Explorations in Complex Analysis

MAA-Complex Analysis back_Layout 1 9/7/12 2:13 PM Page 1
Michael A. Brilleslyper
Michael J. Dorff
Jane M. McDougall
James S. Rolf
MAA
Explorations
Lisbeth E. Schaubroeck
Richard L. Stankewitz
Kenneth Stephenson in Complex
Explorations in Complex Analysis

Michael A. Brilleslyper Michael J. Dorff Jane M. McDougall James S. Rolf Lisbeth E. Schaubroeck Richard L. Stankewitz Kenneth Stephenson
Explorations in Complex Analysis Analysis
Complex analysis is a core area in mathematics: here hard analysis meets
beautiful geometry and many applications find their pure mathematics Michael A. Brilleslyper
footing. It is a standard topic in mathematics curricula and recommended for Michael J. Dorff
the mathematically inclined in science and engineering. Unlike most texts, this Jane M. McDougall
James S. Rolf
book focuses squarely on student-driven, technology-enhanced, and visually-
oriented investigations. Students can explore topics from complex dynamics
and minimal surfaces to fluid flow and circle packing with exercises and Kenneth Stephenson
explorations explicitly linked to computer applets supplied with the text.
ISBN: 978-0-88385-355-9
9 780 883 85 355 9
Classroom Resource Materials
Classroom Resource Materials

i i
“ECA˙Book” — 2012/8/28 — 13:58 — page i — #1

i i
Explorations
in
Complex Analysis
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page ii — #2

i i

c 2012 by the Mathematical Association of America, Inc.
Library of Congress Catalog Card Number 2012943816

Print edition ISBN 978-0-88385-778-6
Electronic edition ISBN 978-1-61444-108-3
Printed in the United States of America
Current Printing (last digit):
10 9 8 7 6 5 4 3 2 1
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page iii — #3

i i
Explorations in Complex Analysis
Michael A. Brilleslyper
United States Air Force Academy
Michael J. Dorff
Brigham Young University
Jane M. McDougall
Colorado College
James S. Rolf
Yale University
United States Air Force Academy
Ball State University
Kenneth Stephenson
University of Tennessee, Knoxville
Published and Distributed by

The Mathematical Association of America
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page iv — #4

i i
Council on Publications and Communications

Frank Farris, Chair
Committee on Books
Gerald M. Bryce, Chair
Classroom Resource Materials Editorial Board
Gerald M. Bryce, Editor
Michael Bardzell
Jennifer Bergner
Diane L. Herrmann
Philip P. Mummert
Barbara E. Reynolds
Susan G. Staples
Philip D. Straffin
Cynthia J Woodburn
Holly S. Zullo
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page v — #5

i i
CLASSROOM RESOURCE MATERIALS

Classroom Resource Materials is intended to provide supplementary classroom material
for students—laboratory exercises, projects, historical information, textbooks with unusual
approaches for presenting mathematical ideas, career information, etc.
101 Careers in Mathematics, 2nd edition edited by Andrew Sterrett

Archimedes: What Did He Do Besides Cry Eureka?, Sherman Stein
Calculus: An Active Approach with Projects, Stephen Hilbert, Diane Driscoll Schwartz,
Stan Seltzer, John Maceli, and Eric Robinson
Calculus Mysteries and Thrillers, R. Grant Woods
Conjecture and Proof, Miklós Laczkovich
Counterexamples in Calculus, Sergiy Klymchuk
Creative Mathematics, H. S. Wall
Environmental Mathematics in the Classroom, edited by B. A. Fusaro and P. C. Kenschaft
Excursions in Classical Analysis: Pathways to Advanced Problem Solving and Undergrad-
uate Research, by Hongwei Chen
Explorations in Complex Analysis, Michael A. Brilleslyper, Michael J. Dorff, Jane M. Mc-
Dougall, James S. Rolf, Lisbeth E. Schaubroeck, Richard L. Stankewitz, and Kenneth
Stephenson
Exploratory Examples for Real Analysis, Joanne E. Snow and Kirk E. Weller
Geometry From Africa: Mathematical and Educational Explorations, Paulus Gerdes
Historical Modules for the Teaching and Learning of Mathematics (CD), edited by Victor
Katz and Karen Dee Michalowicz
Identification Numbers and Check Digit Schemes, Joseph Kirtland
Interdisciplinary Lively Application Projects, edited by Chris Arney
Inverse Problems: Activities for Undergraduates, Charles W. Groetsch
Keeping it R.E.A.L.: Research Experiences for All Learners, Carla D. Martin and Anthony
Tongen
Laboratory Experiences in Group Theory, Ellen Maycock Parker
Learn from the Masters, Frank Swetz, John Fauvel, Otto Bekken, Bengt Johansson, and
Victor Katz
Math Made Visual: Creating Images for Understanding Mathematics, Claudi Alsina and
Roger B. Nelsen
Mathematics Galore!: The First Five Years of the St. Marks Institute of Mathematics, James
Tanton
Methods for Euclidean Geometry, Owen Byer, Felix Lazebnik, Deirdre L. Smeltzer
Ordinary Differential Equations: A Brief Eclectic Tour, David A. Sánchez
Oval Track and Other Permutation Puzzles, John O. Kiltinen
A Primer of Abstract Mathematics, Robert B. Ash
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page vi — #6

i i
Proofs Without Words, Roger B. Nelsen

Proofs Without Words II, Roger B. Nelsen
Rediscovering Mathematics: You Do the Math, Shai Simonson
She Does Math!, edited by Marla Parker
Solve This: Math Activities for Students and Clubs, James S. Tanton
Student Manual for Mathematics for Business Decisions Part 1: Probability and Simula-
tion, David Williamson, Marilou Mendel, Julie Tarr, and Deborah Yoklic
Student Manual for Mathematics for Business Decisions Part 2: Calculus and Optimiza-
tion, David Williamson, Marilou Mendel, Julie Tarr, and Deborah Yoklic
Teaching Statistics Using Baseball, Jim Albert
Visual Group Theory, Nathan C. Carter
Which Numbers are Real?, Michael Henle
Writing Projects for Mathematics Courses: Crushed Clowns, Cars, and Coffee to Go,
Annalisa Crannell, Gavin LaRose, Thomas Ratliff, Elyn Rykken
MAA Service Center

