) The ﬂow is fully characterised by the adimensional sink strength η and the ratio of vortex to sink strength ξ. We select a counterclockwise rotation. a stable attractor could be present. y) = (−a. if only one sink is used. as well as numerical solutions of the general equations of ﬂow. usually set to 1. and a is the distance of each sink from the centre of coordinates. In recent years.5 and ξ = 10. we assume the superposition of the potential ﬂows of a point sink and a point vortex. 0) and . chaotic ﬂow appears for parameter values η ≥ 0. We apply Aref’s blinking vortex system ﬁrst of all to a rotating ﬂuid.7) The parameter T is the ﬂow period. in order for chaotic mixing to take place in the course of the process. both theoretically and experimentally. a e K´ rolyi et al. the complex potential for a sinking vortex point is w(z) = −(Q + iK) ln z − z s  where z s = (±a. To model the velocity ﬁeld caused by a sink. 1997. several studies have appeared investigating this phenomenon both theoretically and experimentally. a More precisely. 0).
10) If a particle is located at (x0 . then after time t = 1 it will be located at (x1 . In accordance with the theory just discussed. y0 ) = (−a. 0) appears at (a. by using the theory of diﬀerence equations and discrete systems. We will use the form: d ηξ (10. 0) and the time period is T = 1.3: The two symmetric sinks model (x. We next need to specify the form of the angle φt .210 Chaotic Modelling and Simulation FIGURE 10. the ﬂow is not stationary and there are jumps in the velocity ﬁeld at each half period T/2.8) The system of iterative diﬀerence relations for x and y is obtained by equating the real and imaginary parts of (10.9) (10. like Aref’s system.8) can be written as xt+1 + iyt+1 = a + b (xt + a) + iyt (cos φt + i sin φt ) (10. According to this system. In other words. 0) in the next time period. We propose to analyse a discrete time system. a particle located at (−a. 0). 0). y1 ) = (a.9): xt+1 = a + b(xt + a) cos φt − byt sin φt yt+1 = b(xt + a) sin φt + byt cos φt (10. a rotationtranslation model of this type is expressed by the diﬀerence equation: zt+1 = a + b(zt − z s )eiφt Equation (10.3. It is more straightforward to consider the geometry of this system.11) φt = φ0 + 2 = c + 2 r r . y) = (a. as illustrated in Figure 10. The model we search must be a rotationtranslation one with a contraction parameter b < 1 expressing the gradual shortening of the radius r that causes the particles to follow logarithmic spiral trajectories around the sinking vortices.
The ﬁrst of these points is the centre of the righthandside sink. .7(a) illustrates this case. the chaotic region is wider. d = π and a = 5. Another possibility is to give high values to the parameter d expressing the vortex strength. 2. The two main vortex forms can be separated when the parameter d expressing the vortex strength is relatively small. y) = (a. for the parameter c. The value d = 2π leads to a more complicated vortex form. FIGURE 10. The positions of the four sinks are located on the corners of a square. There are two main vortex forms counterbalancing each other. One can also ﬁnd the form of the vortices in the case of three sinks located in an equilateral triangle. c = π and d = 3. Such a case is presented in Figure 10. The simulation of this situation is achieved by selecting a value c = 2π for the sink strength. for times t = 1. The ﬁrst form is located at the righthandside sink at (x. The attractor is now completely separated into two chaotic vortex forms (attractors). 2ab sin φ).6. where φ = d/(4a2 ). b = 0. 0). This is achieved by assuming a special value. The other parameters are b = 0. 3 Figure 10. There are three equilibrium points.95. 0). while the other parameters remain the same as in the previous example.5. However.4: Chaotic attractor in the twosink problem The chaotic attractor in Figure 10. ξ = 10 and thus d = 5. thus dividing the total circle in 4 four sectors. as illustrated in the following ﬁgures. namely: r= (x + a)2 + y2 For the experiments presented in the literature.Chaotic Advection 211 where c = φ0 = π is chosen to account for the halfcycle rotation.5.7(b). The second vortex form is centred at (x. as presented in Figure 10. The parameter d here is set to 1.8. A form with four vortices is presented in Figure 10. y) = (a+2ab cos φ. η = 0. d = ηξ is the vortex strength. c = 2π .4 illustrates the twosink case for parameter values a = 1. 3. and r = rt is the distance from the vortex (−a.
5: Two distinct vortex forms (d = 1) FIGURE 10.212 Chaotic Modelling and Simulation FIGURE 10.6: The chaotic attractor with strong vortex strength parameter d = 2π .
