Discovering

Dynamical Systems
Through Experiment
and Inquiry
Textbooks in Mathematics
Series editors:
Al Boggess, Kenneth H. Rosen

Advanced Calculus
Theory and Practice, Second Edition
John Srdjan Petrovic

Advanced Problem Solving Using Maple

Applied Mathematics, Operations Research, Business Analytics, and Decision Analysis
William P Fox, William Bauldry

Nonlinear Optimization
Models and Applications
William P. Fox

Linear Algebra
James R. Kirkwood, Bessie H. Kirkwood

Train Your Brain

Challenging Yet Elementary Mathematics
Bogumil Kaminski, Pawel Pralat

Real Analysis
With Proof Strategies
Daniel W. Cunningham

Contemporary Abstract Algebra, Tenth Edition

Joseph A. Gallian

Geometry and Its Applications, Third Edition

Walter J. Meyer

Linear Algebra
What You Need to Know
Hugo J. Woerdman

Introduction to Real Analysis, Third Edition

Manfred Stoll

Discovering Dynamical Systems Through Experiment and Inquiry

Thomas LoFaro and Jeff Ford

For more information about this series, please visit: https://www.routledge.com/

Textbooks-in-Mathematics/book-series/CANDHTEXBOOMTH
 Discovering
Dynamical Systems
Through Experiment
and Inquiry

Thomas LoFaro and Jeff Ford

First edition published 2021
by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-2742

and by CRC Press

2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN

© 2021 Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, LLC

Reasonable efforts have been made to publish reliable data and information, but the author and pub-
lisher 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 stor-
age or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, access www.copyright.
com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA
01923, 978-750-8400. For works that are not available on CCC please contact mpkbookspermis-
[email protected]

Trademark notice: Product or corporate names may be trademarks or registered trademarks and are
used only for identification and explanation without intent to infringe.

ISBN: 978-0-367-90394-7 (hbk)

ISBN: 978-1-003-02413-2 (ebk)

Typeset in Computer Modern font

by KnowledgeWorks Global Ltd
In memory of
J.R. and T.J.
Contents

Preface xi

1 An Introduction to Dynamical Systems 1

1.1 What Is a Dynamical System . . . . . . . . . . . . . . . . . . 1
1.2 Numerical Iteration and Orbits . . . . . . . . . . . . . . . . . 2
1.3 Graphical Iteration . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Modeling Using Discrete Dynamical Systems . . . . . . . . . 8

2 Sequences 11
2.1 Introduction to Sequences . . . . . . . . . . . . . . . . . . . . 11
2.2 Convergence of Sequences . . . . . . . . . . . . . . . . . . . . 12
2.3 The Squeeze Theorem . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Arithmetic Limit Theorems . . . . . . . . . . . . . . . . . . . 18
2.5 Bounded and Unbounded Sequences . . . . . . . . . . . . . . 19
2.6 Subsequences . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.7 Liminfs and Limsups . . . . . . . . . . . . . . . . . . . . . . 24
2.8 Cauchy Sequences . . . . . . . . . . . . . . . . . . . . . . . . 29

3 Fixed Points and Periodic Points 33

3.1 Fixed Points . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.1 Fixed Points of Linear Systems . . . . . . . . . . . . . 34
3.1.2 Attracting Fixed Points of Nonlinear Systems . . . . . 37
3.1.3 Repelling Fixed Points of Nonlinear Systems . . . . . 41
3.1.4 Neutral Fixed Points of Nonlinear Systems . . . . . . 44
3.2 Periodic Points . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.2.1 Stability of Periodic Points . . . . . . . . . . . . . . . 48
3.2.2 New Periodic Orbits from Old . . . . . . . . . . . . . 50

4 Analysis of Fixed Points 53

4.1 Fixed Point Existence Theorems . . . . . . . . . . . . . . . . 53
4.2 The Inverse and Implicit Function Theorems . . . . . . . . . 59
4.2.1 The Inverse Function Theorems . . . . . . . . . . . . . 59
4.2.2 The Implicit Function Theorem . . . . . . . . . . . . . 62
4.3 Hyperbolic Periodic Points . . . . . . . . . . . . . . . . . . . 64

vii
viii Contents

5 Bifurcations 69
5.1 What is a Bifurcation? . . . . . . . . . . . . . . . . . . . . . 69
5.2 Introduction to Bifurcation Diagrams . . . . . . . . . . . . . 71
5.3 The Tangent Bifurcation . . . . . . . . . . . . . . . . . . . . 72
5.4 The Period Doubling Bifurcation . . . . . . . . . . . . . . . . 76

6 Examples of Global Dynamics 81

6.1 Local Dynamics vs. Global Dynamics . . . . . . . . . . . . . 81
6.2 The Logistic Map with a = 4 (Part 1) . . . . . . . . . . . . . 83
6.3 The Doubling Map . . . . . . . . . . . . . . . . . . . . . . . 85
6.3.1 Basic Dynamics of the Doubling Map . . . . . . . . . 86
6.3.2 The Doubling Map in Binary . . . . . . . . . . . . . . 88
6.4 The Logistic Map with a > 4 (Part 1) . . . . . . . . . . . . . 91

7 The Tools of Global Dynamics 93

7.1 How to study Global Dynamics . . . . . . . . . . . . . . . . 93
7.2 The Cantor Set . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.3 The Shift Map (Part 1) . . . . . . . . . . . . . . . . . . . . . 96
7.3.1 The Sequence Space on 2 Symbols . . . . . . . . . . . 96
7.3.2 Dynamics on the Sequence Space on 2 Symbols . . . . 99

8 Examples of Chaos 103

8.1 Introduction: The Definition of Chaos . . . . . . . . . . . . . 103
8.2 The Shift Map (Part 2) . . . . . . . . . . . . . . . . . . . . . 107
8.3 Topological Conjugacy . . . . . . . . . . . . . . . . . . . . . 108
8.4 Return to The Doubling Map . . . . . . . . . . . . . . . . . . 111
8.5 The Logistic Map with a > 4 (Part 2) . . . . . . . . . . . . . 111
8.6 The Logistic Map with a = 4 (Part 2) . . . . . . . . . . . . . 115

9 From Fixed Points to Chaos 117

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
9.2 Computing a Bifurcation Diagram . . . . . . . . . . . . . . . 119
9.3 Period-doubling to Chaos . . . . . . . . . . . . . . . . . . . . 120
9.4 Windows of Stable Periodic Behavior . . . . . . . . . . . . . 122

10 Sarkovskii’s Theorem 125

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
10.2 The Intermediate Value Theorem . . . . . . . . . . . . . . . 125
10.3 Review of Two Fixed Point Theorems . . . . . . . . . . . . . 127
10.4 Sarkovskii’s Theorem . . . . . . . . . . . . . . . . . . . . . . 127
10.4.1 Discovering Sarkovskii’s Theorem . . . . . . . . . . . . 128
10.4.2 Using Sarkovskii’s Theorem . . . . . . . . . . . . . . . 130
Contents ix

11 Dynamical Systems on the Plane 135

11.1 Linear Algebra Foundations . . . . . . . . . . . . . . . . . . 135
11.2 Linear Systems with Real Eigenvalues . . . . . . . . . . . . . 138
11.3 Linear Systems with Complex Eigenvalues . . . . . . . . . . 144
11.4 Fixed Points of Nonlinear Systems . . . . . . . . . . . . . . . 148
11.5 Periodic Points . . . . . . . . . . . . . . . . . . . . . . . . . . 153
11.6 Chaos in the Hénon map . . . . . . . . . . . . . . . . . . . . 155

12 The Smale Horseshoe 159

12.1 Motivating the Horseshoe Map . . . . . . . . . . . . . . . . . 159
12.2 The Horseshoe Map . . . . . . . . . . . . . . . . . . . . . . . 160
12.3 More Symbolic Dynamics . . . . . . . . . . . . . . . . . . . . 163
12.3.1 Two-Sided Sequence Space . . . . . . . . . . . . . . . 163
12.3.2 The Two-Sided Shift Map . . . . . . . . . . . . . . . . 164
12.4 A Horseshoe in the Hénon Map . . . . . . . . . . . . . . . . 166

13 Generalized Symbolic Dynamics 169

13.1 Topology Foundations . . . . . . . . . . . . . . . . . . . . . . 169
13.2 Shift Dynamical Systems . . . . . . . . . . . . . . . . . . . . 171
13.2.1 One-Sided Shift Spaces . . . . . . . . . . . . . . . . . 171
13.2.2 Two-Sided Shift Spaces . . . . . . . . . . . . . . . . . 172
13.2.3 Shifts of Finite Type . . . . . . . . . . . . . . . . . . . 178
13.3 Representing Shift Spaces with Graphs . . . . . . . . . . . . 181
13.3.1 Higher Edge Graphs . . . . . . . . . . . . . . . . . . . 184
13.4 Markov Partitions . . . . . . . . . . . . . . . . . . . . . . . . 188

Bibliography 193

Index 195
Preface

This textbook was written primarily for undergraduate mathematics students

who have taken the entirety of calculus and have completed an introduction
to proofs course. Ideally, students would also have taken linear algebra and
an introductory analysis course, but this is not necessary. For the most part,
the required material from these disciplines is included in the text.
This book differs from most texts on dynamical systems by blending the
use computer simulations with inquiry-based learning (IBL). Computer sim-
ulations have long been used in the study of dynamical systems to visualize
chaotic behavior, and we believe that these tools provide students with a
platform from which they can begin to see more general mathematical truths.
Inquiry-based learning is an excellent tool to move students from merely
remembering the material, to deeply understanding and analyzing the math-
ematics. The method teaches students to think, to question, and to create,
rather than memorize. The method relies on asking students questions first,
rather than presenting the material in a lecture. Students are expected to work
out problems on their own or in groups, and then the material is discussed in
class. The instructor is less of a “sage-on-the-stage,” and more of a “mentor-
in-the-middle,” giving students the room to explore the content, while gently
guiding them toward the correct solutions.
Inquiry-based learning has been shown to improve student learning and to
be a more equitable method of content delivery. Students in an IBL classroom
tend to have longer retention of the material [5]. Even the students, who
may have trouble recalling specifics about the material have learned how to
learn, and that skill carries over to all disciplines of study. We also see that
underrepresented groups perform better in the IBL classroom [6]. IBL gives
students a chance to proceed at their own pace, to make mistakes, to learn
from them, and to succeed on their own terms.
There are two main ideas that we find helpful as IBL instructors. The first
is to be less helpful. It can be tough to watch a student struggle at the board,
but our jobs as instructors is to make that struggle productive. Ask questions
to lead them to the right solutions, without giving away the solutions. We
have attempted to scaffold the activities in this book so that students need
only make small leaps in their understanding as they progress. Small hints
are occasionally provided when the proof requires something unusual or uses
ideas they learned in other courses.
The second idea is that of productive failure. Students will have many
false starts and wrong turns in their understanding of the content. It is our

xi
xii Preface

job to ensure that they see those failures as learning opportunities, rather than
disappointments. Be patient, be kind, but hold the students accountable. You
can expect some resistance to the method for a few weeks. We have had
students complain: “He doesn’t teach anything; he just stands there asking
questions.” This is normal, but once a student experiences the joy of self-
discovery, there is no turning back.
If you are new to using IBL methods in the classroom, some help-
ful resources are available from the Academy of Inquiry Based Learn-
ing (http://www.inquirybasedlearning.org/) and the Legacy of R.L. Moore
(http://legacyrlmoore.org/). Both of these organizations offer workshops to
help get instructors involved with IBL. The Academy of Inquiry Based Learn-
ing, in particular has online resources and many active communities for in-
structors new to the method.
We refer to the unique approach to teaching mathematics used in this book
as ECAP for Explore, Conjecture, Apply, and Prove. We believe that ECAP
provides students with a deeper understanding of dynamical systems because
it mimics the actual practice of mathematics. In general, each section begins
with exercises guiding students through explorations of the featured concept in
dynamical systems and concludes with exercises that help the student formally
prove the result.
We chose to put the explorations first (as opposed to end-of-chapter
projects) because in our experience, computer explorations provide students
with an enhanced understanding of the mathematical concepts by allowing
improved visualization that is both dynamic and engaging. Moreover, the
ability to manipulate the equations and/or underlying mathematical aspects
allows students to grasp what aspects are essential for a given phenomenon
and what are not. This exploration, combined with some guided questions,
gives students the confidence to conjecture the formal theorems being formed.
The proof subsection then completes the student’s journey from the discovery
of the mathematics to the final step of proving the validity of the result.
Let’s look at this structure in more detail to understand how this text
might be used during the semester. Most sections of the text are divided into
Explore, Conjecture, Apply, and Prove portions. We find this to be an
effective approach to guide the young mathematician toward the depth of
understanding expected in an IBL classroom.
The Explore sections ask leading questions to introduce a concept and
help students see what may be true. Many of the explorations use nu-
merical activities that we have designed specifically to highlight the con-
cepts and ideas being explored. Each accompanying activity is available on
IBLdynamics.com . There you will find a collection of interactive apps de-
signed to get students thinking about the topics at hand. Students can discover
examples and counterexamples through manipulations built into the software.
This is a unique feature of the text, and one that has proven essential for stu-
dents taking the course in the past.
Preface xiii

We recommend assigning as many of the Explore problems as you think

are feasible for your course. These will mostly be done outside of class, but it is
worthwhile to do some of these activities in pairs or small groups during class-
time. This is especially true early in the term when students are unfamiliar
with inquiry-based learning. There is value in having multiple students share
different approaches to solving a problem, especially when some are incorrect.
Following the explorations, we move on to the Conjecture sections. These
activities are most effective when used in the classroom as prompts for discus-
sion about the mathematics. Encourage multiple students to share their ideas.
Look for commonalities and disagreement in their responses. Prompt the stu-
dents to share how the explorations led them to believe that certain things are
true. Look for students who have reached different conclusions. Keep asking
questions until students question the validity of a false conjecture, or are more
deeply convinced of a true conjecture.
The Apply portions include many problems to help students confirm or
disprove their conjectures and to give them the opportunity to practice the
technique at hand. Assign what you feel is reasonable to your students. It
is important to remember that students may also be doing Exploration or
Proof exercises and that all of this work needs to be carefully balanced.
Once the students have explored the content, conjectured the generalities,
and applied their results, they should move on to the proofs in the Proof
portion of the section. We have had success assigning some of these proofs
to be presented in class, and some to be turned in as homework. You will
need to balance your students’ needs with the relevance of the mathematics
to determine what proofs should be assigned as homework vs. presented and
discussed in class. When the proofs are presented in class, students should
be encouraged to present their proofs. These should be again used to foster
mathematical discussions. It is also worthwhile to have a student “recorder,”
who transcribes this work and writes a final version of the proof to be shared
with the class. As you move into more difficult proofs, more scaffolding is
provided to help guide the students.

Advice for Using this Text

When determining what material to include in this text, we focused primarily
on how that topic would or would not lend itself to the ECAP structure.
For example, there is no discussion of complex dynamics because we felt that
going from explorations of Julia sets to conjectures and proofs about their
connectivity would be unreasonable for most undergraduate students. There
are certain topics that are included that we felt were essential to explore and
conjecture, but chose not to prove either because of its difficulty or because a
complete proof would require deeper ideas from other areas of mathematics.
A prime example of this is the period-doubling bifurcation theorem.
xiv Preface

On the other hand, we did see a topical theme emerge that we believe
is unique to a dynamical systems text: symbolic dynamics. While symbolic
dynamics is a fairly standard topic in an undergraduate dynamics text, we
have tried to emphasize it in a way that is more detailed and inclusive than is
typically the case. This is partly because the ideas and proofs in this area are
accessible to undergraduates and approachable using the ECAP methodology.
Thus we chose to conclude this book with a chapter dedicated to a general
discussion of the topic.
Finally, we have chosen to include multiple sections on important ideas
from analysis and topology independent from their application to dynamics.
We feel that most students need a refresher (at the least) on these topics
and the study of dynamical systems provides a wonderful motivation and/or
introduction to these ideas.
In piloting this material, we were able to cover the first 10 chapters of this
book in a single semester. If you don’t need to cover the analysis material, you
will most likely be able to cover at least some of the material in Chapters 11–
13. We believe that the pairings of Chapters 11 and 12 or of Chapters 12 and 13
work well. Chapter 11 is not a prerequisite for Chapters 12 and 13 though the
material concerning the Hénon map in Chapter 11 nicely illustrates the theme
of Chapter 12.
Chapter 1 defines discrete dynamical systems and discusses numerical
methods for tracking the orbits of points in a system. It also introduces the
students to some of the tools on IBLdynamics.com . Students can also work
on graphical iteration, and see how dynamics can apply to population models.
In our experience, the upper-level students taking this class have little
experience with the analysis of sequences and other tools from analysis that
arise when studying dynamical systems. Thus, we have chosen to include much
of this material in this book. Chapter 2 provides a fairly complete coverage of
sequences for this reason. You may choose to skip this chapter if that is not
the case with your students.
Chapter 3 defines fixed and periodic points. The number of explorations
on iteration increases here, as we work to build a students’ intuition about
attracting and repelling points. They are challenged to make conjectures about
attracting and repelling fixed points, studying linear, quadratic, and logistic
maps. Population models are revisited and some important theorems about
fixed and periodic points are proven.
Much of Chapter 4 covers analysis topics with the primary focus being the
implicit function theorem and fixed point theorems. The students are given
more opportunity to come up with counterexamples, and the explorations
reflect this. The chapter concludes with a proof of the hyperbolic fixed points
theorem. Section 4.2 can be safely omitted if your students are familiar with
the implicit function theorem.
Bifurcations are covered in Chapter 5. A large number of explorations are
included at the beginning of the chapter to drive students toward conjecturing
Preface xv

important theorems about bifurcations. Tangent and period-doubling bifur-

cations are covered, as well as bifurcation diagrams.
Chapters 6 and 7 introduce the difference between local and global dynam-
ics. Students explore tools to catalog all of the different behaviors that can
occur in a system. The logistic map and the doubling map are studied. Sym-
bolic dynamics is introduced in the form of binary sequences to study orbits
in the doubling map. The logistic map is revisited with coefficients greater
than or equal to 4.
In Chapter 7, we delve further into symbolic representations, particularly
the shift space on two symbols. The Cantor Set is constructed to aid with mo-
tivation, and the definition of a metric is introduced. If Chapter 2 was omitted,
portions may need to be included here, depending on student background.
Chaos is defined in Chapter 8. Interactive portions of the website help
illustrate sensitive dependence on initial conditions and topological transitiv-
ity. The shift map, the doubling map, and the logistic map are returned to as
examples.
Chapter 9 presents bifurcation diagrams for the logistic map to demon-
strate the transition to chaotic behavior. Interactive tools help students find
windows of stable periodic behavior, and a study of the ordering of periodic
windows is introduced. This lays the foundation for Chapter 10.
Sarkovskii’s theorem and its proof is the focus of Chapter 10. There are
many explorations and interactive activities to motivate the theorem. The
chapter begins with a short review of the intermediate value theorem and two
fixed point theorems from Chapter 4. Then Sarkovskii’s theorem is discussed.
The student is guided through a partial proof at the end of the chapter.
Chapter 11 introduces dynamical systems in the plane. Section 11.1 may
be omitted if the class has sufficient background in linear algebra. The chapter
starts with linear systems, moves to systems with complex eigenvalues, and
then on to nonlinear systems. Interactive explorations on the website illustrate
the gingerbread attractor and the Hénon map. Instructors may wish to review
multivariable calculus or basic complex arithmetic as necessary.
The Smale horseshoe is motivated and constructed in Chapter 12. As a
tool to study the horseshoe, two-sided symbolic spaces and their associated
shift maps are introduced. The chapter concludes by relating the horseshoe
to the Hénon map.
The final Chapter, 13, goes into more detail about symbolic dynamical
systems, and the tools to study them. Symbolic systems on n-symbols are
defined, as are shifts of finite type. We introduce graph theory as a way to
study conjugate systems. Applications are given to tiling spaces and Markov
chains. The chapter concludes with an exploration of Markov partitions and
Arnold’s cat map.
xvi Preface

Message to Students
We could have structured the content of this text like a typical mathematics
book, but we’ve chosen not to, and for a specific reason. This book is designed
to help you learn to think like a mathematician. We want you to go beyond
memorizing theorems and calculating results. We want you to explore, to
discover, and to prove the results for yourself. Most students are brought up
with the idea that mathematics is entirely about showing what is true. It’s
actually the case that the much more important step is showing when and
why something is true. In this text, we plan for you to achieve exactly that.
Most every section starts with exploration activities. You will be asked to
work things out on paper or use activities at IBLdynamics.com to explore
some topic in dynamical systems. It is expected that you try everything. Take
notes, notice patterns, and don’t be afraid to make mistakes. If you see some-
thing that seems to be generally true, be ready to make that conjecture. Be
prepared to share your insight with the class. And be ready to explain why you
think your insight is true. Your instructor will guide you through the process,
but ultimately, the work must be your own. You will learn that it is acceptable
to fail, and that making a false conjecture, or an error in explanation, is just
part of learning to be a mathematician. You will learn that these failures can
be productive. If we learn from our errors and move forward, an error is not
a failure, it is an opportunity to truly understand something new.
A course like this returns dividends proportional to your efforts. It is much
easier to be given an algorithm, and to execute that algorithm, than to struggle
for the results yourself. It is difficult, but struggling is essential for understand-
ing the mathematics and for personal growth. We sometimes tell our students
to “embrace the stuck” because all new and important ideas originate here.
The joy of discovery, the frustration of exploration, and the satisfaction of
collaboration are all important aspects of mathematics. You will explore. You
will conjecture. You will prove results. And you will apply results to a variety
of related problems. All of these aspects combine when one is creating new
mathematics. It is the intention of the authors that this course will prepare
you, not just for an understanding of dynamical systems, but for exploration
into any area of mathematics you find appealing.

Acknowledgments
We have received the help and support of many. But foremost on that list are
our colleagues in the MCS Department at Gustavus Adolphus College who
have listened to our musings, critiqued our writing, and occasionally covered
Preface xvii

our classes. We would be remiss if we didn’t single out Mike Hvidsten, who
encouraged us to take this project from a collection of classroom activities
and demonstrations to a full-fledged textbook.
We had the help of two wonderful Gustavus students, Emilee Mason and
Zach Dawson. Emilee, who “beta-tested” this project in our Discrete Dynami-
cal Systems course, was invaluable in her proofreading and editing duties. She
was especially helpful in finding all of the places where we omitted commas! If
any are still missing, it is our fault and not hers. Zach Dawson was a lifesaver!
A computer science major, Zach translated all of Tom’s poor Mathematica
code into clean and functioning JavaScript. We would still be struggling with
the basics of JavaScript if it weren’t for Zach.
Tom would also like to thank Francis Su and his colleagues at Harvey
Mudd College. During the infancy of this project, Tom spent a week visiting
Francis picking his brain about writing, inquiry-based learning, and life. As is
often the case, these latter conversations were the most valuable.
Jeff would like to thank Dr. Steven Clontz at University of Southern Al-
abama, and the late Dr. Frank Sturm, for their endless late-night discussions
about the value of inquiry-based learning and the nuts and bolts of how to
make it effective.
Various aspects of this work, including funding for Emilee and Zach, were
possible through a generous gift to Gustavus Adolphus College by the family
of Clifford M. Swanson.
1
An Introduction to Dynamical Systems

1.1 What Is a Dynamical System

At the most basic level, a dynamical system is simply something that evolves
deterministically over time. In this context, deterministic means that the sys-
tem changes by rules that are fixed and not random. Dynamical systems arise
in an incredibly wide variety of applications. For example, the motion of a
pendulum is a continuous dynamical system with the angular position and
angular momentum determined at every time. If we know the initial position
and momentum, then the rules of the dynamical system determine these quan-
tities at every time in the future. In ecology, models of population growth are
often discrete dynamical systems. The model is a function that uses the pop-
ulation at one generation to compute the population of the next generation.
Once again, if we know the initial population then by repeatedly applying this
rule we can compute the population at any time in the future.
Ordinary differential equations are often the vehicle for modeling contin-
uous dynamical systems like the pendulum model. In this book we focus on
discrete dynamical systems.

Definition 1.1 A discrete dynamical system is a function f : X → X

that generates a sequence

{xn }∞
n=0 = {x0 , x1 , . . . }

from an initial condition x0 ∈ X using the rule xn+1 = f (xn ).

The space X is called the state space of the dynamical system, and for most of
this book, X will be the real numbers R. What is important here is that it is
very easy to compute the sequences generated by discrete dynamical systems.
Starting with an initial condition x0 , we just apply the function f to compute
x1 . In other words,
f (x0 ) = x1 .
The subscript can be thought of as a discrete time measurement. If time is
measured in years, then x0 is the value of x in year 0, x1 is the value of x in
year 1, and so on.

1
2 An Introduction to Dynamical Systems

Once we have computed x1 we then compute x2 by applying f to x1 so

that
f (x1 ) = x2 .
By continuing this process indefinitely, we generate the sequence {xn }∞ n=0 .
This process of starting with an initial condition x0 and repeatedly applying
the function f is called iteration.
One would think that such a simple system would only exhibit relatively
straightforward behaviors. Maybe solutions can only approach an equilibrium
or at worst be periodic. This is far from true. The kinds of behaviors that
can occur are quite remarkable; so remarkable that we often refer to these
dynamics as chaotic dynamics. Even more remarkably, this wild and chaotic
behavior, which at first seems almost beyond description, can often be easily
described using just sequences of zeros and ones!
The purpose of this text is to guide you on a journey where you discover
these behaviors, you conjecture mathematical statements about them, and
ultimately you prove the truth of these mathematical statements. Let’s pack
our gear and begin exploring.

1.2 Numerical Iteration and Orbits

Iteration is best illustrated with a simple example. Let’s consider the function
f (x) = x/2 and take as an initial condition x0 = 12. Then

x1 = f (x0 ) = f (12) = 6

and
x2 = f (x1 ) = f (6) = 3
and so on. Tables, such as the one inset into figure 1.1, are a natural way
to show sequences generated by iteration. The first column is the iteration
number, and the second column is the sequence of iterates of f with x0 = 12.
The plot in figure 1.1 shows the same iterations but in a graphical format.
Here the horizontal axis is the iteration number (n), and the vertical axis, the
sequence element (xn ).
Notice that each entry in the second column of the table in figure 1.1 is
f applied to the entry immediately above it. This is iteration and the focus
of our study. Given a dynamical system, we would like to describe all of the
possible sequences that are generated by the given function. We call these
sequences orbits. The example in figure 1.1 shows the first six terms of the
orbit of x0 = 12 under iteration by f (x). We might describe this particular
orbit by saying

the orbit of x0 = 12 under iteration by f (x) = x/2 converges to x = 0.

Numerical Iteration and Orbits 3
f(x) x

FIGURE 1.1
The first six iterates of f (x) = x/2 with an initial condition of x0 = 12
displayed as both a table and a graph.

You probably already see that every orbit of this dynamical system, re-
gardless of initial condition, converges to 0 and by doing so you’ve completely
described this dynamical system.

Explore

Exploration 1. Consider the dynamical system

xn+1 = x2n .

For each of the initial conditions listed below, compute the first 10 iterates of
their orbit. Make both a table and a plot like those in figure 1.1. In each case,
describe the orbit in your own words. Look for commonalities and differences
in the different orbits and try to categorize the different types. Suggestion:
Doing this by hand gets old pretty quick. However, doing this with a spread-
sheet is very easy. After doing the first couple by hand, think about how you
might implement this dynamical system in a spreadsheet program. This can
also be used to create plots such as that show in figure 1.1.

a) x0 = −2.0 b) x0 = −1.5 c) x0 = −1.0

d) x0 = −0.5 e) x0 = 0.0 f) x0 = 0.5
g) x0 = 1.0 h) x0 = 1.5 i) x0 = 2.0

Exploration 2. Repeat exploration 1 using the dynamical system

xn+1 = x2n − 1.
4 An Introduction to Dynamical Systems

Use the same initial conditions that you used in exploration 1. Describe each
of the orbits in your own words. Look for commonalities and differences in the
different orbits and try to categorize the different types.

Exploration 3. Using the same initial conditions as exercise 1, find the

orbits of
xn+1 = x2n + 1.
Describe the orbits in your own words. Look for commonalities and differences
in the different orbits and try to categorize the different types.

Exploration 4. Explore the dynamical system

xn+1 = 1 − |xn |.
Use the initial conditions on the following table and describe the orbits.
a) x0 = −1.0 b) x0 = −0.5 c) x0 = −0.4
d) x0 = −0.2 e) x0 = 0.0 f) x0 = 0.2
g) x0 = 0.4 h) x0 = 0.5 i) x0 = 1

Exploration 5. For each of explorations 1 through 3, find an initial condition

where the orbit only contains one value, or show that no such point exists.
Before going much further, we need to introduce some basic terminology
that we will be using to describe orbits. The first and most important type of
orbit is one that is constant or fixed. For example, the orbit of x0 = 0 is fixed
in exploration 1.
Definition 1.2 Let f : X → X. A point x̃ is a fixed point of f if
f (x̃) = x̃. (1.1)
Fixed points are also referred to as equilibrium points. It follows directly from
this definition that if x̃ is a fixed point, then the orbit of x̃ is the sequence
{x̃, x̃, x̃, . . . } since each application of the function f leaves x̃ unchanged.
One almost always begins the study of a given dynamical system by com-
puting the fixed points. This is such an important concept that a good portion
of chapter 3 is dedicated to the subject.
In exploration 2 you saw some other interesting behavior when the initial
condition x0 = 0. You observed that
f (0) = −1, f (−1) = 0, f (0) = −1, . . . .
We can restate this another way. The orbit of x0 = 0 under iteration by f is
{0, −1, 0, −1, . . . }.
We call this a period 2 orbit or we might say that x0 = 0 is a period 2 point
of this dynamical system. Of course x0 = −1 is also a period 2 point.
Numerical Iteration and Orbits 5

Definition 1.3 Let f : X → X. A point x0 is a period 2 point of the

dynamical system xn+1 = f (xn ) if
f (f (x0 )) = x0 .
If f (x0 ) 6= x0 , then we say that x0 is a prime period 2 point.
Let’s discuss this definition a bit further. First, note the two-fold compo-
sition in the definition. Iteration is ultimately the repeated composition of a
function with itself since the output of one step becomes the input to the next.
If f (x0 ) = x1 and f (x1 ) = x0 , then
f (f (x0 )) = f (x1 ) = x0 .
In other words, and this is important, a period 2 point of f (x) is a fixed point
of f (f (x)). Also note that a fixed point x̃ of f must also be a period 2 point
of f since in this case
f (f (x̃)) = f (x̃) = x̃.
To distinguish fixed points from “true” period 2 points, we use the adjective
“prime.”
Dynamical systems can also have period 3 points, period 4 points, and so
on. A point x0 is a period 3 point if
f (f (f (x0 ))) = x0 .
Obviously this gets a bit cumbersome notationally. Imagine what you would
have to write down to describe a period 243 point! To deal with this problem,
we use superscripts on the function name. More formally, f n (x) denotes the
n-fold composition of f with itself. So
f 3 (x) = f (f (f (x)))
and so on.

A Warning about Composition Notation

It is natural to read f 3 (x) as “f (x) cubed” or “f (x) to the third
power.” This is wrong! Always remember that the superscript refers to
function composition and not exponentiation.

Exploration 6. What is f n (f m (x))?

Using this notation, we can now define periodic points in general.
Definition 1.4 Let f : X → X. A point x0 is a period p point of the
dynamical system xn+1 = f (xn ) if
f p (x0 ) = x0 .
If f k (x0 ) 6= x0 for all k < p, then we say that x0 is a prime period p point.
6 An Introduction to Dynamical Systems

Exploration 7. IBLdynamics.com The tool for this exploration on the

website gives the first 11 elements of orbits to several dynamical systems.
Determine whether they are periodic or not. If so, determine the period.

1.3 Graphical Iteration

There is a very useful way of thinking about the iteration of the discrete
dynamical system xn+1 = f (xn ) that uses the graph of the function y = f (x)
and the diagonal line y = x. Recall that the basic idea of iteration is that the
output from step n becomes the input to step n + 1. If we have a plot showing
the graph of y = f (x) and the line y = x, then we can find x1 = f (x0 ) by
drawing a vertical line from x0 on the x-axis to the graph of y = f (x) to find
the point (x0 , f (x0 )) = (x0 , x1 ). Note that the output x1 is the y-coordinate
of this pair. We next need to convert the output x1 from a y-coordinate to
an x-coordinate so that we can input it to f to find x2 = f (x1 ). Thus the
next step in this process is to locate x1 on the x-axis. We do this by drawing
a horizontal line from the point (x0 , x1 ) to the line y = x. The coordinate of
this point is (x1 , x1 ) since all points on a horizontal line all have the same
y-coordinate and we stopped drawing where y = x. Now going vertically to
the x-axis gives us the location of the value x1 on the x-axis. Finally, going
vertically again to the graph of y = f (x) gives us x2 as a y-coordinate. We
repeat this process to iterate the dynamical system and in so doing graphically
compute the orbit of x0 .
Note that at one point in the process, we go vertically to the x-axis and
then vertically to the graph of y = f (x). This seems wasteful and is in fact
unnecessary. Here is a step-by-step description of this process that eliminates
this wasteful step.
1. Draw a vertical line to the graph of y = f (x).
2. Draw a horizontal line to the diagonal y = x.
3. Repeat, repeat, repeat, . . .
This process gives a picture like the one in figure 1.2 that shows the it-
eration of xn+1 = x2n with an initial condition of x0 = 0.9. Note how the
“stairsteps” start at the point (0.9, 0.81) on the graph of y = x2 and go hor-
izontally to the diagonal y = x and then vertically to y = x2 . Each vertical
line represents one iteration. Hence this figure shows a total of 6 iterates (the
first vertical line from the x-axis to the graph is not shown which is why you
only count 5 of them).
Graphical iteration is sometimes referred to as graphical analysis and
the pictures generated in this way are called either stair-step diagrams or
cobweb diagrams.
Graphical Iteration 7

FIGURE 1.2
A plot showing the graphical analysis of xn+1 = x2n with an initial condition
of x0 = 0.9.

Explore

Exploration 8. Explain why a fixed point of a function f is the intersection

of the graph of y = f (x) and the line y = x.

Exploration 9. IBLdynamics.com Print out the graphs in figure 1.3 from

the IBLdynamics.com website. For each graph, approximate the values of the
fixed points (if any). Practice generating stair-step diagrams for each of these
dynamical systems. For each one, describe in your own words the orbits that
you find. Look for commonalities and differences in the different orbits and
try to categorize the different types.

Exploration 10. IBLdynamics.com Adjust the graph at IBLdynamics.com

for each of the following parameters (i.e., k values). Draw the graph on paper
and construct a stair-step diagram with the indicated initial conditions.

k 0.5 0.25 0.1 0.1 −0.5

x0 0.5 0.5 0.4 0.6 0.5
8 An Introduction to Dynamical Systems

- -
- -

-
-

- -

- -

- -
-
-

- -

FIGURE 1.3
Figures for practicing graphical analysis.

1.4 Modeling Using Discrete Dynamical Systems

Discrete dynamical systems are incredibly useful for modeling real-world phe-
nomena that occur in discrete time steps. In this section you will develop a
population dynamics model that falls into this category. Let pn denote the
population of some species at generation n.

Apply

Assumption: Generations are non-overlapping, and the population at gen-

eration n + 1 depends deterministically on only the population at generation
n. In other words, there exists some function f such that pn+1 = f (pn ).
Modeling Using Discrete Dynamical Systems 9

Framework: Assume that the function f has the form

f (p) = λ(p)V (p)p

where
• V (p) is called the viability and measures the percentage of newborns
in the species that survive to reproductive age, and
• λ(p) is the mean reproductive rate and measures the mean number
of newborns per adult.

Application 11. Given the above descriptions of the functions λ and V,

what mathematical properties do you think each of these functions should
have? You might want to first think about what values they should take. Then
about what shapes the graphs should or should not have. For each property
you list, explain how and why that property is related to the modeling of a
population. This might mean listing additional modeling assumptions.

Application 12. Give one or two examples of possible viability functions

V (p). This function may also use some parameters (i.e., constants), but you
do not need to assign specific values to these parameters. Are there any un-
derlying assumptions inherent in your choice of V that you did not specify in
application 11? Suggestion: Start with the simplest example you can think
of and then build from that.

Application 13. Give one or two examples of a possible reproductive rate

function λ(p). This function may also use some parameters (i.e., constants),
but you do not need to assign specific values to these parameters. Are there
any underlying assumptions inherent in your choice of λ that you did not
specify in application 11? Suggestion: Start with the simplest example you
can think of and then build from that.

Application 14. Use the methods described in this chapter to play with
your model. Try a variety of reasonable parameter values for the unassigned
parameters. We suggest using a spreadsheet to do this, but feel free to use
whatever tool you find appropriate. Describe some of the behaviors that your
model produces. Below are a few questions that might be helpful to you as
you do this. Do not limit yourself to these questions. Be creative!
1. Is the behavior that you observe realistic? For example, does it ever
lead to negative population numbers. If it is not realistic, think
about modifying it and repeating this process.
2. Does the population approach some fixed value? Does it oscillate
between multiple values? Does it do something more complicated?
10 An Introduction to Dynamical Systems

3. Does the behavior depend on the initial population size in some

way? For example, does the population go extinct for some starting
values but not for others?
4. How does the behavior change, if at all, when you change some of
the parameters in the model?
5. Graph your function f (p) = λ(p)V (p)p and the diagonal line as you
did in section 1.3. Describe how this graph changes as you adjust
one or more of the parameters in the model.
2
Sequences

2.1 Introduction to Sequences

The study of discrete dynamical systems is ultimately the study of infinite
sequences generated via iteration as described in chapter 1. You have probably
spent some time in your mathematics career learning about sequences. For
example, in calculus 2, you spent several days learning about infinite sequences
and how they are used to study infinite series. Or maybe you played with the
Fibonacci sequence as a child. And if you had an analysis class, maybe you
learned how sequences of rational numbers are used to define the real numbers.
In this class, we are studying sequences not as a vehicle to understand some
other mathematical concept, but rather as objects that are interesting in their
own right.
This chapter provides some of the fundamental concepts for studying se-
quences and while it may be review for many of you, we believe that spending
some time on the fundamentals is necessary before returning to discrete dy-
namical systems.

Explore

Let’s begin with a fun example, the Fibonacci sequence. The first two
terms of the Fibonacci sequence are F0 = 1 and F1 = 1. Each subsequent
term is determined by the recurrence
Fn = Fn−1 + Fn−2
so that each term is simply the sum of the previous two terms.
Exploration 1. Compute Fn for n = 0, . . . , 10.
It is natural (and useful) to think of a sequence as simply an ordered list of
things (usually numbers). As mathematicians, we need to turn this intuitive
idea into something much more precise.