P.O. Box 91112
Washington, DC 20090-1112
1-800-331-1MAA FAX: 1-301-206-9789
i i
i i
i i
“ECA˙Book” — 2012/9/5 — 14:07 — page vii — #7

i i
Contents
Introduction xi
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
Using Java Applets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi
Using Links in the Electronic Book . . . . . . . . . . . . . . . . . . . . . . . . xviii
1 Complex Dynamics: Chaos, Fractals, the Mandelbrot Set, and More 1

Richard L. Stankewitz (text and software design)
James S. Rolf (software coding and design)
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Newton’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Iteration of an Analytic Function . . . . . . . . . . . . . . . . . . . . . . 19
1.4 Critical Points and Critical Orbits . . . . . . . . . . . . . . . . . . . . . . 35
1.5 Exploring the Mandelbrot Set . . . . . . . . . . . . . . . . . . . . . . . . 40
1.6 Transcendental Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1.7 The Mandelbrot Set is Universal . . . . . . . . . . . . . . . . . . . . . . 61
1.8 Concluding Remarks and New Directions . . . . . . . . . . . . . . . . . 62
1.9 Additional Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
1.10 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
1.A Appendix: Definitions and Properties of the Julia and Fatou Sets . . . . . 78
1.B Appendix: Global Conjugation and Möbius Map Dynamics . . . . . . . . 80
1.C Appendix: Code for Drawing Random Dynamics Pictures . . . . . . . . . 81
2 Soap Films, Differential Geometry, and Minimal Surfaces 85

Michael J. Dorff (text)
James S. Rolf (software)
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.2 Differential geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
2.3 Minimal surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
2.4 Weierstrass representation . . . . . . . . . . . . . . . . . . . . . . . . . 116
2.5 The Gauss map, G, and height differential, dh . . . . . . . . . . . . . . . 125
2.6 Minimal surfaces and harmonic univalent mappings . . . . . . . . . . . . 139
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
2.8 Additional exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
2.9 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
vii
i i
i i
i i
“ECA˙Book” — 2012/9/5 — 14:07 — page viii — #8

i i
viii Contents
3 Applications to Flow Problems 161

Michael Brilleslyper (text)
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

3.2 Background and Fundamental Results . . . . . . . . . . . . . . . . . . . 164
3.3 Complex Functions and Vector Fields . . . . . . . . . . . . . . . . . . . 167
3.4 Complex Potential Functions . . . . . . . . . . . . . . . . . . . . . . . . 172
3.5 Uniform Flows in the Plane and other Regions . . . . . . . . . . . . . . . 174
3.6 Sources and Sinks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
3.7 Flow in a Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
3.8 Flows in Other Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
3.9 Flows inside the Disk . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
3.10 Interval Sources and Sinks . . . . . . . . . . . . . . . . . . . . . . . . . 183
3.11 Steady State Temperature Problems . . . . . . . . . . . . . . . . . . . . 186
3.12 Flows with Source and Sinks not on the Boundary . . . . . . . . . . . . . 190
3.13 Vector Fields with Other Types of Singularities . . . . . . . . . . . . . . 192
3.14 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
4 Anamorphosis, Mapping Problems, and Harmonic Univalent Functions 197

Michael J. Dorff (text)
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

4.2 Anamorphosis and Möbius Maps . . . . . . . . . . . . . . . . . . . . . . 198
4.3 The Family S of Analytic, Normalized, Univalent Functions . . . . . . . 208
4.4 The Family SH of Normalized, Harmonic, Univalent Functions . . . . . . 214
4.5 The Shearing Technique . . . . . . . . . . . . . . . . . . . . . . . . . . 221
4.6 Properties of the Dilatation . . . . . . . . . . . . . . . . . . . . . . . . . 233
4.7 Harmonic Linear Combinations . . . . . . . . . . . . . . . . . . . . . . 240
4.8 Convolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
4.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
4.11 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
5 Mappings to Polygonal Domains 271

Jane McDougall and Lisbeth Schaubroeck (text)
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271

5.2 Schwarz-Christoffel Maps . . . . . . . . . . . . . . . . . . . . . . . . . 272
5.3 The Poisson Integral Formula . . . . . . . . . . . . . . . . . . . . . . . . 284
5.4 Harmonic Function Theory . . . . . . . . . . . . . . . . . . . . . . . . . 291
5.5 Rado-Kneser-Choquet Theorem . . . . . . . . . . . . . . . . . . . . . . 296
5.6 Star Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
5.7 Dilatations of Polygonal Maps are Blaschke Products . . . . . . . . . . . 304
5.8 An Important Univalence Theorem . . . . . . . . . . . . . . . . . . . . . 309
i i
i i
i i
“ECA˙Book” — 2012/9/5 — 14:07 — page ix — #9

i i
Contents ix
5.9 The Dilatation for Star Mappings . . . . . . . . . . . . . . . . . . . . . . 311

5.10 Open Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
5.11 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
6 Circle Packing 317

Ken Stephenson (text and software)
6.1 First Impressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318