As there is no radial or space contraction.3 for Figure 10.98. Similarly. i. we use the classical rotationtranslation model with a rotation angle expressed by an inverse function of the square of the radius r: θn = c + d 2 xn + y2 n . 10.e.4 Chaotic Forms without Space Contraction Chaotic forms similar to those produced earlier in this chapter.8(c)) two distinct symmetric chaotic forms arise.5 Other Chaotic Forms In the following simulation. Figure 10. b = 1.7(c)). The axis y = 0 is the symmetry axis.3. when b = 1.8(a) illustrates the case where a = 1.7: Vortex forms 10. As the vortex strength parameter d is decreasing (d = 0. The chaotic form is symmetric. but without any space contraction. Figure 10. but the resulting chaotic attractors are not classiﬁed as main vortex forms.4 for Figure 10. the ﬂow could be chaotic.Chaotic Advection 213 The parameters selected for the simulation are b = 0. Interesting chaotic details are present in this attractor. a ﬁfth order vortex form results when c = 2π and the other parameters remain 5 unchanged (Figure 10.8(d) gives an enlargement of the righthandside chaotic attractor produced when d = 0. c = π and d = 1. resulting in a Jacobian of J = 1. (a) Third order (b) Fourth order (c) Fifth order FIGURE 10. d = π and a = 5. are of particular interest. A division of the main chaotic form is starting.8(b) and d = 0.
y) = (a. 0). with the other parameters unchanged.99). d = 10 and c = 1. 0) and (x. y) = a + ba cos (θa ) . as illustrated in Figure 10. which could express low speeds directed towards the sink.9(a)). (x. When a = 10. six rotating vortexlike spiral objects appear (Figure 10.9(b). ba sin (θa ) where: θa = c + d a2 . The space contracting parameter b takes a value close to 1 (b = 0. These chaotic forms illustrate twoarmed spiral galaxies located at (x.3) (d) Enlargement of the previous righthandside attractor FIGURE 10.4) (c) Division into two images (d = 0. (a) One chaotic image (d = 1) (b) Starting division into two images (d = 0. three objects appear.8: Chaotic forms without space contraction When c = 2. y) = (0.214 Chaotic Modelling and Simulation We examine the eﬀect of varying the rotation parameter c while keeping the other parameters ﬁxed.
Chaotic Advection 215 (a) Sixth order (b) Third order (c) Second order (d) Fifth order FIGURE 10.9: Higher order vortices .
4 and d = 6. the following function for the rotation angle is considered: θn = 1 sin c − d 2 1+xn +y2 n The parameters are set to a = 1. The chaotic attractors are located at (x.10(b) appears.9(d)).11 illustrates the resulting chaotic attractor. Five chaotic attractors appear when c = 5 (Figure 10. for instance c = 0.10(a).10: Spiralling towards the sink 10. 0).5. y) = (a.6 Complex Sinusoidal Rotation Angle Very interesting chaotic attractors arise by assuming that the rotation angle of the rotationtranslation model is a complex sinusoidal function.10(c). the chaotic form of the attractor is two onearmed spirals. The original rotating disk has radius R = 3. b = 0. Figure 10. y) = (0. The original rotating disk has radius R = 1. When c takes smaller values. but more details are present in Figure 10. The number of spirals remains the same. Figure 10. as seen in Figure 10. The parameters remain the same as in . multiarmed spirals are present.216 Chaotic Modelling and Simulation For c = 3. The image shows a circular ring along with an internal vortex form and an external part consisting of three main vortex forms and a smaller fourth one. A change in the function for the rotation angle results in the interesting chaotic rotation image presented in Figure 10. (a) R = 1 (b) R = 2 (c) R = 3 FIGURE 10.12(a). illustrated in Figure 10.83. 0) and (x.9(c). When R = 2. c = 0. For the ﬁrst application.
12: Sinusoidal rotation angle chaotic attractors .Chaotic Advection 217 FIGURE 10. and a smaller one in the outside region. while the rotation angle function now is: −1 = sin c − d 2 + y2 rn xn n d θn = sin c − −1 The resulting image has a circular ring with two main vortex forms.4.11: A sinusoidal rotation angle chaotic attractor (c = 0. (a) θn = 1/ sin (c − d/rn ) (b) d = 16 FIGURE 10. interconnected with a central rotating part. as well as vortex forms inside. d = 6) the preceding example.
b = 0. whereas the inside part is larger. including a central circular bulge and vortex forms rotating clockwise and counterclockwise. b = 0. The standard rotationtranslation model is also applied with parameters a = 1. A large number of concentric rings appear.218 Chaotic Modelling and Simulation A relatively simpler function for the rotation angle.13(b)). . while several vortex forms rotate clockwise and counterclockwise inside the ring. (a) θ = 1/ sin dr2 . The parameters are a = 1. is: d 2 1 + xn + y2 n −1 θn = sin However.13: Sinusoidal rotation angle chaotic attractors A completely diﬀerent image results by introducing another function for the rotation angle: θn = 1 2 sin d xn + y2 n = 1 sin dr2 The same rotationtranslation case with parameters a = 1. b = 0. b = 0.8 FIGURE 10.13(a)) with surrounding vortex forms.85 and d = 16 provides also a circular ring image (Figure 10. with only one parameter.85 and d = 16. the resulting chaotic image in Figure 10. a = 0. b = 0.85.2.12(b) is also quite complicated.8 and d = 16 (Figure 10. As before.2. but in this case the outside ring vortex forms are small.7 (b) θ = 1/ sin dr2 . there is a circular ring and three connected vortex forms outside of the ring.1.