11
12 Sequences

Definition 2.1 A sequence {x0 , x1 , x2 , . . . } = {xn }∞

n=0 with xn ∈ X for all
n is a function from the natural numbers N to the space X.

It is common practice to omit the n = 1 and the ∞ from this notation and
simply write {xn } when referring to a sequence when no confusion will arise.
In this more sophisticated view, a sequence is a function that assigns a
value in the space X to each natural number via n 7→ xn . This gives a list of
numbers with the position of the number in the list corresponding to n. The
space X is the set of all outputs of this function (or codomain) and for us X
will almost always be the real numbers R.
Exploration 2. Compute the first 5 elements of each of the sequences below.
Describe the behavior of each sequence in your own words.
1
1. xn = 1 −
n
2. yn = (−1)n
 n
4
3. zn =
5
4
4. w0 = 0, wn+1 = wn
5
n2 + n − 1
5. an =
n+4
2
n +n−1
6. bn =
n2 + 4

2.2 Convergence of Sequences

The most basic question you can ask about a sequence is whether it converges
or not. You have an intuitive understanding of what this means. A sequence
converges to some value L if the terms of the sequence get “closer and closer
to L.”
But because logical precision is essential in mathematics, this intuitive idea
needs to be formalized. Here is the formal definition where we assume that
xn ∈ R for all n.

Definition 2.2 A sequence {xn } converges to a limit L < ∞ if for every

ε > 0 there exists an N > 0 such that n > N implies that |xn − L| < ε. When
the sequence {xn } converges to a limit L we write

lim xn = L.
n→∞

If a sequence does not converge to a limit L, we say that the sequence diverges.
Convergence of Sequences 13

This definition should sound very familiar to you. It is almost exactly the
definition of a limit that you learned in calculus. In particular, it is almost
word for word the definition of a “limit at infinity.” How does definition 2.2
formalize our intuitive concept of convergence? If ε is how closely you want to
approximate the limit L, then N is how far you have to go out in the sequence
to obtain this level of precision.

Understanding Absolute Values

One of the main difficulties for many students when using the defini-
tion of a limit is dealing with absolute values. Many of us were told that
when seeing an absolute value we simply “drop the minus sign.” Unfortu-
nately, this simple rule is not at all helpful when dealing with expressions
like the one in definition 2.2.
Here is a more helpful way of interpreting the expression

|A − B|. (2.1)

Read expression 2.1 as “the distance between A and B.” The inequality

|x − 4| < 3

then reads “the distance between x and 4 is less than 3.” When thought
of this way, it is clear that this is the same as

1<x<7

where the left-hand term is 1 because 4 − 3 = 1 and the right-hand term

is 7 because 4 + 3 = 7.
Using this idea, the inequality

|xn − L| < ε

in definition 2.2 reads “the distance between xn and L is less than ε.”
This naturally leads to

L − ε < xn < L + ε

and we now have the expression written in a form that is amenable to

algebra.
14 Sequences

Explore

Exploration 3. Interpret each of the following absolute value statements

using the distance interpretation described in the box above. Express each
of them as a double inequality with only x in the center. You may need to
manipulate some of them to put it in the necessary form.
a) |x − 3| < 0.1 b) |x + 3| < 0.1 c) |x| < 0.1
d) |3x − 6| < 1 e) |10x − 5| < 2 f) |6x − 4| < 10

Exploration 4. IBLdynamics.com The tool on the website for this exercise

shows the elements of the sequence
1
xn = 1 − ,
n
the limit value is L = 1, and the “ε-band” is the interval (L − ε, L + ε) about
L = 1. The tool allows you to change the value of ε. Doing this causes the
image to zoom in or out vertically (by changing the limits on the vertical axis)
and to zoom in or out horizontally (by changing the limits on the horizontal
axis). Use the tool to estimate the value of N needed to ensure that all further
elements of the sequence are in the ε-band for 5 different values of ε. Do you
see a relationship between your chosen value of ε and N ?

Exploration 5. Reread definition 2.2. For the values of ε that you chose in
exploration 4, directly compute the value of N so that
|xn − L| < ε.
How do each of these computed values compare with your answers in Explo-
ration 4?

Apply

It is often the case that the techniques you used in calculus to compute
limits at infinity apply to sequences. That is the case with many of the ex-
ercises in this section. It might be helpful to review this material in general,
and L’Hopital’s rule in particular.
 n
4
Application 6. Compute the limit of the sequence zn = and use
5
definition 2.2 to prove that it converges to this limit value.
The Squeeze Theorem 15

Application 7. Compute the limit of the sequence zn = an for 0 < a < 1

and use definition 2.2 to prove that it converges to this limit value.

2n + 4
Application 8. Compute the limit of the sequence zn = and use
5n + 1
definition 2.2 to prove that it converges to this limit value.

2
Application 9. Compute the limit of the sequence zn = and use
5n + 1
definition 2.2 to prove that it converges to this limit value.

Application 10. Compute the first 10 terms of the sequence

1
zn = (−1)n + .
n
Does the sequence converge or diverge? Why?

Application 11. Compute the first 10 terms of the sequence

(−1)n + 1
zn = .
n
Does the sequence converge or diverge? Why?

L’Hopital’s Rule and Sequences

L’Hopital’s rule can be helpful when computing the limit of sequences,
but you must be careful. L’Hopital’s rule applies to limits of differentiable
functions on R, but a sequence is a function on the natural numbers N.
The only time you can use L’Hopital’s rule to determine the limit of a
sequence {xn } is when
xn = f (n)
for some differentiable function f . Thus, you can use it for the limit in
application 8 but you cannot for the limit in application 11. If you apply
L’Hopital’s rule to a sequence, you should always justify its use.

2.3 The Squeeze Theorem

There are a variety of theorems that make the computing of limits (as well
as the proofs) much easier. In what follows, we explore some of the basic
theorems concerning the convergence of sequences. Again, many of these are
almost identical to calculus theorems about limits.
16 Sequences

Explore

Exploration 12. IBLdynamics.com Begin by using the application for the

exercise to choose a random number. Next, create two different sequences that
each converge to your random number. Call one of them {an } and the other
{bn } and enter the formulae for these sequences you’ve found into the boxes
provided. Visually confirm that the sequences converge to the value you were
given.

Exploration 13. Create a third sequence cn as follows. Click the button on

this website activity to generate a random weight value w. Then, let

cn = wan + (1 − w)bn .

What can you say about the relationship of the values cn relative to the values
of an and bn ?

Exploration 14. IBLdynamics.com Plot all three sequences on the same

set of axes. Does this picture agree with your answers to exploration 13?

Conjecture

Conjecture 15. Based on your work above, complete the following statement
of the Squeeze Theorem.

If an ≤ cn ≤ bn for all n and then .

Apply

Application 16. Use the Squeeze Theorem to compute the following limits.
cos n
1. lim
n→∞ n
2
n + sin n cos n
2. lim
n→∞ n2
The Squeeze Theorem 17

Application 17. Use the Squeeze Theorem to show that the following se-
quences converge.
cos(n2 )
1. xn =
n4
sin(n)
e
2. xn =
n2
(−1)n
3. xn =
n
Follow the steps below to prove that

lim n1/n = 1. (2.2)

n→∞

Application 18. Let sn = n1/n − 1. To what value should the sequence {sn }
converge in order to prove equation 2.2?

Application 19. Manipulate the formula for sn in application 18 to show

that
n
n = (sn + 1) .
Application 20. Look up the Binomial Theorem if you don’t already know
it. It tells you how to expand (a + b)n . Use this to write out at least the first
3 terms of your work in application 19.

Application 21. Explain why the result of application 20 implies that

n(n − 1) 2
n> sn .
2
Application 22. Solve the inequality in application 21 for sn .

Application 23. Explain why 0 ≤ sn for all n.

Application 24. Apply the Squeeze Theorem to prove the result in equa-
tion 2.2.

Prove

Proof 25. Use Definition 2.2 to prove the Squeeze Theorem as stated below.

Theorem 2.1 If an ≤ cn ≤ bn for all n and

lim an = lim ab = L
n→∞ n→∞

then
lim cn = L.
n→∞
18 Sequences

2.4 Arithmetic Limit Theorems

All of the standard arithmetic operations can be done with sequences to cre-
ate new sequences. In other words, sequences can be added, subtracted, mul-
tiplied, etc. There are theorems about the convergence of these new sequences
in terms of the convergence of the original sequences. In what follows, suppose
that both {an } and {bn } are convergent sequences with

lim an = A
n→∞

and
lim bn = B.
n→∞

Conjecture

Conjecture 26. Write a theorem about the convergence of the sequence

{cn } if cn = an ± bn .

Conjecture 27. Write a theorem about the convergence of the sequence

{cn } if cn = Kan where K is a constant.

Conjecture 28. Write a theorem about the convergence of the sequence

{cn } if cn = an bn .

Conjecture 29. Write a theorem about the convergence of the sequence

{cn } if cn = 1/bn . Be careful with your hypothesis!

Conjecture 30. Write a theorem about the convergence of the sequence

{cn } if cn = an /bn . Be careful with your hypothesis!

Apply

Application 31. Find the values to which the following sequences converge.
Do these results agree with your conjectures 26 through 30?
3n2 − 1
1. an =
10n + 5n2
Bounded and Unbounded Sequences 19
ln(n + 1)
2. bn =
ln(1 + 4n)
3. cn = e−n

Application 32. Assume {an } and {bn } are both real sequences. Provide
counterexamples to each of the following statements.
1. If limn→∞ an = 0, then limn→∞ (an bn ) = 0.
2. If limn→∞ (an bn ) = 0, then either limn→∞ an = 0 or limn→∞ bn =
0.
3. If limn→∞ (an +bn ) does not exist, then either {an } diverges or {bn }
diverges.
an+1
4. If limn→∞ an = 1, then {an } converges.
Do these problems suggest revising any of your conjectures 26 through 30?

Application 33. Recall that the Fibonacci sequence is given by F0 = 1, F1 =

1 and Fn = Fn−1 + Fn−2 . It can be shown that the sequence
 
Fn
Fn−1
converges to√some finite value φ. Assume that the sequence converges to prove
that φ = 1+2 5 . This constant is known as the golden ratio.

Prove

Proof 34. Prove each of the statements in conjectures 26 through 30.

2.5 Bounded and Unbounded Sequences

Sequences that do not exceed some maximum value have some useful conver-
gence properties that we will use in our study of dynamical systems. Thus, it
is helpful to explore the convergence properties of such sequences. As usual,
we begin with some definitions.

Definition 2.3 A sequence {xn } is bounded above if there exists M such

that xn ≤ M for all n. A sequence {xn } is bounded below if there exists
m such that xn ≥ m for all n. A sequence that is both bounded above and
bounded below is said to be bounded.
20 Sequences

Definition 2.4 A sequence {xn } is said to have an infinite limit if for every
M > 0 there exists N > 0 such that n > N implies xn > M. We denote this
by
lim xn = ∞.
n→∞

Definition 2.5 A sequence {xn } is nondecreasing if xn+1 ≥ xn for all n.

A sequence {xn } is nonincreasing if xn+1 ≤ xn for all n. We often refer to
sequences having either of these properties as monotone sequences.

A Note on Monotone Terminology

Mathematicians are sometimes a little careless with terminology as it
relates to monotone sequences. We sometimes call a sequence monotone
increasing even when it is technically monotone nondecreasing. That is,
we might call a sequence where xn+1 ≥ xn monotone increasing even
though two consecutive terms of the sequence might be equal. I suppose
the only reason for this is that it is a lot easier to say “increasing” than
“non-decreasing.”

Explore

Exploration 35. List some techniques that you might use to show that a
sequence is monotone?

Exploration 36. Construct several different examples of monotone increas-

ing sequences that are bounded above. Repeat with monotone decreasing se-
quences. What do you notice about each of these sequences?

Conjecture

Conjecture 37. Based on your work in exploration 36, conjecture a the-

orem about the limit of a monotone increasing sequence that is bounded
above. State this conjecture in formal mathematical language. Repeat this
for a monotone decreasing sequence bounded below.
Bounded and Unbounded Sequences 21

Apply

Application 38. Use definition 2.4 to prove that

n2 + 3
lim = ∞.
n→∞ n − 1

Application 39. Use definition 2.4 to prove that

n2 + 3
lim = ∞.
n→∞ n + 1

Be careful! The proof is not identical to the proof of application 38.

Application 40. Let x1 = 1 and define

n
xn+1 = x2 .
n+1 n
1. Compute the terms x1 through x4 .
2. Show that the sequence {xn } is monotone decreasing and bounded
below. Conclude that the sequence converges.
3. Challenge: Prove that the sequence {xn } converges to 0.

Application 41. Let x1 = 1 and define

1
xn+1 = (xn + 1).
3
Use the techniques of this section to prove that the sequence {xn } converges.
Challenge: What does it converge to and why?

13n
Application 42. Is the sequence an = monotonic?
n!
Application 43. If a sequence of positive real numbers is strictly decreasing,
must it converge to zero?

Application 44. Can a monotone sequence which is unbounded converge?

Application 45. Can a bounded sequence which is not monotone converge?

22 Sequences

Prove

At this point, you should be fairly comfortable with monotone sequences

and hopefully you conjectured that a monotone increasing sequence that is
bounded above converges. The formal statement of that theorem is as follows.

Theorem 2.2 If {xn } is a monotone increasing sequence bounded above then

{xn } converges. Similarly, if {xn } is a monotone decreasing sequence bounded
below then {xn } converges.

Before turning to the proof of this theorem, we need to discuss an impor-

tant property of the real numbers called the Completeness Axiom.
Axiom 2.1 (Completeness Axiom) If S ⊂ R is bounded above, then S
has a least upper bound called a supremum and denoted sup S. Similarly, if
S is bounded below, it must have a greatest lower bound called an infimum
and denoted inf S.
Infimums and supremums are like minimums and maximums except they do
not necessarily need to be contained in the set S.
Exploration 46. Find the infimums and supremums of the following sets if
they exist. Determine, if possible, whether each is included in the given set or
not.

1. S = (0, 1)
2. S = [0, 1]
3. S = {x ∈ (0, 1) | x ∈ Q}
4. S = {x ∈ R | x = 1/n, n ∈ Z}

Notes on the Completeness Axiom

It is important to understand that the Completeness Axiom is an
axiom and not a theorem. That means that this property is one of the
fundamental (and hence non-provable) assumptions used to define the
real numbers. It is akin to one of the axioms of Euclidean geometry such
as “through any 2 distinct points there is one and only one line.” All of ge-
ometry is built up from these axioms. Similarly, all of the properties of the
real numbers are built from a set of axioms one of which is Completeness.
Also note that the Completeness Axiom applies to sets of real numbers,
not just sequences. However, a sequence is a set if we ignore the ordering.
Subsequences 23

With this background in mind, we can proceed with the proof of this
theorem. You will prove it for monotone increasing sequences with the proof
for monotone decreasing sequences being similar.
Proof 47. Let {xn }∞n=1 be a monotone increasing sequence bounded above.
Explain why the set {xn } has a supremum u. Let sup{xn } = u.
Our goal is to show that in fact the sequence {xn } converges to u.
Proof 48. Fix ε > 0 and explain why u − ε is not a supremum of {xn }. Use
this to conclude that there exists N > 0 such that if n > N then xn > u − ε.

Proof 49. Conclude the proof by using the result of proof 48 to show that
|xn − u| < ε for all n > N and thus

lim xn = u.
n→∞

2.6 Subsequences
It is sometimes important to consider only a selection of terms in a sequence,
rather than all of the terms. For example, you might want to look at the se-
quence of even indexed terms or the sequence of odd indexed terms because
doing so makes it easy to determine whether the sequence converges or di-
verges. A new sequence taken from a given sequence is called a subsequence
and is formally defined below.

Definition 2.6 Given a sequence {xn }∞

n=1 and an infinite set of natural num-
bers v1 < v2 < v3 < · · · , a subsequence {x̂k }∞
k=1 is the sequence where
x̂k = xvk .

Explore

Exploration 50. Let’s start with a basic example to illustrate this concept.
Consider the sequence
∞
(−1)n

{xn }∞
n=1 = .
n n=1

1. What is the subsequence {ak } = {x2k }?

2. List properties of this subsequence.
3. What does the subsequence {ak } converge to?
24 Sequences

4. What is the subsequence {bk } = {x2k−1 }?

5. List properties of this subsequence.
6. What does the subsequence {bk } converge to?
7. What does this tell you about the original sequence? What sequence
theorem are you using here?

Exploration 51. In applications 6–9, you computed the limits of various

sequences. For each of these sequences, perform the following:
1. Construct a subsequence that converges to the same value as the
original sequence.
2. Construct a subsequence that diverges, or show one does not exist.

Exploration 52. Consider the sequence xn = cos( nπ

2 )

1. Does this sequence converge or diverge?

2. Does the subsequence {x2n+1 } converge or diverge?
3. Does the subsequence {x4n } converge or diverge?

Exploration 53. Construct an unbounded sequence with a bounded subse-

quence.

Exploration 54. Consider the sequence xn = (−1)n . Find two disjoint

subsequences that converge to −1 and 1, respectively.

Prove

Proof 55. Prove that, if a sequence is convergent, its subsequences are also
convergent, and moreover, they converge to the same value.

2.7 Liminfs and Limsups

In proving theorem 2.2, you used either the supremum or the infimum concept
to show that a bounded monotone sequence converges. In this section, we will
take this type of analysis one step further by creating sequences of these
quantities and then computing their corresponding limits.
Liminfs and Limsups 25

Let {xn } be a sequence of real numbers. Define

uN = inf {xn : n > N } (2.3)

and
vN = sup {xn : n > N } . (2.4)
Carefully note how both n and N are used in these definitions. For a given
uN or vN , the number N is fixed while n indicates all of the subscripts greater
than this fixed value. Thus, each uN is an infimum of the set that is composed
of all elements of the sequence {xn } that have an index greater than N. You
might say that each uN is an infimum of the “tail” of the original sequence.
Similarly, with vN except that these are supremums.
These uN and vN values form sequences in their own right. We can ask
whether these new sequences converge and, if so, what are the limits. Addi-
tionally, it is natural to ask how these limits do, or do not, relate to the limit
of the original sequence when one exists.

Definition 2.7 The limit infimum of a sequence {xn } (denoted lim inf) is

lim inf xn = lim uN = lim inf {xn : n > N } . (2.5)

N →∞ N →∞

Similarly,

lim sup xn = lim vN = lim sup {xn : n > N } . (2.6)

N →∞ N →∞

Explore

Exploration 56. IBLdynamics.com Use the sequence {xn } with

(−1)n
xn = 1 +
n
to answer the following questions.
1. Make a list of the sequence elements x1 through x6 . Plot these points
on a graph with n on the horizontal axis and xn on the vertical axis.
2. Let’s start with the infimums. Compute u1 through u6 where

u1 = inf {xn : n > 1}

u2 = inf {xn : n > 2}
u3 = inf {xn : n > 3}

etc. Remember, an infimum can be a minimum!

26 Sequences

3. Now, let’s turn to the supremums. Compute v1 through v6 where

v1 = sup {xn : n > 1}

v2 = sup {xn : n > 2}
v3 = sup {xn : n > 3}

etc. Remember, a supremum can be a maximum!

4. What are some of the things you notice about these two sequences?
(a) Are they monotone or not?
(b) Are they subsequences of the original sequence?
(c) Do either appear to converge?
(d) If so, to what?
5. The tool on the website for exploration 56 shows the points of the
sequence xn , along with uN and vN . Does this agree with what
you observed in answering the preceding questions? Explain.

Exploration 57. IBLdynamics.com Repeat the activities you did in explo-

ration 56 but with the sequence
 
n 1
xn = (−1) 1 + .
n

Again, there is a tool on the website that you should use to verify your think-
ing.

Exploration 58. IBLdynamics.com There is one final tool on the website

that generates a random sequence {xn } and again shows the sequences {uN }
(green dots) and vN (red dots). What do these dots (i.e., sequences) do as you
scroll through?

Exploration 59. Give an example of a sequence {xn } where either the

sequence {uN } or the sequence {vN } is not a subsequence of {xn }.

Conjecture

Let’s see if we can turn these observations about specific examples into
formal mathematical statements.
Liminfs and Limsups 27

Conjecture 60.

Given any sequence {xn } (convergent or not), the sequence of infimums

{uN } is . Similarly, the sequence of infimums {vN } is
.

Conjecture 61.

If {xn } is a bounded sequence, then the sequence {uN } is

and the sequence {vN } is .

Conjecture 62.

Suppose that the sequence {xn } converges to L. Then the sequence {uN }
and the sequence {vN } .

Conjecture 63.

Suppose that for a sequence {xn } the sequences {uN } and {vN } converge
to the same value L. Then the sequence {xn } .

Conjecture 64. Combine conjectures 62 and 63 into a single theorem.

Apply

Application 65. Find the lim sup and lim inf of the sequence
nπ
an = nsin( 2 ) .

Application 66. Find the lim sup and lim inf of the sequence
 nπ 
an = sin .
3

Application 67. Find the lim sup and lim inf of the sequence
 1
2 n+1 if n is even
an = .
1 if n is odd
28 Sequences

Prove

The examples that you have worked through in this section have led you
to discover the important relationships between the lim sup and lim inf of a
sequence and how these quantities relate to the limit of the given sequence.
In particular, you conjectured the following theorem.

Theorem 2.3 A sequence {xn } on R converges to a finite value L if and only

if
lim inf xn = lim sup xn = L.

We will approach the proof of this in several steps.

Lemma 2.1 Given any sequence {xn }, the sequence of infimums {un } is
monotone increasing. Similarly, the sequence of supremums {vn } is monotone
decreasing.

Proof 68. Prove Lemma 2.1.

Lemma 2.2 If the sequence {xn } is bounded, then the sequence of infimums
{un } converges. Similarly, the sequence of supremums {vn } converges.

Proof 69. Prove Lemma 2.2.

The following sequence of proofs will lead you through a proof of Theo-
rem 2.3.
Proof 70. Let’s begin by assuming that

lim xn = L (2.7)
n→∞

and prove that lim inf xn = lim sup xn = L.

1. Explain why we can apply Lemma 2.2 to conclude that both
lim inf xn and lim sup xn exist.
2. Let U = lim inf xn and V = lim sup xn . Explain why U ≤ V .
3. Assume that U 6= V and use equation 2.7 to reach a contradiction
and thus prove this part of the theorem.
Now let’s prove the other direction of Theorem 2.3. For the next two proofs,
assume that
lim inf xn = lim sup xn = L.
Cauchy Sequences 29

Proof 71.
1. Prove that for every ε > 0 there exists N0 such that

sup {xn : n > N0 } < L + ε.

2. Explain why this implies that xn < L + ε for all n > N0 .

Proof 72.
1. Prove that for every ε > 0 there exists N1 such that

inf {xn : n > N1 } > L − ε

and explain why this implies that xn > L − ε for all n > N1 .

Proof 73. How should you choose N so that both proof 71 and 72 are
true? Use this to explain why |L − xn | < ε for all n > N . Finish the proof of
Theorem 2.3.

2.8 Cauchy Sequences

Up to this point, the only tools we have to determine whether a sequence con-
verges or not is to either compute its limit or show it is a bounded monotone
sequence. However, for many sequences it is difficult to compute the actual
limit of the sequence directly, even though we may suspect that it converges
to something. Additionally, there are convergent sequences that are not mono-
tone such as the sequence
(−1)n
xn = .
n
It would be helpful to have another tool that allows us to determine whether
a sequence converges or not, but does not actually force us to compute the
limit of the given sequence. This idea is not new to you. In calculus, you used
a variety of tools to determine whether an infinite series converges or not
(remember the ratio test). None of these tools actually computed the value
of the infinite series. Additional work was needed to complete this task when
that was even possible.
You have probably already noticed that if {xn } is a convergent sequence
then as you compute terms with large n values, the distance between terms
gets really small. This idea is summarized in the following definition.

Definition 2.8 A sequence {xn } is a Cauchy sequence if for every ε > 0

there exists N > 0 such that if m, n > N , then |xm − xn | < ε.
30 Sequences

Explore

Exploration 74. Discuss the similarities and differences of definition 2.8 to

that of convergence given in definition 2.2.
Explorations 75 through 78 illustrate a common mistake when working
with Cauchy sequences. It seems reasonable to assume that if the distance
between consecutive terms in the sequence goes to 0, then the sequence is a
Cauchy sequence. This is not the case as the following set of exercises demon-
strates.
√ √
Exploration 75. Show that the sequence n + 1 − n converges to 0.
√
Exploration 76. Show that sequence where xn = n diverges.

Exploration 77. Show that for the sequence in exploration 76,

lim |xn+1 − xn | = 0.
n→∞

Exploration 78. Show that sequence in exploration 76 is not a Cauchy

sequence by choosing ε = 1 and showing that for all N > 0, if N < m = k 2
and N < n = (k + 1)2 then

|xm − xn | = ε.

Exploration 79. What phrase in definition 2.8 allows us to choose a specific

value of ε to show that the sequence {xn } is not a Cauchy sequence?

Conjecture

Conjecture 80. Suppose that {xn } is a convergent sequence. Do you think

that it must also be a Cauchy sequence? Why or why not?

Conjecture 81. Suppose that {xn } is a Cauchy sequence on R. Do you think

that it must also converge to a number in R? Why or why not?

Conjecture 82. Suppose that {xn } is a Cauchy sequence on the open interval
(0, 1). Do you think that it must also converge to a number in (0, 1)? Why or
why not?
Cauchy Sequences 31

Conjecture 83. Suppose that {xn } is a Cauchy sequence on the closed

interval [0, 1]. Do you think that it must also converge to a number in [0, 1]?
Why or why not?

Conjecture 84. Suppose that {xn } is a Cauchy sequence where xn a rational

number for all n. Do you think that it must also converge to a rational number?
Why or why not?

Apply

P∞ 1
Application 85. Prove that the series n=1 n2 converges by showing that
the sequence of partial sums is Cauchy.

Application 86. Prove that, if {an } is Cauchy, then the sequence {a2n } is
also Cauchy. Show the converse is not true.

Application 87. Suppose {an } is a Cauchy sequence and that there exists
a subsequence of {an } that converges to L. Prove that {an } converges to L
as well.

Prove

Before proceeding with a proof, a discussion of the Conjecture questions

above is in order. They point towards an important analysis concept called
completeness (that is different from the Completeness Axiom). Ideally, one
wants every Cauchy sequence on some metric space X to converge to an
element of X (a metric space is basically just something where there is a well-
defined notion of distance). Conjectures 82 and 84 demonstrate that this is
not necessarily true. It is true for the real numbers and the closed interval
[0, 1], but not for the open interval (0, 1) or the rational numbers. We say
that a metric space X is complete if every Cauchy sequence in X converges
to a point in X. We won’t pursue the concept of completeness any further
except to note that the real line R is complete and any closed interval [a, b] is
complete. This gives us the following theorem.
Theorem 2.4 A sequence {xn } converges on R if and only if {xn } is a
Cauchy sequence.
As we have done in the past, we will guide you through the proof of this
theorem.
Proof 88. One direction of this proof is fairly straightforward. Prove that if
{xn } is a convergent sequence on R then {xn } is a Cauchy sequence.
32 Sequences

We can now work on showing that if our sequence {xn } is Cauchy then it
converges.
Proof 89. Prove that if {xn } is a Cauchy sequence on R, then {xn } is
bounded. Hint: What is the contrapositive statement?

Proof 90. Fix ε > 0 and use the fact that {xn } is a Cauchy sequence to
show that xm + ε is an upper bound for xn for all n > N.

Proof 91. Now show that

vN = sup {xn : n > N } ≤ xm + ε

and conclude that vN − ε ≤ xm for all m > N.

Proof 92. Explain why vN − ε ≤ xm for all m > N implies that vN − ε is a

lower bound for {xm : m > N }. Conclude that

vN − ε < inf {xm : m > N } = uN .

Proof 93. Explain each of the following inequalities.

lim sup xn ≤ vN ≤ uN + ε ≤ lim inf xn + ε

Proof 94. Explain why

lim sup xn ≤ lim inf xn

and
lim inf xn ≤ lim sup xn .
Be careful with the subscripts!

Proof 95. Conclude that lim inf xn = lim sup xn and explain why this implies
that the original sequence converges.
3
Fixed Points and Periodic Points

3.1 Fixed Points

The study of almost every dynamical systems begins with determining the
values of the fixed points. Fixed points were introduced in section 1.2 where
you learned that a fixed point x̃ satisfies

f (x̃) = x̃. (3.1)

This equation tells us that fixed points can be found algebraically by solving
the equation f (x) = x for x. We call this special value x̃. This is often a
fairly straightforward process. For example, it is easy to see that the fixed
points of f (x) = x2 are x = 0 and x = 1 since these are the only solutions
to x2 = x. Sometimes, we are unable to compute the exact values of fixed
points. Can you algebraically find the fixed points of cos x? This definition
also tells us something important about the graphical nature of fixed points.
It says that fixed points are the points of intersection of the graphs of y =
f (x) and the diagonal line y = x. You saw this when you were doing the
graphical analysis problems in chapter 1. Figure 3.1 illustrates this with the
two functions discussed above.
Once we know the values of a fixed point, the next thing we need to deter-
mine is its relationship to other orbits of the dynamical system. In particular,
do the orbits of nearby starting values converge to the fixed point, go away
from the fixed point, or exhibit some other behavior less easily categorized?

Definition 3.1 A fixed point x̃ is attracting (or stable) if there exists an

open interval U containing x̃ such that if x0 ∈ U , then

lim f n (x0 ) = x̃.

n→∞

In short, a fixed point is attracting if nearby initial conditions converge to

it under iteration. There are also repelling (or unstable) fixed points and
while the formal definition is a bit more complicated, the basic premise is that

33
34 Fixed Points and Periodic Points
y y

x x

FIGURE 3.1
Fixed points are the intersection points of the graphs of y = f (x) and y = x.
(Left) The function f (x) = x2 has two fixed points. (Right) The function
f (x) = cos x has a single fixed point.

nearby initial conditions move away from the fixed point under iteration. We
will formally define this concept later. Finally, fixed points that are neither
attracting nor repelling are called neutral.

3.1.1 Fixed Points of Linear Systems

We begin the exploration into fixed points by investigating the dynamics of
linear dynamical systems of the form

xn+1 = mxn + b (3.2)

where m and b are fixed parameters.

Explore

Exploration 1. Find the values of the fixed point(s) in terms of the param-
eters m and b. Are there any values that either of these parameters cannot
take?

Exploration 2. IBLdynamics.com Use the tool on the website to explore

the dynamics of linear systems when b = 0. You should pay particular atten-
tion to how the value of m affects the fixed point and/or its stability. Describe
what you see.
Fixed Points 35

Exploration 3. IBLdynamics.com Use the tool on the website to explore

the dynamics of linear systems for other values of b.
1. Describe how the value of m affects the fixed point and/or its stabil-
ity. Is this different from what you observed in the previous exercise?
2. How does b change the fixed point and/or its stability?
3. Does the stability of the fixed point depend in any way on the initial
condition?

Conjecture

Conjecture 4. The work in the previous two problems suggests a theorem

about linear dynamical systems. Fill in the blanks below to complete it.

Consider the linear dynamical system

xn+1 = mxn + b.

If , then there is a unique fixed point at . If

, then the fixed point is attracting. If , then the
fixed point is repelling.

Apply

In this section, you will model and analyze the population of walleye in
Teelo Lake where there is a strict, fixed limit on the total number of fish
caught per year. Let n measure the number of years since 2020 (i.e., when
n = 10 it is 2030) and let wn denote the walleye population at year n.
Application 5. Model the walleye population assuming that there is no
fishing whatsoever. Assume that the rate of change of the walleye population
from year n to year n + 1 is proportional to the walleye population at year n
with proportionality constant r > 0. Show that

wn+1 = (1 + r)wn .

Application 6. What are the conditions on r that guarantee that the walleye
population will not go extinct in Teelo Lake? Describe the dynamics of the
walleye population in this scenario.
36 Fixed Points and Periodic Points

Application 7. Suppose that exactly h fish are caught per year. Modify
equation 5 to model this situation. In this harvesting model, what is the
equilibrium walleye population and for what values of r and h (if any) is it
attracting?

Application 8. Are there situations where the walleye population will go

extinct? Describe them in terms of the model parameters and the initial con-
ditions.

Application 9. Consider a different harvesting model. Instead of assuming

a fixed number of fish caught per year, assume that the number of fish caught
in year n + 1 is proportional to the walleye population of year n. Let h denote
the constant of proportionality. What is the model equation in this case?

Application 10. For the model in application 9, what is the equilibrium

walleye population and for what values of r and h (if any) is it attracting?
Are there situations where the walleye population will go extinct? Describe
them in terms of the model parameters and the initial conditions.

Application 11. Discuss the pros and cons of these two models.

Prove

In this section, you will prove the theorem about fixed points of linear
dynamical systems from conjecture 4.
Theorem 3.1 Consider the linear dynamical system
xn+1 = mxn + b.
If m 6= 1 then there is a unique fixed point at x = b/(1 − m). If |m| < 1 then
the fixed point is attracting. If |m| > 1 then the fixed point is repelling.

Proof 12. Use mathematical induction to prove that the solution to equa-
tion 3.2 with b = 0 and initial condition x0 is xn = mn x0 .

Proof 13. Prove that the fixed point of equation 3.2 when b = 0 is attracting
if |m| < 1 and repelling if |m| > 1.

Proof 14. Assume that m 6= 1 and let

b
yn = xn − .
1−m
Rewrite equation 3.2 in terms of yn and yn+1 .

Proof 15. Use the results of proof 13 to prove that the fixed point of equa-
tion 3.2 is attracting if |m| < 1 and repelling if |m| > 1.
Fixed Points 37

3.1.2 Attracting Fixed Points of Nonlinear Systems

We’ve now completely described the stability of fixed points in linear dynam-
ical systems on R. Although the dynamics of nonlinear systems can be much
more complicated, the insights gained from studying fixed points of linear
systems is valuable. Keep these insights in mind as we turn our attention to
analyzing fixed points of non-linear systems.
Consider the dynamical system

xn+1 = f (xn ) (3.3)

where we assume that f : R → R is a differentiable function with a continuous

derivative.
Your goal in this section is almost identical to what you did in subsec-
tion 3.1.1; find conditions on the function f that determine the stability of a
fixed point.
Almost every exploration in this book from this point on will focus on the
dynamics of the logistic family of functions

fa (x) = ax(1 − x) (3.4)

and the associated dynamical system

xn+1 = fa (xn ) = axn (1 − xn ). (3.5)

This is another population dynamics model. In the Apply exercises below,

you will be guided through the development of this model. We choose to
use this dynamical system to demonstrate almost every concept that follows
because
• the algebra and calculus required for most calculations is straightforward,
• the dynamics can be readily interpreted in a modeling context, and most
importantly,
• the lessons learned by studying it are transferable to almost every other
nonlinear dynamical system on R.

Explore

We begin with a few exercises that highlight the graphical properties of

the logistic family of functions of equation 3.4. Knowing these properties will
be essential when you begin exploring the dynamics.
Exploration 16. Plot the graph of y = fa (x) for a > 0 and answer the
following questions.
38 Fixed Points and Periodic Points

1. What are the x-intercepts of y = fa (x)? Do they depend on the

value of a?
2. What is the critical point of y = fa (x)? Does it depend on the value
of a? Is it a maximum or a minimum?
3. What is the y-value of the critical point (this is sometimes called a
critical value)? Does it depend on the value of a?
With those fundamentals in hand, we can begin exploring the dynamics of
equation 3.5.
Exploration 17. Show that if a 6= 0 then the fixed points of equation 3.5
are x = 0 and x = (a − 1)/a.

Exploration 18. IBLdynamics.com Use the tool on the website to deter-

mine the values of a for which the fixed point at x = 0 is attracting, repelling,
or neither. Pay attention to the relationship between the graphs of y = fa (x)
and y = x at the fixed point x = 0. What is this relationship when the fixed
point is attracting versus repelling?

Exploration 19. IBLdynamics.com Use the tool on the website to deter-

mine the values of a for which the fixed point at x = (a − 1)/a is attracting,
repelling, or neither. Pay attention to the relationship between the graphs of
y = fa (x) and y = x at the fixed point x = (a − 1)/a. What is this rela-
tionship when the fixed point is attracting versus repelling? This is harder to
see than the last problem! For some parameter values where the fixed point
is attracting, you will see almost the exact same relationship as you did in
exploration 18. But for others, the relationship will be harder to see. Here is
a clue if you struggle to see this: What is the “slope” of fa (x) at the fixed
point when the fixed point is just to the left of the critical point? What is the
“slope” of fa (x) at the fixed point when the fixed point is just to the right of
the critical point?
Just to convince you that what you just observed is not totally dependent
on the choice of function, let’s repeat these exercises with a different family
of functions.
Exploration 20. IBLdynamics.com For the family of functions sa (x) =
a sin(x), use the sine iteration tool on the website to determine the values of a
where the fixed point at x = 0 is attracting, repelling, or neither. What is this
relationship between the two graphs shown on the website near x = 0 when
the fixed point is attracting versus when it is repelling?

Exploration 21. Think about the explorations above and what you observed
about the stability of the fixed points. Reflect on the relationship between the
graph of the dynamical function and the diagonal line. Is there a computable
quantity that determines whether a fixed point is attracting or repelling?
Fixed Points 39

Conjecture

Conjecture 22. Now let’s put this together as a formal mathematical state-
ment.

Assume that f (x̃) = x̃ and that f is differentiable with a continuous

derivative in a neighborhood of x̃. If then x̃ is an attracting
fixed point.

Apply

Now seems a good point to discuss why the logistic model in equation 3.5
is a reasonable population dynamics model. Recall from section 1.4 that a
general form of many population models is

pn+1 = λ(pn )V (pn )pn , (3.6)

where n denotes the generation, pn is the population at generation n, λ is the

reproductive rate function, and V is the viability function that measures the
survival percentage of newborns.
Application 23. Let’s begin by assuming that λ(p) = a. What is the popu-
lation dynamics interpretation of this assumption?

Application 24. Let’s turn to the viability function and suppose that
p
V (p) = 1 − .
K
1. What is V (0) and V (K)? What is the interpretation of these prop-
erties?
2. More generally, what can you say about the value of V (p) for 0 ≤
p ≤ K? Why is this important?
3. Is V (p) increasing or decreasing? What is the interpretation of this
property?
4. The parameter K is a population size. What does it signify in this
model?
5. If we assume that K = 1, then how can we interpret the variable
p?
40 Fixed Points and Periodic Points

Application 25. Show that the assumptions of the previous two exercises
give the logistic model of equation 3.5.
Now let’s put together the things we’ve learned about fixed point stability
to analyze the fixed points of this model.
Application 26. For equation 3.5, use conjecture 22 to determine the values
of a where the fixed point at x = 0 is attracting. Your answer should be an
interval. Interpret this in terms of population dynamics.