6.2 Basics of Circle Packing . . . . . . . . . . . . . . . . . . . . . . . . . . 322
6.3 Circle Packing Manipulations . . . . . . . . . . . . . . . . . . . . . . . . 325
6.4 Discrete Function Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 327
6.5 Function Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
6.6 Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
6.7 Wrapup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
6.9 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
A Background 345
A.1 Functions of a Complex Variable as Mappings . . . . . . . . . . . . . . . 345
A.2 Continuity and Analyticity in C . . . . . . . . . . . . . . . . . . . . . . 348
A.3 Contour Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
A.4 Taylor Series and Laurent Series . . . . . . . . . . . . . . . . . . . . . . 350
A.5 Key Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
A.6 More Advanced Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
A.7 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
B The Riemann Sphere 357

B.1 Stereographic Projection and Spherical Geometry . . . . . . . . . . . . . 357
B.2 The Spherical Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
B.3 Topology in C and C . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
B.4 Continuity on the Riemann Sphere . . . . . . . . . . . . . . . . . . . . . 361
B.5 Analyticity on the Riemann Sphere . . . . . . . . . . . . . . . . . . . . . 362
B.6 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
Index 365
Index of Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Index of Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
About the Authors 371
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page x — #10

i i
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page xi — #11

i i
Introduction
This book is written for undergraduate students who have studied some complex analysis
and want to explore additional topics in the field. It could be used as
a supplement for an undergraduate complex analysis course allowing students to
explore a research topic;
a guide for undergraduate research projects for advanced students;
a resource for senior capstone courses; or
a portal for the mathematically curious, a hands-on introduction to the beauties of
complex analysis.
This book differs from other mathematics texts. It focuses on discovery, self-driven
investigation, and creative problem posing. The goal is to inspire students to investigate,
explore, form conjectures, and pursue mathematical ideas. Students are taken on a guided
tour of the topics and are given many opportunities to pursue their own investigations.
Interlaced in the reading are exercises, explorations using computer applets, and projects.
They are an essential part of the learning process. For this reason, most of them end with
the phrase Try it out! to remind the student that the activity needs to be done before going
on. Activities include:
Examples—Students should be sure that they can follow the arguments and provide
small details when needed.
Exercises—These have a well-defined goal and should be done before going on to
the next paragraph in the text. Skipping them would result in the reader missing a
fundamental skill or idea.
Explorations—These also should be done before going on. Generally, they do not
have a well-defined problem to solve. Some may include undirected investigating or
playing with applets. There is no specific outcome expected, but much will be gained
from engaging with the material. Such activities are at the heart of what this book is
for: getting students to explore mathematics on their own.
Small Projects—These are optional activities that may take up to a few weeks to
complete.
Large Projects—These could be a semester-long project, a capstone project, or an
honors thesis.
Additional Exercises—Additional exercises may include any of the previous activ-
ities. They appear at the end of chapters and are optional.
xi
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page xii — #12

i i
xii Introduction
During these activities, students should consider such questions as: Why was this prob-
lem posed? Why is it interesting? If I changed the problem slightly, does it make it eas-
ier? harder? impossible? What does it say about the general theory? Thinking about such
questions is part of mathematics research and investigating the unknown. This can make
for slow reading. Progress should be measured not by the number of pages read but by the
amount of independent thought given to the material. If students read just a few pages of
a chapter and then become motivated to work on a problem or set of problems devised on
their own, the authors of this book would be delighted. As Albert Einstein said, quoting
from Philipp Frank, Einstein: His Life and Times, Knopf, New York, 1947:
It is not so very important for a person to learn facts. For that he does not really
need a college. He can learn them from books. The value of an education in a
liberal arts college is not learning of many facts but the training of the mind to
think something that cannot be learned from textbooks.
This book delves into six current research topics, providing numerous directions for
student investigation. Each topic is presented in a self-contained chapter with mathemat-
ical background, new material, exercises, explorations, problems suitable for projects,
and computer software for exploring the topic. Two appendices provide background. Ap-
pendix A recaps key definitions and results from undergraduate complex variables, while
Appendix B discusses the Riemann sphere. Results from appendices will be cited in var-
ious chapters, but students can take these as known and continue reading, or look to the
appendix for further details. There is also an Index of Notation on page 365 for quick
reference to symbols employed throughout.
Here are descriptions of the chapters in this book:
Chapter 1: Complex Dynamics This chapter investigates chaos and fractals as they re-
late to dynamical systems that come from iterating complex-valued functions, i.e., given
an initial value z0 we consider the values z1 D f .z0 /; z2 D f .z1 / D f .f .z0 //, z3 D
f .z2 / D f .f .f .z0 ///; : : : , and how the sequence fzn g behaves. This kind of iteration
arises in Newton’s method for approximating roots, and so our chapter begins by ask-
ing: Which initial values will work for Newton’s method (i.e., converge to a root)? If the
initial value z0 is changed slightly, will similar or drastically different behavior result?
The questions are considered computationally, visually, and experimentally with the aid of
computer applets. We then consider the iteration of any complex analytic map, which leads
to the mathematics behind the Mandelbrot set, and much more.
Chapter2: Soap Films, Differential Geometry, and Minimal Surfaces Dipping a wire
frame into soapy water produces an iridescent soap film that clings to the frame. Such soap
films are related to minimal surfaces, which are beautiful geometric objects that minimize
surface area locally. Visually, minimal surfaces can be thought of as saddle surfaces that
bend upward in one direction in the same amount that they bend downward in the perpen-
dicular direction. In this chapter, we present the necessary background from differential
geometry, a field of mathematics in which the ideas and techniques of calculus are applied
to geometric shapes, to give an introduction to minimal surfaces. Then we use ideas from
complex analysis to present a nice way to describe minimal surfaces and to relate the ge-
ometry of the surface with this description. This allows us to begin investigating some of
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page xiii — #13