To see this. For r > 1/(2b) area expansion takes place. b = 0.7 A Special RotationTranslation Model 2 xn+1 a b xn − y2 n = rot θn + yn+1 0 2bxn yn 2 a + b xn − y2 cos θn − 2bxn yn sin θn n = 2 2 b xn − yn sin θn + 2bxn yn cos θn A special rotationtranslation model is given by the following set of relations where the rotation angle is given by the relation θn = d rn This map deserves special attention. where the the rotation takes place on ellipses where the ratio of the one side to the other is h. ¯ ¯ ¯ and yn = yn . They arise from the rotationtranslation equations. xn+1 = a + b yn+1 = b xn cos θn − hyn sin θn (10. the map is area preserving.8 for the translation parameter.14(a) illustrates the simulation results. hence we have: ¯ xn+1 = hxn+1 ¯ = hxn cos(θ) − hyn sin(θ) = xn cos(θ) − h¯ n sin(θ) ¯ y .8.12) 1 xn sin θn + yn cos θn h This system results is an analogue of the classical rotationtranslation system. Figure 10. yn ) the new coordinates. Then xn = hxn . When r = 1/(2b). There is a rotation centre and vortexlike forms. after a rescaling on the x direction by h.8 10.6 for the area contraction parameter and d = 5 for the rotation parameter. 10.1 Other RotationTranslation Models Elliptic rotationtranslation The following system of two diﬀerence equations provides elliptic forms when the ellipticity parameter h is not 1. denote by ( xn . The parameter values selected for the application are a = 0.Chaotic Advection 219 10. whereas when r < 1/(2b) a space contraction takes place. as the Jacobian is J = 4b2 r2 .
8. are of particular interest when studying chaotic advection cases. y) = (a.9. but now the rotation angle obeys the function θn = c + 2 rn d −e where. 10.52 and d = 1. A space contraction parameter e is inserted in the function expressing the rotation angle. and the parameters are a = 0. b = 0.2 Rotationtranslation with special rotation angle This case is modelled by using the classical rotationtranslation model with space contraction.5.014. The resulting image is illustrated in Figure 10. and the ones following it. The elliptic rotation equations give a somewhat deformed image. b = 0. The main part of the image is a circular ring centred at (x.15(a).12). This case is illustrated in Figure 10.14: Chaotic images of rotationtranslation maps and yn+1 = yn+1 ¯ = = xn sin(θ) + yn sin(θ) 1 xn cos(θ) − yn sin(θ) ¯ ¯ h Adding the translation and contraction parameters results in (10.5 and the other parameters are a = 3. for our example. where the rotation angle has the form θ = dr2 .8. compared to the previous circular type cases presented in this chapter. Three vortex forms appear outside the circular ring and two smaller vortex forms appear inside the .220 Chaotic Modelling and Simulation (a) A special map (b) An elliptic map FIGURE 10.14(b).9. This image. d = 3 and h = 0. c = 1. 0). e = 4.
but now with c = 17.15(c) has the same parameters as the previous case. As the parameter c changes. There are now two outer vortex forms. However the image is similar to that of Figure 10. but changes appear in the vortex forms. All the vortex forms originate from the periphery of the circular ring. except for c = 8.15: Chaotic images with area contraction rotation angles . In the same ﬁgure. Figure 10. (a) c = 1. The two vortex forms inside the circular ring take up more space. Figure 10.52.15(a).Chaotic Advection 221 ring.5.15(d) is similar. the two axes of coordinates and the part of the circle expressing the space curvature due to the space contraction parameter b and the translation parameter a are illustrated. whereas the other parameters remain unchanged. d = 1.3 and d = 1. the resulting image shows the same circular ring located at the same centre and having similar magnitude. 0 . while the outer vortex forms create a main vortex image with vortex forms inside. d = 1 (c) c = 8.5 (d) c = 17 FIGURE 10. This a ab circle is of radius R = 1−b2 centred at (x. Figure 10.014 (b) c = 4.3. y) = 1−b2 . Smaller vortex forms appear inside every main vortex form.15(b) illustrates the case where c = 4.
Discuss the mixing of ﬂuids when the chaotic region is reached. Use the rotation angle φt = c + d/rt2 .10).1) to their reﬂection analogues. Consider the system (10. What is the disadvantage with the presence of chaos in mixing? How can we avoid chaotic phenomena in ﬂuid mixing? 4. Change the rotation equations (10. 2. . Change the rotation equations (10. and draw chaotic images when c = π/2 and d is varying. Add a translation parameter “k” at the second equation. and ﬁnd a symmetry axis for this new system. and ﬁnd the resulting chaotic and nonchaotic forms for various values of the parameters b. c and d. 3. 5.10). c and d.10) to their reﬂection analogues. Chaotic advection is closely related to chaotic mixing.222 Chaotic Modelling and Simulation Questions and Exercises 1. Find a symmetry axis for the system (10. and ﬁnd the resulting chaotic and nonchaotic forms for various values of the parameters b.