Application 27. For equation 3.5, use conjecture 22 to determine the values
of a where the fixed point at x = (a − 1)/a is attracting. Your answer should
be an interval. Interpret this in terms of population dynamics.

Application 28. Let’s tie things together by considering the fixed point at
x = 0 of fa (x) = a sin x from exploration 20. Determine the values of a where
this fixed point is attracting.

Prove

Theorem 3.2 (Attracting Fixed Point Theorem) Assume that f (x̃) =

x̃ and that f is differentiable with a continuous derivative in a neighborhood
of x̃. If |f 0 (x̃)| < 1 then x̃ is an attracting fixed point.

The following exercises lead you to one proof of this theorem. There are
other approaches to proving this fundamental result and we encourage you to
think about other methods of proof.
In this proof we assume, without loss of generality, that the fixed point is
x = 0. If it is not, we would simply change variables in a manner similar to
proof 14 in section 3.1.1.
Let’s begin by looking at the case where 0 ≤ f 0 (0) < 1.
Proof 29. Show that there exists ε > 0 such that if 0 < x < ε then
0 ≤ f (x) < x.

Proof 30. Show that if 0 < x0 < ε and xn+1 = f (xn ), then the sequence
{xn } is decreasing and bounded below. Cite the reason why this implies that
the sequence converges.

Proof 31. Show that sequence {xn } converges to 0 by

1. proving that it cannot converge to a value of x strictly less than 0,
and then
2. assuming that it converges to a value of x strictly greater than 0
and then reaching a contradiction.
Fixed Points 41

Proof 32. Repeat exercises 29 through 31 modified appropriately for −ε <

x < 0 to complete the proof.
With this done, the case where −1 < f 0 (x) < 0 is straightforward.
Proof 33. Let g(x) = −f (x). What can you say about g 0 (0)? Why does the
work above imply that g(x) has an attracting fixed point at x = 0? Why does
this imply that f (x) has an attracting fixed point at x = 0?

3.1.3 Repelling Fixed Points of Nonlinear Systems

In definition 3.1, we defined an attracting fixed point but delayed the definition
of repelling fixed point. So let us define that now.

Definition 3.2 A fixed point x̃ of f (x) is repelling (or unstable) if there

exists an open interval U containing x̃ such that for all x0 ∈ U, other than x̃,
there exists N > 0 such that f N (x0 ) ∈
/ U.

This definition, while a bit complicated sounding, simply says that a fixed
point is repelling if nearby points iterate away from the fixed point. More
precisely, it says that if we start off at any point x0 in some small neighborhood
U of the fixed point, then at some time (the N th iterate) the orbit of x0 leaves
the interval U. Note that it does not say that the orbit leaves and never comes
back. This can happen. We will see instances of this type of behavior later.

Explore

Your goal in this section is almost identical to what you did in section 3.1.2;
to determine conditions that insure that a fixed point is either attracting or
repelling.
Exploration 34. IBLdynamics.com Review the explorations from sec-
tion 3.1.2 and note when each of the fixed points is repelling. On the website
this is broken up into two tools, one for the logistic function, and one for the
sine function.

Conjecture

Conjecture 35. Reflecting on both the explorations and Theorem 3.2 com-
plete the following to conjecture a theorem on repelling fixed points.
42 Fixed Points and Periodic Points

Assume that f (x̃) = x̃ and that f is differentiable with a continuous

derivative in a neighborhood of x̃. If , then x̃ is a repelling
fixed point.

Apply

Application 36. Return to the logistic equation (3.5) and note the interval
of a-values for which the fixed point at x = 0 is attracting. Now determine
the interval of a-values where the fixed point at x = 0 is repelling.

Application 37. Again for the logistic equation, review the interval of values
of a where the fixed point at x = (a − 1)/a is attracting. Determine the
interval(s) of a-values where it is repelling. Restrict your work to the case
where a > 0.

Application 38. IBLdynamics.com Construct a fixed point graph for the

logistic function. This is a graph with the parameter a on the horizontal axis
and the location of the fixed points on the vertical axis. There will be two
curves on this graph. Indicate the stability of each fixed point by coloring
or dashing each curve (for example, you might color attracting fixed points
aquamarine and repelling fixed points ruby).

1. Remember that the logistic equation models population dynamics.

Let’s interpret the fixed point graph that you just created in this
context.
(a) For what values of a will the population go extinct?
(b) For what values of a will the population approach a non-zero
equilibrium state?
(c) Do all initial conditions x0 ∈ [0, 1] approach this equilibrium
population? Why or why not?
(d) How do the equilibrium values change as a function of the pa-
rameter a?
2. Let’s look at some other mathematical features of this fixed point
graph. What is the relationship between the two curves on the fixed
point graph at the value(s) of a where the stability changes from
attracting to repelling or vice versa?
3. Go to the tool on the website for this exercise. Describe what hap-
pens to the graph of f (x) = ax(1 − x) as you increase a? Is this
consistent with your observations in exploration 16?
Fixed Points 43

4. Now, look at the fixed point at x = 0 and manipulate the graph

through the value of a where this point changes from being attract-
ing to being repelling. Describe what happens both in terms of the
graph and in terms of both fixed points. Write a few sentences to
relate this behavior to the fixed point graph that you created above.

Application 39. If you need any additional practice on finding fixed points
and determining their stability, here are few more problems. Find all of the
fixed points of the following functions and determine whether each is attracting
or repelling.
1. f (x) = x2 − 1/8
2. f (x) = x2 − 2
3. f (x) = ex − 1
4. f (x) = x3
√
5. f (x) = x

Prove

Theorem 3.3 (Repelling Fixed Point Theorem) Assume that f (x̃) = x̃

and that f is differentiable with a continuous derivative in a neighborhood of
x̃. If |f 0 (x̃)| > 1 then x̃ is a repelling fixed point.

The following exercises lead you to one proof of this theorem. As in the
attracting fixed point proof, we assume without loss of generality that the
fixed point is x = 0 and begin by considering the case where f 0 (0) > 1.
Proof 40. Show that there exists ε > 0 such that if 0 < x < ε, then x < f (x).

Proof 41. Show that if 0 < xn < ε and xn+1 = f (xn ), then xn+1 > xn .

Proof 42. Show that if 0 < xn < ε, then there exists N > 0 such that
xN > ε.

Proof 43. Repeat proofs 40 through 42, modified appropriately, for −ε <
x < 0 to complete the proof.

Proof 44. Now prove the theorem assuming that f 0 (0) < −1 by modifying
the argument of proof 33.
44 Fixed Points and Periodic Points

3.1.4 Neutral Fixed Points of Nonlinear Systems

We now turn our attention to analyzing so-called neutral fixed points of non-
linear systems.

Definition 3.3 A fixed point x̃ of f (x) is neutral if |f 0 (x̃)| = 1.

As you will begin to see in the explorations below, neutral fixed points are
important when thinking about families of dynamical systems. This is because
when a fixed point is neutral, a small change to the function being iterated
can lead to qualitatively different types of dynamics. This change in behavior
is called a bifurcation and will be the subject of chapter 5.
Our immediate goal is not to explore bifurcations, but simply to categorize
some of the dynamics that occur in the neighborhood of a neutral fixed point.
The dynamical ideas presented here are not a complete list of the properties
of neutral fixed points. Instead, we focus on the most common dynamical
behaviors that can occur. These are easily understood through a combination
of calculus ideas and graphical analysis.

Explore

Exploration 45. In each of the graphs of figure 3.2, x = 0 is the only

fixed point and is a neutral fixed point. What is the value of f 0 (0) in each?
Use graphical analysis to iterate the function and describe orbits with initial
conditions on either side of the fixed point. A version of this figure is available
on the website if you wish to print out a copy for doing graphical analysis.

Exploration 46. Describe the concavity of each of the functions shown

above. What quantity determines concavity?

Exploration 47. Pretend that you could reach into the top two plots ((a)
and (b)) and slide the functions up and down. Describe what happens. Pay
particular attention to fixed points. Something different happens to the lower
two plots ((c) and (d)) when you do this? Describe the difference.

Exploration 48. Pretend that you could reach into the lower two plots and
twist them around x = 0 in one direction or the other (by this I mean you can
push one side up and one side down but the graph is pinned to x = 0). Again,
describe what happens. Pay particular attention to fixed points. Something
different happens to the upper two plots when you do this? Describe the
difference.
Fixed Points 45

y y

y y

x x

FIGURE 3.2
Plots of 4 functions that have neutral fixed points.

It will be helpful if we give some names to the dynamics that you observed
in the explorations above. The easiest is captured in the bottom two figures.
We call a fixed point weakly attracting if it is neutral fixed point and there
exists an interval U containing the fixed point such that initial conditions in
U converge to the fixed point under iteration. Weakly repelling is defined
similarly. There is no standard name for the dynamics in the top two figures.
I’m tempted to combine the words “repelling” and “attracting” and call them
retracting points but that has a different meaning. So let’s use the less clever
term of directionally attracting.

Conjecture
46 Fixed Points and Periodic Points

Keeping in mind the examples above, let’s put together all of the conditions
needed for a fixed point to be directionally attracting or weakly attracting.
Conjecture 49. Let’s start with directionally attracting (graphs (a) and (b)
of figure 3.2).

If
• (the condition that implies that x = 0 is a fixed point)

• (the condition that implies that x = 0 is neutral) and

• (the condition that implies that x = 0 is directional)
then x = 0 is a directionally attracting fixed point.

How does the sign of f 00 (0) determine which side is attracting and which side
is repelling?

Conjecture 50. Now let’s turn to the graphs (c) and (d) in figure 3.2.

If
• (the condition that implies that x = 0 is a fixed point)
• (the condition that implies that x = 0 is neutral) and
• (the condition that implies that x = 0 is attracting)
then x = 0 is a weakly attracting fixed point.
• If then x = 0 is a weakly repelling fixed point.

Apply

When studying the dynamics of a family of functions that depend on a

single parameter, it is often important to find both the variable and parameter
values where that families of functions fa (x) have neutral fixed points. There
is a system of two equations that need to be solved simultaneously to do this:
fa (x) = x
|fa0 (x)| = 1.
Let’s return to our investigation of the logistic equation for exercises 51
and 52.
Fixed Points 47

Application 51. IBLdynamics.com For what value of the parameter a is

the fixed point x = 0 neutral? This value of a is called a bifurcation value.
Describe the stability of this fixed point for values of a on either side of the
bifurcation value. The logistic iteration tool on the website can be used to
help you with this but the calculations should be done by hand as well.

Application 52. IBLdynamics.com For what value of the parameter a

is the non-zero fixed point x = (a − 1)/a neutral? Describe the stability of
this fixed point for values of a on either side of the bifurcation value. Use the
logistic iteration tool on the website to help you with this, but the calculations
should be done by hand as well.

Prove

We won’t be proving either of the following theorems. However, we think

that it is important to list them for future reference. Note that in conjecture 49,
we assumed for simplicity that the fixed point was at x = 0. In what follows,
the fixed point is x = x̃.
Theorem 3.4 If
• f (x̃) = x̃,
• f 0 (x̃) = 1, and
• f 00 (x̃) 6= 0
then x = x̃ is a directionally attracting fixed point.
Theorem 3.5 If
• f (x̃) = x̃,
• f 0 (x̃) = 1,
• f 00 (x̃) = 0, and
• f 000 (x̃) < 0
then x = x̃ is a weakly attracting fixed point. If f 000 (x̃) > 0, then the fixed point
is weakly repelling.

Proof 53. Explain why the last condition of Theorem 3.5 (f 000 (x̃) < 0) is
necessary.
We have focused on neutral fixed points where f 0 (x̃) = 1 in the examples
above but neutral fixed points also occur when f 0 (x̃) = −1. This is why we
needed to solve |fa0 (x)| = 1 in the Apply section above.
48 Fixed Points and Periodic Points

3.2 Periodic Points

Periodic points and orbits were introduced in chapter 1. In this section, we
discuss them further. Our primary emphasis will be determining the stability
of a periodic point. Recall that in definition 1.4 we said that a point x0 is a
period p point of a dynamical system xn+1 = f (xn ) if

f p (x0 ) = x0 . (3.7)

If p is the smallest integer for this equality to hold, then x0 is a prime period
p point. (Note that the adjective prime here does not modify the number p
and thus we are not saying that the number p is prime! A function might have
a prime period 4 orbit for example.) Additionally, if x0 is a period p point,
then the associated periodic orbit is

{x0 , x1 , . . . , xp−1 , x0 , . . . }

where f (xk ) = xk+1 for k = 0, 1, . . . , p − 2 and f (xp−1 ) = x0 .

In addition to the stability question, there is a second important question
about periodic orbits that we will begin to explore here. If a given dynamical
system has a period p point, what other periodic points must also exist? The
answer to this question will ultimately be answered in chapter 10, but we will
be able to give a partial answer in this section.

3.2.1 Stability of Periodic Points

Explore

The following questions explore some of the basic properties of periodic orbits.
Exploration 54. Suppose that x0 is a prime period p point. What are all of
its non-prime periods? Explain.

Exploration 55. Suppose that x0 is a period p point (not necessarily prime).

What periods must it also be? What periods could it also be?

Exploration 56. Suppose that you know that a dynamical system f has a
prime period 2 orbit and you are given a graph of y = f 2 (x). How could you
locate those points using the graph? How do you make sure that they are the
prime period 2 points? How does this concept generalize to prime period p
points?
Periodic Points 49

Exploration 57. IBLdynamics.com The tool on the website generates a

pair of stair-step diagrams for the logistic equation with an initial condition
of x0 = .5. Both y = fa (x) and y = fan (x) are shown.
For each of the a-values given below, the logistic equation has an attracting
periodic orbit. Determine its period. How many periodic points are there of
this period? How many periodic orbits are there of this period?

a) a = 1.75 b) a = 3.25 c) a = 3.46 d) a = 3.63 e) a = 3.84

Now let’s turn our attention to determining stability of periodic orbits. If

x0 is a prime period p point of a function f , then the definition of periodicity
implies that x0 is a fixed point of f p (x). This is a really important fact! Keep
it in mind as you complete the exercise below.
Exploration 58. IBLdynamics.com Use the parameter values from the
previous exercise to answer these questions. Writing down the period of each
attracting periodic orbit you found, use the tool on the website to look at
the graph of fan (x) for the correct value of a and period n. What can you
say about the derivative of the right graph at the attracting periodic points?
What about the repelling ones?

Conjecture

Conjecture 59. You know that if a differentiable function f has a fixed point
x0 and |f 0 (x0 )| < 1, then x0 is attracting (and if this value is greater than 1,
it is repelling). Now, suppose that f has a prime period p point x0 . Complete
the following theorem that gives conditions on when a periodic point x0 is
attracting or repelling?

If f is differentiable, f p (x0 ) = x0 , and then x0 is an attract-

ing period p point. Similarly, if , then x0 is a repelling period
p point.

Apply
50 Fixed Points and Periodic Points

Application 60. Let f (x) = x2 − 2. What is f 2 (x) and what degree polyno-
mial is it? What is f 3 (x) and what degree polynomial is it? In general, what
is the degree of f n (x)? What does this tell you about the task of computing
periodic points?

Application 61. The function f (x) = x2 − 1 has both fixed points and
period 2 points. Find them and determine their stability.

Prove

Theorem 3.6 If f is differentiable, f p (x0 ) = x0 , and

0
(f p ) (x0 ) < 1,
then x0 is an attracting period p point. Similarly, if
0
(f p ) (x0 ) > 1,
then x0 is a repelling period p point.
The proof of this is very simple. Simply apply the attracting fixed point
theorem to the function f p . However, we will work through a slight restate-
ment of this theorem.
Proof 62. Suppose that f is differentiable and f (x0 ) = x1 . Show that

0
f 2 (x0 ) = f 0 (x0 ) f 0 (x1 ) .

Proof 63. Suppose that f is differentiable and {x0 , x1 , . . . , xp−1 } is a prime

period p orbit. Show that
0
(f p ) (x0 ) = f 0 (x0 ) f 0 (x1 ) . . . f 0 (xp−1 ) .
Conclude that if the product of the derivatives evaluated on the periodic orbit
is less than 1 in absolute value, then the orbit is attracting.

3.2.2 New Periodic Orbits from Old

Back in chapter 1, we discussed that one of our overarching goals is to com-
pletely describe the dynamics of a given dynamical system. One of the things
we want to include in this description is the types of periodic orbits that do
or do not arise in a given dynamical system. It turns out that for continuous
functions of a real variable, we can deduce the existence of some periodic or-
bits simply by knowing the existence of a periodic orbit of some other period.
We aren’t quite ready to answer this question completely, but the tool we’ve
been playing with and the fixed point theorems from above can be used to
start answering this question.
Periodic Points 51

Explore

Exploration 64. IBLdynamics.com In exploration 57, you determined at-

tracting periodic orbits for the logistic map. This is summarized in the table
below.

a-value period
1.75 1
3.25 2
3.46 4
3.63 6
3.84 3

TABLE 3.1

For each of these parameter values, complete the following using the peri-
odic orbit tool on the website.
1. Determine whether there are prime period n orbits for n = 1, . . . , 10.
As n gets bigger, it may be difficult to see (and count) all intersec-
tions of the functions graph with the diagonal. Do the best you
can. For each parameter, make a table that summarizes what other
prime periodic orbits exist for that particular value.
2. Which of these parameter values gave you the most periodic orbits?
What was the period of the attracting orbit for that parameter
value?

Conjecture

Conjecture 65. Based on the explorations above, do you believe that if f

has a periodic point, then f must also have a fixed point? Why or why not?
Do you believe that if f has a fixed point then it must have a periodic point
of prime period greater than 1? Why or why not?
52 Fixed Points and Periodic Points

Prove

Now we turn to proving some of the things that you observed in the ex-
plorations above. Your main tool for these proofs is Theorem 4.1 and using
the fact that f k (f m (x)) = f k+m (x). You first encountered this fact in explo-
ration 6 of chapter 1.
Proof 66. Prove the following. If f is continuous and f has a prime period 2
orbit, then f has a fixed point. (You’ve already done this! In what problem?)
There are two nice corollaries to the theorem you just proved.
Proof 67. Prove the following. If f is continuous and has a prime period 2n
orbit, then f has a period n orbit. Hint: Let G(x) = f n (x). What is f 2n (x)
in terms of G.

Proof 68. Prove the following. If f is continuous and has a prime period 2n
orbit, then f has periodic orbits of period 2k for all k ≤ n. Hint: How can
you use proof 67?
4
Analysis of Fixed Points

4.1 Fixed Point Existence Theorems

There is a rich mathematical history of fixed point theorems. It should be
clear from an application perspective why this is so: a fixed point corresponds
to an equilibrium state of the system being modeled and a natural question
to ask about any natural phenomenon is “does the system reach equilibrium
and if so, what is it?”
Many of these fixed point theorems apply to a broad array of underlying
state spaces. For example, there are fixed point theorems for functions whose
domain and range is the real numbers; for functions whose domain and range
is a closed ball; and on and on. Probably, the two most celebrated fixed point
theorems are the Brouwer fixed point theorem and the contraction mapping
theorem.
As you have learned in the exercises of chapter 3, there are basically two
fixed point questions to be answered about a given dynamical system. We first
need to determine whether the dynamical system has at least one fixed point.
This is naturally called an existence question. Once we determine that a
fixed point does exist, then we turn to questions of stability. Is it attracting or
repelling, and if it is attracting, what values converge to the fixed point (this
is called the basin of attraction)?
The existence question is generally the easiest to answer. However, it is
often more difficult to determine either the number of fixed points or their
stability. For example, the Brouwer fixed point theorem can be used to deter-
mine the existence of fixed points for a surprisingly broad array of dynamical
systems. However, it says nothing about whether there is one, two, or two hun-
dred fixed points. And if it can’t do that, it certainly can’t tell us anything
about their stability!
In this section, you will explore two different fixed point theorems for
functions on R. The first one is like the Brouwer fixed point theorem in that
it can only be used to determine whether a fixed point exists. The second
one, like the contraction mapping theorem, tells us there is exactly one fixed
point and it is attracting. This additional information does, however, come at
a cost.

53
54 Analysis of Fixed Points

The Brouwer Fixed Point Theorem

The Brouwer fixed point theorem states that every continuous function
from a closed ball of Rn to itself has at least one fixed point. On the real
line, a closed ball is more commonly called a closed interval and this is
what you will be proving below. In R2 , a closed ball is a closed disk. One
aspect of this theorem that makes it powerful is that it that it is true in
Rn and not just R.
The wonderful book Differential Topology [4] provides a very nice proof
of this theorem and demonstrates it application in some very interesting
examples.

Explore

Both of the theorems in this section concern functions that map a closed
interval [a, b] into itself. In other words, the output interval of the function is
a subset of the input interval to the function. Here is a simple, yet surprisingly
useful, way of visualizing such function that begins with drawing a rectangle
using the directions below.

• Sketch a set of axes and include on it the diagonal line y = x. On the x-axis,
mark the two endpoints of the interval [a, b].
• Use the line y = x to mark the points (a, a) and (b, b). Also, mark the values
a and b on the y-axis.
• Use the points (a, a) and (b, b) to draw a rectangle. When you are finished,
you should have picture similar to figure 4.1.

Exploration 1. Suppose that f maps the interval [a, b] into the interval
[a, b]. What is the relationship between the graph of f (x) and the rectangle
that you constructed?

Exploration 2. Sketch a function f (x) that maps the interval [a, b] into
the interval [a, b] that has exactly one fixed point. Sketch a function f (x) that
maps the interval [a, b] into the interval [a, b] that has multiple fixed points.

Exploration 3. Sketch a function f (x) that maps the interval [a, b] into the
interval [a, b] that has no fixed points. What property must the function f lack
to make this possible?
Fixed Point Existence Theorems 55
y

FIGURE 4.1
The line y = x and the rectangle created using the two corner points (a, a)
and (b, b).

Conjecture

Conjecture 4. Based on these exercises, what is the missing hypothesis and

conclusion in the following theorem?

If f ([a, b]) ⊂ [a, b] and f is , then there exists a point c ∈ [a, b]

such that .

Apply

The following exercises explore some of the hypotheses and conclusions of

conjecture 4 (which is given in Theorem 4.1 below). You are given functions,
asked to determine whether the theorem does or does not apply, and then to
determine whether there is or is not a fixed point. Think carefully about each
problem because together they illustrate the strengths and limitations of this
theorem.
56 Analysis of Fixed Points

Application 5. Verify that Theorem 4.1 applies to

1 + sin (kx)
fk (x) =
2
on the interval [0, 1] for each k = 1, 2, 3, . . . . The theorem implies that each fk
has a fixed point in [0, 1]. Does each have a unique fixed point in the interval
[0, 1]? Plot several of these and discuss what you see and how that does or
does not relate to the theorem.

Application 6. Explain why Theorem 4.1 does not apply to

gk (x) = 1 + sin (kx)

on the interval [0, 1] for each k = 1, 2, 3, . . . ? Does this mean that there is no
fixed point in the interval [0, 1]? Plot several of these and discuss what you
see and how that does or does not relate to the theorem.

Application 7. Explain why Theorem 4.1 does not apply to


x + .5 if 0 ≤ x < .5
h(x) = .
x − .4 if .5 ≤ x ≤ 1

Does this mean that there is no fixed point in the interval [0, 1]? Plot this
function and discuss what you see and how that does or does not relate to the
theorem.

Application 8. Explain why Theorem 4.1 does not apply to


.25x + .25 if 0 ≤ x < .75
q(x) = .
x − .75 if .75 ≤ x ≤ 1

Does this mean that there is no fixed point in the interval [0, 1]? Plot this
function and discuss what you see and how that does or does not relate to the
theorem.

Prove

Theorem 4.1 If f ([a, b]) ⊂ [a, b] and f is continuous, then there exists a
point c ∈ [a, b] such that f (c) = c.

After drawing the pictures, the proof of this statement probably seems so
obvious that it hardly seems worth doing. However, there is a bit of work to
do.
Proof 9. Find your old calculus book and look up the Intermediate Value
Theorem. Use it to prove this theorem.
Fixed Point Existence Theorems 57

Explore

Theorem 4.1 is an existence theorem in that it provides a condition that

guarantees that at least one fixed point exists in a given interval. It does not
tell us how many nor does it tell us anything about the stability of these fixed
points. In the next set of explorations, we will try and take this idea a bit
further in an attempt to address these limitations.
Exploration 10. Make additional copies of figure 4.1 (two or three will
suffice). Sketch several functions that map the interval [a, b] into the interval
[a, b]. They should be increasing and “less steep” than the line y = x.
1. How many fixed points do these functions have?
2. Use graphical analysis to determine the stability of the fixed points
in this case. What initial conditions (if any) in the interval [a, b]
converge to the fixed points?
3. What is the formal mathematical way of saying that a function f
is everywhere increasing and “less steep” than the line y = x?

Exploration 11. How many fixed points must a decreasing function that
maps the interval [a, b] into the interval [a, b] have? Repeat the last two items
of exploration 10 in this case.

Conjecture

Conjecture 12. Thinking about the pictures you drew above, complete the
following statement.

If f ([a, b]) ⊂ [a, b] and , then f has a fixed point

in [a, b] that is .

Apply

Application 13. Suppose that a point x = a is a prime period 2 point of f

with f (a) = b, f (b) = a and a < b. Prove that if f is continuous and has a
prime period 2 orbit, then f has a fixed point as well.
58 Analysis of Fixed Points

Prove

Below is a slightly more formal version of the theorem of conjecture 12.

Theorem 4.2 If f is differentiable with a continuous derivative, f ([a, b]) ⊂

[a, b], and |f 0 (x)| < 1 for all x ∈ [a, b], then f has a unique fixed point x̃ ∈ [a, b].
Moreover, if x0 ∈ [a, b], then f n (x0 ) → x̃ as n → ∞.

Below are the exercises that lead you to the proof of this theorem. But
before completing those steps, it is instructive to think about a “non-proof”
that seems easy, but is unfortunately incorrect.
Proof 14. The following “proof” is wrong for at least two reasons! What
are they and why?
Since f ([a, b]) ⊂ [a, b], we know by the Theorem 4.1 that f has a fixed
point x̃ in the interval [a, b]. By assumption, |f 0 (x̃)| < 1. This implies that
the fixed point is attracting. Thus, if x0 ∈ [a, b], then

lim f n (x0 ) = x̃.

n→∞

Now, let’s do it right in proofs 15 through 18.

Proof 15. Adapt the proof of Theorem 4.1 and use the given derivative
restriction to prove that f has a unique fixed point x̃ ∈ [a, b].

Proof 16. Look up the Mean Value Theorem in your calculus book. Explain
why that theorem and the derivative condition given here implies that there
exists λ ∈ (0, 1) such that

|f (x) − f (y)| ≤ λ|x − y|

for all x and y in the interval [a, b].

Proof 17. Now fix an initial condition x0 ∈ [a, b] that is not the fixed point.
Show that the sequence of iterates {f n (x0 )} is a Cauchy sequence (this requires
using the Mean Value Theorem inequality from proof 16 and the geometric
sum formula). Explain why this sequence must converge to a point x0 ∈ [a, b].

Proof 18. Complete the proof by using the continuity of f to show that x0
is a fixed point of f.
The Inverse and Implicit Function Theorems 59

The Contraction Mapping Theorem

Theorem 4.2 is a special case of the contraction mapping theorem;
a fixed point theorem that applies to functions not just on R but on
any complete metric space. One application of the contraction mapping
theorem that you may have encountered elsewhere is in the proof of the
existence and uniqueness theorem in a differential equations course.
A metric space is a topological space where there is a way of measuring
the distance between any two points in the space. Let’s call this distance
function (or metric) d so that d(x, y) is the distance between x and y. As
you read in chapter 2, a metric space is complete if every Cauchy sequence
in the space converges to a point in the space.
A function f is a contraction if there exists a λ ∈ (0, 1) such that

d(f (x), f (y)) ≤ λd(x, y)

for all points x and y in our metric space. The contraction mapping the-
orem says that if f is a contraction on a complete metric space, then f
has a unique fixed point. The proof of this much more general theorem is
almost identical to the proof that you completed in the exercises above.

4.2 The Inverse and Implicit Function Theorems

The implicit function theorem is the key tool in understanding basic bifurca-
tion theory, the topic of chapter 5. In this section, you will complete a sequence
of explorations aimed at understanding this theorem. We will not prove this
theorem here, but note that it is often covered in an advanced calculus or
multivariable analysis course. Because you encountered it previously in both
calculus and linear algebra, we begin with an exploration of the an equivalent,
but more familiar theorem: the inverse function theorem.

4.2.1 The Inverse Function Theorems

The inverse function theorem is presented in calculus about the time you learn
the derivative formulas for ln(x), arctan(x), and other inverse functions. You
may not realize it, but you also encountered this theorem in linear algebra
(although it is usually not named this). The goal of this section is two-fold.
First, we hope to re-familiarize you with invertibility and think more carefully
about this concept. Second, we want to connect the calculus and linear algebra
versions of this theorem. This should set the stage for understanding the
implicit function theorem. So, let’s start with the calculus version.
60 Analysis of Fixed Points

Explore

Exploration 19. When is a function f : R → R invertible?

Exploration 20. Describe some of the relationships between the original

function f and its inverse f −1 .
1. What happens when you compose f and f −1 in either order?
2. What are their domains and ranges and how are they related to
each other?
3. How are their graphs related to each other?

Exploration 21. Let’s look at a concrete example. Suppose that f (x) =

5x + 4. What is f −1 (y)? Describe a general process for computing f −1 (y)
given a function f (x). Is this process always “doable”?

Exploration 22. Now, graph the function f (x) = x2 . Is it invertible given

your answer to exploration 19?
2
√ 23. What do we mean when we write “if f (x) = x then
Exploration
f −1 (x) = x”? What unsaid assumption did we make about the domain of
f?

Exploration 24. If f (x) is differentiable and invertible, what is a formula

for (f −1 )0 (y)? If you don’t remember this from calculus, use the chain rule to
differentiate
f f −1 (y) = y

and solve for (f −1 )0 (y). Remember that since f (x) = y we have f −1 (y) = x.
Now, let’s switch gears and think about the idea of invertibility from linear
algebra.
Exploration 25. If A is an n×n matrix, when is A invertible? There are quite
a few equivalent properties so you should list as many as you can remember.

Exploration 26. If L(x) = Ax where x ∈ Rn , when is the function L

invertible? How does this differ, if at all, from your answer above?
Finally, let’s try and relate these two ideas with an example. Consider the
function F : R2 → R2 defined by
F (x, y) = (x2 + xy + y 2 , 3x − y).

Exploration 27. The derivative DF of F is a 2 × 2 matrix of partial deriva-

tives with the first row being the partial derivatives of the first function and
the second row being the partial derivatives of the second function. Compute
DF.
The Inverse and Implicit Function Theorems 61

Exploration 28. Compute DF (1, 1) and DF (0, 0). One of these two matrices
is invertible and one is not. Identify which is which.

Exploration 29. What do you think these calculations imply about the
invertibility of the original function F ?

Prove

Unlike other Prove sections, you will not be lead through a proof of this
theorem. Instead, our goal is to simply present to you the theorem and to
take some time to explain it and relate it to the work you did in the Explore
section. We’ll begin with the version of the inverse function theorem that you
learned in calculus that applies to functions of 1 variable.

Theorem 4.3 (Inverse Function Theorem (single variable)) Suppose

that f : R → R is differentiable, f (x0 ) = y0 , and f 0 (x0 ) 6= 0. Then there exists
open intervals U 3 x0 and V 3 y0 and a differentiable function f −1 : V → U
such that
• f −1 (y0 ) = x0 ,
• f −1 (f (x)) = x for all x ∈ U and f (f −1 (y)) = y for all y ∈ V , and

0 1
• f −1 (y0 ) = .
f 0 (x0 )
This version of the inverse function theorem says that near a non-critical
point (sometimes called a regular point) there is an inverse function to f. In
other words, there is a unique solution to y = f (x) in that neighborhood.
It does not say that you can compute the inverse function. We call this an
existence result because it says such a thing exists, but gives us no method of
constructing it. Nor does it say that this solution works everywhere; only in
some possibly small interval about the given point. We call this a local result
simply because it applies only near a given regular point. This is why this
theorem and those like it, are called local existence theorems.
Theorem 4.3 generalizes to functions F : Rn → Rn . In this version, we
explicitly see the connection between the invertibility of the function F and
the invertibility of the derivative matrix DF .

Theorem 4.4 (Inverse Function Theorem (multivariable)) Suppose

that F : Rn → Rn is differentiable, F (x0 ) = y0 , and DF (x0 ) is an invertible
matrix. Then there exists open sets U 3 x0 and V 3 y0 and a differentiable
function F −1 : V → U such that
• F −1 (y0 ) = x0 ,
62 Analysis of Fixed Points

• F −1 (F (x)) = x for all x ∈ U and F (F −1 (y)) = y for all y ∈ V , and

−1
• DF −1 (y0 ) = (DF (x0 )) .

Like the first version of this theorem, this is a local existence theorem. It
says that if the Jacobian matrix DF of F is invertible at the point x0 , then
the function F is locally invertible. In exploration 25, you were asked to list a
number of conditions that are equivalent to a matrix being invertible. One of
those conditions is that the determinant should be non-zero. So, this theorem
could have read “if det(DF (x0 )) 6= 0 then . . . ”.
This is helpful in understanding why the second theorem is a generalization
of the first. If n = 1, then DF is a 1 × 1 matrix, which is simply a number, and
the determinant of a number is just that number. We know that numbers are
invertible (i.e., have a reciprocal) as long as they are not equal to 0, which is
exactly the condition given here.

4.2.2 The Implicit Function Theorem

Explore

This theorem is probably completely new to you although the underlying

algebra is something that you’ve known for a while. The goal of these exercises
is to help you see the connection between the algebra that you know well and
this new theorem, which might sound quite intimidating.
For the following exercises, consider the function G : R2 → R defined by

G(x, y) = x2 + y 2 − 1.

Exploration 30. What is {(x, y) | G(x, y) = 0}? Describe this set both in
words and in a picture.

Exploration 31. Can you solve G(x, y) = 0 for y?

Exploration 32. Can you use this solution to define a function? How?
√ √
Exploration 33.√Does your
√ answer work for the point ( 2/2, 2/2)? What
about the point ( 2/2, − 2/2)?

Exploration 34. Use the function G to compute dy/dx using implicit differ-
entiation. At what values of (x, y) is dy/dx undefined? Identify these points
on your picture from exploration 30.
The Inverse and Implicit Function Theorems 63

Exploration 35. Can you find a single solution to G(x, y) = 0 that works
for all (x, y) in a neighborhood of the point (1, 0)? If so, what is it and what
is the neighborhood? If not, what went wrong?

Exploration 36. Can you find a single solution √ to √G(x, y) = 0 that works
for all (x, y) in a neighborhood of the point ( 2/2, 2/2)? If so, what is it
and what is the neighborhood? If not, what went wrong?

Exploration 37. Compute the partial derivatives of G and evaluate them

at both of the points above. What happened in the case where you could not
solve for y?

Exploration 38. Name the function that you found in problem 35 or 36

above γ(x). What is G(x, γ(x))?

Prove

The inverse function theorem (4.3) tells us when we can solve y = F (x)
for x. The implicit function theorem is very similar. It tells us when we can
solve G(λ, x) = 0 for x. We will usually be interested in cases where λ is a
real number, but the theorem is the same if λ is a list of real numbers. Thus,
we will state the theorem in that setting.

Theorem 4.5 (Implicit Function Theorem) Suppose that G : Rk × R →

R; (λ, x) 7→ G(λ, x), is a C 1 function satisfying
• G(λ0 , x0 ) = 0, and
∂G
• (λ0 , x0 ) 6= 0.
∂x
Then there exists δ > 0 and a C 1 function

ψ : λ ∈ Rk | |λ − λ0 | < δ → R


such that ψ(λ0 ) = x0 and G(λ, ψ(λ)) = 0 for all λ with |λ − λ0 | < δ.

Again, note that the implicit function theorem is a local existence theorem.
It does not tell us how to solve for x, simply that a solution does exist in a
neighborhood of a given regular point. The function ψ of the theorem is that
solution.
There are a few more things to note about the implicit function theorem.
First, this is actually a special case of the theorem. There is a version of it
that allows x to be of dimension greater than one similar to the second version
of the inverse function theorem. In this more general version, the ability to
64 Analysis of Fixed Points

solve for x again depends on a linear algebra condition of a matrix of partial

derivatives. We will not be needing this version here. Second, the implicit
function theorem does give us a formula for the derivative of the function
ψ. We have chosen to omit it here because we will not need it to prove the
upcoming bifurcation theorems.
Finally, and maybe this is not a surprise, the inverse and implicit function
theorems are equivalent. That means you can prove one of them by assuming
that the other one is true. This is actually not that difficult and I encourage
you to try and prove that the single variable inverse function theorem (4.3) is
equivalent to the implicit function theorem (4.5) with λ ∈ R.

4.3 Hyperbolic Periodic Points

In dynamical systems, we are often interested in describing more than the dy-
namics of a single system. Often we want to describe the range of dynamical
behaviors that occurs in a family of equations that depend on parameters. A
typical question to ask, from a mathematical modeling perspective, is “do the
dynamics observed for one specific parameter value persist for other nearby
parameter values?” If not, then the model is probably not a very good de-
scription of the observed phenomenon.
Sometimes, this persistence of dynamics with respect to changes to a pa-
rameter is referred to as structural stability. Here is a simple example that
you have already explored. You know that in the logistic family that x = 0 is
an attracting fixed point when 0 < a < 1. Thus for a parameter value in this
range, the logistic equation is structurally stable. A small change to the pa-
rameter does not significantly alter the dynamics. Similarly, when 1 < a < 3,
the system is structurally stable. For a value of a in this range, a small change
to the parameter moves the fixed point slightly, but does not change the overall
dynamics.
The following definition will help us quantify the idea of structural stability.

Definition 4.1 A fixed point x0 of a differentiable dynamical system xn+1 =

f (xn ) is hyperbolic if |f 0 (x0 )| 6= 1. If |f 0 (x0 )| = 1, then the fixed point is
non-hyperbolic. A period p point x0 is hyperbolic if |(f p )0 (x0 )| = 6 1.