i i
Introduction xiii
the interesting properties and new research questions that can be explored with the help of
the applets.
Chapter 3: Applications to Flow Problems Two dimensional vector fields are used to
model and study a wide range of phenomena. Of particular interest are those that are irrota-
tional and incompressible, which can be used to model the velocity of an ideal fluid flowing
in a region or the electric field in a region free of charges. Modeling two dimensional fluid
flow is a standard application of conformal mappings in complex variables. This chapter
takes a geometric and visual approach to this topic and then extends it to several other ap-
plications. Fields of interest include those generated by combinations of sources and sinks
that add or remove fluid from the flow. Throughout the chapter, we use examples, theory,
and exercises to develop methods that allow these fields and many others to be modeled
and analyzed. Also, the included applet FlowTool displays the streamlines and equipoten-
tial lines for a wide variety of fields, and permits real-time dynamical experimentation with
sources and sinks in several pre-selected regions. Students with an interest in using tech-
nology to visualize mathematical objects will find many opportunities to explore their own
ideas in extending this material.
Chapter 4: Anamorphosis, Mapping Problems, and Harmonic Univalent Functions
Anamorphosis is a method of distorting an image that appears normal when viewed from
a different perspective. Think of a warped mirror that makes you look taller and thinner or
shorter and fatter. This method has been used in the past in painting and is still used today
for dramatic effect. In complex analysis, analytic maps such as Möbius transformations
distort images in a systematic way. We begin this chapter by discussing geometric proper-
ties of Möbius transformations and other analytic functions. But analytic maps have been
around for a long time. Is there anything new to investigate? Yes, there is. The real and
imaginary parts of analytic functions are harmonic and satisfy the Cauchy-Riemann equa-
tions. What happens if we remove the condition of satisfying the Cauchy-Riemann equa-
tions? We then create a new collection of functions known as complex-valued harmonic
functions. Recently, there has been a resurgence in the study of one-to-one complex-valued
harmonic functions also known as harmonic univalent functions. We will explore geomet-
ric properties of these functions and see some of the bizarre behaviors they have that form
the basis of an exciting new area of research.
Chapter 5: Mappings to Polygonal Domains A rich source of problems in analysis
is determining when, and how, we can create a univalent (one-to-one) function from one
region onto another. In this chapter, we consider the problem of mapping the unit disk
onto a polygonal domain by two classes of functions, analytic and harmonic functions.
For analytic functions we give an overview of the Schwarz-Christoffel transformation, that
leads to some rich mathematics, the study of special functions. We then use the Poisson
Integral Formula to find harmonic functions that map onto polygonal domains. Proving
that they are univalent requires us to explore the theory of harmonic functions and uses
some new techniques.
Chapter 6: Circle Packing Circle packings are configurations of circles with prescribed
patterns of tangency. They exist in quite amazing and often visually stunning variety, but
what are they doing in a book on complex analysis? Complex analysis is, at its heart, a
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page xiv — #14

i i
xiv Introduction
geometric topic. The reader will see this in the global geometry on display in Chapters 1–5,
but the foundation lies at the local level where, as the saying goes, “analytic functions map
infinitesimal circles to infinitesimal circles.” In Chapter 6 this geometry will come to life
in the theory of discrete analytic functions when we map actual circles to circles. Using
the Java application CirclePack, we will create, manipulate, and display maps between
circle packings that are the discrete analogues of familiar maps between plane domains,
including some of those encountered earlier in the book. Direct access to the underlying
geometry gives insight into fundamental topics like harmonic measure, extremal length,
and branching. Moreover, we will see that our discrete functions not only mimic their
continuous counterparts, but also converge to them under refinement. In short, Chapter 6
is about quantum complex analysis.
Acknowledgements
We sincerely thank all those colleagues and their students who read through early drafts of
the manuscript and provided very valuable feedback. We will list these people individually
with the chapters they reviewed. We also thank Melissa Mitchell who provided valuable
comments on the Introduction and Appendices. Of course, any errors that remain are the
full responsibility of the authors.
Further, we thank the National Science Foundation for their support of this project
through grants No. DUE-0632969, No. DUE-0632976, No. DUE-0633125, and No. DUE-
0633266.
We offer special acknowledgement and sincere gratitude to Jim Rolf. Jim has managed
to design, code, and maintain fourteen separate applets for the first five chapters of this text.
His work to keep things well organized while dealing with a seemingly unlimited stream
of requests from so many authors has been astounding. His work turned out to be well
beyond what any of us initially thought we would need or want from him, yet he adapted
and produced professional applets tailored to our specific desires. Since active use of these
applets is critical to this text, we especially thank Jim for his tremendous efforts which
were most critical to improving the overall project.
Chapter 1: Complex Dynamics As a text that hopes to inspire students by showing

them the beauty of complex variables research, I would like to dedicate this chapter to the
professors who have inspired me, namely, Juha Heinonen, Joe Miles, and Aimo Hinkkanen.
Also, I thank Neal Coleman, Stephanie Edwards, Dan Lithio, Kevin Pilgrim, and Irina
Popovici, who provided very useful feedback after reading through early drafts of this
chapter. I also thank Bob Devaney and Phil Rippon for helpful conversations where they
shared ideas and insights, and gave encouragement.
Chapter 2: Soap Films, Differential Geometry, and Minimal Surfaces