Chapter 11
Chaos in Galaxies and Related Simulations
11.1
Introduction
A large variety of chaotic forms appears in stellar observations. These forms exhibit astonishingly large varieties of shapes. At ﬁrst glance, there is not a simple method to handle and organise the entire material of stellar observations based on a few rules. The related cosmology theories for star and galaxy creation and formation are of considerable interest. Related laws help us describe the evolution of stellar shape over time. Already from the earliest observations, it was evident that chaos was the main characteristic of the shape and form of many stellar formations, including galaxies and chaotic nebulae. A great number of stellar formations develop under high temperatures, the socalled ‘ﬁre’ that the ancient Greek philosopher, Heracleitos of Ephesus (Heraclitus), had emphasised as the basis of genesis. All these “ﬁre” formations are mainly chaotic forms. A fundamental problem in nonlinear dynamics is that of exploring how the properties of orbits change and evolve as one or more parameters of a dynamical system change, in order that the system exhibit chaotic behaviour. Recent advances in nonlinear physics have resulted from such an approach. The transition to chaos and the routes to chaos are extensively explored in the case of onedimensional maps and, mainly, for the logistic map. Of considerable interest however are also the chaotic dynamics of twodimensional maps, such as the catmap and the H´ non map. e Another very important transition, related to physics, is one in which the system is in a chaotic state and, as parameters change, the system transitions from one type of chaos to another. Of particular interest is the case where the chaotic attractors that correspond to a system change as a parameter gradually takes higher or lower values. Many stellar and galactic systems may ﬁt to chaotic attractor phenomena. Chaotic attractors in twodimensional and threedimensional space are quite stable objects and can be useful in exploring and simulating the forms of galaxies, clusters of galaxies, nebulae, black holes and other stellar objects. There are several approaches regarding the handling of chaotic situations occurring under the inﬂuence of a central force as in the case of stellar systems. An approach is to consider a system of masses moving in diﬀerent orbits under gravity. Early studies of the subject have been
223
224
Chaotic Modelling and Simulation
undertaken by Contopoulos and Bozis (1964); LyndenBell (1969); Toomre (1963, 1964); Hunter and Toomre (1969); Contopoulos et al. (1973). A detailed approach of chaotic dynamics in astronomy can be found in Contopoulos (2002). Ostriker and Peebles (1973) explore the stability of ﬂattened galaxies while Raha et al. (1991) study the dynamical instability of bars in disk galaxies. The box and peanut shapes generated by stellar bars are discussed in Combes et al. (1990). Extensive work on how one can separate chaotic from ordered domains is found in the works of Contopoulos and Voglis (Patsis et al., 1997). Objects of very complicated morphology are found and simulated in Xray clusters (Navarro et al., 1995). The discovery of new ultraluminous IRAS galaxies give rise to new studies on how galactic systems interact. A number of works regarding chaos and chaotic attractors and the interaction with chaotic dynamics in astronomy are of interest. Most notable among them are the work of White et al. (1998) on the anomalous transport near threshold of the standard map and the approach by Arrowsmith and Place (1990); Acheson (1997) in their books on dynamical systems and calculus and chaos. H´ non and Heiles (1964) e developed a Hamiltonian system which developed into a prototype for computer experiments of chaotic dynamics in astronomy. Of considerable interest is also the areapreserving quadratic map developed by H´ non (1969). Period doubling bifure cations of this map appear in recent studies (Murakami et al., 2002). It is very complicated to express chaos analytically, namely so that every stage is explained and deﬁned in a deterministic way. However, the forms resulting from billions of chaotic intermediate steps are frequently of considerable ‘attracting’ shape. These attractors are in many cases reproduced in the laboratory by using very simple rules. Following the studies on chaos in the last decades, it is evident very few main characteristics or parameters of the chaotic system play a critical role in the formation of attractors. In stellar systems the main characteristics are: 1) gravity, the attracting force between bodies (masses), and 2) the ﬁrst BigBang. Geometrically, stellar systems move, grow or decline by: a) rotating and b) translating (following linear movements). According to the established theories, expansion of the universe (translation) follows the original BigBang, whereas gravity, the attracting force, causes translation in the opposite direction. Attracting forces in diﬀerent directions cause rotations and translations in stellar formations. In chaotic dynamics, the main eﬀort is usually centred on the formulation of relatively simple models with only a few chaotic parameters. If the main characteristics of the phenomenon are included in the ﬁnal model, the chaotic character of many phenomena in nature could be well expressed. Chaotic formations in galaxies could appear when rotations, reﬂections and translations (axial movements) are present along with the inﬂuence of gravitational forces that play an important role. In this chapter, simple iterative models are presented. They are the simplest possible models, including rotation, reﬂection, translation and the inﬂuence of gravitational forces on the angle of rotation. These models present interesting aspects that are studied analytically and are presented in several graphs following the simulation. Several spiral forms of various formations can be modelled by changing the parameters of the mod
Chaos in Galaxies and Related Simulations
225
els. The study of galactic forms by using diﬀerential equations and Hamiltonians is presented in Chapter 12.