Explore

Recall that a fixed point of a function f (x) is simply the intersection of

the graph of y = f (x) and the line y = x. If we have a family of functions
fa , then the fixed points are again the intersections of the graph of y = fa (x)
Hyperbolic Periodic Points 65

and the line y = x but the exact location of these fixed points may change
depending on the value of parameter a. The following exercises explore what
happens to fixed points as a parameter changes. Use the apps on the website
to answer the questions below.
Exploration 39. IBLdynamics.com Before you use any of the website tools,
look at the graph on the site. Use it to estimate the value of the fixed point of
this function and the value of the derivative of the function at the fixed point.
1. According to definition 4.1, is the fixed point of this function hy-
perbolic or non-hyperbolic?
2. As you interact with the graph, does the fixed point move? Do any
new fixed points appear? Is this new fixed point(s) hyperbolic or
non-hyperbolic?
3. Is this system structurally stable?
4. Can you manipulate the graph in such a way that there are multiple
fixed points? Do you have to manipulate the graph a little or a lot?
Let’s repeat this exercise but with a slightly different starting figure.
Exploration 40. IBLdynamics.com Before you use any of the website tools,
look at the graph on the site. Use it to estimate the value of the fixed point of
this function and the value of the derivative of the function at the fixed point.
1. According to definition 4.1, is the fixed point of this function hy-
perbolic or non-hyperbolic?
2. As you interact with the graph, does the fixed point move? Do any
new fixed points appear? Is this new fixed point(s) hyperbolic or
non-hyperbolic?
3. Is this system structurally stable?
4. Can you manipulate the graph in such a way that there are multiple
fixed points? Do you have to manipulate the graph a little or a lot?
The exercises above should have given you a good intuitive understanding
of the relationship between hyperbolic fixed points and structural stability.
Wiggling the graph a little bit (that is changing the parameter slightly) does
not cause a new fixed point to emerge or the existing one to vanish. However,
we need to be more precise than this and that degree of precision is the main
difficulty in discovering the statement of this theorem. One needs to precisely
state the hypotheses and conclusions using formal mathematical language and
notation. The following exercises are designed to walk you through the process
of expressing this concept precisely in a step-by-step fashion. In the final
exercise, you are prompted to put all of this together in a formal statement
of the Hyperbolic Fixed Point Theorem.
66 Analysis of Fixed Points

Exploration 41. Express the statement

when c = c0 the family of functions Fc (x) has a hyperbolic fixed point
x = x0
in formal mathematical terms and notation.

Exploration 42. Now suppose that along with the conditions above, Fc (x)
is continuous and differentiable with respect to both c and x. Informally, what
can you say about fixed points of Fc (x) for c 6= c0 but nearby? What can you
say about the derivative at the fixed point for c 6= c0 but nearby?

Exploration 43. How can we make the statement “for c = 6 c0 but nearby”
formal using mathematical language and notation? (Hint: Remember, abso-
lute value means distance!)

Exploration 44. How can we express the statement “there is a fixed point
that depends on the value c” using mathematical notation?

Conjecture

Conjecture 45. Use your answers from above to write the Hyperbolic
Fixed Point Theorem.

Suppose that x0 is a hyperbolic fixed point of Fc0 (x) (i.e., ).

Then there exists a δ > 0 and a C 1 function γ : (c0 − δ, c0 + δ) → R such
that

• (conclusion about a fixed point), and

• (conclusion about the derivative of the fixed point).

Apply

The logistic function of population growth can easily be modified to in-

corporate terms that take harvesting into account. In the case of proportional
harvesting, the model equation becomes

xn+1 = ax(1 − x) − hx
Hyperbolic Periodic Points 67

and if the harvesting rate is constant, we get a second model

xn+1 = ax(1 − x) − h.

In both models, the parameter h > 0.

Exploration 46. Complete the following for both models.
1. Determine the fixed points. These will depend on both a and h.
2. You know that when h = 0, there is a non-zero, attracting, hyper-
bolic fixed point for 1 < a < 3. What happens to the location of this
fixed point as h increases from 0? What happens to the stability of
this fixed point as h increases from 0?
3. Sketch the regions in the (a, h) plane where each of the fixed points
is attracting. If you take a value on the boundary of one of these re-
gions, is there still a fixed point? Is it hyperbolic or non-hyperbolic?

Exploration 47. Modify the definition of a hyperbolic fixed point to define

a hyperbolic periodic point.

Prove

Theorem 4.6 (Hyperbolic Fixed Point Theorem) Suppose that x0 is a

hyperbolic fixed point of Fc0 (x) (i.e., Fc0 (x0 ) = x0 and |Fc00 (x0 )| 6= 1). Then
there exists a δ > 0 and a C 1 function γ : (c0 − δ, c0 + δ) → R such that
• Fc (γ(c)) = γ(c) for all c ∈ (c0 − δ, c0 + δ) (i.e., γ(c) is a fixed point), and
• |Fc00 (x0 )| =
6 1 (i.e., γ(c) is a hyperbolic fixed point.)

Before proceeding with the proof of this theorem, it will be helpful to

discuss what exactly it says. First, the function γ(c) is what we might call
a “fixed point function.” The input is the parameter c and the output is a
fixed point of Fc (x). Review the first bullet point to see why this is so. This is
again a local existence theorem. It only tells us that this fixed point exists for
parameter values near the original parameter value c0 . That is the role of the
δ in the theorem. In the language of exploration 43, the interval (c0 − δ, c0 + δ)
formally defines what we mean by “for c 6= c0 but nearby.”
Now on with the proof.
Proof 48. Define a new function G(c, x) = Fc (x) − x. What is G(c0 , x0 )?
What is
∂
G(c, x) .
∂x (c0 ,x0 )
68 Analysis of Fixed Points

Proof 49. Use the Implicit Function Theorem to show that you can solve
G(c, x) = 0 for x in terms of c near c = c0 . Call this function γ(c).

Proof 50. Interpret this result in terms of the original family of functions
Fc .
5
Bifurcations

5.1 What is a Bifurcation?

In the previous chapters, you were given some tools to analyze a given dynam-
ical system. You know how to compute fixed points and determine whether
they are attracting, repelling, or neither. You can do the same with periodic
points although the algebra is generally much more tedious. In this section,
our focus changes from considering a single dynamical system to consider-
ing families of dynamical systems that depend on parameters. You’ve been
introduced to two different families that illustrate many of the fundamental
concepts of bifurcation theory. Most of your work has focused on the logistic
family
xn+1 = fa (xn ) = axn (1 − xn ) (5.1)
where the parameter is a. We indicate the function’s dependence on a with
a subscript to the function name. Less time has been spent studying the
quadratic family
xn+1 = Qc (xn ) = x2n + c. (5.2)
Here, c is the parameter and again denoted with a subscript.
When working with these families in the previous chapters, you were usu-
ally given a parameter value and then asked to find periodic points and de-
termine their stability. Now, we will build on that work by asking these same
questions; but now the answer will depend on the parameter. Let me give you
a few example questions for the logistic family (though the same questions
could be asked for the quadratic family or any other family of functions).
• For what values of a is the fixed point x = 0 attracting? How about repelling?

• There is non-zero fixed point as well. How does it depend on the parameter
a? When is it attracting and when is it repelling?
• For what values of a are there period 2 orbits? How many? Attracting or
repelling?

I think you get the basic idea here. We want to compute the values of the
fixed points and periodic points in terms of the parameter. We also want to
determine the stability of these points in terms of the parameter. Ultimately,

69
70 Bifurcations

our goal is to paint a picture (literally) that will describe the dynamics of fa
for every parameter value a. This picture is called a bifurcation diagram and
you will begin to draw one in section 5.2.
But before we do that, we should first think about not how the parameter
affects the dynamics, but instead how the parameter affects the graph of the
function that determines the dynamics.

Explore

Consider the graphs of the family of quadratic functions defined in equa-

tion 5.2 when answering the following questions. There is a tool on the website,
which may be helpful when answering these questions, but you should be able
to answer all of them without using it by applying techniques from algebra
and calculus.
Exploration 1. IBLdynamics.com How does changing the parameter c
affect the graph of y = Qc (x)?

Exploration 2. IBLdynamics.com What is the critical point of Qc (x) as a

function of the parameter c? Is it a max or a min? What is the critical value
(i.e., the y-value of the critical point)?

Exploration 3. IBLdynamics.com What are the x and y intercepts as a

function of c? For what values of c are there x-intercepts? What happens to
the x-intercepts as c increases?

Exploration 4. IBLdynamics.com We know that the intersections of the

graph of y = Qc (x) and the line y = x correspond to fixed points of the
dynamical system. For what values of c are there fixed points? What are
formulas for them in terms of c? Describe what happens to the fixed points
as c decreases. In particular, how does the distance between them change, if
at all?
Let’s repeat these exercises, but this time with the graphs of the logistic
family defined in equation 5.1. As in the last explorations, the tool on the
website may be helpful, but you should be able to answer all of these questions
without using it by applying techniques from algebra and calculus.
Exploration 5. IBLdynamics.com How does changing the parameter a
change the graph of y = fa (x)?

Exploration 6. IBLdynamics.com What is the critical point as a function

of the parameter a? Is it a max or a min? What is the critical value (i.e., the
y-value of the critical point)?
Introduction to Bifurcation Diagrams 71

Exploration 7. IBLdynamics.com What are the x and y intercepts of fa (x)

in terms of a? For what values of a are there x-intercepts? What happens to
the x-intercepts as a increases?

Exploration 8. IBLdynamics.com For what values of a are there fixed

points? What are formulas for them in terms of a? Describe what happens to
them as a increases.

5.2 Introduction to Bifurcation Diagrams

A bifurcation diagram is a plot that summarizes how the dynamics of a system
change as the parameter changes. Ideally, it should tell us about all periodic
points, both attracting and repelling, and any other complicated behavior that
may occur. In practice, this is not possible by hand calculation alone. Later, we
will learn how to numerically generate bifurcation diagrams that completely
describe all of the attracting dynamics (at least for the quadratic and logistic
families and dynamical systems that are similar to those two families). Right
now, the goal is simply to introduce the concept of bifurcation diagrams in
the context of fixed points only.

Explore

The following exercises will lead you through the construction of a “fixed
point only” bifurcation diagram for the logistic family with the parameter
a > 0. You may want to refer to your answers from exercises 26 and 27 in
subsection 3.1.2.
Exploration 9. Get a sheet of paper and two colored pens (say red and
blue). The colors will be used to differentiate between attracting (red) and
repelling (blue) fixed points.
1. On the sheet of paper, draw a set of axes with the horizontal axis
labelled a (the parameter) and the vertical axis labelled x (the fixed
point). Mark a = 0, 1, 2, 3, 4 on the horizontal axis.
2. You’ve already shown that one of the fixed points of the logistic
function is x = 0. You also showed that this fixed point is attracting
when 0 < a < 1 and repelling when a > 1. Plot this curve (i.e.,
x = 0) on your axes using the blue pen for the attracting region
and the green for the repelling.
72 Bifurcations

3. You also showed that the other fixed point is given by x = (a − 1)/a
and that it is attracting when 1 < a < 3 and repelling for a > 3.
Plot this curve on the same set of axes using the coloring scheme
above.

Exploration 10. IBLdynamics.com Verify that you got the “fixed point
only” bifurcation diagram for the logistic family that is on the website.

5.3 The Tangent Bifurcation

Probably, the most common bifurcation is the tangent or saddle-node bifur-
cation. Generally speaking, a tangent bifurcation occurs when an attracting
fixed point and a repelling fixed point merge at some parameter value. Past
this parameter value, there is no longer a fixed point. This description might
seem a bit odd, but the following explorations should give you a good idea of
how and why this happens. It is not as magical as it sounds.

Explore

The name “tangent bifurcation” gives a pretty good hint as to what hap-
pens in this kind of bifurcation. The first exercise illustrates the phenomenon.
Exploration 11. IBLdynamics.com There is a tool on the website to ex-
plore the tangent bifurcation. As you manipulate the graph on the site, pay
particular attention to the appearance of fixed points. Adjust the value of the
parameter c to answer the following questions.
1. Identify the value of c where the graph is tangent to the line y = x.
How many fixed points are there at this value of c? What are their
stabilities?
2. How many fixed points are there prior to this value of c? What are
their stabilities?
3. How many fixed points are there after this value of c? What are
their stabilities?
4. Make a rough sketch of the bifurcation diagram or “fixed point
graph.” Remember that the parameter c is the horizontal axis and
the location of the fixed points is on the vertical axis. When you
did this in exploration 9, you used color to indicate stability.
The Tangent Bifurcation 73

Exploration 12. IBLdynamics.com The two examples for this problem on

the website are not tangent bifurcations. Use them as you did in exploration 11
to answer the following questions for each graph.
1. Identify the value of the parameter c where the graph is tangent to
the line y = x.
2. How many fixed points are there prior to this value of c? How many
fixed points are there after this value of c? What are their stabilities?
3. Draw the bifurcation diagram.
4. Explain what the difference is between these graphs and the graph
of exploration 11. Can you use a single mathematical computation
to explain why these are not tangent bifurcations?

Exploration 13. IBLdynamics.com The example for this problem on the

website is again not a tangent bifurcation. Use it to answer the following
questions:
1. Identify the value of the parameter c where the graph is tangent to
the line y = x.
2. How many fixed points are there prior to this value of c? How many
fixed points are there after this value of c? What are their stabilities?
3. Draw the bifurcation diagram.
4. Explain what the difference is between this graph and the graph of
exploration 11. Can you use a single mathematical computation to
explain why this is not a tangent bifurcation?
In exploration 11, you saw that a tangent bifurcation happens in a family
of functions Fc (x) when, at a bifurcation value c = c0 , there are no fixed
points for values of c less than c0 , one fixed point at c0 , and two fixed points
for values of c greater than c0 . Of course, this could also happen in the other
direction; no fixed points for c > c0 , etc. Exercises 12 and 13 pointed to other
necessary conditions for this bifurcation.
Exploration 14. The first condition is that

Fc0 (x0 ) = x0 .

Explain this equation in plain English.

Exploration 15. The second condition is that at c0 , the graph of y = Fc0 (x)
must be tangent to the line y = x at the fixed point x = x0 . Express this
condition using formal mathematical notation.
74 Bifurcations

Exploration 16. The third condition, illustrated in exploration 12, is that

as the parameter changes through the bifurcation value, the graph of y =
Fc (x) must actually pass through the diagonal. In other words, you can’t
just “kiss” the diagonal. Express this condition using formal mathematical
notation (Hint: It is again a derivative condition).

Exploration 17. Finally, the fourth condition implies that when the graph
passes through the diagonal, exactly two fixed points appear. An example of
this not occurring was exploration 13. Express this condition using formal
mathematical notation (Hint: Reviewing subsection 3.1.4 might be helpful).
Take another look at the bifurcation diagram that you created in explo-
ration 11. The curve you drew is not a function of the parameter c since it
fails the vertical line test. But if we rotate this picture ninety degrees, it is a
function. In other words, this “fixed point graph” can be viewed as function
whose inputs are fixed point values (x) and whose outputs are the correspond-
ing parameter value (c). This is different than what we did in the hyperbolic
fixed point theorem. Let’s call this function p(x) so that at the bifurcation
value we have p(x0 ) = c0 .
Exploration 18. Look at the fixed point graph again. What can you say
about p0 (x0 ) and p00 (x0 )?

Conjecture

Conjecture 19. Use your answers from above to write the tangent bifur-
cation theorem.

Consider the family of dynamical systems

xn+1 = Fc (xn )
where Fc : R → R is a C 2 function in terms of both x and c. Additionally,
assume that there exists a c0 and x0 such that
• Fc0 (x0 ) = x0 ,
• (derivative condition from 15),
• (derivative condition from 16), and
• (derivative condition from 17).
Then there exists intervals I 3 x0 and J 3 c0 and a C 1 function p : I → J
with p(x0 ) = c0 such that (fixed point conclusion) for all
x ∈ I. Moreover, (derivative conclusions about p).
The Tangent Bifurcation 75

Apply

Application 20. The graph of the tangent bifurcation in exploration 11 is

of the quadratic family of functions Qc (x) = x2 + c. Use the hypotheses of
the tangent bifurcation theorem to find the exact value of c where the tangent
bifurcation occurs. Draw the bifurcation diagram up to this point.

Application 21. Consider the family of dynamical system xn+1 = ex + c.

For what value of c is there a tangent bifurcation? Verify all of the hypotheses
of the tangent bifurcation theorem.

Prove

Theorem 5.1 (Tangent Bifurcation Theorem) Consider the family of

dynamical systems
xn+1 = Fc (xn )
where Fc : R → R is a C 2 function in terms of both x and c. Additionally,
assume that there exists a c0 and x0 such that
• Fc0 (x0 ) = x0 ,
∂Fc
• (c0 , x0 ) = 1,
∂x
∂ 2 Fc
• (c0 , x0 ) 6= 0, and
∂x2
∂Fc
• (c0 , x0 ) 6= 0.
∂c
Then there exists intervals I 3 x0 and J 3 c0 and a C 1 function p : I → J
with p(x0 ) = c0 such that
Fp(x) (x) = x
for all x ∈ I. Moreover, p0 (x0 ) = 0 and p00 (x0 ) 6= 0.
We now proceed with the proof of the tangent bifurcation theorem. Let’s
simplify things a little bit by assuming that at the bifurcation value c0 , the
fixed point is x0 = 0. In other words, assume Fc0 (0) = 0. As we did in the
proof of the hyperbolic fixed point theorem, let
G(c, x) = Fc (x) − x.

Proof 22. Show that G(c0 , 0) = 0.

76 Bifurcations
∂
Proof 23. Show that G(c, x) 6= 0.
∂c (c0 ,0)

Proof 24. Use the implicit function theorem to conclude that there exists a
smooth function p(x) such that G(p(x), x) = 0 near x = 0. Interpret this in
terms of the original function Fc .
Note that you used the implicit function theorem here to solve for the
parameter c in terms of the variable x. This is just the opposite of what was
done in the proof of the hyperbolic fixed point theorem!
Proof 25. Differentiate G(p(x), x) = 0 with respect to x and solve for p0 (x).
Use this to show that p0 (0) = 0.

Proof 26. Differentiate the expression for p0 (x) again with respect to x. Use
this to show that p00 (0) 6= 0.

Proof 27. Explain what the graph of c = p(x) represents in the bifurcation
diagram. What do the derivative conditions above tell you about the shape of
this graph?

5.4 The Period Doubling Bifurcation

To this point, we have explored two fundamental results about bifurcations.
First, the hyperbolic fixed point theorem states that if the derivative at a
fixed point is not equal to 1 in absolute value, then a bifurcation cannot occur
at that particular parameter value. Second, the tangent bifurcation theorem
describes what often happens when the derivative at the fixed point is equal
to 1. So, what generally happens when the derivative at a fixed point is equal
to -1?
If you go back and review the proof of the hyperbolic fixed point theorem
(theorem 4.6), you will see that the proof still works if the derivative is equal
to −1. Thus, a fixed point must persist for parameter values nearby. This is
not surprising. The function illustrated in figure 5.1 has f 0 (x0 ) = −1 at the
fixed point x0 . Clearly, if we wiggle the graph around a little, the fixed point
does not suddenly disappear. What does happen, however, is that the fixed
point might change from being attracting to repelling or vice versa. But does
anything else interesting happen when this occurs?
The Period Doubling Bifurcation 77

FIGURE 5.1
A plot of the graphs y = f3 (x) = 3x(1 − x) and the line y = x. At the fixed
point, x0 = 2/3, f30 (x0 ) = −1. If we increase the parameter through a = 3, the
fixed point remains though its stability changes from attracting to repelling.

Explore

Exploration 28. IBLdynamics.com Let’s consider the logistic family of

functions for values of a ∈ [2.5, 3.3]. The tool on the website generates stair-
step diagrams in this parameter range, for initial condition x0 = 1/2.

1. Show by direct calculation that when a = 3 the fixed point is x0 =

2/3.
2. Show by direct calculation that f30 (2/3) = −1.
3. Use the tool on the website to describe what happens to the dy-
namics as a increases from a = 2.5 to a = 3.3. In particular, what
happens to the stability of the fixed point and why? Describe the
dynamics near the fixed point for values of a less than 3 and for
values of a greater than 3.
It is not at all clear from the exploration above where the period 2 orbit
comes from. It seems to just emerge from the fixed point as the parameter
passes through the bifurcation value at a = 3. If we want to understand
something involving a period 2 orbit, it makes sense to look at the graph of
y = fa2 (x). The following two explorations do that to illustrate what happens
in a period-doubling bifurcation.
78 Bifurcations

Exploration 29. IBLdynamics.com The explorations on the website show

two different plots of y = fa2 (x). Use these tools to answer the following
questions.

1. Describe in general terms what happens to the non-zero fixed points

of fa2 (x) as a increases through the bifurcation value of a = 3.
How many fixed points are there when a < 3 and what are their
stabilities? How many are there when a > 3 and what are their
stabilities? The first tool will be most helpful for this exercise.
2. Recall that fixed points of fa2 (x) are period 2 points of fa (x). Are
any of the fixed points of fa2 (x) actually fixed points of fa (x) (i.e.,
not prime period 2 points)? If so, are they attracting or repelling?
Are any of the fixed points of fa2 (x) prime period 2 points of fa (x)?
If so, are they attracting or repelling?
3. In exploration 28, you showed that f3 (2/3) = 2/3 and f30 (2/3) = −1.
What is f300 (2/3)? Look at the figure first to determine f300 (2/3) from
the properties of the graph. Now calculate it exactly. How is this
different from the second derivative condition in the tangent bifur-
cation theorem? Why do you think this is important in explaining
the bifurcation that you see when looking at the graph of y = fa2 (x)?

Exploration 30. IBLdynamics.com The tool on the website for this ex-
ploration looks more closely at the function from exploration 29. Explore how
the fixed point moves and the graph itself looks like it is rotating about the
fixed point. Focus on the rotation. Do you see how the graph of y = fa2 (x)
“rotates” through the line y = x as a passes through the bifurcation value of
a = 3?
1. Explain why this property is important for the creation of three
fixed points of fa2 (x) as a passes through the bifurcation value.
2. What can you say about
∂
fa2 (x)


∂x
for x < 3? For x = 3? For x > 3?
3. What does this exploration imply about the value of

∂2
 
∂ ∂
fa2 (x) = fa2 (x)



∂a ∂x ∂a∂x

evaluated at a = 3, x = 2/3?
The Period Doubling Bifurcation 79

Conjecture

Conjecture 31. Use your answers from above to write the Period Doubling
Bifurcation Theorem.

Consider the family of dynamical systems

xn+1 = Fc (xn )

where Fc : R → R is a C 2 function in terms of both x and c. Additionally,

assume that there exists a c0 and x0 such that
• (fixed point condition),

• (derivative condition), and

• (derivative condition from 30).
Then there exists intervals I 3 x0 and J 3 c0 and a C 1 function p : I → J
with p(x0 ) = c0 such that

• (x is a period 2 point and p(x) is a corresponding parameter

value), but
• (x is not a fixed point)
for all x ∈ I. Moreover, p0 (x0 ) and p00 (x0 ) .

Apply

Application 32. When does a period doubling bifurcation occur in the

logistic family? Verify the hypotheses of Theorem 5.2.

Application 33. When does a period doubling bifurcation occur in the

quadratic family? Verify the hypotheses of Theorem 5.2.

Prove
80 Bifurcations

Theorem 5.2 Consider the family of dynamical systems

xn+1 = Fc (xn )

where Fc : R → R is a C 2 function in terms of both x and c. Additionally,

assume that there exists a c0 and x0 such that
• Fc0 (x0 ) = x0 ,

• Fc00 (x0 ) = −1, and

∂2
Fc2 (x) |(c0 ,x0 ) 6= 0.


•
∂c∂x
Then there exists intervals I 3 x0 and J 3 c0 and a C 1 function p : I → J
with p(x0 ) = c0 such that
2
• Fp(x) (x) = x, but

• Fp(x) (x) 6= x
for all x ∈ I. Moreover, p0 (x0 ) = 0 and p00 (x0 ) 6= 0.

Instead of proving this theorem, let’s spend some time thinking about what
it says and some of the meanings of the hypotheses and the conclusions.
Proof 34. In part 3 of exploration 29 above, you noted from the graphs
that (Fc20 )00 (x0 ) = 0. Use the hypotheses of the Period Doubling Bifurcation
Theorem to prove that this is true in general.

Proof 35. What do you think the purpose of the third bullet point in the
hypothesis of Theorem 5.2 is? In particular, what do you think it says about
how the graph changes with c?
The function p(x) in the conclusion does come from the implicit function
theorem, but not applied directly. That is basically why we are not proving
this theorem here. The last two questions ask you to think about what the
conclusions of this theorem mean.
Proof 36. What do the two bullet points in the conclusion tell us about the
function p(x)? What does p(x) graph in a bifurcation diagram?

Proof 37. What do the derivative conditions on p imply about this graph?
Can you sketch the bifurcation diagram from this?
6
Examples of Global Dynamics

6.1 Local Dynamics vs. Global Dynamics

Your work up to this point has been learning the tools to understand what
is called local dynamics. For example, given a dynamical system, you can
compute the fixed points and describe what happens near those fixed points.
We don’t have tools to define exactly what we might mean by “near” because
the implicit function theorem doesn’t give us a way to do that.
You’ve also spent some time learning about bifurcations in families of func-
tions. And while this work is about families of dynamical systems as opposed
to a single dynamical system, the work is still local. A tangent bifurcation,
for example, tells us about the emergence of fixed points in pairs, but gives us
no information about these dynamical systems away from these points. Nor
does it tell us about all the bifurcations that might occur in the given family
of dynamical systems.
To fully describe the global dynamics of a dynamical system, we would like
to completely describe what happens to every orbit of every dynamical system
in a family of dynamical systems. This would be truly global information be-
cause we would have a complete description of all possible dynamic behaviors.
Not surprisingly, it is usually not possible to attain this level of detail.
So what seems reasonable to do?

1. Identify the points whose orbits stay bounded (i.e., don’t go off to
infinity).
2. Determine what periodic orbits do and do not exist. If possible,
describe their stability.
3. Develop a method for describing the kinds of non-periodic orbits
that occur.
In short, we would like to catalog all of the different dynamical behaviors
that occur, even though we may not be able to pinpoint the exact orbits that
exhibit these behaviors.
In this section, we will introduce an incredibly helpful tool for describing
and cataloging the global dynamics of some dynamical systems that exhibit
dynamics that initially seem beyond description. This tool boils down to de-

81
82 Examples of Global Dynamics

scribing orbits with infinite sequences of zero’s and one’s. And these sequences
are pretty straightforward to understand.
But before we do that, let’s look back on one of our earlier examples and
paint a global portrait of its dynamics to get a feel for our goal.

Explore

All of the problems in this section are about the logistic family of dynamical
systems
xn+1 = fa (xn ) = axn (1 − xn ).
Our goal here is to describe the fate of every possible initial condition in
R, but we don’t limit our analysis to the interval [0, 1] as we have done for
much of the earlier work. We are not concerned with the population dynamics
interpretation of this dynamical system. We are simply thinking about it as a
mathematical object, and thus are not limited to only positive values of x.
Application 1. IBLdynamics.com Start with the parameter interval 0 <
a < 1. Create a graph of the logistic function y = fa (x) for a value of a between
0 and 1. You can do this using the tool on the website, using graphing software
of your choice, or by hand. Be sure to include the line y = x in your figure.
You probably want a plotting domain of −1 < x < 2 or slightly bigger. Use
this graph to answer the following questions. Some of them review properties
that you have learned previously. What is different here is that this sequence
of exercises guides you in creating a picture that summarizes global dynamics.
For applications 2 through 8, use the graph from application 1.
Exploration 2. What are the fixed points and what are their stabilities?

Exploration 3. Describe the orbits of all initial conditions that are less than
the smallest of the two fixed points.

Exploration 4. Describe the orbits of all initial conditions that are between
the two fixed points.

Exploration 5. On your graph, locate the two points that are eventually
fixed points. There is one that lands on each of the two fixed points.

Exploration 6. Describe the orbits of all initial conditions that are between
the fixed point x = 0 and the eventually fixed point x = 1.

Exploration 7. Describe the orbits of all initial conditions that are between
the eventually fixed point x = 1 and the other eventually fixed point.
The Logistic Map with a = 4 (Part 1) 83

Exploration 8. Describe the orbits of all initial conditions that are greater
than the largest of the two eventually fixed points.

Application 9. IBLdynamics.com Repeat applications 2 through 8 with

the parameter interval 1 < a < 3. The tool on the website might be helpful, but
you may need a more complete graph to help with initial conditions outside
the interval shown there. You can zoom out on the graph on the website or
use a different graphing program.

Application 10. You’ve done it! Color a pair of number lines in some way
to summarize the work that resulted from the work you did in applications 1
through 9. Your colors should signify the different types of behaviors that you
observed.

6.2 The Logistic Map with a = 4 (Part 1)

Now, let’s turn to studying the logistic map when a = 4. Ultimately, we will
need to fill in what happens when a is between a = 3 and a = 4, but that will
come in a later chapter.

Explore

When a = 4, the logistic equation is given by

xn+1 = f4 (xn ) = 4xn (1 − xn ). (6.1)

Some of the work that you did in applications 1 through 9 will be relevant
here. However, much of this is quite different, very interesting, and somewhat
counterintuitive. Most importantly, the dynamics that you will discover here
and the techniques you will use to discover them illustrate some fundamental
concepts in dynamical systems.
Understanding the properties of the graph of y = f4 (x) is essential because
these ultimately determine the dynamics. Figure 6.1 shows the graph of y =
f4 (x). As usual, the line y = x is also shown. There is also something new in
this figure: the lines x = 1 and y = 1 are superimposed in gray. We’ll see why
shortly.
Exploration 11. Prove that the maximum value of f4 (x) is y = 1. Prove that
the fixed points are x = 0 and x = 3/4 and that both of them are repelling.
84 Examples of Global Dynamics
y

FIGURE 6.1
The graph of the logistic function with a = 4.

Exploration 12. Show, using graphical analysis, that the orbits of initial
conditions x0 with x0 < 0 or x0 > 1 tend to negative infinity.
Remember that our ultimate goal is to describe the orbits of all initial
conditions in R that occur in equation 6.1. At this point, we are only left with
the initial conditions in the unit interval.
Exploration 13. Consider an initial condition x0 ∈ [0, 1]. Prove that the
orbit of this point remains in the interval [0, 1] for all time. More precisely,
prove that if x0 ∈ [0, 1], then f4n (x0 ) ∈ [0, 1] for all n ≥ 0.

Exploration 14. Use the graph in figure 6.1 to identify points that are
eventually fixed to x = 0. Identify points that are eventually fixed to x = 3/4?

Exploration 15. IBLdynamics.com Print out a copy of this graph (one

is available on the website) and perform graphical analysis for several ini-
tial conditions x0 ∈ [0, 1]. Do you observe any periodic behavior? Is there a
pattern?

Exploration 16. IBLdynamics.com Repeat exploration 15 using the logis-

tic function iteration tool on the website.

Exploration 17. Thinking about the graph of y = f4 (x), sketch the graph
of y = f42 (x) on the interval [0, 1] by hand. Note: The formula of f42 (x) will
not help you very much. Instead, think about how the properties of the graph
of y = f4 (x) determine the graph of y = f42 (x).
The Doubling Map 85

Note: The tool on the website for explorations 18 through 23 will plot the
graphs of y = f4n (x) for n = 1 to 7. It will also perform graphical analysis. Use
it to check your answer to exploration 17 and then to answer explorations 18
through 23.
Exploration 18. IBLdynamics.com How many critical points does y =
f4n (x) have? Is there a general formula for the number of critical points in
terms of n? What are the critical values (i.e., y-values of these critical points)?
Describe the orbits of these points under iteration by f4 .

Exploration 19. IBLdynamics.com What happens to the number of crit-

ical points of y = f4n (x) and the distance between them as n goes to infinity?

Exploration 20. IBLdynamics.com How many fixed points does f4n (x)
have? Is there a general formula for the number of fixed points of f4n (x) in
terms of n? Describe the orbits of these points under iteration by f4 .

Exploration 21. IBLdynamics.com Explain why f4 (x) has periodic points

of all periods. Does f4 (x) have prime periodic points of all periods? Why or
why not?

Exploration 22. IBLdynamics.com What happens to the distance between

fixed points of y = f4n (x) as n goes to infinity? What does this tell you about
the distance between periodic points of f4 ?

Exploration 23. IBLdynamics.com When you set n = 1 using the tool

on the website and do the graphical analysis, can you find any periodic orbits
(other than the obvious fixed points)? Can you find any eventually fixed points
(other than the obvious ones)?

Exploration 24. Looking at explorations 18 through 23, can you explain

the apparent inconsistencies here? You showed that there are infinitely many
eventually fixed points that are arbitrarily close together. You also showed that
there are infinitely many periodic points that are arbitrarily close together.
This seems to imply that periodic and eventually fixed points are almost
everywhere. Yet when you or the computer performs graphical analysis, you
don’t see any of these orbits! What are some reasons why this might be so?

6.3 The Doubling Map

When you think back on the explorations in section 6.2, you should see that
there were two related properties that lead to the infinitude of periodic points
86 Examples of Global Dynamics

(as well eventually fixed points), and to the non-periodic dynamics observed
in the stair-step diagrams. The first of these is that the function f4 maps the
interval [0, 1] onto the interval [0, 1]. This implies that every orbit remains in
the interval [0, 1] for all n. The second property is that the function is “two-
to-one.” Every point in the interval [0, 1] (except 1/2) has exactly 2 values
that map onto it. We call these two points preimages. These two properties
combine to cause f4n (x) to have 2n − 1 critical points with critical values of
either 0 or 1, and for there to be 2n period-n points for all n > 0.
Another property of f4 makes computing these critical points and periodic
points difficult, if not impossible. That is its non-linearity. While we can use
the quadratic formula to find the fixed points, this tool goes out the window
very quickly since f4n (x) is a polynomial of degree 2n .
Clearly we can’t construct a continuous, linear, two-to-one function since
the graphs of linear functions are lines! However, there is a simple piecewise
linear function that has these important properties. This function is called the
doubling map and is defined by

2x if 0 ≤ x < 1/2
D(x) = 2x (mod 1) = . (6.2)
2x − 1 if 1/2 ≤ x ≤ 1
The graph of the doubling map is shown in figure 6.2. Iterating the doubling
map is easy; simply multiply by 2 each time and if the number ever becomes
greater than 1, ignore the integer part (that is what the notation mod 1
indicates in equation 6.2). So, for example
5 5 5 2 4 1 2
7→ 7→ ≡ 7→ ≡ 7→ . . .
12 6 3 3 3 3 3
So, 5/12 is an eventually period 2 point as it lands on the period 2 orbit of
{1/3, 2/3}.

6.3.1 Basic Dynamics of the Doubling Map

Explore

Exploration 25. Describe the orbits of the initial conditions

1/4, 1/3, 1/5, 1/6, 3/16

under iteration by the doubling map D.

Exploration 26. What are the fixed points of D? What are the prime period
2 points? Period 3 points?
The Doubling Map 87
y

FIGURE 6.2
The graph of the doubling map D(x).

Exploration 27. Describe all eventually fixed points of D.

Exploration 28. Now let’s think about the graph of D(x) and its iterates.
1. Carefully plot the graph of y = D2 (x) by hand. How many pieces
are there? What are the points of discontinuity? Identify on your
graph the period 2 points of D and eventually fixed points of D.
2. Repeat for y = D3 (x).
3. Do you see a pattern here? What can you say about the graph of y =
Dn (x). How does this translate to periodic points and eventually
fixed points of D?

Exploration 29. IBLdynamics.com The doubling map tool on the website

generates stair-step diagrams for D. Do you see any of the periodic behavior
that you described above? Why the discrepancy?

Conjecture

Conjecture 30. Complete the following theorem about the number of period
n points of the doubling map D.
88 Examples of Global Dynamics

The doubling map D has periodic points for each n > 0.

Conjecture 31. Complete the following theorem about the distance between
consecutive period n points of the doubling map D.

The distance between consecutive period n points of the doubling map D

is less than or equal to for each n > 0.

Prove

Theorem 6.1 The doubling map D has 2n periodic points for each n > 0.

Proof 32. Prove Theorem 6.1.

Theorem 6.2 The distance between consecutive period n points of the dou-
bling map D is less than or equal to 1/2n−1 for each n > 0.

Proof 33. Prove Theorem 6.2.

6.3.2 The Doubling Map in Binary

The best way to fully understand and appreciate the dynamics of the doubling
map is to represent the numbers in the unit interval in binary. Some of the
material covered here will be review for you, but I encourage you to work
through this carefully. The ideas and techniques presented in this section are
fundamental to this work and will reappear frequently as we move forward.
Recall that the geometric series is
∞
X 1
1 + x + x2 + · · · = xk = , if |x| < 1. (6.3)
1−x
k=0

As we will see shortly, the ability to work with geometric series is extremely
important in understanding the complicated dynamics that we are beginning
to explore. If it has been a while since you worked with geometric series in
calculus, I suggest you take some time to go back and review them. One thing
to note is that it does matter where the summation index (k in this case)
starts and ends. In particular, we will often encounter geometric series that
The Doubling Map 89

start at k = 1 instead of k = 0. By starting at k = 1, the first term of 1 is

omitted from the sum in 6.3 and thus we get
∞
X 1 x
xk = −1= . (6.4)
1−x 1−x
k=1

So, why do we need this? We are so comfortable with decimal expansions

that we rarely think about what an expression like 0.26723 . . . actually means.
So, it makes sense to begin reviewing decimal expansions. A decimal expansion
0.d1 d2 . . . is short hand for the series
∞
X dk
0.d1 d2 · · · = ,
10k
k=1

where each coefficient dk takes an integer value between 0 and 9. Take for ex-
ample the decimal 0.333 . . . . We know this equals 1/3 but why? Well, because
∞ ∞  
X 3 X 1 1/10 1
k
= 3 k
= 3 = . (6.5)
10 10 1 − 1/10 3
k=1 k=1

Exploration 34. A common misconception is that the decimal 0.999 . . . is

“really close to 1.” Show that in fact 0.999 · · · = 1.
A binary expansion is almost identical except that the denominators are
powers of 2, and the numerators bk can be only 0 or 1. More formally, a binary
expansion is
∞
X bk
0.b1 b2 · · · = (6.6)
2k
k=1

with bk = 0 or 1. Let’s find the value of the binary expansion s = .010101 . . .

as an example . We use the same method as we did for the decimal expansion
in equation 6.5:
∞ ∞
1 1 1 X 1 X 1 1/4 1
s = .010101 · · · = 2
+ 4 + 6 + ··· = 2k
= k
= = .
2 2 2 2 4 3/4 3
k=1 k=1

It is interesting and helpful to think about what the binary expansion of

a number s tells us. It gives us directions to find s in the unit interval. If the
first digit is 0, we know that the number is between 0 and 1/2. If the first
digit is 1, then the number is between 1/2 and 1. I find it helpful to think of
0’s as “lefts” and 1’s and “rights.” The second digit tells us if we are on the
left or right side of the interval identified by the first digit. And so on.
Let’s return to the example above. Because the first digit of s is 0, we know
s ∈ [0, 1/2]. The second digit is 1, so s is in the right half of this interval; that
is s ∈ [1/4, 1/2]. The third digit is now 0, so s ∈ [1/4, 3/8] and so on. This is
illustrated in the binary expansion tool on the website.
90 Examples of Global Dynamics

Exploration 35. IBLdynamics.com This tool shows how we can locate

point s through its binary expansion. Each n takes a successive term of the
binary expansion to refine the interval further.