Chapter 4: Anamorphosis, Mapping Problems, and Harmonic Univalent Functions
I would like to thank my colleagues Casey Douglas and Michelle Hackman who read
through the chapter on minimal surfaces and provided me with very helpful feedback.
Also, I would like to thank my 2009 and 2010 REU students who worked through both
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page xv — #15

i i
Introduction xv
chapters and gave me lots of advice from the students’ perspective. These students include
Valmir Bucaj, Sarah Cannon, Amanda Curtis, Sam Ferguson, Laura Graham, Jamal Law-
son, Rachel Messick, Jessica Spicer, Ryan Viertel, and Melissa Yeung. Finally, I would
like to thank Lonette Stoddard who used Adobe Acrobat to draw many of the images in
the chapters.
Michael Dorff
Chapter 3: Applications to Flow Problems I am particularly indebted to Harrison Pot-

ter who used ideas from an early draft of this work in writing his undergraduate honors
thesis at Marietta College. He inspired several of the chapter’s exercises and projects. I
would also like to acknowledge Professor Tristan Needham and his beautiful text Visual
Complex Analysis, Oxford University Press, 1997, for giving me a greater appreciation of
the geometric aspects of complex analysis. I offer a special thank you to my colleagues
Beth Schaubroeck and Jim Rolf, who agreed to participate in this crazy scheme. Finally, I
wish to thank my co-authors for sticking with a project that lasted far longer than any of us
thought it would.
Michael Brilleslyper
Chapter 5: Mappings to Polygonal Domains We gratefully acknowledge the numerous

students, colleagues, and reviewers who gave us insight into the material and the writing
of this chapter. We especially thank Peter Duren, both for the inspiration through his work
in harmonic functions, and also for detailed and helpful feedback on an earlier draft of the
chapter. We would like to dedicate this work to our husbands and children, whose support
and love made our efforts possible.
Jane McDougall and Lisbeth Schaubroeck
Chapter 6: Circle Packing I would first thank William Thurston for bringing this won-
derful topic into existence. Next, my PhD students, Tomasz Dubejko, G. Brock Williams,
Matt Cathey, Gerald Orick, Chris Sass, and James Ashe, along with the undergraduate and
REU students who have shared in circle packing discoveries. Finally, thanks to those who
helped in programming CirclePack, especially Fedor Andreev, Sam Reynold, Benjamin
Pack, Chris Sellers, and Alex Fawkes. I gratefully acknowledge support from the National
Science Foundation over many years. I dedicate this chapter to my lovely wife Dee: thanks
for your patience, dear.
Kenneth Stephenson
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page xvi — #16

i i
xvi Introduction
Using Java Applets

Software is essential in using this text. Java applets are provided to explore complex func-
tions in Chapters 1–5, while the Java application CirclePack is provided to explore Chapter
6. All software provided for use with this text may be found at www.maa.org/ebooks/
EXCA/applets.html. The applets can be run over the internet in a web browser or
downloaded onto your computer and run locally, while CirclePack must be downloaded
and run locally.
We now demonstrate how to use the applet ComplexTool, which is used in several
chapters and which shares many features with the other applets. We can graph the image of
a variety of standard domains under numerous familiar functions. We begin by examining
how to draw the image of the unit disk D under the function f .z/ D .1 C z/2 . Open
ComplexTool (see Figure 0.1).
F IGURE 0.1. The applet ComplexTool
In the middle section near the top there is a box that has f(z)= before it. In this box,
enter (1+z)^ 2. (In general, you will enter functions into ComplexTool using the same
syntax you would in a standard computer algebra system.) Below this, there is a window
that reads No grid. Click on the down arrow H and choose the option Circular
grid; an image of a circular grid appears on the left to show the domain being graphed.
Next, click on the button Graph which is in the middle section below the function you
entered earlier. The image of the circular grid under the action of the function appears
on the right. To change the size of the image window, click on the down arrow H above
the image and choose a different size, such as Re: [-3,3] Im: [-3,3]. You can
also zoom in and out by left- or right-clicking either image, by dragging the slider in the
bottom left of the center panel, or by turning the scroll wheel on the mouse. Also, you
can move the axes so that the image is centered by positioning the cursor over the image
and depressing the mouse button while dragging the image (see Figure 0.2). To go back
to the original region you graphed, push the Default View button above either the
domain or range. If you want to change the domain that is being graphed, you can do so
manually by changing Center, Interior circles, Rays, the boxes surrounding
, and the boxes surrounding radius. Alternatively, you can check the box that says
Vary or Vary radius to see sliders that dynamically change the -range and radius
range. (You may have to push the Graph button to regraph after you make changes.) You
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page xvii — #17

i i
Using Java Applets xvii
F IGURE 0.2. The image of the unit disk under the map f .z/ D .1 C z/2 .
can also position the cursor over the domain and drag it around while watching the range
dynamically change. This will not change the size of the domain, but will change its center.
To go back to where you had originally centered the object, you can push the Snap to
button.
To figure out exactly which points in the domain correspond to points in the range, you
can check the box in front of Sketch. Then you can scribble in the domain window and
see its image appear in the range. When the Sketch function is turned on, you cannot
dynamically move the domain, as a mouse-click in the domain window is interpreted as
sketching instead.
Exploration 0.1. Use ComplexTool to graph several functions. Some functions to try are:
(a) Graph the image of D under the function f .z/ D 1 z z .
(b) Consider the two analytic functions f1 .z/ D 2z 2 z and f2 .z/ D 21 z 2 z each
defined on D. One is one-to-one while the other is not. Graph the image of D under
the function f1 .z/ and under f2 .z/. Determine how you can tell by looking at the
applet which function is one-to-one.
(c) Determine the smallest set of values that you can use for theta in a circular wedge
domain to get an image of a complete disk under f .z/ D z 3 . (The Vary option is
good to use for this exercise.)
(d) For the function f .z/ D e z , plot the images of vertical and horizontal lines. Explain
your result mathematically.
(e) Determine if there exists a domain D so that the image of D under f .z/ D log.z/
covers the plane.
(f) Demonstrate the periodicity of sin.z/ using ComplexTool. (Find a region to graph and
drag it horizontally to watch its image change.)
Try it out!
In addition to the action of the applet, you can use the menus across the top to change
settings. Perhaps most helpful is the option to export screen shots from the applet. The
Export Settings menu allows you to choose which file type to export, and then you
can choose to export the domain, range, or the whole screen. As you do the exercises and
explorations in this text, you will find it helpful to document the work that you have done
by taking hand-written or electronic notes.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page xviii — #18