11.2
Chaos in the Solar System
Complicated and even chaotic orbits and paths of the solar system are studied and simulated in related chapters of this book. The bodies in the solar system mainly follow cyclic or elliptic orbits around a centre of mass. In this case, the simulation can follow a purely geometric approach. To simplify the problem, the centre of mass is located in the centre of the large mass. The attracting force is, of course, gravity, and it is presented by an inverse square law for the distance. The nondimensional equations of motion for a body with small mass m rotating around a large body with mass M, which we consider to stay in equilibrium position, are, in Cartesian coordinates: x x=− ¨ 2 + y2 3/2 x (11.1) y y=− ¨ x2 + y2 3/2 The initial conditions are x = 1, y = 0, x = 0 and y = ν. The escape velocity is ˙ ˙ √ ν = 2, and the circular orbit is obtained when ν = 1. The circular orbit is illustrated in Figure 11.1(a). In the same ﬁgure an elliptic orbit is presented, where the initial √ velocity is 1 < ν = 1.2 < 2. A hyperbolic orbit√ also presented. The initial is velocity is higher than the escape limit (ν = 1.45 > 2) and the particle ﬂies oﬀ to inﬁnity. Figure 11.1(b) illustrates a large number of elliptic paths for the same value of the initial velocity as before. The paths remain in the same plane, but deviate from the original position, forming the shape presented in the ﬁgure. The motions of two attracting point masses are relatively simple. The equations of motion for the point mass m1 in Cartesian coordinates are: ¨ x1 = − y1 = − ¨ (x1 − x2 )m2
3/2
(x1 − x2 )2 + (y1 − y2 )2 (y1 − y2 )m2 (x1 − x2 )2 + (y1 − y2 )2
(11.2)
3/2
The case for the point mass m2 is analogous. Without loss of generality, the gravity constant is G = 1 and the total mass of the rotating system is M = m1 + m2 . If we give the two masses initial positions (x1 , y1 ) and (x2 , y2 ), and initial velocities x1 , y1 , x2 and y2 , the masses revolve around each ˙ ˙ ˙ ˙ another while drifting together, as a pair, in space. Their centre of mass drifts at a constant velocity. Relative to an observer moving with the centre of mass, their orbits are ellipses and, in special cases, circles.
226
Chaotic Modelling and Simulation
(a) Circular and elliptic orbits
(b) Elliptic paths
FIGURE 11.1: Elliptic orbits
The motion of two attracting points in space is presented in Figure 11.2(a). The heavy line represents the movement of the centre of mass. The masses revolve around and drift together. The points have masses m1 = 0.4, m2 = 0.6, and initial velocities x1 = −0.15, x2 = 0, y1 = −0.45 and y2 = 0.25. The same motion, viewed relative ˙ ˙ ˙ ˙ to the centre of mass, is illustrated in Figure 11.2(b). Each pointmass follows an elliptic path.
(a) Attracting bodies in space
(b) Relative movement
FIGURE 11.2: Attracting bodies in space and relative movement
The threebody problem reduced in the plane still gives very complicated and chaotic orbits. To simplify the problem, we assume that the third body, a satellite,
Chaos in Galaxies and Related Simulations
227
is of negligible mass related to the other two bodies. The other two, the planets (primaries), travel about the centre of the combined mass in circular orbits in the same plane. The satellite rotates in the same plane. The total mass of the planets is m1 + m2 = 1. The equations of motion for the satellite are x=− ¨ y=− ¨ m1 (x − x1 )
3/2 r1
− −
m1 (y − y1 )
3/2 r1
3/2 r2 m2 (y − y2 ) 3/2 r2
m2 (x − x2 ) (11.3)
where (x, y) is the position of the satellite, r1 , r2 are the distances of the satellite from the two planets m1 , m2 respectively, and (x1 , y1 ) and (x2 , y2 ) are the positions of the planets at the same time t. In the following example, the primaries have masses m1 = 0.9 and m2 = 0.1, and the position and initial velocity of the satellite are (−1.033, 0) and (0, 0.35) respectively. The complicated and chaotic paths that arise are illustrated in the following ﬁgures. In Figure 11.3(a), the movement is viewed from the outside of the system, from “space.” It shows that the satellite can move between the two revolving primaries. Figure 11.3(b) illustrates the orbit of the satellite viewed in a coordinate system that rotates with the revolution of the planets (rotating frame), with the xaxis always passing through the planets, according to the rotation formula: xnew = x cos t + y sin t ynew = −x sin t + y cos t (11.4)
(a) Motion in space
(b) Rotating frame
FIGURE 11.3: Motions in space
By using diﬀerence equations to model solar systems, it is possible to simulate
228
Chaotic Modelling and Simulation
chaotic patterns like, for instance, the rings of Saturn and of other planets. The simplest approach is to use an equation such as the logistic, and to assume that the system rotates following elliptic orbits. Two realisations appear in the next ﬁgures. The parameter of the logistic model is b = 3.6 for Figure 11.4(a) and b = 3.68 for Figure 11.4(b). The rotation formula is given by: xt = 2rt cos t yt = 2rt sin t where rt+1 = brt (1 − rt ). (11.5)
(a) b = 3.6
(b) b = 3.68
FIGURE 11.4: Ring systems
11.3
Galaxy Models and Modelling