Explore

Let’s now return to the dynamics of the doubling map, but this time
let’s do it not with traditional fractions or decimals, but with their binary
representation.
Exploration 36. Above, we showed that the binary expansion of 1/3 is
.010101 . . . . Write this out as a series and then compute D(1/3) and D2 (1/3).
Does this agree with the work you did in exploration 25?

Exploration 37. Repeat exploration 36 except with the initial condition

s = .011010101 . . . Review the first example at the beginning of this section
where we computed the orbit of 5/12 under iteration by the doubling map.
Are the orbits the same? Use a geometric series calculation to determine the
value of s.

Conjecture

Conjecture 38.

Suppose that s = .b1 b2 b3 . . . in binary. Then D(s) = in

binary?

Conjecture 39. What are the fixed points of the doubling map D in binary?
The period 2 points? The period 3 points?

Dn (s) = s if and only if .

Conjecture 40. Give several examples (in binary) of eventually fixed points
of D. Similarly, give examples of eventually period 2 points of D.
The Logistic Map with a > 4 (Part 1) 91

s is an eventually fixed point of D if and only if .

s is an eventually period n point of D if and only if .

Conjecture 41. Suppose that the binary representation of a point s is

neither repeating nor eventually repeating. What can you say about the orbit
of s under D?

Prove

Proof 42. Suppose that a point s = .b1 b2 b3 . . . in binary. Prove that D(s) =
.b2 b3 . . . in binary.
We say that a binary sequence s = .b1 b2 b3 . . . is a periodic binary se-
quence of period n if it repeats blocks of length n. For example
s = .011011011 . . .
is a period sequence of period 3.
Proof 43. Prove that x is a periodic point of the doubling map D if and
only if the binary expansion of x is a periodic binary sequence.

6.4 The Logistic Map with a > 4 (Part 1)

The logistic map exhibits some important and interesting behavior when a >
4. Remember that our goal is to describe all possible dynamical behavior
regardless of initial condition.

Explore

All of the questions in this section are about the iteration of the logistic
function when a = 4.5 as shown in figure 6.3. As usual, we’ve included the
graph of the line y = x and again have superimposed the lines x = 1 and
y = 1. Printing out at least one copy of figure 6.3 (it is on the website) may
be helpful in answering the questions in this section.
Exploration 44. Show, using graphical analysis, that the orbits of points
with initial conditions less than 0 or greater than 1 go to negative infinity.
92 Examples of Global Dynamics

FIGURE 6.3
The graph of the logistic function with a = 4.5.

Exploration 45. Now, consider the interval [0, 1]. Graphically, identify the
values of x in this interval where f4.5 (x) > 1. What are the orbits of these
points under iteration?

Exploration 46. Identify the two points xL and xR that are eventually fixed
to x = 0 in exactly two iterations? How do these points relate to your answer
from the previous question?

Exploration 47. Graphically, identify the intervals in [0, 1] that get mapped
into the interval (xL , xR ). What are the orbits of these points under iteration?
How many of these intervals are there, and approximately where are they
located?

Exploration 48. IBLdynamics.com Repeat explorations 46 and 47 using

the tool on the website for higher iterates. In addition to the graph of the
n
function, the x-axis is color coded to indicate the value of f4.5 (x). Here are a
few things to think about as you do this.
n
1. How do the intervals relate to properties of the graphs of y = f4.5 (x)
for each n?
2. How many intervals of each color are there for each n? Is there a
general formula in terms of n? Why?
3. Are there points in the interval [0, 1] whose orbits remain in that
interval for all n? If so, describe them. If not, explain why.
4. Does the process demonstrated here remind you of anything else
that you have encountered in mathematics? If so, what?
7
The Tools of Global Dynamics

7.1 How to study Global Dynamics

In chapter 6, you explored dynamical systems that exhibited some surprising
properties. In particular, that work suggested that some dynamical systems
on R

• have infinitely many periodic orbits,

• have prime periodic orbits of all periods,
• have infinity many eventually fixed points (and presumably eventually pe-
riodic points), and
• sometimes exhibit these dynamics on an invariant set having a complicated
topological structure.

All of these things and more are true for the examples that you explored in
chapter 6. Moreover, these properties are universal. They are the fundamental
characteristics of what mathematicians refer to as chaotic dynamical systems.
But before we can prove these facts about the examples of chapter 6, we
need to take a brief detour to introduce two essential concepts: the Cantor set
and the shift map. We begin with the Cantor middle-thirds set that exemplifies
the fractal structure that often accompanies chaotic dynamics. Next, we will
introduce a function called the shift map that captures the essence of the
dynamics that you discovered in chapter 6.
Once we understand these two tools, we can then use them to prove that
the dynamical systems of chapter 6 are in fact chaotic. We begin this process
by using the dynamics to understand the structure of the invariant set and
in many cases, this allows us to prove that this set has the structure of the
Cantor set. We then can use this structure to construct a new function that
puts the dynamics of the given dynamical system into correspondence with the
dynamics of the shift map. In this way, we can prove that a given dynamical
system has the properties listed in the bullet points above. This work will be
done in chapter 8.

93
94 The Tools of Global Dynamics

7.2 The Cantor Set

Hopefully, your answer to part 4 of exploration 48 in chapter 6 is that logistic
map with a > 4 has a property reminiscent of the Cantor Set. You may have
already encountered the Cantor middle-thirds set in an analysis or topology
course. However, we should review some of its remarkable properties here
not just because they are interesting, but because they provide us a tool for
understanding dynamical systems like the ones in chapter 6.
The construction of the classic Cantor middle-thirds set K is an iterative
process that begins with the closed interval [0, 1]. In the first step of the
construction, remove the middle-third open interval (1/3, 2/3). This leaves us
with the set    
1 2
K1 = 0, ∪ ,1 .
3 3
The next step is to remove the open middle-thirds of each of these two inter-
vals. This leaves a set K2 consisting of 4 closed intervals:
       
1 2 3 6 7 8
K2 = 0, ∪ , ∪ , ∪ ,1 .
9 9 9 9 9 9

We imagine continuing in this fashion infinitely many times. The result of this
infinite process is the Cantor middle-thirds set K. Formally, we define
∞
\
K= Kn . (7.1)
n=0

This set has some remarkable properties that you will prove below. Probably,
the most confusing result that you will prove is that in one sense the Cantor
set is “infinitely small,” while in another sense, it is “infinitely large.” The
exercises below will make this more precise and lead you through the required
proofs.

Prove

The first thing we need to do is very basic: we need to show that this
infinite process of removing open middle-thirds intervals leaves something in
the set K.
6 ∅.
Proof 1. Prove that K =
Now that we know that K is non-empty, we aim to do two different things
to get a feel for its size. On the one hand, we want to “count” the number of
elements in K. On the other hand, we want to compute the “length” of K.
The Cantor Set 95

Proof 2. Let’s be sure that we understand the structure of each of the

preliminary sets that are used in constructing K.

1. How many intervals are in K1 ? What is the length of each of them?

What are the endpoints of these intervals?
2. How many intervals are in K2 ? What is the length of each of them?
What are the endpoints of these intervals?
3. How many intervals are in K3 ? What is the length of each of them?
What are the endpoints of these intervals?
4. In general, how many intervals are in Kn ? What is the length of
each of them?

Proof 3. Prove that if x is an endpoint of Kn for some n, then x ∈ K.

A set X is countably infinite (or countable) if the elements in X can be
put in one-to-one correspondence with the natural numbers N. One way of
doing this is to develop a method or algorithm to list the points in X. Once
you have such a list, then you can use it to specify the first point of K, the
second point of K, the hundredth point of K, and so on.
Proof 4. Prove that the set of all endpoints of K is countably infinite by
developing a procedure to list them.

Proof 5. Prove, using a geometric series, that the total length of the intervals
removed to construct K sums to 1. What does this imply about the “length”
of K? Hint: Proof 2 suggests how to set up this series.

Proof 6. Prove that K contains no intervals.

So, now you know that the Cantor set is “really small” in the sense that
its total length equals 0. You started out with an interval of length 1 and
removed a collection of intervals totaling length 1. So in the sense of length
(technically measure), K is “really small.”
Proof 7. Prove that x ∈ K if and only if
∞
X sn
x= n
, where sn = 0 or sn = 2. (7.2)
n=1
3

You might be a little concerned about this last statement since we know that
1 1 0 0
= + 2 + 3 + ··· ∈ K
3 3 3 3
and this is not of the form given in equation 7.2. This is not the problem that
it seems at first.
96 The Tools of Global Dynamics

Proof 8. Show that

∞
1 X 2
= .
3 n=2 3n
We can deal with other problematic endpoints of K in a similar fashion.
You probably noticed the similarity between equation 7.2 in proof 7 and
the binary expansions discussed in chapter 6. They are both based on geo-
metric series, but in equation 7.2, there is a 3n in the denominator instead
of a 2n . Expansions such as these, with sn = 0, 1, or 2, are called ternary
expansions. They can be thought of in the same way we thought of binary
expansions except instead of a left and right, we now have a left (0), middle
(1), and right (2).
It is reasonable to think that the process of creating the Cantor set only
leaves the endpoints of each Kn in K. We also know from proof 4 that the set
of endpoints is countable. So, it seems reasonable to guess that the Cantor set
K is countably infinite. Surprisingly, this is not the case.
Proof 9. Show that the Cantor set K is uncountable by constructing a
function F : K → [0, 1] that is one-to-one and onto. Hint: Use ternary and
binary expansions. F is really simple if you do it this way.

Proof 10. Explain why proof 9 implies that there are points in K that are
not endpoints of intervals removed.

Proof 11. Explain why proof 9 implies that K is “really large” in the sense
of counting. Discuss why it is remarkable that K is both “really large” in one
sense and “really small” (proof 5) in another. Compare the properties of the
Cantor set to other, more common, subsets of R.

7.3 The Shift Map (Part 1)

In section 6.3.2, you learned that the easiest way to describe the global dy-
namics of the doubling map D is to represent numbers in the interval [0, 1] in
binary. Doing this allowed you to easily list all of the periodic points and all
of the eventually periodic points. It turns out that this idea can be general-
ized to the other functions that we studied in chapter 6. In this section, we
generalize the lessons of section 6.3.2 in preparation for applying them to the
other dynamical systems discussed in chapter 6.

7.3.1 The Sequence Space on 2 Symbols

Define the sequence space on two symbols by
Σ2 = {.s0 s1 s2 . . . | si = 0 or 1 for all i} . (7.3)
The Shift Map (Part 1) 97

In other words, Σ2 is a space whose points are all infinite sequences of zeros
and ones. We begin by defining a metric (i.e., a distance function) d on Σ2 .
Let s = .s0 s1 s2 . . . and t = .t0 t1 t2 . . . be any two elements of Σ2 . Define a
metric d : Σ2 → R by
∞
X |sk − tk |
d(s, t) = . (7.4)
2k
k=0

Metrics, like the one defined here, must satisfy 3 properties:

1. d(s, t) = d(t, s). This is called commutativity.

2. d(s, t) ≥ 0 for all s and t with equality if and only if s = t. This is
called positivity.
3. d(s, t) ≤ d(s, u) + d(u, t) for all s, u, and t. This is called the triangle
inequality.

Explore

Let’s begin by computing some distances to get comfortable with Σ2 and

the metric d. Observe that equation 7.4 looks very much like a geometric series,
except the numerator is not constant. Because of that, all of these calculations
boil down to using the geometric series formula 6.3.
Exploration 12. Let s = .0 and t = .1 (the overbar indicates that the
sequence repeats). Show that d(s, t) = 2. Prove that d(s, t) ≤ 2 for all s, t ∈
Σ2 .

Exploration 13. Let s = .01 and t = .1. Compute d(s, t).

Exploration 14. Let s1 = .11110 and t = .1. Compute d(s1 , t). Let s2 =
.111110 and compute d(s2 , t). Which of s1 and s2 is nearer to t?

Exploration 15. Let s1 = .10100 and t1 = .10101. Compute d(s1 , t1 ). Let

s2 = .1010100 and t2 = .1010101 and compute d(s2 , t2 ). Which pair of points
are closer to together?

Exploration 16. In general terms, when do you think that two points s and
t are “close together”?
98 The Tools of Global Dynamics

Conjecture

Use your observations from explorations 23 through 16 to formalize the

notion of “closeness” in Σ2 .
Conjecture 17.

Let s, t ∈ Σ2 and suppose that si = ti for i = 0, 1, . . . , n. Then

d(s, t) ≤ .

Conjecture 18.

1
Let s, t ∈ Σ2 . If d(s, t) < then .
2n

Apply

Application 19. The ideas presented here can be generalized to more than
just infinite sequences of 0s and 1s. Can you define Σ3 , the sequence space on
3 symbols? What is a metric on this space? What about Σn ?

Application 20. Let s = .0, t = .1 ∈ ΣN . Compute d(s, t).

Application 21. Let 0 ≤ k ≤ n − 2 and define s = .k, t = .(k + 1) ∈ ΣN .

Compute d(s, t).

Application 22. Show that if s, t ∈ ΣN then d(s, t) ≤ N .

Exploration 23. Let 0 ≤ k ≤ n − 1. If s1 = .kkkk0 and t = .1 compute

d(s1 , t). If s2 = .kkkkk0 compute d(s2 , t). Which of s1 and s2 is nearer to t?
The Shift Map (Part 1) 99

Prove

Proof 24. Prove that d(s, t), as defined in equation 7.4, is a metric by showing
that it has each of the three listed properties.

Proof 25. Prove the following theorem.

Theorem 7.1 Let s, t ∈ Σ2 and suppose that si = ti for i = 0, 1, . . . , n. Then
1
d(s, t) ≤ .
2n

Proof 26. Prove the following theorem.

1
Theorem 7.2 Let s, t ∈ Σ2 . If d(s, t) < , then si = ti for i = 0, 1, . . . , n.
2n

7.3.2 Dynamics on the Sequence Space on 2 Symbols

Now let’s return to dynamical systems, but this time our state space will be
Σ2 instead of R. Define the function σ : Σ2 → Σ2 by

σ(.s0 s1 s2 . . . ) = .s1 s2 . . . .

The function σ is called the shift map because it simply shifts the sequence
left one space and deletes the first entry. For example,

σ(.011100 . . . ) = .11100 . . . .

The shift map defines a dynamical system on Σ2 . We iterate it in the same

way we have been doing for functions on R. The only difference here is that
instead of using the real numbers as the state space, we are using infinite
sequences of zeros and ones.

Explore

Exploration 27. What are the fixed points of σ?

Exploration 28. What are the period 2 points of σ? Period 3 points? Gen-
eralize for period n points?
100 The Tools of Global Dynamics

Exploration 29. Let s = .11110 and t = .1.

1. What is d(s, t) (you answered this in a previous question)? Are they
“close” together?
2. What is d(σ(s), σ(t))?
3. What is d(σ 4 (s), σ 4 (t))? Are they “close” together?

Exploration 30. Let s = .0. Give an example of a point t such that

1
d(s, t) < 5 and d(σ n (s), σ n (t)) = 2 for n > 6.
2
Exploration 31. Repeat the previous exercise, but this time start with a
point s that is a prime period 2 point.

Exploration 32. Give an example of an eventually fixed point. Give an

example of an eventually period 2 point.

Exploration 33. Can you think of any other interesting dynamical behavior
that this function might have? Give an example and explain what happens
and why you think that it is interesting.

Exploration 34. For each n, how many points of period n does σ have?
Note: Count all period n points, not just those of prime period n.

Exploration 35. Are there prime periodic points of all periods?

Exploration 36. Suppose that I gave you some point s in Σ2 and some
ε > 0. Describe how to find a periodic point p such that d(s, p) < ε?

Prove

While you are familiar with the idea of a continuous function on R, you
may not have seen that the idea of continuity extends to any metric space
X. The definition of continuity below is almost identical to the definition of
continuity that you first learned in calculus. The only difference is that we
replace the absolute value (the distance function on R) with the more general
distance function d.
Definition 7.1 Let X be a metric space with metric d. A function f : X → X
is continuous if for every ε > 0 there exists δ > 0 such that if d(x, y) < δ,
then d(f (x), f (y)) < ε.

Proof 37. Prove that σ is continuous.

The Shift Map (Part 1) 101

In mathematics, we often need to describe ways in which one set U relates

to another set V. There are a few ways to do this that you already know.
Phrases such “subset of” and “intersects” describe basic relationships between
different sets. However, sometimes we want to describe a relationship between
sets in more detail. Are the points of U only in one part of V, or are they
uniformly spread out? Are there lots of points of U in V, or just a few? The
concept of density, defined below, is one tool that can be used to answer the
first of these questions.

Definition 7.2 Let X be a metric space with metric d and suppose U, V ⊂ X.

A set U is dense in a set V if for every ε > 0 and every point u ∈ U , there
exists a v ∈ V such that d(u, v) < ε.

To illustrate this definition let’s begin with proofs involving familiar sets.
Proof 38. Show that the rational numbers are dense in the real numbers.

Proof 39. Show that the set

 
1
U= : n∈N
n

is not dense in the interval [0, 1].

Now let’s return to our sequence space Σ2 and the shift map σ.
Proof 40. Prove that periodic points of σ are dense in Σ2 .

Proof 41. Prove that eventually fixed points of σ are dense in Σ2 .

8
Examples of Chaos

8.1 Introduction: The Definition of Chaos

All of the dynamical systems that you have explored in the past two chapters
are called chaotic dynamical systems. Chaos, in the mathematical sense, was
first encountered in the 1960s independently by a variety mathematicians and
mathematically inclined scientists. Possibly, the first person to describe this
phenomenon was the meteorologist Edward Lorenz in his 1963 paper Deter-
ministic non-periodic flow published in the Journal of Atmospheric Science. [7]
The story of Lorenz’s discovery of chaos and the stories of the other pioneers
in this field are remarkable. We highly recommend the book Chaos: The Mak-
ing of a New Science by James Gleick [3] for those who may be interested.
It vividly tells these stories, describes the underlying mathematical ideas for
non-specialists, and explains why chaos was such a surprising and important
mathematical and scientific discovery.

Definition 8.1 A dynamical system F : X → X is chaotic if

• Periodic points of F are dense in X,
• F displays sensitive dependence on initial conditions on X, and
• F is topologically transitive on X.

You should already understand the meaning of the first bullet point in the
definition. The other two need definitions and explanations.

Explore

Let’s begin with sensitive dependence.

Definition 8.2 A dynamical system F : X → X displays sensitive depen-
dence on initial conditions on X if there exists b > 0 such that for every
ε > 0 and every x ∈ X, there exists a y ∈ X with d(x, y) < ε and an N > 0
such that d(F N (x), F N (y)) > b.

103
104 Examples of Chaos

There are a lot of quantifiers to parse in this definition, and we should take
some time to get a feel for what it says. In short, it says that F displays sensi-
tive dependence on initial conditions if two close points (x and y), eventually
get separated by a distance b under iteration by F. We need to discuss this
more carefully to fully understand all of the nuances of the definition. The
best way to do this is to slowly and carefully analyze each phrase in the defi-
nition and then see how they work together to capture the concept of sensitive
dependence.
First, definition 8.2 says “for every x,” which means that this phenomenon
happens everywhere in X.
Next, it says “there exists a b > 0.” The constant b is a “separation con-
stant” as it gives us a measure of how far apart the orbits of x and y will
become at later time N . This is stated in the conclusion that there exists an
N such that d(F N (x), F N (y)) > b.
Now on to the ε phrase in the definition. This phrase says “for every ε.”
We should think of ε as a measure of initial closeness. This is expressed in the
inequality d(x, y) < ε.
Finally, let’s talk about the y. This time the definition says “there exists
a y,” which implies that the separation might not happen for all points close
to x, but only for at least one point close to x.
Hopefully, this careful description of the definition is helpful. However, it
is probably best to see what happens when we iterate a system that displays
sensitive dependence on initial conditions. The tool on the website for explo-
rations 1 through 5 allows you to explore this phenomenon when iterating the
logistic map with a = 4.
Exploration 1. IBLdynamics.com Using the tool on the website, set ε =
0.0001. Try to find an initial condition where the distance between iterates
remains less than 1/2 for all iterates displayed on the tool. Repeat this with
ε = 0.00001.

Exploration 2. IBLdynamics.com Try to find a value of ε > 0, where the

distance between orbits is always less than 1/2 for the 20 iterates shown?

Exploration 3. IBLdynamics.com If the distance between orbits ever be-

comes greater than 1/2 , does this distance remain so for all iterates thereafter?

Exploration 4. IBLdynamics.com Let x0 = 0.2 and ε = 0.0001. Track the

orbits by making a list using the letters L and R. Write an L (for left) every
time an iterate is less than 1/2 and write an R (right) every time an iterate is
greater than 1/2 .
1. How do these lists relate to one another, if at all?
2. What is the first entry where one list has an L but the other list
has an R? Call this value N for future reference.
Introduction: The Definition of Chaos 105

Exploration 5. IBLdynamics.com Convert your sequences of letters from

Exploration 4 to a sequence of numbers by replacing each L with 0 and each
R with 1. Call these sequences s and t, respectively.
1. Find an upper bound for d(s, t) in Σ2 using the metric given in equa-
tion 7.4. This is an estimate on how close the points are together
initially.
2. Use the value of N that you found in exploration 4 to find a lower
bound on d(σ N (s), σ N (t)). This is an estimate on how far apart
these iterates are after N iterates.
3. How do you think this relates to sensitive dependence from defini-
tion 8.2?
The third characteristic of chaos, as defined in definition 8.1, is topological
transitivity.
Definition 8.3 A function F : X → X is topologically transitive if for
any pair of open sets U and V in X there exists k > 0 such that

F k (U ) ∩ V 6= ∅.

This definition basically says that if a dynamical system is topologically tran-

sitive on a set X, then the open sets of X get all mixed up under iteration.
More specifically, it says that given any two open sets U and V , an iterate of
U will eventually intersect V .
To understand definition 8.3, we need to first understand what is meant
by F (I) where I is an interval. Formally, we define

F (I) = {y | y = F (x), for all x ∈ I} .

The first topological transitivity tool on the web site uses the logistic function
f4 (x) to illustrate this basic idea. You define an interval I and the tool displays
its image f4 (I).
Exploration 6. IBLdynamics.com Use the tool on the website to choose
the left and right endpoints (xmin and xmax) of an input interval I so that
both endpoints have values less than 1/2 .
1. How are the endpoints of the image interval f4 (I) determined?
2. For a point in the image interval, how many points in the domain
interval I get mapped onto that point?
Answer these questions again when xmin and xmax are both greater than 1/2 .

Exploration 7. IBLdynamics.com Now, take xmin < 1/2 < xmax and again
use the tool on the website to observe f4 (I).
1. How are endpoints of the image interval f4 (I) determined?
106 Examples of Chaos

2. For a point in the image interval, how many points in the domain
interval I get mapped onto that point?

Exploration 8. The endpoints of the image interval f4 (x) were determined

differently in exploration 7 than they were in exploration 6. What caused this
difference?

Exploration 9. Your answers to item 2 in explorations 6 and 7 also differed.

What caused this difference?

Exploration 10. How do your answers to explorations 8 and 9 relate to each

other?
The tool on the website for exploration 11 shows sequences of interval
iterates and how these image intervals relate to a fixed target interval.
Exploration 11. IBLdynamics.com Play with this tool a bit and try to find
examples where the iterated interval does not eventually intersect the target
interval. How does this relate to the concept of topological transitivity?

Prove

Directly proving topological transitivity for a given dynamical system can

sometimes be tricky. However, for some functions, there is a much easier way
to prove this that uses the concept of denseness given in definition 7.2. Recall
that for a dynamical system F, the orbit O(x) is the set of iterates of the
initial condition x and is formally defined by
∞
O(x) = {F n (x)}n=0 .
Theorem 8.1 Let X be a metric space with metric d. If F : X → X is
continuous and has a dense orbit, then F is topologically transitive.
The following steps guide you through a proof of this theorem. Let U and V
be any two open sets in X. Our goal is to show that there exists N > 0 such
that
F N (U ) ∩ V 6= ∅.

Proof 12. Explain why we can assume that that U ∩ V = ∅.

The assumption of Theorem 8.1 tells us that there exists a point x ∈ X such
that O(x) is dense in X.
Proof 13. Show that there exists an n1 > 0 such that
F n1 (x) ∈ U.
Let y = F n1 (x).
The Shift Map (Part 2) 107

Proof 14. Show that there exists an n2 > n1 such that

F n2 (x) ∈ V.

Proof 15. Show that

F n2 −n1 (y) ∈ V.

Proof 16. Explain why this implies that

F n2 −n1 (U ) ∩ V 6= ∅.

8.2 The Shift Map (Part 2)

We now return to the study of the shift map σ and prove that it is chaotic on
Σ2 .

Prove

Proof 17. Prove that periodic points of σ are dense in Σ2 .

Proof 18. Prove that σ exhibits sensitive dependence on initial conditions

by completing the following steps.
1
1. Fix ε > 0. Explain why we can choose N > 0 such that < ε.
2N
2. Fix any point s = .s0 s1 · · · ∈ Σ2 . Define t so that d(s, t) < (1/2 )N .
Hint: What must the sequence t look like if s and t are close?
3. Further, define t to ensure that

d σ N (s) , σ N (t) ≥ 1.

4. Explain why this proves the result.

Proof 19. Consider a point s in Σ2 constructed in the following manner.

1. Let the first two entries of s be 0 and 1.
2. Append to that all sequences of 0’s and 1’s having length 2. (there
are 4 of them.)
3. Append to that all sequences of 0’s and 1’s having length 3. (how
many are there?)
108 Examples of Chaos

4. Keep appending sequences in this manner. Note: You can’t write

out s completely, but you should be able to explain the process of
doing so in detail.
Prove that the orbit of s under iteration by σ is dense in Σ2 .

Proof 20. Devise a different method for constructing another point s0 that
also has a dense orbit.

8.3 Topological Conjugacy

Now that you have shown that the shift map σ is chaotic on Σ2 , we want to
use that fact to prove that many of the other systems that we have explored
are also chaotic. The basic methodology for doing this is to
• construct a function that creates a correspondence between a carefully cho-
sen subset X of R and Σ2 , and

• use this function to show that the dynamics of the original function are
equivalent to the dynamics of the shift map σ. This equivalence is known as
a conjugacy.
Let S be the function that defines the correspondence between X and Σ2 .
Generally, we need S to create a one-to-one correspondence between points of
our dynamic set X and points of our sequence space Σ2 and thus S must be
• one-to-one,
• onto, and
• continuous with a continuous inverse S −1 .

A function that has these properties is called a homeomorphism. Below

are the definitions of one-to-one and onto. You have likely encountered these
concepts in other mathematics courses.

Definition 8.4 A function f : X → Y is one-to-one if for all x1 and x2

in X, f (x1 ) = f (x2 ) implies that x1 = x2 . Equivalently, if x1 6= x2 then
f (x1 ) 6= f (x2 ).

Definition 8.5 A function f : X → Y is onto if for every y ∈ Y there exists

an x ∈ X such that f (x) = y.
Topological Conjugacy 109

A Note on Function Nomenclature

The terminology used by mathematicians to describe different classes
of functions might seem a bit confusing to you. This is partly because
different mathematical disciplines want to describe different functional
properties and this has led to a plethora of similar sounding terms de-
scribing different functional properties.
One common nomenclature is to use the suffix “morphism” when de-
scribing a function. A one-to-one function is called a monomorphism.
An onto function is called an epimorphism. A function that is both one-
to-one and onto is called a homeomorphism. A diffeomorphism is a
homeomorphism that is also differentiable. And the names keep coming.
One important warning is in order. You may have encountered homo-
morphisms in an abstract algebra class. This is a type of function that
is unrelated to the homeomorphisms that we are using here. It is unfor-
tunate that these two largely unrelated classes of functions are described
by words differing in only a single letter.

The method used to establish that two different functions possess equiv-
alent dynamics is summarized in the diagram of figure 8.1 where arrows are
used to represent the action of a function. Along the top row, we represent
the dynamical function F that we are studying. Along the bottom row is a
representation of the shift map σ. The homeomorphism S is represented by
the two vertical arrows on the left and right of the diagram. These connect
the underlying phase spaces. In this case, S maps values of X to values of
Σ2 . Because S is a homeomorphism, we say that the spaces X and Σ2 are
topologically equivalent.
Using the homeomorphism S to establish the topological equivalence of the
spaces is only half of the problem. We also need to establish dynamical equiva-
lence of the functions being iterated. This is done by looking at the interaction
between the dynamical functions and the homeomorphism S as represented
in the commutative diagram. The diagram is about function composition and
the easiest way to understand such a diagram is to follow the functions from
the upper left of the diagram to the bottom right of the diagram.
• If we go “right then down,” the diagram tells us to first apply F (top arrow
right) and then apply S (right arrow down). Thus, that pathway is S(F (x)).
• If we go “down then right,” the diagram tells us to first apply S (left ar-
row down) and then apply σ (bottom arrow right). Thus, that pathway is
σ(S(x)).
When this diagram applies, it says that

S(F (x)) = σ(S(x)). (8.1)

110 Examples of Chaos
F
X X
S S
σ
Σ2 Σ2

FIGURE 8.1
A commutative diagram for the topological conjugacy between a function F
and the shift map σ.

This relationship, combined with the fact that S is a homeomorphism, estab-

lishes the dynamical equivalency between F and σ. This is how one proves
that a function F is chaotic.
The hardest part of using this method to establish that a dynamical system
is chaotic is not defining the function S, but showing that it satisfies all of the
required properties discussed above. Because of that, we will not complete all
of those proofs. One good resource for the proofs required is An Introduction
to Chaotic Dynamical Systems by Robert L. Devaney [2].

Apply

Diagrams such as the one shown in figure 8.1 are known as commuta-
tive diagrams and are used widely in mathematics where the composition
of multiple functions in multiple ways are being compared. We have labeled
the upper and lower topological spaces as X and Σ2 simply because that is
what they will be in our context. These spaces could be R, Rn , C or almost
anything else.
Application 21. You have encountered the concept embodied in the com-
mutative diagram above in linear algebra when you studied the concept of
similar matrices. A matrix A is similar to a matrix B if there exists an in-
vertible matrix P such that

B = P AP −1 . (8.2)

1. Rewrite equation 8.2 so that it is in the form of equation 8.1.

2. What are the spaces at the corners of the commutative diagram in
this setting?
3. Draw the commutative diagram expressed in equation 8.2.

Application 22. What linear algebra concept does the “commutative dia-
gram” below represent?
Return to The Doubling Map 111
A
~v λ~v
P P
D
~e λ~e

8.4 Return to The Doubling Map

In this section, you will prove the doubling map D(x) = 2x (mod 1) is chaotic
on the interval [0, 1]. Despite the warning at the end of the previous section,
we can complete all the tasks needed to do this for this example.

Prove

Proof 23. Review section 6.3 on the doubling map D. What dynamical
properties of D suggest that this function is chaotic on [0, 1]?

Proof 24. There is a very natural way to define a function S : [0, 1] → Σ2

that is hinted at in the doubling map exercises of section 6.3.2. What is it?

Proof 25. Prove that S is one-to-one.

Proof 26. Prove that S is onto.

Proof 27. Prove that S is continuous. Note: This is almost as easy as it

seems, but you need to be careful with ε and δ. Also, remember that the
domain of S is [0, 1] and the range of S is Σ2 .

Proof 28. Prove that S −1 is continuous.

At this point, you have proved that S is in fact a homeomorphism. The
final task is to use this homeomorphism to show that the dynamics of D and
σ are equivalent.
Proof 29. Prove that S(D(x)) = σ(S(x)) and conclude that the doubling
map D is chaotic on [0, 1].

8.5 The Logistic Map with a > 4 (Part 2)

Now its time to return to the logistic map to prove that the logistic map
with a > 4 is chaotic. It will be helpful to review the material covered in
112 Examples of Chaos

section 6.4 where you first explored these dynamics. Figure 8.2 shows the
2 3
graphs of y = f4.5 (x), (left) y = f4.5 (x), (center), and y = f4.5 (x), (right) to
help you remember some of the essential properties. In particular, because the
central portion of the graph of y = f4.5 (x) is above the line y = 1, we know
that the orbits of points in the center iterate to negative infinity. Similarly,
there is an interval to the left of the center and one to the right of the center
that get mapped into the center and hence also iterate to negative infinity. In
fact, you showed that for each n there are 2n−1 open intervals such that if x
is in one of these intervals, then
n−1
2
f4.5 (x) > 1.

Hence, these values of x iterate to negative infinity. This leaves very few points
whose orbits remain in the interval [0, 1] for all time.

n=1 n=2 n=3

FIGURE 8.2
n
The graphs of y = f4.5 (x) with n = 1, 2, 3.

Several steps are required to prove that this function is chaotic and it will
be helpful to outline the process before working through the actual proofs.
1. You have already shown that almost every point has an orbit that
goes to negative infinity. However, there is a set of points (that we
will call Λ) whose iterates remain in the unit interval for all time.
Our first step will be to determine the topological structure of this
set Λ.
2. The second step is to develop a method for describing the orbits of
points in Λ. This method (actually a function) is called an itinerary
because it describes the orbit of a point x by telling us where it is
at every time in its orbit.
3. Finally, we will use the itinerary to create a topological conjugacy
between the logistic function restricted to Λ and the shift map σ on
Σ2 . This will prove that the logistic function is chaotic on Λ.
The Logistic Map with a > 4 (Part 2) 113

Prove

Let’s begin by introducing some notation. Let

   
1
I0 = x ∈ 0, f (x) ≤ 1 (8.3)
2

and    
1
I1 = x ∈ ,1 f (x) ≤ 1 . (8.4)
2
The intervals I0 and I1 are highlighted along the x-axis in figure 8.3. Note that
I0 ∪ I1 is simply the set of all x-values such that f (x) ∈ [0, 1]. In mathematical
notation
I0 ∪ I1 = {x ∈ [0, 1] | f (x) ∈ [0, 1]} .

FIGURE 8.3
The graph of y = f4.5 (x) with the intervals I0 and I1 highlighted.

All of the results that you are about to prove are√

true when the parameter
a > 4. However, the proofs are easier if a > 2 + 5. We will assume this
property for the remainder of this section. The following exercise motivates
this choice of a.
Proof 30. Prove that if x ∈ I0 ∪ I1 , then |fa0 (x)| > 1. Hint: For what values
of x is |fa0 (x)| smallest for x ∈ I0 ∪ I1 ? Look at figure 8.3.

Proof 31. Prove that I0 has two disjoint subintervals I00 and I01 such that
• if x ∈ I00 then fa (x) ∈ I0 and fa2 (x) ∈ I0 , and
• if x ∈ I01 then fa (x) ∈ I0 and fa2 (x) ∈ I1 .
Identify these intervals on figure 8.3.
114 Examples of Chaos

Proof 32. Repeat proof 31 for subintervals of I1 . Specifically, prove that I1

has two disjoint subintervals I10 and I11 such that
• if x ∈ I10 then fa (x) ∈ I1 and fa2 (x) ∈ I0 , and
• if x ∈ I11 then fa (x) ∈ I1 and fa2 (x) ∈ I1 .
Identify these intervals on figure 8.3.

Proof 33. We can naturally continue to define subintervals in this way.

Describe the orbits of points in I001 and I101 . How many subintervals of type
Is0 s1 s2 are there if si = 0 or 1 for i = 0, 1, 2? Where is each located?

Proof 34. In general, how many subintervals of type Is0 s1 s2 ...sk are there if
si = 0 or 1 for i = 0, 1, . . . , k?

Proof 35. What kind of set is being constructed using this process?
Now let’s define the invariant set Λ that we are interested in. Let

Λ = {x ∈ [0, 1] | fan (x) ∈ [0, 1] ∀ n} . (8.5)

Proof 36. Show that Λ is nonempty.

Proof 37. What is the relationship between Λ and I0 ∪ I1 ?

Proof 38. What is the relationship between Λ and

[
Is0 s1 ?
si ∈{0,1}

Proof 39. In general, what is the relationship between Λ and

[
Is0 s1 ...sn ?
si ∈{0,1}

Proof 40. Show that Λ contains no intervals by following the steps below.
1. Assume, by way of contradiction, that there exists an interval
[α, β] ∈ Λ. Explain why [α, β] is contained in either I0 or I1 .
2. One implication of the mean value theorem is that if m = min|fa0 (x)|
on [α, β], then
|fa (α) − fa (β)| ≥ m |α − β| .
Use this fact to reach a contradiction.

Proof 41. How can we use the subintervals defined in proof 33 to track the
orbit of a point in x ∈ Λ using a sequence of 0’s and 1’s?
The Logistic Map with a = 4 (Part 2) 115

Proof 42. Define a function S : Λ → Σ2 based on the result of proof 41.

S(x) is called the itinerary of x.

Proof 43. Assume for the moment that S is a homeomorphism (i.e., one-to-
one, onto, and continuous) to answer the following questions. (You will prove
some of these properties shortly).
1. Prove that Λ is uncountable and conclude that Λ must be topolog-
ically equivalent to the Cantor set.
2. Prove that
S (fa (x)) = σ (S (x))
for all x ∈ Λ. Conclude that fa is chaotic on Λ.
As mentioned above, the itinerary function S is a homeomorphism. Veri-
fying some of the required properties is tricky. In the exercises below, we focus
on two properties that are fairly straightforward to prove.
Proof 44. Prove that S is one-to-one. Hint: Assume that it is not. What
must be true about points in the interval connecting two points with identical
itineraries? Use the mean value theorem again.

Proof 45. Prove that S is onto. Hint: Use the itinerary to create a nested
sequence of closed intervals.

8.6 The Logistic Map with a = 4 (Part 2)

We complete our exploration into chaotic dynamics by returning to the logistic
map f4 (x). You will show that this dynamical system is chaotic on the interval
[0, 1] by showing that it is topologically conjugate (technically semi-conjugate)
to the doubling map D(x).

Prove

Proof 46. Let h1 (x) = cos(2πx) and q(x) = 2x2 − 1. Show that

h1 (D(x)) = q(h1 (x)).

Hint: What is the double angle formula?

Proof 47. Draw the commutative diagram like that shown in figure 8.1 to
illustrate the result of proof 46.
116 Examples of Chaos

Proof 48. Find a function h2 (x) = Ax + B such that

h2 (q(x)) = f4 (h2 (x)).