i i
xviii Introduction
Using Links in the Electronic Book

The electronic version of this book comes with links to assist with its use. Each applet or
computer application is colored red and is linked directly to a web page where the software
can be found. To illustrate, click on the red link ComplexTool. In addition to these links to
software, there are links to each numbered referenced item, e.g., Definition A.1, Theorem
A.8, and Bibliography item [1] but these links are not colored. Simply click on the number
of the reference, e.g., “A.1.” in the above Definition A.1., to be taken to the item.
When using Adobe Reader X, use the Page Navigation buttons called Previous
View and Next View to return to the location in the text where the link was clicked. If
you right-click on the tool bar, under Page Navigation you can select Previous
View and Next View for these buttons to appear on the tool bar. Alternatively, click
on the VIEW drop down menu, then click Page Navigation, then select Previous
View. A backward arrow will then appear in the toolbar at the top. Repeat for Next
View to add a forward arrow.
PDF readers other than Adobe Reader X will likely have buttons with similar functions.
We encourage you to use these to make the best use of the links in this text.
Adobe Reader X can be downloaded for free from get.adobe.com/reader/.
i i
i i
i i
“ECA˙Book” — 2012/9/5 — 14:07 — page 1 — #19

i i
Complex Dynamics: Chaos, Fractals,

1
the Mandelbrot Set, and More
Richard L. Stankewitz (text and software design),
James S. Rolf (software coding and design)
1.1 Introduction
This chapter introduces complex dynamics, an area of mathematics that has inspired and
continues to inspire research and experimentation. Its goal is not to give a comprehen-
sive description of the topic, but to engage you with the general notions, questions, and
techniques of the area and to encourage you to actively pose as well as pursue your own
questions.
Dynamics, in general, is the study of mathematical systems that change over time, i.e.,
dynamical systems. For example, consider a Newtonian model for the motion of the planets
in our solar system. Here the mathematical system is a collection of variables correspond-
ing to the location and velocity of the planets relative to the sun, and this system changes
over time according to Newton’s laws of motion. We can describe the process by which
the system evolves (i.e., the rules of how the system changes over time) by differential
equations relating the system variables to each other and to Newton’s laws of motion.
Many dynamical systems can be described similarly, for example the population of
bacteria in a petri dish, the weather in Ann Arbor, Michigan (temperature, pressure, and
wind velocity, to be more precise), the average temperature of Earth’s atmosphere and
oceans, and the flight of a paper airplane that you might toss across the room. The models
have a set of system variables, and some rule or set of equations that describes how they
change over time. Their values at time t is called the state of the system at time t.
Knowing the initial state of the system (e.g., today’s location and velocity of each
planet), we often try to analyze the equations that describe how the system variables change
over time in order to answer such questions as: What will the state of the system be tomor-
row? next week? next year? one hundred years from now? Will the system in the long run
settle into some sort of equilibrium? Will small changes or errors in our knowledge of the
initial state only lead to small changes or errors in the system at some future time, or could
they lead to huge changes? Such natural questions about the future states of a dynamical
system have led to results of practical importance, and to some beautiful mathematics.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 2 — #20

i i
2 Chapter 1. Complex Dynamics: Chaos, Fractals, the Mandelbrot Set, and More
The famous physicist Richard P. Feynman said ([15, p. 9]): “Physicists like to think that
all you have to do is say, these are the conditions, now what happens next?” Unfortunately
there are many systems which we cannot adequately solve for the purposes of making use-
ful predictions, and so What happens next? is a question that we cannot answer. Frequently
this is due to our inability to find the right pattern, or to solve some differential equation.
Perhaps someone (maybe you!) will come along and find a clever solution, so we can then
predict the future of these systems. But for a large class of systems there is no hope of ever
being able to predict, with any useful accuracy, how it will behave. What’s astounding is
that, for these systems, it is not a matter of finding the right solution. In fact, we sometimes
even have what we thought was a great solution, a formula even, but there is a problem
with applying it.
The heart of the problem is that we have only solved a model, an approximation to
the real world, and so there will be some error built into the model, which we hope is
small. The problem is that: (1) we can never pinpoint the initial conditions EXACTLY, and
(2) ANY approximation (or error) to the initial conditions leads to errors so large that we
cannot have any confidence at all in applying our prediction to the real world. Such systems
are called chaotic, a term you will explore and even be asked to mathematically define in
this text. Though chaotic systems do not allow for precise answers to some of the questions
scientists like to ask, much can be gained from studying them.
In this chapter we will study simple chaotic systems that can be analyzed using complex
analysis. We do this because they are of interest in their own right, and because it will
lead to understanding fundamental principles of complicated systems, like those mentioned
above. The dynamical systems we consider are discrete iterative systems, which are in
some sense the easiest. For them time is represented by a natural number n. Also, there is
no need to solve a differential equation to determine the system’s state; we just repeatedly
apply a function. Furthermore, the states of our systems are described not by a large number
of variables (as are needed to represent the positions and velocities of the planets), but just
a single complex variable. The system is called iterative because the state zn of the system
at time n will evolve according to the rule znC1 D f .zn / where f is a complex-valued
function. Thus, given an initial state, computing the future state zn is just a simple matter of
computing the values z1 D f .z0 /; z2 D f .z1 / D f .f .z0 //, z3 D f .z2 / D f .f .f .z0 ///,
and so on. However, predicting the behavior of the sequence of states fzn g (e.g., deciding
if fzn g converges or not), an altogether different problem, is by no means simple!
This chapter uses tools from complex function theory to investigate several types of
discrete iterative systems, including Newton’s method, polynomial iteration, exponential
iteration, and trigonometric iteration. We also consider what happens when these systems
are perturbed by changing a parameter, leading us into bifurcation theory. There are many
ways to perturb a system and in the concluding section we describe two more: perturba-
tion with a pole and random dynamics. We begin with perhaps the most familiar discrete
dynamical system, Newton’s method.
Although we are not studying real world dynamical systems directly, we should keep
in mind that they exhibit many of the same behaviors as the systems we do study.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 3 — #21