Studies on chaotic dynamics in galaxies and in other complicated stellar objects indicate that order and chaos coexist in these structures. It is also quite diﬃcult to simulate complicated stellar objects, especially in the limits where chaos is present. Experience on chaotic simulations during the last decades indicates that the simpler the chaotic model, the more successful it is. Remarkable solutions of the ﬂow equations were provided by Lorenz (1963). He transformed a highdimensional system into a threedimensional one, retaining the chaotic character of the system. R¨ ssler o (1976d) proposed the simplest model of threedimensional ﬂow. Since, even nowadays, the subject of chaos in astronomy has many unexplored aspects, a convenient way of dealing with the problem is to explore the qualitative and quantitative aspects
Chaos in Galaxies and Related Simulations
229
by considering simple chaotic models, in which only the main characteristics of the system under consideration are retained. Such characteristics in galaxies and galactic formations are of course: 1) rotationreﬂection, 2) translation, and 3) area contraction due to gravitation (expressed by a parameter b < 1). Rotation is connected with the theory of a central force by assuming that the rotation angle obeys a function of the distance r from the origin. Translation is assumed to take place when a strong gravitation source is located in a large distance from the origin or, even, when gravity waves are present. To simplify the problem, only the twodimensional approach in the (x, y) plane is presented here. A convenient formulation of the iterative map expressing rotationreﬂection and translation is given by the following diﬀerence equation: fn+1 = a + b fn eiθ where: fn = xn + iyn fn = xn − iyn (for rotation) (for reﬂection)
For rotation, the determinant of this map is det J = b2 = λ1 λ2 . This is an area contracting map if the Jacobian J is less than 1. The eigenvalues of the Jacobian are λ1 and λ2 . For reﬂection the Jacobian gives: det J = −b2 . The translation parameter is a in both cases. In the cases discussed above, rotation and reﬂection are followed by a translation along the xaxis. As discussed earlier, this has the advantage, over a general translation, of reducing the number of parameters by one, without aﬀecting the results in the majority of the cases studied. The angle θ is assumed to be a function of the distance r from the origin. In the cases studied here, r = x2 + y2 , the Euclidean metric. The general forms studied can be classiﬁed as follows: θ = g(rm ), where m = 1, 2, −1, −2, −3/2, are the most convenient values for the exponent. By expanding the function g(rm ) in a Taylor series and retaining the ﬁrst terms to the right the following forms for the angle θ for reﬂection and rotation cases arise: θn = crn
2 θn = crn
θn = c +
d rn d θn = c + 2 rn d θn = c + 2/3 rn
Both rotation and reﬂection generate chaotic attractors for appropriate values of the parameters a, b, c, and d.
located at the circle K.8. 0). An interesting property of the above simulation is that the original single simple rotating form splits into two forms. c and d.5(f).6.5(c) and Figure 11.5(f). .6(a). a very interesting property of the map is the existence of ab a disk F of radius 1−b . Further.5(a)) and taking the values a = 0. a. as in Figure 11. 0). that show the same direction of rotation and are located at places depending on the displacement a.5(b)). the parameters of the model are b = 0. In these ﬁgures. 0) in the (x.5(d). and at d d a + a cos c + 2 . 0) for the attractor on the right side of Figure 11. for rotation: cos θn = 1 a2 1 + b2 − 2 2b rn a 1 − b2 2 + y2 = n a2 b2 (1 − b2 )2 ab a So the ﬁxed points lie on a circle K with radius R = 1−b2 . The second attractor is the one on the right side of the ﬁgure. varies.5(f). the parameter a seems to have very little eﬀect in the development of the phenomenon. while the other attractors continue in a clockwise direction until the ﬁnal point. leads to the formation of two distinct attractors.5(a). centred at ( 1−b2 . located at (a. the ﬁxed points obey the equations xn − and.5(e)). This is illustrated in Figure 11.6(b). produce chaotic forms in which a centre of revolution appears in (a. equal to b2 = 0. For the ﬁrst application we choose the following function for the rotation angle: θn = c + d r2 The simulation results are illustrated in Figures 11. The parameters are: a = c = d = 5. c = 3 and d = 6. More complicated chaotic forms arise for appropriate values of the parameters a. three attractors appear. This is an example where an area contraction takes place. centred at (a. b. When a = c = d = 4 and b = 0.5(a) through 11. a = 4 and a = 5 in the subsequent ﬁgures. starting from very low values. 0) in the (x. In the case of Figure 11. The centres of the two attractors are located at (a.45. 0).8.01 in Figure 11. The ﬁrst attractor. the points of which remain inside of it under the map. In Figure 11.4. one of which is located at (a.230 Chaotic Modelling and Simulation A discussed in Chapter 8. a sin c + 2 a a for the attractor on the left side. when a takes higher values (Figure 11. The other parameter.9. The point masses are distributed around the centre of coordinates and their distribution follows an inverse law. is at the top of the ﬁgure.25 and b = 0. Another example appears in Figure 11. a rotating form appears. This chaotic form is restricted inside the two circles F and K that were discussed earlier. Higher values of a. In the second case (Figure 11. (a = 0.64. but further a quite large rotating form appears which. y) plane. these attractors are ﬁnally separated in two forms. presenting a cluster of chaotic attractors. a = 1. a = 2. y) plane.