Proof 49. Append this conjugacy to the commutative diagram that you
made in proof 47.

Proof 50. Look at your two-level commutative diagram. What is the conju-
gacy function between the doubling map D and the logistic function f4 ?

Proof 51. Briefly explain why this implies that f4 is chaotic on [0, 1].

Proof 52. Recall that a conjugacy function must be one-to-one, onto, and
continuous. The one that you just created violates one of these properties and
that is why it is called a semi-conjugacy. What property is violated and why?
9
From Fixed Points to Chaos

9.1 Introduction
Let’s take a few minutes to review what we know about the logistic family of
functions. In chapter 3, you showed that the logistic family of functions
• has a fixed point at x = 0 and it is attracting if 0 < a < 1, and
• has a fixed point at x = (a − 1)/a and it is attracting if 1 < a < 3.
You summarized this by creating the plot of the value of the attracting fixed
point as a function of the parameter a like the one shown in figure 9.1.

FIGURE 9.1
A fixed point diagram for the logistic family for 0 < a < 3.

Additionally, in chapter 5 where you studied bifurcations, you learned that

at a = 1 a tangent bifurcation occurs and at a = 3, a period doubling bifurca-
tion occurs. So, this figure neatly summarizes everything we know about the
asymptotically stable dynamics of the logistic family for 0 < a < 3.
Then, in chapter 6, you explored chaotic behavior in great detail and
learned that the dynamics of the logistic family of functions is chaotic when
a ≥ 4. So, the natural question to ask is “how do the dynamics go from the
very tame dynamics seen when a < 3 to incredibly complicated dynamics
when a ≥ 4?” That is the goal of this chapter.

117
118 From Fixed Points to Chaos

That transition is summarized in a bifurcation diagram for the logistic

family of functions. Bifurcation diagrams, like the one shown in figure 9.2,
are often computed numerically, not analytically. The horizontal axis is the
parameter value a and the vertical axis gives a representation of the asymptotic
or long term dynamics for each parameter value. For 0 < a < 3, this figure is
a numerically calculated version of figure 9.1. Take a few minutes to convince
yourself that this is indeed the case.

FIGURE 9.2
A bifurcation diagram for the logistic family for 0 < a < 4.

In this section, you will

• learn how to compute bifurcation diagrams like this,
• learn how to read and interpret the diagram, and
• explore some of the patterns that take a system from equilibrium dynamics
to chaotic dynamics.
Remember, the mathematical ideas that we are exploring for the logistic family
are ubiquitous! Many other dynamical families have almost identical dynamics
and bifurcations and in that sense this diagram is universal.
Unlike in previous chapters, your exercises of the in this chapter will be
primarily numerical explorations using apps on the website and conjectures
based on these explorations. There are proofs of the ideas that you will explore
in this section, but they require some sophisticated mathematical methods.
However, there are some remarkable patterns hidden in figure 9.2 to be dis-
covered with a good computer program and a little patience.
Computing a Bifurcation Diagram 119

9.2 Computing a Bifurcation Diagram

The basic idea behind computing a bifurcation diagram is fairly simple.
1. Make a list of parameter values {a1 , a2 , . . . , aM }.
2. For each parameter value ak , choose an initial condition x0 and then
plot the N + 1 points

{(ak , x0 ) , (ak , x1 ) , . . . , (ak , xN )} .

This basic outline will need some refinement to get an accurate picture, but
it captures the essence of the algorithm. The next few exercises and theorems
will tell you how to make these refinements to display an accurate bifurcation
diagram.

Explore

Exploration 1. Remember that you only want to see the attracting orbits
(when one exists). Modify the plotting of the points in step 2 of the method
outlined above so that you are only plotting approximations of the attracting
orbits and not the iterates that are “far away” from it?

Exploration 2. Take another look at the bifurcation diagram in figure 9.2.

Notice that near the bifurcation values a = 1 and a = 3, the figure isn’t very
crisp. There is a thick black smear of points and not a single, fine curve. Why
do you think that happened? What could be done to improve this?
A more complicated question is how to pick an initial condition or initial
conditions to generate a figure that is an accurate representation of the bifur-
cation diagram. You want to be sure that the initial condition you choose will
converge to an attracting orbit when one exists and that you have accounted
for all possible attracting orbits that might exist. In other words, if there are
two attracting periodic orbits, we want to see both of them in the bifurcation
diagram. It turns out that the critical points of a function are the key to doing
this in general. We won’t go into detail explaining why this is so (if you are
interested in the details, see one of the two Devaney books [1, 2]). However,
we will present the important definitions and theorems that tell us this is in
fact so.

Definition 9.1 The Schwarzian derivative of a C 3 function F : R → R is

2
F 000 (x)
 00
F (x)
SF (x) = 0 − . (9.1)
F (x) F 0 (x)
120 From Fixed Points to Chaos

And you always wondered the third derivative might be good for! An impor-
tant fact related to the iteration of functions is that the Schwarzian derivative
remains negative when we compose a function with itself.
Lemma 9.1 If SF < 0, then SF n < 0 for all n ≥ 0.
The other important concept that we need to introduce is that of a basin
of attraction.
Definition 9.2 The basin of attraction of a fixed point x0 of a function F
is the set of all points whose orbit converges to x0 . The immediate basin of
attraction of x0 is the maximal interval containing x0 that lies in the basin
of attraction.
These two definitions and lemma 9.1 are used to prove the following theorem.

Theorem 9.1 Suppose SF < 0. If x0 is an attracting periodic point of F ,

then either the immediate basin of attraction of x0 extends to +∞ or −∞ or
else there is a critical point of F whose orbit is attracted to x0 .

Prove

Proof 3. Explain why theorem 9.1 implies that a polynomial of degree n

with negative Schwarzian derivative can have at most n + 1 attracting orbits.

Proof 4. Show that if F is a quadratic polynomial, then SF < 0.

Proof 5. Prove the following for the logistic family of functions. If fa has
an attracting periodic orbit Ω in [0, 1], then f n (1/2) → Ω as n → ∞. Hints:
The points of a period k orbit are fixed points of what function? How many
critical points must this function have?

Proof 6. Modify the algorithm for computing the bifurcation diagram of the
logistic function using this new idea.

9.3 Period-doubling to Chaos

Now that we understand how to construct bifurcation diagrams like the one
shown in figure 9.2, we can begin to explore the details of the diagram and
interpret them in terms of dynamics and bifurcations.
Period-doubling to Chaos 121

Explore

Let’s begin our exploration of this bifurcation diagram by looking at what

happens to the dynamics as we increase the parameter a past 3 to about 3.5.
Figure 9.3 is a magnification (or zoom) of the bifurcation diagram in this
region.

FIGURE 9.3
The bifurcation diagram for the logistic family for 2.95 < a < 3.6.

Exploration 7. What attracting periodic orbit exists for 3.1 < a < 3.4?

Exploration 8. What attracting periodic orbit exists for a = 3.5?

Exploration 9. Using figure 9.3, describe what you think happens to the
periods of the attracting orbits as a increases from 3 to about 3.6.

Exploration 10. IBLdynamics.com Use the tool on the website to zoom in

on the bifurcation diagram for the logistic function. Start with the parameter
range 3.5 < a < 3.6. What is the highest periodic orbit you can find? You
should keep zooming in on this region to explore what is happening in this
region.

Exploration 11. IBLdynamics.com You know that an attracting period 2

orbit exists for a slightly bigger than 3 because we have shown that a period-
doubling bifurcation occurs at a = 3. Use the logistic bifurcation tool on the
website to answer the following questions.
1. At approximately what value of a does the period 2 orbit bifurcate
to a period 4 orbit?
122 From Fixed Points to Chaos

2. At approximately what value of a does the period 4 orbit bifurcate

to a period 8 orbit?
3. At approximately what value of a does the period 8 orbit bifurcate
to a period 16 orbit?
What is happening to the lengths of these intervals?
You are seeing what is called the period-doubling cascade.

period 1 −→ period 2 −→ period 4−→ period 8 −→ · · · .

Exploration 12. Based on these explorations, do you think that this cascade
stops at some finite period 2n or not?

9.4 Windows of Stable Periodic Behavior

Now let’s focus on the right half of figure 9.2 past the first period-doubling
cascade.

Explore

Exploration 13. IBLdynamics.com Using the tool on the website, give an

example of a parameter value where there does not appear to be any attracting
periodic orbit. Do you think the dynamics are or are not chaotic?

Exploration 14. IBLdynamics.com Use the tool on the website to find a

value of a where there is an attracting period 3 orbit. An attracting period 5
orbit. An attracting period 7 orbit.
You have certainly noticed that in this diagram, there are “windows” of
attracting dynamics. Let’s look at these a bit more closely.
Exploration 15. IBLdynamics.com Use the tool to zoom in on the period
3 window (use 3.8 < a < 3.87 at first but you may need to narrow this range
as you work through this exercise).
1. What happens to the period 3 orbit as a increases through this
range? In other words, what attracting periodic orbits are “de-
scended” from the original periodic orbit.
2. How does this relate to what you observed in section 9.3?
Windows of Stable Periodic Behavior 123

3. Find some other windows of stable periodic behavior and again

zoom in on it. What do you notice?
4. Do you have any ideas about why the period 3 orbit “suddenly”
appears from the chaotic behavior?

Exploration 16. IBLdynamics.com Let’s more carefully explore the emer-

gence of the period 3 window using the graphical analysis tool on the website.
You will want to manipulate the parameter a from about a = 3.8 to a = 3.87.
Pay particular attention to what happens to the graph of y = F 3 (x) as the
period 3 orbit appears. Describe what caused the sudden emergence of this
window. In particular, what bifurcation does this look like?

Apply

There is an interesting pattern in the bifurcation diagram that is related

to the ordering of the periodic windows in the bifurcation diagram. On the left
side is the period 1 window and the biggest window elsewhere in the diagram
is the period 3 window. Use the tool on the website to complete the following
exercises to discover this pattern.
Application 17. IBLdynamics.com Zoom in on the diagram to see that
the biggest two windows between the period 1 and period 3 windows are of
periods 5 and 6 and that the order of them is
1 6 5 3.

Application 18. IBLdynamics.com Repeat this zooming in process to fill

in this listing.
1 A B 6 C D 5 E 3.
For example, the letter E here represents the period of the biggest window
between the period 5 and period 3 window.

Application 19. IBLdynamics.com One more time, but now only between
periods 1 and 6.

1 a b A c d B e 6.
(Note the A and B you determined in application 18.)

Application 20. Do you see a pattern? Could you complete one more step
in this process without actually zooming in?
10
Sarkovskii’s Theorem

10.1 Introduction
One of the overarching goals of this book is to describe, as completely as
possible, the entire range of dynamical behavior that can occur in a given
dynamical system. One aspect of this goal is to describe the types of periodic
behavior that can and cannot occur in a given system. In chapter 6, we did
this primarily by using the shift map to show that those dynamical systems
that are conjugate to the shift map on Σ2 have periodic orbits of all periods
and that those orbits are dense in the invariant, chaotic set defined by the
conjugacy.
The goal of this section is a bit different. We aim to determine what prime
periodic orbits must exist if we already know the prime period of one orbit
already. You have already done a little of this work in chapter 3. In that
chapter, you proved that if f is continuous and there exists a period 2 orbit,
then f must also have a fixed point. But what can we say if somehow we know
that f has a period 13 point? Is there a period 12 point? What about a period
14 point?
The answer to this question turns out to be quite remarkable. And what
makes the answer especially satisfying is that its proof uses no high-powered
mathematical machinery. In fact, it relies almost solely on one often overlooked
theorem of calculus: the intermediate value theorem.

10.2 The Intermediate Value Theorem

Every result in this section is a consequence of the intermediate value theorem.
Informally, the intermediate value theorem says that we can draw a continuous
function without lifting our pencil from the paper. This intuition is important,
but to understand and apply this theorem, we need a deeper and more formal
understanding than the pencil analogy gives us.

125
126 Sarkovskii’s Theorem

Theorem 10.1 (intermediate value theorem) If f (x) is a continuous

function on the interval [a, b] and c is between f (a) and f (b), then there exists
an x ∈ [a, b] such that f (x) = c.
In practice, we often use the following corollaries. The first is simply the case
where we assume that c = 0 and is actually equivalent to the Intermedi-
ate Theorem. In the second corollary, there is an additional hypothesis that
guarantees uniqueness.
Corollary 10.1 Let f (x) be a continuous function on the interval [a, b]. If
f (a) ≤ 0 and f (b) ≥ 0 or vice versa, then there exists an x ∈ [a, b] such that
f (x) = 0.
Corollary 10.2 Let f (x) be a continuous, strictly decreasing function on the
interval [a, b]. If f (a) ≥ 0 and f (b) ≤ 0, then there exists a unique x ∈ [a, b]
such that f (x) = 0.

Prove

You will prove the intermediate value theorem by completing each of the
proofs in this section. More precisely, you will prove Corollary 10.1, which is
equivalent to theorem 10.1.
Assume that f (a) ≤ 0 ≤ f (b). Let x0 = a and y0 = b. Define
x0 + y0
m0 = .
2
If f (m0 ) = 0, then x = m0 and we are done. If f (m0 ) < 0, let x1 = m0 and
y1 = y0 . If f (m0 ) > 0, let y1 = m0 and x1 = x0 . Continue in this fashion
creating two sequences {xn } and {yn }.
Proof 1. Explain why if either xn = 0 or yn = 0, then we are done. Now,
assume that this does not happen. Prove that f (xn ) < 0 and f (yn ) > 0 for
all n.

Proof 2. Prove that the sequence {xn } is monotone non-decreasing and that
the sequence {yn } is a monotone non-increasing.

Proof 3. Show that sequences {xn } and {yn } are bounded. What does that
imply about each of them?

Proof 4. Show that

lim |xn − yn | = 0.
n→∞
What does this imply about the limits of the two sequences?

Proof 5. Use the fact that f is continuous to prove Theorem 10.1

Proof 6. Prove corollary 10.2 using this result.

Review of Two Fixed Point Theorems 127

10.3 Review of Two Fixed Point Theorems

In section 4.1, you proved two important fixed point theorems and it may be
worthwhile to return to that section and review them. Theorem 10.2 below is
a restatement of theorem 4.1 so that it is easily available to you. Theorem 10.3
is new, but its proof follows directly from the results of section 10.2.

Prove

Theorem 10.2 If f is continuous and f ([a, b]) ⊃ [a, b], then there exists a
point c ∈ [a, b] such that f (c) = c.

Proof 7. Draw a picture of a function f having the properties described in

the hypothesis of the above theorem.

Proof 8. Prove this theorem using corollary 10.1 of the intermediate value
theorem.

Proof 9. Prove that if f is continuous and f has a prime period 2 orbit, then
f has a fixed point.
The following theorem will be quite useful in upcoming work.

Theorem 10.3 If f is continuous, decreasing, and f ([a, b]) ⊃ [a, b], then
there exists a unique point c ∈ [a, b] such that f (c) = c.

Proof 10. Draw a picture of a function f having the properties described in

the hypothesis of theorem 10.3.

Proof 11. Prove this theorem using corollary 10.2 of the intermediate value
theorem.

10.4 Sarkovskii’s Theorem

As you will see shortly, Sarkovskii’s theorem [8] answers the question posed
at the beginning of this chapter. It provides a way of deducing the existence
of different periodic orbits based on knowing the existence of a single periodic
orbit.
128 Sarkovskii’s Theorem

10.4.1 Discovering Sarkovskii’s Theorem

We begin with a sequence of numerical explorations that hint at the power of
this remarkable theorem.

Explore

The following exercises look at periodic orbits of the logistic family of func-
tions and serve to motivate Sarkovskii’s theorem. Use the tool on the website
to explore each of the following questions. Keep a chart that summarizes what
prime periodic orbits do and do not exist for each given periodic orbit.
Exploration 12. IBLdynamics.com When a = 2.5, the logistic function has
a fixed point. Determine whether it has a prime period n point for n = 1, ..., 8.

Exploration 13. IBLdynamics.com When a = 3.175, the logistic function

has a prime period 2 point. Determine whether it has a prime period n point
for n = 1, ..., 8.

Exploration 14. IBLdynamics.com When a = 3.834, the logistic function

has a prime period 3 point. Determine whether it has a prime period n point
for n = 1, ..., 8.

Exploration 15. IBLdynamics.com When a = 3.5, the logistic function

has a prime period 4 point. Determine whether it has a prime period n point
for n = 1, ..., 8.

Exploration 16. IBLdynamics.com When a = 3.739, the logistic function

has a prime period 5 point. Determine whether it has a prime period n point
for n = 1, ..., 8.

Exploration 17. IBLdynamics.com When a = 3.628, the logistic function

has a prime period 6 point. Determine whether it has a prime period n point
for n = 1, ..., 8.

Exploration 18. IBLdynamics.com When a = 3.70186, the logistic func-

tion has a prime period 7 point. Determine whether it has a prime period n
point for n = 1, ..., 8.

Exploration 19. IBLdynamics.com When a = 3.557, the logistic function

has a prime period 8 point. Determine whether it has a prime period n point
for n = 1, ..., 8.
Sarkovskii’s Theorem 129

Exploration 20. Consider the dynamical system

1
xn+1 = xn + (mod 1).
3
Show that every point is of period 3 but there does not exist any other prime
periodic orbits. Why is this not a contradiction to Theorem 10.2?

Conjecture

The exploration exercises above point toward a more general truth about
the existence of certain kinds of periodic points and how they may or may not
depend on other types of periodic points. Think about those exercises as you
answer the following questions.
Conjecture 21. If you know that f is continuous and has a fixed point,
what other prime periodic points, if any, must it also have?

Conjecture 22. If you know that f is continuous and has a prime period 2
point, what other prime periodic points, if any, must it also have?

Conjecture 23. If you know that f is continuous and has a prime period 3
point, what other prime periodic points, if any, must it also have?

Conjecture 24. If you know that f is continuous and has a prime period 4
point, what other prime periodic points, if any, must it also have?

Conjecture 25. If you know that f is continuous and has a prime period 5
point, what other prime periodic points, if any, must it also have?

Conjecture 26. If you know that f is continuous and has a prime period 6
point, what other prime periodic points, if any, must it also have?

Conjecture 27. If you know that f is continuous and has a prime period 7
point, what other prime periodic points, if any, must it also have?

Conjecture 28. If you know that f is continuous and has a prime period 8
point, what other prime periodic points, if any, must it also have?
130 Sarkovskii’s Theorem

10.4.2 Using Sarkovskii’s Theorem

Before we state Sarkovskii’s theorem, we need to introduce the Sarkovskii
ordering of the natural numbers which is a bit unusual. When reading state-
ments involving this ordering, I suggest reading the symbol . as “Sarkovskii
greater than.” So for example, 3 . 5 would be read as “3 is Sarkovskii greater
than 5” and this means that 3 comes before 5 in the Sarkovskii ordering. The
Sarkovskii ordering is

3 . 5 . 7 . ···
3·2 . 5·2 . 7·2 . ···
3 · 22 . 5 · 22 . 7 · 22 . ··· (10.1)
..
.
. . . 23 . 22 . 2 . 1.

Let’s look at this peculiar ordering more carefully. The ordering starts with all
of the odd numbers greater than or equal to 3 listed in increasing order (the
first line of equation 10.1). Next, come the odd numbers greater than or equal
to 3 multiplied by 2 (the second line of equation 10.1) also listed in increasing
order. Then, the odd numbers greater than or equal to 3 multiplied by 4 (the
third line of equation 10.1). This continues through all positive integer powers
of 2. The ordering concludes with the powers of 2 but now in decreasing order
and thus ending in 1.

Theorem 10.4 (Sarkovskii’s theorem) Suppose that f is a continuous

function on R and has a prime period n orbit. If n . m, then f has a prime
period m orbit.

Sarkovskii, Yorke, and Chaos

Sarkovskii published theorem 10.4 in the Ukrainian Journal of Math-
ematics in 1962 (an English language reprint can be found here [8]). A
combination of political and language barriers caused this seminal paper
to be unknown to western mathematicians until the late 1970s.
In 1975, the mathematician James Yorke and his graduate student
T.Y. Li published a paper entitled “Period 3 implies Chaos” in the Amer-
ican Mathematical Monthly [5]. In this paper, they prove, amongst other
results, that the existence of a period 3 orbit implies periodic orbits of all
periods. In other words, they proved a small part of Sarkovskii’s theorem.
This paper is also the origin of the term “chaos” to describe the dynamics
that you have been studying here. In the book Chaos [3], Gleick relates
a story of the initial meeting of Sarkovskii and Yorke, and the political
and language barriers that unfortunately severely limited interactions be-
tween Soviet mathematicians and their colleagues in the United States
and Western Europe during that time.
Sarkovskii’s Theorem 131

Explore

Exploration 29. Prove that every natural number n appears once and only
once in equation 10.1.

Exploration 30. Check your answers to the conjectures 21 through 28. Are
they consistent with this theorem?

Apply

Application 31. At the beginning of this chapter, I posed the question “if
f is continuous on R and has a prime period 13 orbit, must it also have a
prime period 12 orbit or a prime period 14 orbit?” What is the answer to this
question?

Application 32. If f is continuous on R and has a prime period 13 orbit,

describe all periodic orbits it must also have. List all periodic orbits it might
not have.

Application 33. Can a continuous function on R have a prime period 56

orbit but not a prime period 24 orbit?

Application 34. Can a continuous function on R have a prime period 24

orbit but not a prime period 56 orbit?

Prove

Before working on a partial proof of Sarkovskii’s theorem, you will explore

a few consequences of this remarkable theorem and some related ideas.
Proof 35. Use Sarkovskii’s theorem to prove that if a continuous function
on R has only finitely many periodic orbits, then all of them are of period 2k
for some k.
132 Sarkovskii’s Theorem

�
� � � � �
�

FIGURE 10.1
(Left) The graphs of a function g(x) that has a period 5 orbit but not a period
3 orbit.

In some sense, Sarkovskii’s theorem is optimal in that there do exist func-

tions that have a period 5 orbit but not a period 3, or have a period 7 orbit
but not a period 5 orbit, and so on. In proofs 36 through 40, you will show
that the function illustrated in figure 10.1 has a period 5 orbit, but not a
period 3 orbit.
Proof 36. Use figure 10.1 (go to the website for a printable version) to show
that the integer values 1, 2, . . . , 5 lie on a period 5 orbit of the function g(x).
Note: Carefully look at the axes! The bottom-left corner is (1,1) not (0,0).

Proof 37. Show that there is not a period 3 point in the interval [1, 2] by
finding g 3 ([1, 2]). This should be done using graphical analysis on the interval
[1, 2].

Proof 38. Repeat exploration 37 for the intervals [2, 3] and [4, 5].

Proof 39. For the function g(x) in figure 10.1, prove that g 3 ([3, 4]) = [1, 5].
Explain why this implies that there is a fixed point of g 3 in [3, 4].

Proof 40. Use the analysis of the interval [3, 4] done in exploration 39 to
show that g 3 ([3, 4]) is decreasing. Use that to conclude that the fixed point of
g 3 in [3, 4] is not of prime period 3. What is it?
As you might imagine from the strangeness of the Sarkovskii ordering, the
complete proof of this statement requires considering many different scenarios.
You will tackle a more manageable piece of the proof. You will prove that the
existence of a prime period 3 orbit implies the existence of prime periodic
orbits of all periods.
Sarkovskii’s Theorem 133

Suppose that f is a continuous function with a prime period 3 orbit {a, b, c}

with a < b < c and f (a) = b, f (b) = c, and f (c) = a.
Proof 41. Draw a number line and place a, b, and c on the line. Draw arrows
to indicate how f maps these points to each other. Label the closed interval
[a, b] as I0 and the closed interval [b, c] as I1 .
1. What is f (I0 )?
2. What is f (I1 )?
We can now begin proving that there exists periodic orbits of all periods.
All of this work will use the theorems you proved in the section 10.3 and thus
ultimately the intermediate value theorem 10.1.
Proof 42. Prove that there exists a fixed point of f in I1 .

Proof 43. Prove there exists a period 2 point of f in I0 . Be sure to explain

why this point cannot be a fixed point of f .
Now, let’s turn to proving the existence of a period 4 point.
Proof 44. Let A0 = I1 . Show there exists an interval A1 ⊂ I1 such that
f (A1 ) = A0 .

Proof 45. Show there exists an interval A2 ⊂ A1 such that f (A2 ) = A1 .

What is f 2 (A2 )?

Proof 46. Show there exists an interval A3 ⊂ A2 such that f (A3 ) = I0 .

Proof 47. Show that f 4 (A3 ) = I1 . Explain why this implies there is a fixed
point off 4 in A3 ?

Proof 48. Why must this fixed point of f 4 be a prime period 4 point of f ?

Proof 49. Discuss how you might generalize the argument of proofs 44
through 48 to prove that f has a prime period n orbit for all n > 3.
11
Dynamical Systems on the Plane

11.1 Linear Algebra Foundations

Now that we have a good understanding of dynamical systems on the real
line, we can use those ideas to study higher dimensional dynamical systems.
In this chapter, we focus on dynamical systems in the plane. However, much
of what you will learn in this chapter generalizes to higher dimensions. We
need to begin by reviewing some important ideas from linear algebra and
multivariable calculus that provide a foundation for studying dynamics in R2 .
Let A be a 2 × 2 matrix with real entries. Then A is a representation of
a function called a linear transformation from R2 to R2 . More specifically, if
(x, y) = x ∈ R2 , then the function L : R2 → R2 defined by

L(x) = Ax (11.1)

is a linear transformation. In section 11.2, you will begin exploring the dynam-
ics of linear transformations on R2 . In this section, we highlight the properties
of linear transformations that will aid you in that work.
The most important properties of the matrix A in the study of dynamical
systems are its eigenvalues and eigenvectors.
Definition 11.1 If A is an n × n matrix, then λ is an eigenvalue of A with
associated eigenvector v if

L(v) = Av = λv. (11.2)

Note that equation 11.2 is similar to a fixed point equation where the eigen-
vector v is “almost fixed” by the linear transformation L. We say that it is
almost fixed because the eigenvalue λ is essentially a scaling factor telling us
by how much the eigenvector v gets lengthened or shortened. We will need
to make this description more precise because the eigenvalues may or may
not be real numbers. But for now, it begins to explain why eigenvectors are
important in understanding linear dynamical systems.
Because this is not a linear algebra course, we won’t discuss all of the
important properties of eigenvalues and eigenvectors and what they tell us
about the linear transformation. Below is a short summary of the relevant
facts that we will need here.

135
136 Dynamical Systems on the Plane

1. Eigenvalues are found by solving the equation

det(A − λI) = 0

for λ. If A is an n×n matrix, then this is a degree n polynomial and

hence there are n eigenvalues counting multiplicity. This polynomial
is called the characteristic equation of A.
2. Because the eigenvalues are found by finding the roots of a polyno-
mial, eigenvalues may be real or complex numbers. If the matrix A
has real entries, then complex eigenvalues come in complex conju-
gate pairs α + iβ and α − iβ.
3. For each eigenvalue λi , solve the equation

(A − λi I)vi = 0

for a non-zero vector vi . This is an eigenvector associated with the

eigenvalue λi . When done correctly, there will be at least one free
variable to choose. This means that eigenvectors are not unique.
4. If the eigenvalue λ is a root of multiplicity one of the characteristic
equation, then any two associated eigenvectors are scalar multiples
of each other.
5. Eigenvectors associated with complex eigenvalues also come in com-
plex conjugate pairs. So if α + iβ has an eigenvector of u + iw, then
α − iβ has an eigenvector of u − iw.
6. If the eigenvalues of A are distinct, then the associated set of eigen-
vectors B = {v1 , . . . , vn } are linear independent and form a basis
for the underlying vector space.
7. If the eigenvalues of A are distinct, then the matrix P, whose
columns are the eigenvectors, diagonalizes A via the equation

P −1 AP = D

where D is a diagonal matrix whose diagonal entries are the eigen-

values λi . We say that the matrices A and D are similar. An im-
portant consequence of this is that if A is a diagonal matrix, then
its diagonal entries are its eigenvalues.
Linear Algebra Foundations 137

Eigenvalues, Eigenvectors, and Linear Transformations

The relationship between a matrix and a linear transformation is a
little bit more complicated than I let on. In particular, equation 11.1 is
imprecise. However, it will suffice for how we are using linear algebra to
study dynamical systems. Let me at least hint at the true nature of this
relationship here.
A linear transformation is a special kind of function from a vector
space V to another vector space W . In our situation, V = W. A matrix A
is a representation of that linear transformation. There are many different
representations of the same linear transformation and these depend on
what basis one chooses for the vector space. The eigenvalues are properties
of the linear transformation and as such, they are the same for all matrix
representations. By contrast, the eigenvectors are properties of the matrix
representation. This is why similar matrices have the same eigenvalues,
but different eigenvectors. They represent the same linear transformation,
but in different bases.

Let’s do a simple example. Let

 
3 14
A= .
0 −4

The first step is to compute the characteristic polynomial

3−λ 14
det(A − λI) = = (λ + 4)(λ − 3).
0 −4 − λ

This implies that the eigenvalues are λ1 = −4 and λ2 = 3.

The next step is to compute the eigenvectors for each of these eigenvalues.
Let’s consider λ1 = −4. We need to solve
    
7 14 a 0
(A − λ1 I)v1 = =
0 0 b 0

for a and b to compute this eigenvector. This is equivalent to the system of

equations

7a + 14b = 0
0 = 0.

Note that the matrix equation has reduced to essentially one equation and
two unknowns. Thus, we have a free choice of either a or b. If we choose a = 2,
then b = −1 and thus our eigenvector is v1 = (2, −1)T .
There are two things to note here. First, we are not allowed to choose
a = 0 because that forces b = 0 and and then v1 = 0. This is not an eigen-
vector because eigenvectors must be non-zero. Second, you might have made
138 Dynamical Systems on the Plane

a different choice for a than I did. You might have chosen to let a = 1, which
forces b = −1/2 giving v1 = (1, −1/2)T . But if we multiply your eigenvector
by 2 we get mine. This is what is meant when we say that eigenvectors are
unique up to scalar multiplication.
Exploration 1. Find the eigenvector v2 in the example above.

Exploration 2. Find the eigenvalues and eigenvectors of the matrix

 
1 −1 7
B= .
2 7 −1

11.2 Linear Systems with Real Eigenvalues

A linear dynamical system on R2 has the form

xn+1 = axn + byn

yn+1 = cxn + dyn

which can be expressed as the matrix equation

xn+1 = Axn (11.3)

where xn = (xn , yn )T and  

a b
A= .
c d
Our goal in this section is to explore the types of dynamics that occur
in equation 11.3. Naturally, the dynamics will depend in some way on the
entries of the matrix A and to keep things relatively straightforward we will
initially consider matrices A that have real eigenvalues. We consider matrices
with complex eigenvalues in section 11.3.

Explore

Exploration 3. Show that 0 is a fixed point of equation 11.3.

Exploration 4. Show that xn = An x0 . In other words, show that iteration

of the function is simply repeated matrix multiplication.
Linear Systems with Real Eigenvalues 139

In the next few exercises, you will look at several examples that illustrate
the most common types of dynamics that can occur in equation 11.3. You
should use what you have learned about linear systems on R to describe the
dynamics of these planar systems.
Exploration 5. Let  
.5 0
A= .
0 .2
1. Rewrite the dynamical system xn+1 = Axn as a system of equations
in terms of the coordinates xn and yn .
2. Describe the orbit of the initial condition (x0 , 0) for all x0 ∈ R.
3. Describe the orbit of the initial condition (0, y0 ) for all y0 ∈ R.
4. Use these results to describe the orbit of the initial condition (x0 , y0 )
for all x0 and y0 ∈ R.
5. Draw a picture in the (x, y) plane that illustrates the dynamics of
this system. Your picture should reflect all of the dynamical prop-
erties that you discovered above.

Exploration 6. Repeat exploration 5 with

 
1.5 0
A= .
0 1.2

Exploration 7. Repeat exploration 5 with

 
1.5 0
A= .
0 .2

The pictures that you sketched in the three explorations 5 through 7 are
called phase portraits. Like the graphical analysis figures that you used
to study real dynamical systems, these pictures illustrate the dynamics of
the system. However, because we are now working in two dimensions, we are
no longer able to use the graph of the function being iterated to help us
understand the dynamics. The best we can do is to plot the iterates as points
in the (x, y) plane.
These three examples illustrate three fundamental behaviors that occur in
linear systems. The phase portrait of exploration 5 illustrates an attracting
fixed point sometimes called a sink. Exploration 6 is a repelling fixed point
or a source. Finally, Exploration 7 has one attracting direction (the y-axis)
and one repelling direction (the x-axis). A fixed point such as this is called a
saddle.
Exploration 8. Let  
−.5 0
A= .
0 .2
140 Dynamical Systems on the Plane

The only difference between this matrix and that given in exploration 5 is that
the first non-zero entry is negative. How does this change the dynamics? In
particular, is the origin still a sink? What would happen if both entries were
negative? What would happen in the other examples if we changed the signs?
In every example above, the axes are invariant in the sense that if an initial
condition is on one of these lines, then the orbit of that point remains on that
line. This motivates the following definition.

Definition 11.2 Consider a linear dynamical system xn+1 = L(xn ) with L :

R2 → R2 a linear transformation. A line ` is invariant if for every x0 ∈ `,
Ln (x0 ) ∈ ` for all n ≥ 0.

Exploration 9. Let  
3/2 1
A= .
0 2

1. Describe the orbit of (x0 , 0) for all x0 ∈ R. In other words, describe

the orbits of points starting on the x-axis.
2. Describe the orbit of (x0 , 3x0 /2) for all x0 ∈ R. In other words,
describe the orbits of points starting on the line y = 3x/2.
3. Is the origin a sink, saddle, or source?
4. In the (x, y) plane, sketch the two lines above and indicate with
arrows whether points on them go toward the origin or away from
it. Use this picture and your intuition to describe orbits of other
initial conditions.
5. Compute the eigenvalues and eigenvectors of the matrix A. How do
these relate to the lines and arrows you just graphed?

Exploration 10. Let  

1/2 1
A= .
0 1/3

1. Describe the orbit of (x0 , 0) for all x0 ∈ R. In other words, describe

the orbits of points starting on the x-axis.
2. Describe the orbit of (x0 , −x0 /6) for all x0 ∈ R. In other words,
describe the orbits of points starting on the line y = −x/6.
3. Is the origin a sink, saddle, or source?
4. In the (x, y) plane, sketch the two lines above and indicate with
arrows whether points on them go toward the origin or away from
it. Use this picture and your intuition to describe orbits of other
initial conditions.
5. Compute the eigenvalues and eigenvectors of the matrix A. How do
these relate to the lines and arrows you just graphed?
Linear Systems with Real Eigenvalues 141

Triangular Matrices
You probably noticed that in the previous two exercises, the eigenval-
ues of A are also the diagonal entries of A. Hopefully, you also realized
that this is true because one of the two off-diagonal entries of the matrix
equals zero. Matrices where all of the entries below (or above) the diag-
onal are equal to zero are called triangular matrices. If a matrix A is
triangular, then the eigenvalues of A are the diagonal entries of A.
If the matrix is not triangular, then the eigenvalues are not the diag-
onal entries!

Conjecture

The exercises above point to a general theorem about linear dynamical

systems with real eigenvalues.
Conjecture 11.

Let A be a 2 × 2 matrix with real, distinct eigenvalues λ1 and λ2 with

corresponding eigenvectors v1 and v2 . Consider the dynamical system

xn+1 = Axn .

• The lines through the origin defined by the eigenvectors v1 and v2 are
.
• If then the origin is a sink.
• If then the origin is a source.

• If then the origin is a saddle.

Apply

For each of the matrices A in applications 12 through 16, consider the

dynamical system
xn+1 = Ax.
There are three tasks for each of these applications.
142 Dynamical Systems on the Plane

• find the eigenvalues and eigenvectors of A,

• classify the phase portrait as a sink, source, saddle, or neither, and
• sketch the phase portrait including the invariant lines.

3 1
 
/8 /8
Application 12. A = 1 3
/8 /8

5 7
 
/6 /6
Application 13. A = 7 17
/12 /12
 
8 −6
Application 14. A =
3 −1
 
6 −6
Application 15. A =
1 −1

5 −4
 
/3 /3
Application 16. A = −2 7
/3 /3

Prove

Theorem 11.1 Let A be a 2 × 2 matrix with real, distinct eigenvalues λ1

and λ2 having corresponding eigenvectors v1 and v2 . Consider the dynamical
system
xn+1 = Axn . (11.4)
• The lines through the origin defined by the eigenvectors v1 and v2 are in-
variant.
• If |λi | < 1 for i = 1, 2, then the origin is a sink.
• If |λi | > 1 for i = 1, 2, then the origin is a source.
• If |λ1 | < 1 < |λ2 |, then the origin is a saddle. Moreover, the eigenvector v1
defines the attracting direction and the eigenvector v2 defines the repelling
direction.

The following exercises will lead you through a proof of theorem 11.1.
Let’s begin by formally defining a “line through the origin defined by a vector
v.” Intuitively, we understand this to mean the line through the origin that
extends in the direction of v. Precisely, a line `(v) defined by a vector v is the
set
`(v) = (x, y) ∈ R2 | (x, y) = tv for some t ∈ R .

Linear Systems with Real Eigenvalues 143

Proof 17. Prove that if an initial condition x0 of equation 11.4 is a member

of `(vi ), then there exists a C ∈ R such that
x1 = Cvi .
Conclude that xn ∈ `(vi ) for all n ≥ 0.
Recall from linear algebra that the modulus or norm of a vector v is
defined by  
vx
q
|v| = = vx2 + vy2 .
vy

Proof 18. Suppose that |λi | < 1 and that x0 ∈ `(vi ). Prove that
lim xn = 0.
n→∞

This proves that the eigenvectors associated with real eigenvalues less than
one in absolute value define attracting invariant lines. Now let’s do a similar
proof for eigenvalues greater than one in absolute value.
Proof 19. Suppose that |λi | > 1 and that x0 ∈ `(vi ), x0 6= 0. Prove that
lim |xn | = ∞.
n→∞

Proof 20. Explain why these last three proofs prove theorem 11.1.
There is more that we can say about the orbits of initial conditions on an
invariant line. We can, in fact, easily compute the sequence of iterates of these
points.
Proof 21. Suppose that λi and vi are real eigenvalues and eigenvectors of
A and that x0 ∈ `(vi ). Prove that
xn = λni x0 .
Now let’s turn our attention to initial condition that are not on either
invariant line.
Proof 22. Explain why for all x0 ∈ R2 there exists unique constants C1 and
C2 such that
x0 = C1 v1 + C2 v2 .

Proof 23. Prove that

xn = C1 λn1 v1 + C2 λn2 v2 .