i i
1.2. Newton’s Method 3
How to Use this Chapter

The sections of this chapter can be worked through in order. However, to proceed to
Section 1.3 quickly, Section 1.2 may be skipped, with the exceptions of Sections 1.2.3
and 1.2.6. Also, Section 1.6 may be skipped by anyone not wishing to investigate the dy-
namics of transcendental entire maps. Reading the entire chapter and working on several
additional exercises as well as projects would suit a three credit hour 15-week semester
reading course. A 2-3 week group or individual project for the end of the semester in an
undergraduate complex variables course is provided by either Section 1.2 or Sections 1.2.3
and 1.2.6 together with Section 1.3.
Because the natural setting for this chapter is the Riemann sphere all of Appendix B
should be worked through, though it is not necessary that it be completed in full before
starting. There are also three chapter appendices that begin on page 78 providing added
information relevant to the chapter. Results from the appendices, when referenced, may be
accepted as stated without interrupting your reading, or their details may be investigated in
the appendix.
In addition to ComplexTool, introduced on page xvi, this chapter will utilize the follow-
ing applets which can be accessed at www.maa.org/ebooks/EXCA/applets.html.
1. Real Newton Method Applet, to visualize the real-valued Newton’s method.
2. Complex Newton Method Applet, to visualize the complex-valued Newton’s method.
3. Real Function Iterator Applet, to iterate a real function, and see its orbit displayed
as a numerical list and as points on a number line.
4. Complex Function Iterator Applet, to iterate any complex function, and see its orbit
displayed as a numerical list and as points in the complex plane.
5. Cubic Polynomial Complex Newton Method Applet, to explore the attracting basins
for Newton’s method applied to the cubic polynomials p .z/ D z.z 1/.z /.
6. Global Complex Iteration Applet for Polynomials, to draw the basin of infinity for a
polynomial.
7. Mandelbrot Set Builder Applet, to illustrate how the Mandelbrot set is constructed.
8. Parameter Plane and Julia Set Applet, to investigate the parameter plane and dy-
namic plane pictures for the families of functions z 2 C c, z d C c, ce z , c sin.z/,
c cos.z/, and z d C c=z m .
1.2 Newton’s Method

Solving equations, finding solutions to ordinary differential equations, finding eigenvalues
of a matrix: all are important mathematical procedures. However, they cannot always be
done exactly (by computing a value exactly we meanpbeing able to express it in terms of
standard mathematical operations and functions, e.g., cos.=12/). When exactness is not
possible, often the solution can be approximated with an iterative numerical method.
For example, consider the problem of finding a root of a complex-valued function f .z/,
i.e., a value ˛ such that f .˛/ D 0. If the function is the quadratic f .z/ D az 2 C bz C c
where
p a ¤ 0, then there are two roots given by the quadratic formula ˛ D . b ˙
b 2 4ac/=2a. If the function is a cubic or quartic polynomial, then there exists a for-
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 4 — #22

i i
mula (or, more precisely, procedure) for finding the roots exactly. However, if f .z/ is a
quintic polynomial, then there is not, in general, a procedure that will exactly find any
of its roots. (Even though any polynomial of degree n has, by the fundamental theorem
of algebra, n roots in the complex plane, this theorem does not help us to actually find
them.) The same is true for many transcendental functions such as h.z/ D cos z z and
g.z/ D e z 4z. In such cases we must give up on finding exact roots and resort to approx-
imation.
For real-valued functions of a real variable, there are approximation methods for root
finding based on the intermediate value theorem, such as the bisection method. When con-
sidering a complex-valued function f , we can sometimes approximate a root of f using
Rouche’s theorem (see [1, p. 294]). Specifically, if f and g are analytic on and inside a
simple closed curve C with jf j > jgj on C , then f and f C g have the same number of
zeroes (counting multiplicities) inside C . Hence, if g can be chosen such that f C g has
a root inside of C , then so must f . However, even if f is a nice function (e.g., a polyno-
mial), this can be a difficult method to implement. So we look for a better method. One of
the best is Newton’s method, for finding both real and complex roots. Often it allows us to
quickly approximate roots with extreme accuracy, if we have access to a computer. In this
section we examine Newton’s method for both real and complex functions, with the goal
of understanding when it succeeds and when it fails.
1.2.1 Real Newton’s Method

If we seek to find a root ˛ of a differentiable real-valued function f .x/ defined for a real
variable x, then we can apply Newton’s method as follows. We start with an initial guess
x0 close to ˛ and define
f .x0 / f .x1 / f .x2 /
x1 D x0 ; x2 D x1 ; x3 D x2 ;
f 0 .x0 / f 0 .x1 / f 0 .x2 /
and, in general,
f .xn /
xnC1 D xn :
f 0 .xn /
The geometry behind Newton’s method, as illustrated in Figure 1.1, is as follows: Given
an approximation x0 to the root ˛, we considers a linear function fQ that approximates f
near x0 . The best linear approximation is the first order approximation fQ.x/ D f .x0 / C
f 0 .x0 /.x x0 / whose graph is the tangent line L to the graph of f at x0 . The root of
fQ.x/, i.e., the x-intercept of L, is then the definition of x1 .
Exercise 1.1. Before reading further, compute fQ.x/ for f .x/ D x 3 2x and x0 D 2 as
in Figure 1.1. Then use fQ.x/ to find the equation of the tangent line L and also check that
x1 , given by the above formula, is the x-intercept of L. Try it out!
Exercise 1.2. Verify that for general f .x/, the function fQ.x/ has a root at x1 where fQ and
x1 are as given above. Try it out!
In general (as illustrated in Figure 1.1), we expect that the root x1 of fQ.x/ will be a
better approximation to ˛, the sought-after root of f .x/, than the initial guess x0 . This
process is repeated using x1 as the initial guess to generate a new approximation x2 . We
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 5 — #23