4) (e) Two interconnected images (a = 4) (f) Two separate images (a = 5) FIGURE 11.5: Chaotic forms of a simple rotationtranslation model .45) (c) Middle stage (a = 1.6) (d) Second phase (a = 2.01) (b) Starting rotation (a = 0.Chaos in Galaxies and Related Simulations 231 (a) Initial stage (a = 0.
The attracting points are solutions of the equations that satisfy the special conditions set for the general rotation equation above. The point masses are trapped in trajectories and are led to the attracting points as when a black hole is present.3. 11. A twoarmed spiral structure appears.7(d). In these cases there are attracting points and no other attractors.232 Chaotic Modelling and Simulation (a) Three chaotic images (b) Multiple chaotic images FIGURE 11. The rotating points inside the circle F are trapped in the attractor region. c = 3 and d = 6.7(a) to 11. Two of the satellites have their location centre at the a ab circumference of the circle K of radius 1−b2 .7(c) the parameters are a = 4.7(d) the parameters are a = 2.9.7(b) and 11. b = 0. c = 1. or they disappear into the attracting point.3.7(a).6: Chaotic forms of a simple rotationtranslation model Rotating forms can also be formed by using rotation equations with a linear expression for the rotation angle: θn = c + drn The translation parameter is a. 0 . The points that oscillate inside the attracting region escape from this region only if they . The chaotic attractor is located at (a. 0).25. and the question is how these points will be distributed after time t corresponding to n iterations.8(a)) where the rotation angle follows an inverse law regarding the distance r from the origin: d θn = c + rn In the ﬁrst case. In Figure 11. and four satellites are 2 located at (lim(xn ). c = 2. b = 0.3. lim (yn = yn+1 ) . The results are summarised in Figures 11. located at a . and an area contracting parameter b < 1 is essential ab to keep the system into a limit included within a circle F of radius 1−b centred at (a.5 and d = 0.03 and d = 0. In Figures 11. the translation parameter is a = 0. Chaotic attractors and attracting points appear in an example of rotation (Figure 11. 0) and the attracting point is located in the periphery of the circle K. The centre of this circle is at 1−b2 . A number of rotating point masses are equally distributed within the circle F.6 and the other parameters are b = 0.9. lim(yn )).9.
Chaos in Galaxies and Related Simulations 233 (a) Chaotic object with satellites (b) Enlargement of chaotic objects (c) A twoarmed spiral with satellites (d) A twoarmed spiral galactic form FIGURE 11.7: Chaotic forms of a simple rotationtranslation model .
In Figure 11. in Figure 11.9 for space contraction. (a) Attractor and attracting point (b) Two attractors and the accompanying attracting points FIGURE 11. The image is quite complicated. The situation examined was that of two galaxies approaching each other.3. Here. the attracting point is the point where xn+1 = xn and yn+1 = yn . The resulting shape is illustrated in Figure 11.9.0055 The parameters for the system are a = 4 for translation.234 Chaotic Modelling and Simulation reach the periphery of the attractor.0121/3 . as well as places with high and low density. Now two attractors appear. and each attractor is connected to the attracting point with trajectories. . while the other parameters remain unchanged.8: Attractors and attracting points in rotationtranslation model 11. The rotation angle has the form: θ = −d 2r − c 1 r2 − cr + c2 1 ln − tan √ 2 6 3 (r + c) c 3 1 0. where the parameters are a = 5 and c = 2. except for the space contraction parameter. There are several attracting points.8(b). and c = 0. the attracting points are the points where xn+2 = xn and yn+2 = yn . The rotation angle and the other parameters. and then they follow the trajectory that leads to the attracting point. The particles follow characteristic trajectories over time. while. The same behaviour is seen in the next example (Figure 11.0122/3 and b = 0.1 A special rotationtranslation image This example comes from Contopoulos and Bozis (1964). are the same as in Contopoulos and Bozis (1964).8(b)). we apply a rotationtranslation model with space contraction. A number of particles are introduced into a small circle near the centre of coordinates. d = 0.8(a).