Proof 24. Prove if |λi | < 1 for i = 1, 2 then for all initial conditions x0
lim |xn | = 0.
n→∞

In other words, prove that if the origin is a sink, then all orbits converge to
it.
144 Dynamical Systems on the Plane

Proof 25. Suppose that |λi | > 1 and that x0 6= 0. Prove that

lim |xn | = ∞.
n→∞

This explains why we call the origin a source.

Proof 26. Why can’t we make a similar statement when |λ1 | < 1 < |λ2 |?
Why do you think this is called a saddle?

11.3 Linear Systems with Complex Eigenvalues

Because eigenvalues of 2 × 2 matrices are found by finding the roots of a
quadratic polynomial, they may either be real or come in complex conjugate
pairs. We will see that complex eigenvalues lead to somewhat different dy-
namics. The goal of this section is to begin the process of understanding the
dynamical similarities and differences of linear systems with complex eigen-
values as compared to those with real eigenvalues. √
Recall that a number z is complex if z = a + ib where we define i = −1.
A quadratic polynomial with real coefficients that has one complex root of
a + ib must have a second root of a − ib. We say that a + ib and a − ib are
complex conjugates. If z = a + ib, then we denote the complex conjugate
of z by z = a − ib.

Complex or Imaginary?
In my experience, students often confuse the terms complex and imag-
inary. A number z = a + ib is a complex number. The value a is called
the real part of z and the value b is referred to as the imaginary part
of z. If z = ib, then we say that z is pure imaginary. In short, 3 + 4i
is a complex number but 5i is pure imaginary.

Explore

Suppose that we want to construct a linear dynamical system

xn+1 = axn + byn

yn+1 = cxn + dyn

that rotates every point in the plane by some fixed angle φ. How do we choose
the constants in equation 11.5 to accomplish this goal? Given that we want
Linear Systems with Complex Eigenvalues 145

to rotate every point, it seems natural to represent our initial points in polar
form so that
xn = r cos θ
(11.5)
yn = r sin θ
for some fixed values of r and θ. We want the point (xn , yn ) to be rotated by
an angle of φ so that means that we need
xn+1 = r cos(θ + φ)
(11.6)
yn+1 = r sin(θ + φ).

Exploration 27. Look up the sum of angles formulas for sin(θ + φ) and
cos(θ + φ) and rewrite equation 11.6 using them.

Exploration 28. Use equation 11.5 to eliminate both r and θ from your
answer to exploration 27 and replace them with xn and yn .

Exploration 29. Finally, express this as the matrix equation

    
xn+1 cos φ − sin φ xn
= . (11.7)
yn+1 sin φ cos φ yn

Exploration 30. IBLdynamics.com Use the tool on the website to plot the
dynamics of equation 11.7 with
√
φ = π/4, φ = π/2, φ = π, φ = π/10, φ = π/ 2
and answer the following questions for each.
1. Do the dynamics that you see depend on the initial condition?
2. Is the fixed point (0, 0) a sink, a source, a saddle, or none of these?
3. Are the dynamics periodic or not? If they are what is the period
and how does this relate to φ?
Linear systems like the ones you’ve just explored are called centers. Points
simply rotate around the origin. These orbits do not tend toward the origin
nor away from the origin as n tends to infinity.
Exploration 31. Show that the eigenvalues of the matrix in equation 11.7
are
λ = cos φ ± i sin φ.
The modulus of a complex number z = α + iβ, denoted |z|, is defined by
p
|z| = α2 + β 2
and generalizes the idea of absolute value on R. To see this note that if z =
α + 0i, then |z| = |α|.
Exploration 32. Show that |λ| = 1 for the eigenvalues of exploration 31.
146 Dynamical Systems on the Plane

Let’s build on what you just learned and consider the dynamical system
    
xn+1 ρ cos φ −ρ sin φ xn
= (11.8)
yn+1 ρ sin φ ρ cos φ yn

where the parameter ρ > 0. Note that when ρ = 1, equation 11.8 reduces to
equation 11.7.
Exploration 33. IBLdynamics.com Use the tool on the website to draw a
variety of phase portraits for systems of the form given in equation 11.8.

1. Describe the dynamics when 0 < ρ < 1.

2. Is the fixed point (0, 0) a sink, a source, a saddle, or none of the
above?
3. Describe the dynamics when ρ > 1.
4. Is the fixed point (0, 0) a sink, a source, a saddle, or none of the
above?
The orbits that you constructed in exploration 33 are called either spiral
sinks or spiral sources depending on whether the orbit spirals in toward the
origin or away from the origin.
Exploration 34. Compute the eigenvalues of the matrix in equation 11.8.
What is their modulus?

Conjecture

Conjecture 35.

Consider the linear dynamical system on R2 defined by

xn+1 = Axn

and let A have complex eigenvalues λi = α ± iβ.

• If , then (0, 0) is a spiral sink.

• If , then is a spiral source.
• If , then (0, 0) is a center.
Linear Systems with Complex Eigenvalues 147

Apply

For each of the matrices A in applications 36 through 40, consider the

dynamical system
xn+1 = Ax.
There are three tasks for each of these applications.

• find the eigenvalues of A,

• classify the phase portrait as a spiral sink, spiral source, or center, and
• sketch the phase portrait including the invariant lines (feel free to use one
of the tools on the website for this).
 
3 −2
Application 36. A =
2 3

7 10
 
/3 /3
Application 37. A = −4 11
/3 /3
 
2 4
Application 38. A =
−2 −2

−3 −10
 
/4 /3
Application 39. A = 1 5
/3 /4
 
3 10
Application 40. A =
−1 −3

Prove

The work that you’ve done in this section points to an interesting fact:
complex eigenvalues correspond to rotational behavior in a dynamical system.
If λ1 and λ2 are complex conjugate pairs of eigenvalues of a dynamical system,
then they can be written so that

λi = ρ(cos φ ± i sin φ)

where ρ = |λi | is the modulus of λi and the angle φ is the argument of λi .

The argument φ determines the amount of rotation in each iteration. The
148 Dynamical Systems on the Plane

modulus ρ is the rate of contraction or expansion and if 0 < ρ < 1, then the
iterates tend toward the origin. We thus call this a spiral sink. If ρ > 1, then
the iterates tend away from the origin and we call this a spiral source. We
call the case where ρ = 1 a center.
The exercises in the apply section above show that this relationship be-
tween complex eigenvalues and the dynamic behavior is true even if the matrix
A is not of the form given in equation 11.8. The proof of this fact is not hard
but relies on some linear algebra ideas that are not discussed above and are
generally not covered in an undergraduate linear algebra course. The basic
idea is that a 2 × 2 real matrix with complex eigenvalues is similar to a ma-
trix of the form given in equation 11.8. This similarity is analogous to the
similarity of a matrix A with real distinct eigenvalues to a diagonal matrix D.
All of the results on linear dynamical systems in R2 discussed in this
chapter are summarized in the theorem below.
Theorem 11.2 Consider the linear dynamical system on R2 defined by
xn+1 = Axn
and let A have eigenvalues λ1 and λ2 .
• If |λi | < 1 for i = 1, 2, then (0, 0) is a sink. Moreover, if the imaginary part
of λi is not equal to 0, then (0, 0) is a spiral sink.
• If |λi | > 1 for i = 1, 2, then (0, 0) is a source. Moreover, if the imaginary
part of λi is not equal to 0, then (0, 0) is a spiral source.
• If |λ1 | < 1 < |λ2 |, then (0, 0) is a saddle.
• If |λi | = 1 for i = 1, 2 and the imaginary part of λi is not equal to 0, then
(0, 0) is a center.

11.4 Fixed Points of Nonlinear Systems

Now, let’s turn to nonlinear dynamical systems. Let (x, y) = x ∈ R2 . Then a
function F : R2 → R2 has the form
F (x, y) = F (x) = (f (x, y) , g (x, y)) . (11.9)
We then define dynamical systems in the usual way by
xn+1 = F (xn ). (11.10)
As usual, fixed points are found by solving F (x) = x. This reduces to solving
the system on equations
f (x, y) = x
(11.11)
g(x, y) = y
Fixed Points of Nonlinear Systems 149

for x and y simultaneously. There may be multiple solutions to these equations

and hence multiple fixed points for the dynamical system 11.10.

Explore

Let’s consider a model of a predator-prey system to illustrate some of

the techniques used to analyze a nonlinear dynamical system in the plane.
Denote the prey population at time n by xn and the predator population
by yn . Assume that in the absence of predators, the prey population grows
logistically. When predators are present, the prey are harvested at a rate
proportional to both the predator and prey population sizes. On the other
hand, we assume that in the absence of prey, the predator population decreases
at a rate proportional to the predator population size. When prey are present,
the predator population increases at a rate proportional to both the predator
and prey population sizes. One possible model that satisfies these requirements
is
xn+1 = axn (1 − xn ) − βxn yn
(11.12)
yn+1 = 45 yn + 3βxn yn
where a > 0 and β ≥ 0. Because this is a population model, we will assume
that x0 ≥ 0 and y0 ≥ 0.
Let’s start by simply connecting this system of equations with the notation
used in equation 11.9. In this example, xn = (xn , yn ) so that

f (x, y) = ax(1 − x) − βxy

4
g(x, y) = y + 3βxy,
5
and
F (x, y) = F (x) = (f (x, y), g(x, y)).
Let’s look at what happens when either the predator or prey population
is absent.
Exploration 41. Show that if y0 = 0, then yn = 0 for all n. Similarly,
show that x0 = 0 implies that xn = 0 for all n. Discuss why this property is
reasonable in this model. Explain why this property implies that the x and y
axes are invariant?
For the remainder of the exercises in this section, let a = 2.
Exploration 42. Use the techniques of one-dimensional dynamical systems
to describe the dynamics on each of the axes. In particular, show that (0, 0)
and (1/2 , 0) are fixed points.
1. Is (0, 0) attracting or repelling for initial conditions on the x-axis?
150 Dynamical Systems on the Plane

2. Is (1/2 , 0) attracting or repelling for initial conditions on the x-axis?

3. Is (0, 0) attracting or repelling for initial conditions on the y-axis?
4. Draw the first quadrant of the (x, y) plane and use arrows to indicate
the dynamics on each of these axes.
Next, we turn our attention to what happens when neither x nor y equals
zero. For simplicity, we will take β = 1/2 in what follows.
Exploration 43. Show that (2/15 , 22/15 ) is the only non-zero fixed point of
equation 11.12.
Let’s take a break here to review what we know about this dynamical
system and what we still need to do yet.
• We know that there are three fixed points at (0, 0), (1/2 , 0), and (2/15 , 22/15 ).
• It seems that the fixed point at (0, 0) is a saddle since initial conditions on
the y-axis have orbits that converge to it, while initial conditions on the
x-axis have orbits that tend away from it.
• We know that the fixed point at (1/2 , 0) is attracting in one direction (along
the x-axis). We don’t know what happens to initial conditions that start
near (1/2 , 0) but off the x-axis.

• We know nothing about the fixed point at (2/15 , 22/15 ).

Our primary goal is to close these gaps. Let’s begin this process by numerically
computing solutions and noting what we see.
Exploration 44. IBLdynamics.com The tool on the website models this
predator-prey system. Use it to answer the following questions.
1. There is a point in the tool that can be moved to choose a variety of
initial conditions. Do this and describe what you see. In particular,
how would you characterize the fixed point at (2/15 , 22/15 ) using the
terminology of theorem 11.2?
2. Use the movable point to choose initial conditions near the fixed
point at (1/2 , 0) and describe these orbits. How would you character-
ize the fixed point at (1/2 , 0) using the terminology of theorem 11.2?
3. Use the movable point to choose initial conditions very near the y-
axis (especially with y ≤ 1/2) and describe these orbits. How would
you characterize the fixed point at (0, 0) using the terminology of
theorem 11.2?
Fixed Points of Nonlinear Systems 151

Conjecture

In the predator-prey model of equation 11.12, you found three fixed points.
Two of them were saddles and the third was a spiral sink. Based on this,
you probably realize that all of the fixed point classifications described in
theorem 11.2 for linear systems also occur in nonlinear systems. This should
not be a surprise. We saw that this is the case for non-linear dynamical systems
on R. We showed that fixed points of linear dynamical systems on R of the
form
xn+1 = ax
are attracting if |a| < 1 and are repelling if |a| > 1. For nonlinear systems of
the form
xn+1 = f (xn )
we have an almost identical condition for a fixed point x̃. Except in that set-
ting, it is the value |f 0 (x̃)| that determines whether x̃ is attracting or repelling.
In short, the derivative evaluated at the fixed point determines stability.
The same thing holds for dynamical systems on R2 . The derivative of
F evaluated at the fixed point usually determines the nature of the fixed
point. But what is the derivative of a function F : R2 → R2 ? You learned in
multivariable calculus that the derivative of a function F : R2 → R2 is a 2 × 2
matrix called the Jacobian matrix. This matrix is denoted DF . The Jacobian
matrix consists of four partial derivatives of F and is given by
!
∂f ∂f
∂x ∂y
DF = ∂g ∂g . (11.13)
∂x ∂y

If x̃ is a fixed point of F , then DF (x̃) determines the stability of x̃.

Conjecture 45.

Suppose that F : R2 → R2 is differentiable and that F (x̃) = x̃.

• If , then the x̃ is a sink.
• If , then the x̃ is a source.
• If , then the x̃ is a saddle.
152 Dynamical Systems on the Plane

Apply

Let’s return to the predator-prey model we have been exploring.

Application 46. Compute the Jacobian matrix DF for equation 11.12.
1. Find DF (0, 0) and compute its eigenvalues to verify that (0, 0) is a
saddle.
2. Find DF (1/2, 0) and compute its eigenvalues to verify that (1/2, 0)
is a saddle.
3. Find DF (2/15, 22/15) and compute its eigenvalues to verify that
(2/15, 22/15) is a spiral sink. (We suggest using a computer algebra
system for this calculation.)

Application 47. IBLdynamics.com Now do another investigation of equa-

tions 11.12 but this time with β = 1/5.
1. Begin by using the predator-prey app on the website to simulate
this dynamical system to get a feel for what happens.
2. Explain why there are still fixed points at (0, 0) and (1/2, 0) and
why the dynamics on each axis is identical to the previous model.
3. Find the value of the third fixed point. How has reducing the value of
β affected the equilibrium predator and prey populations? Interpret
these changes in terms of the model.
4. Compute DF and evaluate it at this new fixed point. Find the eigen-
values. How has the stability changed if at all?

Application 48. Find the fixed points and compute their stability.
xn+1 = 2xn + 4yn − 23
yn+1 = −2xn − 2yn + 21

Application 49. Find the fixed points and compute their stability.
xn+1 = x2n + yn
yn+1 = 2xn − 3yn

Prove

Proving that the eigenvalues of the Jacobian matrix DF (x̃) determine the
stability of a fixed point is beyond the scope of this text. Below is a statement
of the theorem that justifies the use of this matrix to classify fixed points.
Periodic Points 153

Theorem 11.3 Consider the dynamical system on R2 defined by

xn+1 = F (xn )

and suppose that F (x̃) = x̃. Let DF (x̃) have eigenvalues λ1 and λ2 .
• If |λi | < 1 for i = 1, 2, then x̃ is a sink. Moreover, if the imaginary part of
λi is not equal to 0, then x̃ is a spiral sink.

• If |λi | > 1 for i = 1, 2, then x̃ is a source. Moreover, if the imaginary part

of λi is not equal to 0, then x̃ is a spiral source.
• If |λ1 | < 1 < |λ2 |, then x̃ is a saddle.

11.5 Periodic Points

Basically, everything that you already know about periodic points carries over
to dynamical systems on R2 . A point x0 is a period n point if

F n (x0 ) = x0 .

The orbit of x0 is the set of points {x0 , x1 , . . . , xn−1 } with F (xi ) = xi+1 and
F (xn−1 ) = x0 .
It is also true that the stability of this orbit is determined by the product
of the derivatives evaluated along the periodic orbit. But here we need to
be careful. For functions on the real line, these derivative values are scalars
and scalar multiplication is commutative. Thus, we don’t need to worry about
the ordering of the multiplication. However, this is not true on R2 . Matrix
multiplication is not commutative! If {x0 , x1 , . . . , xn−1 } is a periodic
orbit, then the stability is determined by the eigenvalues of

DF (xn−1 ) · DF (xn−2 ) · · · DF (x1 ) · DF (x0 ) (11.14)

in this order.

Apply

As you might imagine, equation 11.14 suggests that it is often quite difficult
to compute the stability of a periodic orbit of a planar dynamical system.
However, the predator-prey model that we have been working with does give
us an opportunity to use this idea in a setting that is fairly straightforward.
So, consider equations 11.12 with a = 3.2 and β = 0.5.
154 Dynamical Systems on the Plane

Application 50. Use what you already know about the logistic equation
to show that there exists a period 2 orbit of the form {(x0 , 0), (x1 , 0)} with
0 < x0 < x1 < 1. Explain why this orbit is attracting when restricted to the
x-axis. (You don’t need to compute the exact values of xi .)

Application 51. Compute DF (xi , 0) and show that it is an upper triangular

matrix for each i.

Application 52. Show, in general, that the product of two upper triangular
matrices is upper triangular. Observe that the diagonal entries of the product
matrix are the products of the diagonal entries.

Application 53. Recall that the eigenvalues of an upper triangular matrix

T are the diagonal entries. Let λ1 be the eigenvalue in the upper-left entry of
T and let λ2 be the eigenvalue in the lower-right entry of T . Show that one of
the eigenvectors of T is (1, 0)T . To which eigenvalue does this correspond to?
Show that the other eigenvector of T has a non-zero second entry.

Application 54. Now let’s put all of this together.

1. Compute M = DF (x1 , 0) · DF (x0 , 0).
2. Explain why the eigenvalue of M having the eigenvector (1, 0)T is
less than 1 in absolute value.
3. Show that the other eigenvalue of M is greater than one.
4. Conclude that this period 2 orbit is a saddle.
The following exercises concern an interesting dynamical system that pro-
duces something called the “gingerbread attractor” (see figure 11.1). The gin-
gerbread attractor is generated by the dynamical system

xn+1 = 1 − yn + |xn |
(11.15)
yn+1 = xn .

We won’t be studying this system in any great detail, but it is a fun example
that illustrates some of the basic ideas of this section.
Application 55. Show that the point (1, 1) is a fixed point. Compute the
Jacobian matrix DF (1, 1) and find the eigenvalues λi . Show that |λi | = 1.

Application 56. Show that the point (0, 0) lies on a period 6 orbit. Identify
the points of this orbit in figure 11.1.

Application 57. Show that the point (−1, −1) lies on a period 5 orbit.
Identify the points of this orbit in figure 11.1.
Chaos in the Hénon map 155
��� ����������� ���
�
�

�
-� � � � �

-�

FIGURE 11.1
The Gingerbread attractor.

Application 58. IBLdynamics.com Use the tool on the website to explore

the orbits of points inside the middle hexagon (the “heart” of the gingerbread
man). Describe these orbits.

Application 59. IBLdynamics.com Use the tool on the website to explore

the orbits of points inside the other hexagons (the “limbs” of the gingerbread
man). Describe these orbits. You might try to prove some of this as well. The
dynamics inside the “heart” would be a good place to start.

11.6 Chaos in the Hénon map

There is much we have left out of the story on dynamical systems in the plane.
For example, there was no discussion of bifurcations. But in this section,
we would like to briefly introduce the chaos part of the story as that most
closely parallels our study of one-dimensional dynamics. You will study chaos
in planar dynamical systems more carefully in chapter 12. There you will build
on the symbolic dynamics work of chapter 7 to develop a fuller picture of what
you will explore in this section.
The story of chaotic dynamics in the plane begins with a closer look at
saddle fixed points. In section 11.4, you learned that when a linear system is
a saddle, then there exists a stable invariant line Es and an unstable invariant
line Eu . When a nonlinear dynamical system has a saddle fixed point, it turns
out that there exists something quite similar. These are called stable and
unstable curves and are denoted by Ws and Wu , respectively. Locally, the
156 Dynamical Systems on the Plane

stable and unstable curves are tangent to the stable and unstable lines Es
and Eu defined by the Jacobian matrix DF evaluated at the fixed point. If
you’ve taken a differential equations course, this is identical to what happens
for saddle equilibrium points in that setting.
However, if we look globally, the stable and unstable curves of a fixed point
can interact in a surprising way. In some systems, the stable and unstable
curves can “cross” each other (this can’t happen in differential equations),
and this leads to a remarkable phenomenon known as a homoclinic tangle.
Homoclinic tangles are the hallmark of chaos in planar dynamical systems.
And although we will not discuss homoclinic tangles in any great detail, the
explorations below point at the complexity of this phenomenon in a dynamical
system known as the Hénon map.

Explore

The Hénon map is given by

xn+1 = 1 + y − ax2
(11.16)
yn+1 = bx.
We will take a = 1.4 and b = 0.3.
Exploration 60. IBLdynamics.com Use the tool on the website allows
you to iterate the Hénon map. Describe in your own words what orbits are
converging to. This object is known as the Hénon attractor.

Exploration 61. Show that the Hénon map has a fixed point at approxi-
mately
(x̃, ỹ) = (0.631355, 0.189405).

Exploration 62. Compute the Jacobian matrix of the Hénon map and
evaluate it at (x̃, ỹ). Compute the eigenvalues and show that this fixed point
is a saddle. Note: This problem should not be done by hand.
Recall that a primary reason that the logistic function is chaotic is that it
is a two-to-one function. A consequence of this is that it is not invertible. In
two dimensions, there is more “real estate” that allows a function to smoothly
stretch and fold a region while still being one-to-one. Each of the next two
exercises shows that the Hénon map is invertible and hence one-to-one. Thus,
something quite different is happening here that causes the apparent chaotic
behavior that you observed in exploration 60.
Exploration 63. Our first method of showing that the Hénon map is in-
vertible uses the inverse function theorem (4.4). Show that if b 6= 0, then the
determinant of the Jacobian matrix of the Hénon map is non-zero for all x
and y. Explain why this implies that the Hénon map is invertible.
Chaos in the Hénon map 157

Exploration 64. Let H(x, y) be the Hénon map of equation 11.16. Show
6 0,
that when b =  
y/b
H −1 (x, y) = .
x − 1 + ay 2 /b2
The fact that H is invertible and that we can compute H −1 gives us a
method of approximating the stable curve Ws and the unstable curve Wu .
The following outlines a method to approximate Wu .
1. Make a list of initial conditions near the equilibrium point (x̃, ỹ) on
Eu . We know these points are close to the unstable curve Wu since
Wu is tangent to Eu at (x̃, ỹ).
2. For each of these initial conditions, compute a list of iterates.
3. Plot all of these points.
We approximate Ws in an almost identical fashion. The first change is that
instead of taking initial conditions on Eu , we take them from Es . The second
is that instead of iterating H, we iterate H −1 .

� �

��� ���

��� ���

� �
-� -� � � -� -� � �

-��� -���

-��� -���

FIGURE 11.2
Images showing a homoclinic tangle in the Hénon map. The plot on the left
gives a local picture of Ws and Wu . The plot on the right is a more global
picture.

Figure 11.2 shows the result of this method. The plot on the left is the
“local” picture. The parabolic shaped curve is approximately the stable curve
Ws and the shorter, more horizontal curve is approximately Wu . Note that
Ws and Wu intersect at the fixed point. The figure on the right is the “global
picture” and gives a glimpse of the homoclinic tangle. Again, the parabolic
shaped curve is an approximation of Ws . Note that now Wu looks very much
like the Hénon attractor that you observed in exploration 60. To generate
this figure, we just computed a few more iterates of each initial condition
than we did for the figure on the left. In the “global figure,” we see multiple
intersections of Ws and Wu . This is the appearance of the homoclinic tangle
that creates the attractor and the associated chaotic dynamics.
158 Dynamical Systems on the Plane

One final observation: If we take any small box centered at a saddle point,
each iteration stretches that box (in the unstable direction Eu ) and compresses
that box (in the stable direction Es ). The Hénon map adds an additional
aspect to this transformation - a folding. If you look at an image of the Hénon
attractor you will note that, while not quite symmetric with respect to the
x-axis, it does appear to be folded across the x-axis. This leads to one final
question for you.
Exploration 65. What property of the Hénon map of equation 11.16 might
be responsible for this folding across the x-axis?
12
The Smale Horseshoe

12.1 Motivating the Horseshoe Map

We ended chapter 11 with a discussion of the Hénon map (equation 11.16),
homoclinic tangles, and saddles. We tried to imply in that discussion that
somehow the Hénon map H stretched, compressed, and folded the plane in a
way that created the Hénon attractor and chaos. The goal of this chapter is
to look at a topological model of this process and to use symbolic dynamics
to prove that this process does, in fact, lead to chaos.
Before doing this in any detail, let’s reflect back on two specific examples
of one-dimensional chaotic dynamics that we studied earlier. The doubling
map D(x) and the logistic function f4 (x) both exhibit chaotic behavior on
the interval [0, 1]. If you think back on your study of those examples, you
might see that the primary reason that both of these functions are chaotic is
that they are two-to-one functions on the entire unit interval. The fact that
almost every point in the unit interval has 2 preimages somehow causes the
interval to get so mixed up under iteration that the dynamics on this interval
are chaotic.
Let’s try and visualize how each of these two functions acts on the unit
interval. The action of the doubling map D can be viewed like this.
Step 1 Stretch the unit interval to twice its length.
Step 2 Cut this new interval in half.
Step 3 Place these two intervals back into the unit interval.
Step 4 Repeat, . . .

The action of f4 can be described similarly.

Step 1 Stretch the unit interval to twice its length.
Step 2 Fold this new interval in half.
Step 3 Place this folded interval back into the unit interval.
Step 4 Repeat, . . .

159
160 The Smale Horseshoe

If you’ve ever seen taffy being pulled, these processes (especially the second
one) are essentially a one-dimensional version of taffy pulling. But making
taffy is different in one fundamental way: you can’t put two different parts of
the taffy in exactly the same spot! It is not a two-to-one process. And that
suggests a deep question: can a one-to-one function be chaotic? If the function
f : R → R is continuous, then the answer is no. You should be able to prove
this fairly easily.
But as we mentioned in section 11.6, the plane has more real estate and a
function F : R2 → R2 can twist and turn a planar region in a lot of different
ways without mapping two points onto the same point. In this chapter, we
introduce the prototypical example of a chaotic dynamical system that is one-
to-one, onto, and continuous on an invariant set Λ ⊂ R2 .

12.2 The Horseshoe Map

The first well-understood example of a chaotic dynamical having the prop-
erties alluded to above was developed by Steve Smale, a professor at the
University of California, while he was on sabbatical in Rio de Janeiro in the
early 1960s. Although the term “chaos” had not yet been defined in the way
we use it now, Smale was essentially interested in constructing an example of a
chaotic dynamical that was both invertible and hyperbolic (that is, persistent
under small perturbations).
It is easiest and sufficient to describe the horseshoe map F in a step-by-
step manner like we did for the doubling map and the logistic function above.
The domain of the function is a rectangle with two half-disks attached to
either end. I’ll refer to this region as a “stadium” and denote it by S (see
figure 12.1). Let F : S → S be defined as follows:
Step 1 Take S and compress it vertically to less than half its height.
Step 2 Take this compressed version of S and stretch it to twice its length.
Step 3 Fold this compressed and stretched version of S into a “horseshoe”
shape and then place it back into S. This is F (S).
Figure 12.1 illustrates the domain S and its image F (S).

Explore

Let’s begin to explore some of the basic properties of the function F .

Exploration 1. Explain in general terms why F is continuous.
The Horseshoe Map 161

( )

FIGURE 12.1
The Smale horseshoe map F . The domain of the function is the stadium-
shaped region S. The image F (S) is shown in gray.

Exploration 2. Explain in general terms why F is one-to-one.

Exploration 3. Explain in general terms why F is not onto S.

Exploration 4. Suppose that x is in the left semicircular region of S. Where

is F (x)? What can you say about the orbit {F n (x)}?

Exploration 5. Show that F has a unique attracting fixed point in the left
semicircular region of S. What theorem did you use here?

Exploration 6. Suppose that x is in the right semicircular region of S.

Where is F (x)? What can you say about the orbit {F n (x)}?

Exploration 7. Are there points in the rectangular center region R whose

orbits converge to the attracting fixed point that you just described?

Exploration 8. Sketch F 2 (S). It should have 4 thin horizontal strips through

the center region R.

Exploration 9. Don’t try and sketch F 3 (S) unless you are very patient.
How many horizontal strips does it have through R? Generalize to F n (S).
These dynamics should remind of the logistic map with a > 4 that you
studied in section 6.4. In that system, you showed the existence of a closed,
invariant set Λ ⊂ [0, 1] by first showing that there was a single open interval
centered around x = 1/2 that left the interval in one iteration. Then you
showed that there were 2 open intervals that mapped into this center interval
and hence left in two iterations. And so on. You showed that what remained
162 The Smale Horseshoe

was a closed subset Λ ⊂ [0, 1] that is a Cantor set and if x ∈ Λ, then fan (x) ∈ Λ
for all n > 0.
The same idea is happening here except instead of open intervals; there
are open vertical strips that get mapped out of the center rectangle R.
Exploration 10. Draw a figure to explain why there are 3 vertical open
strips U` , Uc , and Ur (for left, center, and right) in the rectangle R such that

F (Ui ) 6⊂ R

for i ∈ {`, c, r}. Explain why there are 2 vertical strips V0 and V1 in R such
that F (Vi ) ⊂ R. Label these on your figure.

Exploration 11. Now explain why there are 3 vertical strips in V0 and 3
more in V1 such that if x is in one of these 6 strips, then F 2 (x) 6∈ R. How
many strips are there whose iterates remain in R for 2 iterations?

Exploration 12. How many vertical strips are there in R such that if x is
in one of these strips then F 3 (x) ∈ R? In general, How many vertical strips
are there in R such that if x is in one of these strips then F n (x) ∈ R?
As in section 6.4, this construction continues for all n leaving a set

Λ+ = {x ∈ R | F n (x) ∈ R for all n ∈ N} .

Exploration 13. Describe the topological structure of the set Λ+ .

In exploration 2, you explained why the function F is one-to-one. And
although F is not onto the entire domain (exploration 3), it is of course onto
its image. This implies that F is invertible when its domain and range are
appropriately restricted.
Exploration 14. Think about the original definition of the function F as
described in the three-step process at the beginning of this section and illus-
trated in figure 12.1. Can you come up with a similar description for F −1 ?
The invertibility of F is important. It tells us that the process used to
define Λ+ can be used with F −1 to define a similar set

Λ− = x ∈ R | F −n (x) ∈ R for all n ∈ N .



Exploration 15. Explain why Λ− is a Cantor set of horizontal lines.

The set that we are interested in is the set of points that are in both Λ+
and Λ− . Define
Λ = Λ+ ∩ Λ− .

Exploration 16. If x ∈ Λ, what can you say about both the forward and
backward orbits of x?
More Symbolic Dynamics 163

12.3 More Symbolic Dynamics

Our ultimate goal here is to do exactly what we did with the logistic function
and every other chaotic system that we studied earlier. We want to create a
topological conjugacy between F restricted to the set Λ with a shift map σ
defined on a symbol space Σ. In that way, one can prove that the horseshoe
map F is, in fact, chaotic. We will omit the technical details needed to es-
tablish this conjugacy. If you are interested, you can read these details in An
Introduction to Chaotic Dynamical Systems by Robert Devaney [2].

12.3.1 Two-Sided Sequence Space

You probably noticed that we vaguely said “some shift map” and “some sym-
bol space” and that was intentional. The invertibility of F allows us to use a
variation on σ and Σ2 that is more appropriate and therefore highlights some
additional interesting dynamical properties. The key observation to defining
the conjugacy, and hence the symbol space, is that the invariant set Λ lies
entirely in the two vertical strips V0 and V1 . Thus, just as we did in chap-
ter 8, we want to associate with each x ∈ Λ a sequence of 0s and 1s based
on the itinerary of x. However, the invertibility of F implies that both the
forward and backward orbits of x remain in Λ. In other words, the sequence
should have both negatively and positively indexed entries. This leads to the
following revised definition of the doubly infinite sequence space Σ2 :
Σ2 = {. . . s−2 s−1 .s0 s1 s2 . . . | sj = 0 or 1} . (12.1)
Note that there is a dot (.) in these sequences between the negatively indexed
entries and the non-negatively indexed entries. This dot, referred to as the
center of the sequence, will be important in discussing dynamics on this
space.

Explore

Exploration 17. Give examples of several different points in Σ2 .

This sequence space Σ2 is again a metric space. However, the metric is
slightly different than it was in our original sequence space. The metric d is
defined by
∞
X |sk − tk |
d(s, t) = . (12.2)
k=−∞
2|k|
You have probably not encountered an infinite series that runs from −∞ to ∞
before. The easiest way to use this formula is to break it up into two different
164 The Smale Horseshoe

infinite series and then use the geometric series formula to evaluate each of
them.
Exploration 18. Show that equation 12.2 can be rewritten as
∞ ∞
X |s−k − t−k | X |sk − tk |
d(s, t) = + .
2k 2k
k=1 k=0

Exploration 19. Let s1 = . . . 00.11 . . . , s2 = . . . 11.00 . . . and t =

. . . 00.00 . . . . Compute d(s1 , t) and d(s2 , t). Are they the same or different?
Why?

Exploration 20. What is the maximal distance between any two points in
Σ2 ?

Prove

Proof 21. Prove that d, as defined in equation 12.2, is a metric. The prop-
erties of a metric are given in subsection 7.3.1.
Theorem 7.1 provided us with a natural way to see when two one-sided
sequences s and t are near to each other. It stated that if s and t are two
one-sided sequences such that si = ti for i = 0, 1, . . . , n, then d(s, t) ≤ 21n .
This needs to be modified for two-sided sequences.
Proof 22. Prove the following theorem.
Theorem 12.1 Let s, t ∈ Σ2 and suppose that si = ti for i = −n, . . . , n.
Then
1
d(s, t) ≤ n−1 .
2

Proof 23. Prove the following theorem.

Theorem 12.2 If d(s, t) < 1/2n then si = ti for i = −n, . . . , n.

12.3.2 The Two-Sided Shift Map

As mentioned previously, our old shift map σ doesn’t capture all of the prop-
erties of the horseshoe map. This is because that shift map is two-to-one and
not one-to-one. For example,

σ(.01) = σ(.1) = .1.

More Symbolic Dynamics 165

But, in our new version of Σ2 , we can define a one-to-one version of the shift
map σ : Σ2 → Σ2 by

σ(. . . s−2 s−1 .s0 s1 s2 . . . ) = . . . s−2 s−1 s0 .s1 s2 . . . . (12.3)

Notice that the center (i.e., the “dot”) has moved one unit to the right or
equivalently, the sequence has shifted one unit to the left. Hence, the name
shift map.

Explore

Exploration 24. What is the definition of σ −1 ?

Exploration 25. What are the fixed points of σ? What are the period 2
points of σ? In general, what are the period n points of σ?
Because σ is invertible, the dynamical system defined by it is reversible;
we can iterate both forward and backward. This leads to a refined definition
of an orbit.

Definition 12.1 The forward orbit of a point x0 in a reversible dynamical

system xn+1 = f (xn ) is

O+ (x0 ) = {f n (x0 ) : n ≥ 0} .

The backward orbit of a point x0 is

O− (x0 ) = {f n (x0 ) : n < 0} .

The orbit of x0 is O(x0 ) = O+ (x0 ) ∪ O− (x0 ).

Exploration 26. Construct a point s ∈ Σ2 having a dense forward orbit.

Exploration 27. Construct a point s ∈ Σ2 having a dense backward orbit.

Exploration 28. Construct a point s ∈ Σ2 having a dense forward orbit

that does not have a dense backward orbit.

Exploration 29. Give an example to show that σ has sensitive dependence

on initial conditions.

Exploration 30. Give an example to show that σ −1 has sensitive dependence

on initial conditions.
166 The Smale Horseshoe

Exploration 31. In section 11.6, we mentioned that homoclinic orbits are

often an indicator of chaotic dynamics. Give an example of a point s ∈ Σ2
that has a homoclinic orbit. More specifically, construct s 6= . . . 00.00 . . . such
that
lim σ n (s) = . . . 00.00 . . .
n→∞

and
lim σ n (s) = . . . 00.00 . . . .
n→−∞

Prove

Proof 32. Prove that σ as defined in equation 12.3 is one-to-one.

Proof 33. Prove that σ as defined in equation 12.3 is onto Σ2 .

Proof 34. Prove that periodic points of σ are dense in Σ2 .

Proof 35. Prove that σ has sensitive dependence on initial conditions.

Proof 36. Explain why σ is chaotic.

12.4 A Horseshoe in the Hénon Map

We motivated our investigation into the Smale horseshoe with a brief discus-
sion of Hénon map given in equation 11.16. So let’s finish this chapter by
explicitly seeing this connection.
Exploration 37. Figure 12.2 shows a rectangle R and its image H(R) by
the Hénon map. Compare figure 12.2 with figure 12.1 which we used to define
the Smale horseshoe map F . What are the similarities and differences?

Exploration 38. Figure 12.3 shows a rectangle R and its image H 2 (R) by
the Hénon map. Compare figure 12.3 with the Smale horseshoe figure that
you created in exploration 8. What are the similarities and differences?
A Horseshoe in the Hénon Map 167

H(R)

FIGURE 12.2
A region R (dashed box) and its image H(R) via the Hénon map.

H 2 ( R)

FIGURE 12.3
A region R (dashed box) and its image H 2 (R) via the Hénon map.
13
Generalized Symbolic Dynamics

13.1 Topology Foundations

In this section, we revisit the shift map from chapters 6, 7, and 12. We expand
on those concepts to symbol spaces having more than two symbols. We will see
how symbolic dynamics are not just a tool to study other dynamical systems,
but are interesting dynamical systems in their own right. We also explore
some of the tools use to study these systems. Students interested in a more
complete overview of these topics should consult Ref. [6].
To begin, we introduce a few notions from topology that will be needed
later. Recall from Chapter 7 the definition of a metric. Given a set M , a metric
is a map
d : M × M → R,
which satisfies the following properties:
1. d(s, t) = d(t, s).
2. d(s, t) ≥ 0 for all s and t with equality if and only if s = t.
3. d(s, t) ≤ d(s, u) + d(u, t) for all s, u and t.
The ordered pair (M, d) is called a metric space.
We will be looking at shift dynamical systems similar to those that you
already studied and will require an additional topological property on the sym-
bol space known as compactness. The following definition is not the general
topological definition of compactness, but in a metric space, it is equivalent.

Definition 13.1 A metric space M is compact if every sequence in M has

a convergent subsequence.

We have used the term open set earlier in the text when we referred to open
intervals or open sets in Rn . The following definitions extends this concept to
general metric spaces.

Definition 13.2 Given a metric space (M, d), an open ball of radius ,
centered at a point x is the set

B (x) = {y ∈ M : d (x, y) < } .