The parameters in both cases (Figure 11. Chaotic attractors and other forms that appear in real space (x. where the rotation or reﬂection angle is of the form θn = dr2 . for instance in the following. 2. for the same set of parameters. One would expect that the attractors for the reﬂection case would be a mirror image of the attractors for the rotation case. the particles lead to attracting points where they disappear as if in a blackhole. The chaotic geometric modelling approach presented in this section could be useful in galaxy simulations. b = 0. in the presence of a chaotic attractor. When the space contraction parameter b < 1 three cases appear.10(a) illustrates rotation and Figure 11. the particles follow trajectories leading away from the rotating system if the translation parameter is relatively high. y) simulations indicate that elements of any form can be trapped inside these attractors. and 3. according to the magnitude of the rotation angle θ: 1.4 RotationReﬂection It is interesting to explore the relationships between chaotic attractors in the presence of rotation and reﬂection respectively. a = 2. This can be seen in some particular cases.5 and d = 0.9: A rotation image following a ContopoulosBozis paper (approaching galaxies) 11. the particles will stay or oscillate inside this attractor.10(b) illustrates reﬂection) are. .Chaos in Galaxies and Related Simulations 235 FIGURE 11. The chaotic forms of the attractors verify the assumption that there is a mirrorimage relationship between the two.65.
However. In other words. The usual obstacle in applications is that high relativistic speeds are not possible in real situations. how do the equations of special relativity enter into the rotationtranslation equation system? The introduction of the relativity equations into a rotationtranslation map will change the Jacobian of the system. Some chaotic forms form groups or clusters in real space. i. be estimated. a number of particles follow circular paths. and can thus produce an areacontracting map. The explanatory ability of chaotic models is high and could be useful in many cases.e.10: Comparing chaotic forms: rotationreﬂection Various chaotic geometric forms appear. can. As we derived earlier. computer simulations show that even low relativistic speeds have a considerable in . Numerous chaotic and nonchaotic forms arise.236 Chaotic Modelling and Simulation (a) Rotation image (b) Reﬂection image FIGURE 11.” rather than “elements” in that space. a translation movement takes place. the relativistic speed. in some cases. inﬂuences the system parameters. Some chaotic forms are divided into other similar forms when the parameters of the model take particular values.5 Relativity in RotationTranslation Systems In rotationtranslation systems. and it mainly models “space. Special characteristics of these forms. 11. such as the coordinates of the centre and the geometric interactions of these forms between each other. while. The question we would like to address in this section is how the shape of the attractors changes when the speed of the rotating system relative to another system. This kind of chaotic simulation of galaxy forms has the advantage that it needs limited computing power. at the same time. a twodimensional rotation map is area preserving if the Jacobian determinant of the map is J = 1. These chaotic forms specify places in space that attract masses (particles) from the space around.
A chaotic bulge appears. the relativisticlike coeﬃcient appears as a change in the parameter b. After rearrangement. Figure 11.11: A rotationtranslation chaotic image The rotationtranslation chaotic bulge presented in Figure 11.6) a new form is obtained.11 illustrates the chaotic image from the rotationtranslation map without entering relativistic equations. The equation for the rotation 6 3 angle is: d θn = c + rn 2 where rn = xn + y2 . which now becomes: b∗ = b 1− v2 c2 . FIGURE 11.11 is formed when the space parameter is b = 1. n The relativisticlike approach to the rotationtranslation equations takes into account the length contraction governed by the equations: xrel = x yrel = y 1− 1− v2 c2 v2 c2 where v is the relativistic speed and c is the speed of light. The bulge is symmetric. The translation parameter is π a = π : The other parameters are: c = 12 and d = π . A simple example is presented in the following ﬁgures. Inserting these relativistic forms into the rotationtranslation equations xn+1 = a + b(xn cos 2θn − yn sin 2θn ) yn+1 = b(yn sin 2θn + yn cos 2θn ) (11.Chaos in Galaxies and Related Simulations 237 ﬂuence on the chaotic forms.
1. When v = c. namely values higher than v = 89.12(d). Figure 11.12(a) illustrates the chaotic bulge and the space outside the central chaotic space. The space outside the chaotic bulge is characterised by closed curves and islands.14(b) the relativistic speed is in the upper limit for attractor formation. in which the relativistic speed is v = 2500km/sec. 50000km/sec and b = 0. where the relativistic . We launch particles at x = (−1.13(f)).9860. the inﬂuence on the chaotic bulge and on the space around the bulge is quite evident. when time t is very small. This is mor