169
170 Generalized Symbolic Dynamics

A subset U ⊂ M is open if, for every u ∈ U , there exists some , such that
B (u) ⊂ U .
Additionally, both the metric space M and the empty set ∅ are defined to
be open sets.
If we have an open set, we will also need a definition for a closed set.
Definition 13.3 A subset V ⊂ M is closed if its complement is open in M .
In any metric space, there exists sets that are both open and closed
(unimaginatively called “clopen”). Keep that in mind as you go through this
chapter. Open sets and closed sets aren’t mutually exclusive.
Sometimes, you start with an open set, and you want to make it closed.
Definition 13.4 Given an open set U , its closure U is the smallest closed
set containing U .
The one theorem in this section is commonly proven in a course on real
analysis or topology. If you haven’t seen this proof before, it’s worth giving it
an attempt. The fact that this is the only proof assigned here should give you
some insight into its use later in the chapter.

Prove

Proof 1. Let M be a compact metric space. A subset N ⊂ M is compact if

and only if N is closed.

Topology and Topological Spaces

The primary focus of topology is to determine when one mathematical
object can be continuously deformed into another. For example, a cube
can be continuously deformed into a sphere and so we say that these
objects are “topologically equivalent.” These mathematical objects are
called topological spaces. Topological spaces consist of a set X and a
collection of subsets U of X that are called the open sets of X. This
collection of open sets must satisfy three axioms that specify how the sets
of U relate to each other. All other properties of topological spaces derive
from these axioms and other structures that might be inherent to X. For
example, metric spaces possess an additional property that allows one
to compute the distance between points in X and this gives additional
structure to the space.
If you have not yet taken a topology course, we highly recommend
taking one, especially if you are considering graduate school in mathe-
matics.
Shift Dynamical Systems 171

13.2 Shift Dynamical Systems

In chapter 7, we defined the one-sided sequence space on two symbols (Σ2 )
and the shift map σ : Σ2 → Σ2 . We then expanded this to a two-sided shift
in chapter 12. Metric spaces such as these two are often referred to as shift
spaces. Here, we generalize these ideas to sequences having more than two
symbols and explore some of the uses of this generalization.

13.2.1 One-Sided Shift Spaces

We naturally adapt definition in 7.3.1 to involve n symbols. We still start our
indexing at 0 so that

Σn = {.s0 s1 s2 . . . | si ∈ {0, 1, 2, . . . , n − 1} for all i} . (13.1)

Following the same logic, we extend the metric defined in 7.3.1 to this
space in the natural way. Define d : Σn × Σn → R by
∞
X |sk − tk |
d(s, t) = . (13.2)
nk
k=0

Finally, we define the shift map σ : Σn → Σn by

σ(.s0 s1 s2 . . . ) = .s1 s2 . . . .

Explore

Exploration 2. Prove that for all s, t ∈ Σn , d(s, t) ≤ n.

Exploration 3. Describe the periodic points of the shift map σ in Σ3 . How

many periodic points of period p are there?

Exploration 4. Prove that the periodic points of the shift map σ are dense
in Σ3 .

Exploration 5. Prove that the shift map σ exhibits sensitive dependence on

initial conditions Σ3 .

Exploration 6. Construct a point s ∈ Σ3 that has a dense orbit under

iteration by σ.

Exploration 7. Discuss how these questions do or do not generalize to Σn .

172 Generalized Symbolic Dynamics

Exploration 8. Show that there is a one-to-one correspondence between the

interval [0, 1) ∈ R and the one-sided shift on 10 symbols. What arithmetic
operation in R would be represented by the shift map in this case?

13.2.2 Two-Sided Shift Spaces

We now move on to a two-sided shift, but this time we don’t demand that our
symbols be natural numbers.
Consider a finite set A of symbols, which we call the alphabet.

Definition 13.5 The full shift on A is denoted

AZ = {. . . s−2 s−1 .s0 s1 s2 . . . : si ∈ A} . (13.3)

The dot to the left of the 0-th term is the origin. In the case that the
alphabet is the set {0, 1, . . . , n − 1}, the space is denoted the full n-shift.
A word or a block in A is any finite list of symbols chosen from A. The
number of symbols in a block u is its length, denoted |u|. Given s ∈ AZ , we
use s[i,j] to represent the block of s from the i-th to the j-th entry of s.
Again, you have encountered this concept previously in this book. Consider
the point s = . . . 01.01 . . . in the two-sided shift space Σ2 of chapter 12.
This sequence is constructed by recursively concatenating the word u = 01.
Additionally, |u| = 2. Of course, this is not the only word in s. It might be
helpful to list all of the words of length 2 and of length 3 that appear in s.
We will often want to consider shifts where certain words or blocks do
not appear. To do that, we first let F be a set of words constructed from
the alphabet A that we call the set of forbidden blocks. We then define
XF ⊂ AZ to be the set of two-sided sequences s ∈ AZ which do not contain
any of the words in F. The set XF is also a shift space.
The shift map from chapter 7 can be extended to shifts on spaces with
forbidden blocks in the natural way, with σ : AZ → AZ given by

σ(. . . s−2 s−1 .s0 s1 s2 . . .) = . . . s−1 s0 .s1 s2 s3 . . . .

The action of the shift map is the same as before. What can be different is
the underlying shift space. Because of this, it is common to use a subscript on
the function name to indicate the underlying shift space.
When we use the term shift dynamical system, we mean a shift space
X and an associated shift map σX . We write this as an ordered pair (X, σX ).
It is sometimes the case that it is easier to describe a shift space by what is
allowed, rather than what is not allowed (i.e., the set of forbidden blocks). Let
X be a subset of a full-shift, and Bn (X) the set of all n-blocks which appear
in any element of X. The language of X is the set
∞
[
B(X) = Bn (X).
n=0
Shift Dynamical Systems 173

.
Our next step is to construct a useful metric on these shift spaces. The gen-
eralization to arbitrary symbols, combined with a restricted language, makes
our original metric in equation 13.2 problematic. So let’s consider a different
metric, ρ : XF × XF → R defined by

2−k

6 t[−k,k] and s[−i,i] = t[−i,i] for all i < k}
if s[−k,k] =
ρ(s, t) =
0 if s = t
(13.4)
In proof 28, you will prove that ρ satisfies the properties of a metric. For now,
note that like all other metrics that we have used thus far, points are “close”
if they agree near the origin (i.e., the decimal point).
It can be confusing when we try to write sequences of points in shift spaces
since each point is also a sequence. Going forward, we’ll be using a subscript
to indicate the coordinate in a point and a superscript to indicate the index
in a sequence of points. Thus, s4 gives the 4th entry of the sequence s, s4
gives the 4th element of the sequence {sk }, and s44 gives the fourth entry of
the 4th element of the sequence {sk }. This is consistent with how we denoted
elements in a shift space in chapter 7, but inconsistent with the notation in
chapter 2. We apparently can’t make shift spaces without breaking some eggs,
so we’ve made a choice of which egg to break.
For example, if we define the sequence {xn } in the one-sided shift on two
symbols as the sequence of points of all terms 0, except for a 1 in coordinate
n, we would have

x0 =.10000 . . .
x1 =.01000 . . .
x2 =.00100 . . .
..
.

Then, we would use the notation x20 = x21 = 0, and x22 = 1. Remember, we
start indexing the sequence at 0, starting to the right of the origin.

Explore

Exploration 9. Consider the full 2-shift. What is the alphabet? What are
the forbidden blocks? Are you restricted in what symbols you use?

Exploration 10. Given an arbitrary element s ∈ AZ , how would you calcu-

late |s[i,j] |?
174 Generalized Symbolic Dynamics

Exploration 11. In the shift space {a, b, c}Z , find the distance ρ between
the given points.
1. . . . abbac.abbac . . . and . . . aabac.abbbc . . .
2. . . . abbac.abbac . . . and . . . aabac.abcbb . . .
3. . . . abbac.abbac . . . and . . . abbac.cbbac . . .

Exploration 12. Build a sequence in a shift space on two symbols following

the construction below. Let
x0 = . . . 00.00 . . .
x1 = . . . 001.100 . . .
x2 = . . . 0011.1100 . . .
and so forth. To what does the sequence {xn } converge? Can you prove this
using definition 2.2?

Exploration 13. Consider the shift space {0, 1, 2}Z , and the sequence
x0 = . . . 000.000 . . .
x1 = . . . 001.100 . . .
x2 = . . . 0012.2100 . . .
x3 = . . . 0120.0210 . . .
..
.
List x4 through x6 . Does this sequence converge? Why or why not?

Exploration 14. Is there a convergent subsequence of the sequence in ex-

ploration 13?

Exploration 15. Does the sequence in {0, 1, 2}Z given below converge? Does
it have a convergent subsequence?
x0 = . . . 1010.0101 . . .
x1 = . . . 2121.1212 . . .
x2 = . . . 0202.2020 . . .
x3 = . . . 1010.0101 . . .
..
.

Given a shift dynamical (AZ , σ), and some subset X ⊆ AZ , X is shift

invariant if σ(X) = X. In other words, if x is any element in X, σ(x) ∈ X
as well.
Exploration 16. Are full shifts shift invariant? Are shift spaces shift invari-
ant?
Shift Dynamical Systems 175

Exploration 17. Construct each of the following.

• A subset of the full-shift on 2 symbols which is shift invariant. This must
contain more than 2 points.
• A subset of the full-shift on 3 symbols which is shift invariant. This must
contain more than 3 points.
• A non-empty alphabet A such that every subset of shift space is shift in-
variant.

Exploration 18. Look at the shift space {0, 1}Z . Show that the set
{. . . s−2 s−1 .s0 s1 . . . | s−2 = s−1 = s0 = s1 = 0 and si ∈ {0, 1} otherwise}
is both open and closed.

Exploration 19. Suppose you were to change the values on s−2 through s1
in exploration 18. Pick a few different choices for 0 and 1 for each of these
four terms around the origin. Show that the sets are still always open.

Exploration 20. For any number of the sets you described in exploration 19,
show that their unions are also open.

Conjecture

Conjecture 21.

Suppose you have a shift space on n symbols, and a sequence {xn }. Then
sequence {xn } must/must not have a convergent subsequence.

Conjecture 22.

Suppose you have a shift space on n symbols, and a subset S of that

space. Then S must/must not be shift invariant.

Conjecture 23. What does it mean for a set to be open in a shift space?
Choose a shift we’ve already discussed, describe an open set, and conjecture
a good definition for describing these sets in general.
To check the validity of your conjecture, verify that unions of open sets
are open using your definition. Similarly, verify that finite intersections of
closed sets are closed using your definition. (These are required properties of
a topological space.)
176 Generalized Symbolic Dynamics

Apply

In the course of proving Sarkovskii’s Theorem, we noticed that when there

was a period 3 orbit, there exists a pair of disjoint open intervals I0 and I1
such that f (I0 ) = I1 and f (I1 ) = I0 ∪ I1 . Suppose that we assign a sequence
of 0’s and 1’s to represent the itinerary of a point x ∈ I0 ∪ I1 . Call the set of
sequences constructed in this manner X.
Application 24. Describe the set of forbidden words F.

Application 25. List all words in B1 (X)? List all words in B2 (x)? List all
words in B3 (x)?

Application 26. Describe the language B(X).

Apply

Application 27. In this application, you will prove that a shift space X is
compact. The proof uses a variation of the Cantor diagonalization argument
that is commonly used to prove that the real numbers are uncountable. You
may wish to review that proof before working through the steps outlined
below.
1. Assume you have a shift space X with an alphabet A and some
sequence of points {xn }∞n=0 in X. It will be essential to remember
that a shift space has a finite alphabet.
2. Now, consider the set {xn0 }. This is a sequence of characters in
A, made from the 0-th coordinate of every point in the sequence
{xn }∞
n=0 . Explain why there exists a c0 ∈ A and an infinite set
N0 ⊂ N such that xk0 = c0 for all k ∈ N0 .
3. Next, show that there exists some infinite set N1 ⊂ N0 , and some
c−1 , c1 ∈ A, such that, for all k ∈ N1 , the block between index -1
and 1 satisfies
xk[−1,1] = c−1 .c0 c1 .

4. Continue in this manner to inductively construct the sets Ni and

the associated blocks xk[−i,i] . What is the length around the origin
on which all blocks agree for elements of {xn } with indices n ∈ Ni ?
Shift Dynamical Systems 177

5. Let c = . . . c−2 c−1 .c0 c1 c2 . . . constructed from the inductive steps

described above. Explain why your inductive argument gives the
existence of the point in c ∈ X.
6. Define the sequence
{mi }∞
i=0 ⊂ N

with m0 =min(N0 ), m1 =min(N1 ), etc. Prove that the subsequence

{xmi }∞
i=0 converges to c.
7. Explain why the existence of this convergent subsequence implies
that X is compact.

Prove

Proof 28. Prove that the function ρ defined in equation 13.4 is a metric.

Proof 29. In chapter 7 you proved that the one-sided shift map on Σ2 is
continuous. Using the metric ρ defined in this chapter, prove that the shift
map σ defined on AZ is continuous.

Proof 30. Let X be a shift space, u an element of the language B(X), and
k ∈ Z. A cylinder set in X is defined as the set

CkX (u) = x ∈ X : x[k,k+|u|−1] = u .



We interpret CkX (u) as the set of all points which the block u occurs
starting at index k. We say this cylinder set is centered at u and of radius
k.
1. Prove that a cylinder set in a shift space is open.
2. Prove that a cylinder set in a shift space is closed.

Proof 31. Follow the outline below to show that a subset X of a AZ is a

shift space if and only if it is shift invariant and compact.
1. Begin by assuming that X is a shift space. Use results from earlier
in this section to show it is shift invariant and compact.
2. Now, assume that x is a shift invariant and compact subset of some
full shift. What other property of this set do you have from theo-
rem 1? What does this imply about the set AZ \ X?
3. Show that, for any point y ∈ AZ \ X, there must be a block uy in y,
AZ
of length 2k, such that the cylinder set C−k (uy ) is a subset of AZ .
4. Let F = {uy : y ∈ AZ \ X}. Prove that X = XF .
178 Generalized Symbolic Dynamics

13.2.3 Shifts of Finite Type

Now, that we have discussed shift spaces, we will restrict ourselves to a specific
kind of shift, called a shift of finite type. These show up regularly when we’re
finding shift spaces that are conjugate to dynamical systems.

Definition 13.6 If the set F is finite, the shift space XF is a shift of finite
type (SFT).

Let’s look at an example. Suppose we take a shift space on two symbols,

0 and 1 so that the alphabet is A = {0, 1}. We’ll take our forbidden set of
words to be F = {000, 111}. Note that this rules out the case of more than
three consecutive 0’s or 1’s. So the point . . . 01001.101101 . . . is an element of
XF , but the point . . . 01001.110101 . . . would be forbidden.
We may need to add one additional restriction. When constructing points
in a shift of finite type, we would like to determine what can occur next in a
string, based on the past m elements.

Definition 13.7 A shift of finite type X is an m-step shift if X = XF

where F contains only blocks of length m + 1.

This allows us to consider only potential blocks of the same length as those
in the set of forbidden blocks.

Explore

Exploration 32. What are the forbidden blocks in a full shift on n symbols?

Exploration 33. Is the empty set, ∅, a shift of finite type? Is it a shift space?

Exploration 34. In the example following definition 13.6, find m so that

the shift space that is an m-step shift of finite type.

Exploration 35. Let A = {a, b} and F = {bb}. Consider the 2-step shift of
finite type XF .
1. Find all possible blocks of length 2. Since this is a 2-step shift,
remember that all blocks of length greater than 2 must be made up
of these blocks of length 2.
2. Find all of the blocks of length 3 in this space.
3. Find all of the blocks of length 4 in this space.
4. Could you reconstruct this space, changing the number of steps,
where all blocks in F are of length 3? Of length 4? What do you
think would be the advantage or disadvantage of doing so?
Shift Dynamical Systems 179

Exploration 36. Revisit chapter 8 and proofs 30 through 45 regarding the

logistic map with a > 4. We’ll build a an associated shift space.
1. What is the alphabet?
2. Is the shift one-side or two-sided?
3. What are the forbidden words?
4. How would you describe a general element of this shift space? Relate
the shift map to the application of the logistic map with appropriate
choice of a.

Exploration 37. Consider the shift space on A = {0, 1}, where for any point
in the space, each pair of 1’s must be separated by an even number of zeros.
So . . . 100100.001 . . . would be allowed, but . . . 10010.001 . . . would not. Is this
a shift of finite type?

Conjecture

Recall the definition of a topological conjugacy from section 8.3 and con-
sider this commutative diagram in figure 13.1. Assume that X is and Y is
a shift space. F1 is the set of forbidden blocks in X, and F2 is the set of
forbidden blocks in Y . We assume S is a conjugacy.
σX
XF1 XF1
S S
σY
YF2 YF2

FIGURE 13.1
A commutative diagram between shift spaces XF1 and YF2 .

Conjecture 38. If a shift space is conjugate to a shift of finite type, must it

be a shift of finite type?

Apply

Application 39. Begin with two intervals √

on the real line (which we will
call tiles), labeled a and b. a has length 1+2 5 and b has length 1.
180 Generalized Symbolic Dynamics

1. Place a copy of a to the left of the origin and a copy of b to the

right of the origin.
√
2. Multiply the length of each tile by 1+2 5 . You should think of this
as stretching out the tiles, while keeping the origin between them.
Explain why we can now relabel the expanded a tile as ab and the
expanded b tile as a.
3. Repeat this process a few times. You should now be generating a
two-sided sequence of the symbols a and b.
4. You have generated a tiling of the real line. Putting the origin
between the a and b tiles is not special. It could be in the middle
of a tile. Could it be between two of the same tile?
5. Look at the sequences you generated. Is there a set of forbidden
blocks? Does this resemble a shift space you have already seen?
6. What might the relationship be between this geometric construction
and a shift of finite type?

The Golden Ratio

The shift of finite type in exploration 35 is sometimes called the
golden ratio shift or the Fibonacci shift.
The golden ratio is defined to be the ratio a/b that has the additional
property
a a+b
= .
b a
This ratio is denoted
√
ϕ and called the golden ratio. A little algebra
1+ 5
shows that ϕ = 2 .
The golden ratio shows up in a surprisingly wide variety of math-
ematical applications. For example, a regular pentagram inscribed in a
regular pentagon exhibits the golden ratio. It has long been an important
concept in art, architecture, and other disciplines. It is believed that the
Parthenon in Athens was designed around the golden ratio.
One of the most interesting places that this remarkable constant arises
is in the Fibonacci sequence where it can be shown that the ratio of
consecutive Fibonacci numbers approaches ϕ. This is related to the fact
that the golden ratio, when expressed as a continued fraction, has the
form
1
ϕ=1+ .
1 + 1+ 1 1
1+···

For this reason, the shift space in exploration 35 is usually called the
Fibonacci shift, and the tiling in application 39 is called the Fibonacci
tiling.
Representing Shift Spaces with Graphs 181

13.3 Representing Shift Spaces with Graphs

It is possible to represent a shift of finite type visually with a finite directed
graph. We discuss some of the basic notation here and illustrate their use with
some applications of this technique.
We can think of a graph G as two sets, called vertices (V ) and edges (E),
where each edge has a direction associated with it so that you can “travel”
from vertex to vertex within the graph by moving along connecting edges
only in the indicated direction. If the sets V and E are relatively small, then a
graph can be represented graphically with each vertex a circle and each edge
a curve connecting the appropriate vertices. (Technically, we are describing
what is known as a directed graph or digraph. However, in the context of
dynamics, the directed edges are essential.)
In addition to the sets, we need two functions to keep track of our edges,
which are called t and i. Both are functions from the set of edges to the set of
vertices, and tell us where an edge ends (t for terminal) and where it starts (i
for initial).
Another way of describing the connectivity of a graph is by using an ad-
jacency matrix. An adjacency matrix AG of a graph G is the matrix whose
(m, n)th entry gives the number of edges which start at vertex m and end at
vertex n.
Consider the following basic example. Let G be a graph with vertex set
V = {0, 1} and edge set E = {e0 , e1 , e2 } and adjacency matrix
 
0 1
AG = . (13.5)
1 1

The graph G is shown in figure 13.2. In this example, the initial and terminal
functions of edge e0 are i(e0 ) = 0 and t(e0 ) = 1, respectively.

e0
0 1 e2
e1

FIGURE 13.2
Digraph with adjacency matrix 13.5.

So, what does this have to do with dynamical systems and shift maps? We
have a finite set of edges we’re moving along and we can track which edges
we use, moving forward or backward in time. Thus, tracking the sequence of
edges traversed gives a shift space!
182 Generalized Symbolic Dynamics

Definition 13.8 Given a graph G, we define the edge shift as a set

XG = {. . . e−2 e−1 .e0 e1 e2 . . . | ei ∈ E (G) , t (ei ) = i (ei+1 )} .

The last condition checks that the edges in the sequence actually can follow
each other in the directed graph.

Similarly, we can also track the sequence of vertices visited. This is known
as a vertex shift.

Definition 13.9 Let A be a n × n adjacency matrix with all entries either

0 or 1. Let G be its associated graph. The vertex shift is a shift space with
alphabet {1, 2, . . . , n}, denoted

bG = . . . x−2 x−1 .x0 x1 x2 . . . | Ax ,x = 1 .
X i i+1

Explore

Exploration 40. Draw the graph for each of the following adjacency matri-
ces. Describe the edge shift for each graph. Describe the vertex shift for each
graph.
1.  
1 1
1 1

2.  
0 1
2 0

3.  
1 2 1
1 0 1
1 0 0

Exploration 41. How is the adjacency matrix given in equation 13.5 and
its associated graph in figure 13.2 related to a period 3 orbit of a continuous
dynamical system on R?

Exploration 42. For each edge shift in exploration 40, can you find either
a language or a set of forbidden blocks? What about for the vertex shift?

Exploration 43. Construct an example of a graph that does not have an

edge shift. Find its adjacency matrix.
Representing Shift Spaces with Graphs 183

Exploration 44. For the graph associated with equation 13.5, find the
eigenvalues of the adjacency matrix. What do you suppose is the relation-
ship between these values, the shift space in exploration 35 and the tiling in
application 39?

Exploration 45. We claim there is no edge shift conjugate to the shift of

finite type in exploration 35. If each of the symbols in exploration 35 were an
edge, how many vertices would there be? Could you find a word in the edge
shift that is forbidden in the shift space?

Conjecture

Conjecture 46.

An edge shift with a single vertex and n edges is conjugate to .

Conjecture 47. Take a graph from an earlier exercise, and it’s adjacency
matrix A. We know what information about the graph is available from A.
Calculate A2 , A3 , and A4 . What information about the graph is contained in
Ak ? In particular, what do the diagonal entries compute?

Apply

Application 48. In exploration 41, you made the connection between pe-
riod 3 orbits and the adjacency matrix of equation 13.5. How can you use
conjecture 47 to count the number of period k orbits?

Application 49. The trace of a matrix M is the sum of the diagonal entries
of M . For the adjacency matrix of equation 13.5, compute the trace of Ak for
k = 1, 2, . . . , 5. What is the pattern of these numbers? What does that tell
you about periodic orbits?
184 Generalized Symbolic Dynamics

13.3.1 Higher Edge Graphs

We saw in exploration 45 that an edge shift, which shares all of the same
symbols as a shift of finite type may not account for the entire language or
all of the forbidden words. The higher block code and higher edge graph is a
tool developed to handle this limitation.
We denote a higher block code for a shift X as X [N ] . By this, we mean
a shift constructed on a new alphabet, whose elements are the blocks of length
N in the original shift space X.

Definition 13.10 Given a graph G and N ≥ 1, the N -th higher edge

graph G[N ] is a graph whose vertex set is the collection of paths of length
N − 1 in G. Let e11 e12 . . . e1N −1 and e21 e22 . . . e2N −1 be two vertices in G[N ] . We
define exactly one edge in G[N ] whenever e12 . . . e1N −1 = e21 e22 . . . e2N −2 . In the
case N = 2, this condition is replaced with t(e11 ) = i(e21 ).

Note that in a higher edge graph, there is at most one edge between any
pair of vertices. We can use this when creating a shift directly from an adja-
cency matrix.
Let’s look at an example to motivate this definition. Start with the graph
G in figure 13.3. This should look familiar.

e0
0 1 e2
e1

FIGURE 13.3

It has vertex set 0 and 1. The first higher edge graph G[1] has a vertex set
made up of paths of length 0. We’ll say a path of length 0 is just a vertex so
that by definition G[1] = G.
Now, construct G[2] . The paths of length 1 in G define the new vertex set.
In other words,  
V G[2] = {e0 , e1 , e2 } .

Definition 13.10 tells us how to construct the edges between these vertices.
From the graph of G, we see that after traversing e0 , we could either go on
edge e1 or edge e2 . In other words,

t(e0 ) = i(e1 ) = i(e2 ).

So, we need an edge from node e0 to node e1 and a second edge from node
e0 to node e2 . We label these edges e0 e1 , and e0 e2 . Similarly, we need edges
e1 e0 , e2 e1 , and e2 e2 . This gives the graph G[2] as shown in figure 13.4.
Representing Shift Spaces with Graphs 185
e2 e2

e2

e0 e2 e2 e1

e0 e1

e0 e1
e1 e0

FIGURE 13.4
Second higher edge graph of the graph G in figure 13.2.

Explore

Exploration 50. Using the example from figures 13.3 and 13.4, construct
G[3] . To get you started, the nodes of this graph are the edges of G[2] and so
there are 5 of them.

Exploration 51. Write a few elements from the shift spaces associated with
G, G[2] , and G[3] . Does taking a higher edge graph cause any forbidden words
to appear in elements of the shift space?

Exploration 52. For each graph in exploration 40, find the alphabet of the
associated edge shift. You can use any labeling you like. Find the alphabets
for the 2nd and 3rd higher block codes for each shift you’ve described.

Exploration 53. For each graph in exploration 40, find the adjacency matrix
of the 2nd higher edge graph.

Exploration 54. We saw in exploration 45 that the shift of finite type in

exploration 35 was not conjugate to the edge shift we’d looked at earlier.

1. Call the shift in exploration 35 X. Write out the elements of the

alphabet for the higher block code X [2] .
2. Show that, with careful relabeling, this is conjugate to an edge shift
on the graph G from the start of this section.
186 Generalized Symbolic Dynamics

Exploration 55. Draw a directed graph using at least 3 vertices and at least
6 edges. No vertex should be isolated and you should have at least once edge
leaving each vertex. Self-loops are allowed.
In a graph, a cycle is a path that starts and ends at the same vertex.
1. Write the adjacency matrix A for the graph you’ve drawn. Calculate
the trace of A (the sum of the diagonal entries) and count the
number of self-loops you’ve included.
2. Multiply the matrix by itself to get A2 . For each vertex, count the
number cycles using exactly two edges. Compute the trace of A2 .
3. Try this with a few higher powers on the matrix. For each Ak ,
calculate by hand the number of cycles of length k for each vertex,
then calculate the trace of Ak .
4. How does this relate to periodic orbits?

Conjecture

Conjecture 56.

For N ≥ 2, only the numbers will appear in the adjacency

matrix for G[N ] .

Conjecture 57.

The trace of the k-th power of the adjacency matrix gives you the number
of in the associated edge shift.

Apply

Application 58. Draw the directed graph G for the adjacency matrix
 
2 3
.
1 1

Label the vertex on the left v and the vertex on the right w. Label the two
self-loops on v as a and b and the three edges from v to w as c, d, and e. Label
the edge from w to v as f and the self-loop on w as g.
Representing Shift Spaces with Graphs 187

1. Introduce a probability function P (G) → [0, 1]. Let P (v) = 1/3 and
P (w) = 2/3.
2. Repeatedly use the random number generator on the website to
choose a value between 1 and 3. This process generates an element
of the shift space associated with this graph. Think of this as moving
through the graph. If you get a 1, you’re at v, if you get a 2 or 3,
you’re at w.
3. Generate a few elements of the vertex shift, using these probabili-
ties to choose which vertex is next in the sequence. Are there any
forbidden words?
4. We next want to assign probabilities to each edge. We first introduce
the idea of conditional probability. Let P (E|V ) be the probability
of taking edge E given that you are at vertex V .
(a) Which conditional probabilities need to be zero?
(b) Choose probabilities for each edge such that
X
P (E|V ) = 1 for all V ∈ V (G).
E∈E(G)

(c) Use the random number generator on the website to construct

edge sequences. Are there any forbidden words? Do you see
any relationship between the vertex sequences and the edge
sequences?
5. Define the probability of an sequence of n edges to be

P (E1 E2 . . . En ) = P (i(E1 ))P (E1 |i(E1 ))P (E2 |i(E2 )) · · · P (En |i(En )).

6. Show that the set of possible sequences are exactly the blocks in
the language for the edge shift XG .
What you have constructed is an example of a Markov Chain.

Prove

Proof 59. Let G be a graph. Recalling our notation, XG is the associated

edge shift and (XG )[N ] is the N -th higher block code of that shift. G[N ] is the
N -th higher edge graph and XG[N ] is its associated edge shift. Prove that

(XG )[N ] = XG[N ] .

Proof 60. Prove that, up to renaming of symbols, the 1-step shifts of finite
type are vertex shifts.
188 Generalized Symbolic Dynamics

Proof 61. Prove that, up to a renaming of symbols, every edge shift is a

vertex shift (of a different graph).

Proof 62. Prove that if X is an M -step shift of finite type, there is a graph
G such that X [M ] = XbG and X [M +1] = XG . That is, the M -th higher block
code is the same as the vertex shift of G and the M + 1-st higher block code
is the same as the edge shift of G.

Proof 63. Prove that the i, j entry of the k-th power of an adjacency matrix
gives the number of paths involving k edges when moving from vertex i to
vertex j.

Proof 64. Prove that the trace of the k-th power of the adjacency matrix
gives you the number of points of period k in the associated edge shift.

13.4 Markov Partitions

In this section, we will assume we have a discrete dynamical system on a
metric space X given by some function f : X → X. We will also assume that
the map f is invertible so that the notation f n (x) is defined for any x ∈ X
and any integer n. We write (X, f ) to denote such a dynamical system.
The idea in this section is to divide X into a finite number of parts and keep
track of what part a point in X lands in after n iterations of the dynamical
system. We are tracking the orbit of a point, as in definition 12.1, and looking
at where points land, as we did in sections 8.2 and 12.2.

Definition 13.11 A topological partition P on a metric space X is a

finite collection of disjoint open sets P = {P1 , P2 , . . . , Pn } , whose closures
cover X. That is, X = P1 ∪ · · · ∪ PN .

Once we have such a partition, we can use it to build a symbolic represen-

tation of a dynamical system.

Definition 13.12 Suppose that we have a dynamical system (X, f ), a topo-

logical partition P on X, and indexing set {1, 2, . . . , n} for the open sets in
the partition. A word a1 a2 a3 . . . ar is allowed for P if
r
\
f −1 (Pj ) 6= ∅.
j=1

Definition 13.13 Let LP,f be the set of all allowed words for partition P on
(X, f ). The shift space XP,f with language LP,f is the symbolic dynamical
system corresponding to P.
Markov Partitions 189

There is an issue in definition 13.12 associated with the use of f −1 . In partic-

ular, the inverse images of single points may not be unique. However, we will
restrict our work here only to invertible functions to avoid this complication.
For a given invertible dynamical system (X, f ) and a topological partition
P for each x ∈ XP,f , define
n
\
Dn (x) = f −1 (Pxj ) ⊆ X.
j=−n

Remember that XP,f is a symbolic dynamical system, so the coordinate

xj indicates which set in the topological partition the point x ended up after
j iterations of f .

Definition 13.14 For an invertible dynamical system (X, f ), the topological

partition P gives a symbolic representation of (X, f ) if for every x ∈ XP,f
the set
\∞
Dn (x)
n=0

is a single point. If, in addition, XP,f is a shift of finite type, P is a Markov

partition.

Explore

Exploration 65. Let M be the unit circle. We think of this as just the
interval [0, 1), but where any function acting on a point in M has its result
taken mod 1 as we did with the doubling map D.
Define the dynamical system φ : M → M by φ(x) = 10x (mod 1).
i i+1
• Show that subdividing M into the intervals ( 10 , 10 ) for i ∈ 0, 1, . . . , 9 is a
topological partition.
• For a given point in M , show that tracking which set in the partition a point
lands under this map gives an element of a one-sided shift. Where do you
encounter an issue?
• Is this a Markov partition?
• Relate this to the shift you described in exploration 8.

Exploration 66. Given a symbolic representation of a dynamical system

(X, f ), define π : XP,f → X to be the map which sends a point in the
symbolic representation to the unique point in ∩∞
n=0 Dn (x). Prove that the
following diagram commutes.
190 Generalized Symbolic Dynamics
σ
XP,f XP,f
π π
f
X X

Apply

Application 67. If you have not done so, we recommend you review the
materials in section 11.1 prior to working out this application.
Define the torus T2 = R2 /Z2 . You can think of this as the subset of the
plane [0, 1) × [0, 1) where any linear transformation applied to points in the
set is resolved by taking each coordinate modulo 1.
1. Define a linear transformation on R2 with the matrix
 
1 1
A= .
1 0
Compute the eigenvalues and eigenvectors of this matrix.
2. Let  
x
φ(x, y) = A ,
y
with addition modulo 1. Let π : R2 → T2 be given by

π(x, y) = (x (mod 1), y (mod 1)).

Prove that the diagram below commutes.

A
R2 R2
π π
φ
T2 T2

3. Let L1 be the span of the vector

 
1√
v1 =
− 1−2 5

and L2 , the span of the vector

 
1√
v2 = .
− 1+2 5
Markov Partitions 191

L2 (0, 1)
P2 (1, 1) L1

P1

P3

(−1, 0)

4. Begin with the three rectangles P1 , P2 , P3 in the figure below. De-

scribe how they were drawn.
5. Let Pi0 = π(Pi ). This gives us the following figure. Explain how
this figure was drawn and prove that this is a Markov partition for
dynamical system (T2 , φ).

(0, 1) (1, 1)

P20
P10

P10

P30 P30
P20

(0, 0) (1, 0)

Application 68. Perform the same process as above, using the matrix
 
2 1
.
1 0
192 Generalized Symbolic Dynamics

Construct the Markov partition and the symbolic representation of the map
on the torus. Show that this map is chaotic.
This is more commonly known as Arnold’s cat map and is an example
of a hyperbolic toral automorphism.
Bibliography

[1] Robert Devaney. A first course in chaotic dynamical systems: Theory and
Experiment. Perseus Books, 1992.

[2] Robert Devaney. An introduction to chaotic dynamical systems. CRC

Press, 2018.
[3] James Gleick. Chaos: Making a new science. Open Road Media, 2011.
[4] Victor Guillemin and Alan Pollack. Differential topology, volume 370.
American Mathematical Soc., 2010.
[5] TY Li and JA Yorke. Period 3 implies chaos. Am. Math. Mon., 82:985–992,
1975.
[6] Douglas Lind and Brian Marcus. An introduction to symbolic dynamics
and coding. Cambridge University Press, 1995.
[7] EN Lorenz. Deterministic nonperiodic flow. J. Atmos. Sci., 20:130–141,
1963.
[8] AN Sharkovskiı̆. Coexistence of cycles of a continuous map of the line into
itself. Int. J. Bifurcat. Chaos, 5(05):1263–1273, 1995.

193
Index

basin of attraction, 53 eigenvalue, 135

eigenvector, 135
adjacency matrix, 181 equilibrium point, 4
alphabet, 172
argument, 147 Fibonacci sequence, 11
Attracting Fixed Point Theorem, 40 fixed point, 4, 33
attracting, 33
basin of attraction, 120 center, 148
bifurcation, 47 directionally attracting, 45
diagram, 118 hyperbolic, 64
saddle node, 72 neutral, 34, 44
tangent, 72, 75 non-hyperbolic, 64
bifurcation diagram, 70 repelling, 33, 41
binary, 89 saddle, 139
periodic sequence, 91 sink, 139
Brouwer fixed point theorem, 53 source, 139
spiral sink, 148
Cantor Set, 94 spiral source, 148
center, 145, 148 stable, 33
chaos, 103 unstable, 33, 41
characteristic equation, 136 weakly attracting, 45
closed set, 170 weakly repelling, 45
closure, 170 full shift, 172
cobweb diagram, 6
compact, 169 geometric series, 88
complete, 31, 59 gingerbread attractor, 154
Completeness Axiom, 22 global dynamics, 81
continuity, 100 golden ratio, 19
contraction, 59 graph, 181
contraction mapping theorem, 53, 59 graphical analysis, 6
converge, 12 graphical iteration, 6
cylinder set, 177
Hénon attractor, 156
dense, 101 Hénon map, 156
doubling map, 86 higher block code, 184
dynamical system, 1 higher edge graph, 184
homeomorphism, 108
edge shift, 182 homoclinic tangle, 156

195
196 Index

hyperbolic fixed point, 64 periodic point, 5

hyperbolic fixed point theorem, 67 attracting, 50
prime, 5
Implicit Function Theorem, 63 repelling, 50
intermediate value theorem, 126 phase portrait, 139
invariant line, 140, 143 population dynamics, 8
Inverse Function Theorem, 61 predator-prey, 149
multivariable, 61 predator-prey model, 149
single variable, 61
iteration, 2 quadratic family, 69
itinerary, 115
regular point, 61
Jacobian matrix, 151 Repelling Fixed Point Theorem, 43

language, 173 Sarkovskii’s theorem, 130

lim inf, 25 Schwarzian Derivative, 119
lim sup, 25 sensitive dependence, 103
limit, 12 sequence, 12
linear transformation, 135 bounded, 19
local dynamics, 81 Cauchy, 29
logistic function, 37 monotone, 20
logistic model, 39 sequence space, 96
doubly-infinite, 163
Markov Partition, 189 shift dynamical system, 172
mean reproductive rate, 9 shift map, 99
metric, 97 two-sided, 164
metric space, 169 shift of finite type, 178
complete, 59 shift space, 171, 172
modulus, 143 shift-invariant, 174
complex, 145 Smale Horseshoe, 160
spiral sink, 146, 148
norm, 143 spiral source, 146, 148
Squeeze Theorem, 17
one-to-one, 108
stable curve, 155
onto, 108
stair-step diagram, 6
open set, 170
state space, 1
orbit, 2, 106
structural stability, 64
backward, 165
subsequence, 23
dense, 106
forward, 165 Tangent Bifurcation Theorem, 75
ternary, 96
period 2 point, 5
topological conjugacy, 108
prime, 5
topological partition, 188
Period Doubling Bifurcation
topological transitivity, 105
Theorem, 80
triangular matrix, 141
periodic orbit, 48
Index 197

unstable curve, 155

vertex shift, 182

viability, 9