i i
5
f (x)
a x0
x
1.5 x1 2 2.5
L
–5
F IGURE 1.1. An illustration of the first step in Newton’s method where f .x/ D x 3 2x and
x0 D 2. Newton’s method can be described by saying from x0 move to the graph of f .x/, slide
along the tangent to the x-axis, and repeat.
then iterate to generate successive approximations xn , for n D 0; 1; 2; : : : . We express this

in terms of iteration of the following function.
f .x/
Definition 1.3. The function F .x/ D x is called the Newton map for f .
f 0 .x/
Thus, the orbit, i.e., sequence of iterates F .x0 /; F .F .x0 //; F .F .F .x0 ///; : : : , is the
same as the sequence fxn g generated above, and it will be proven (in Proposition 1.20) that
xn converges to the root ˛ whenever our original guess x0 is close enough to ˛. What it
means to be close enough is important, and we will come back to this in Section 1.2.4, but
we first explore Newton’s method with some examples.
Example 1.4. Consider f .x/ D x 2 3x C 2 D .x 1/.x 2/, which has roots at 1 and
at 2. Let’s apply Newton’s method. The Newton Map for f is
x2 3x C 2 x2 2
F .x/ D x D :
2x 3 2x 3
Making an initial guess x0 D 0:5 and using your calculator (do this now) you can compute
x1 D F .x0 / D 0:875; x2 D F .F .x0 // D F .x1 / D 0:9875, and x3 D F .F .F .x0 /// D
F .x2 / D 0:99984756. Since using a calculator is drudgery and computers are efficient at
such tasks, we have created the Real Newton Method Applet for you to use. Use this now to
confirm the calculations above and then, by taking many iterates of the Newton map using
this applet, convince yourself that with the initial value (a term we use interchangeably
with seed value and starting point) x0 D 0:5, we have xn ! 1: Try it out!
Now use the Real Newton Method Applet to determine the behavior using an initial
guess x0 D 3. Try it out!
What you learned above is that some initial guesses for x0 find the root 1 (i.e., have the
corresponding fxn g converge to 1) while others find the root 2. This suggests the question:
Given an initial guess x0, how do we know which root it will find?
Exploration 1.5. Make a prediction about which seed values for x0 in Example 1.4 will
find the root 1 and which will find the root 2. Are there seed values for which Newton’s
method fails to find either root? Experiment with the Real Newton Method Applet to test
your prediction. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 6 — #24

i i
Exploration 1.6. Analyze Newton’s method for f .x/ D x.x 1/.x C 1/ using different
initial guesses for x0 in the Real Newton Method Applet. Describe (as best as you can) the
set of seed values for x0 which find the root 1, and then do the same for the roots 0 and
1. Try it out!
1.2.2 Global Picture of Real Newton Method Dynamics

Identifying how the orbits of all seed values behave, as you attempted in Explorations 1.5
and 1.6, is what we mean by looking at the global dynamics of Newton’s method. In some
cases this can be done without too much work, but in other cases it can be complicated.
To help with this problem, we use the Graph basins of attraction feature in
the Real Newton Method Applet to illustrate which initial guesses will find which roots of
f .x/ when Newton’s method is applied.
Exercise 1.7. For each of Explorations 1.5 and 1.6, use the Real Newton Method Applet to
get a snapshot of what the dynamics of Newton’s method are for all starting values. Try it
out!
What you see when using the Real Newton Method Applet with f .x/ D .x 1/.x 2/
is what you might expect. The initial values that are closer to the root at 1 will find 1,
and those that are closer to the root at 2 will find 2. One should also check that Newton’s
method fails for the initial value x0 D 1:5. Analytically we see this in the formula because
f 0 .1:5/ D 0 leads to a zero in the denominator in the calculation of x1 . Geometrically we
see this because the tangent line at x0 D 1:5 is horizontal, never crosses the x-axis, and
thus does not determine a value for x1 . We can also understand this dynamically, at least
in a heuristic way. The value 1.5 separates those values that are pulled or attracted to 1 and
those that are attracted to 2, and so by an informal use of symmetry, it would seem that
something must fail to work out at x0 D 1:5.
Though this reasoning is informal, it captures an important idea. As a rule, we encour-
age you to use and create your own heuristic ideas to explain or describe mathematics.
Sometimes it’s hard to be formal with all of your mathematical ideas. But don’t let that
stop you from thinking of and sharing great mathematical thoughts, even if you can’t make
them precise or formal. Some of the best mathematics, if not all mathematics, starts off as
raw unformed ideas with no foundation in formalism. Later, you can (and should) try to be
formal with you ideas.
The case when f .x/ D x.x 1/.x C 1/ is much different from the quadratic case.
A seed for Newton’s method will not always find the root it is nearest to. For example,
x0 D 0:55 will find the root at 1, even though it is closer to the roots at 0 and 1. Indeed,
in this case the set of initial guesses on the real line is divided into intricate regions that
find the roots. In fact, if you zoom in near the point x D 0:4472135951871958 you will
see a cascade of ever shrinking and alternating colored intervals of blue and turquoise (see
Figure 1.2). It turns out that this pattern goes on forever. Use the Zoom feature of the Real
Newton Method Applet to observe this.
Small Project 1.8. Prove the existence of the infinite cascade of ever shrinking and alter-
nating colored intervals of blue and turquoise found in Figure 1.2. Hint: First use the applet
to understand what each of the colored intervals means dynamically, and then try to give a
proof for what you observe.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 7 — #25

i i
y y
x x
F IGURE 1.2. On the left is a cascade of intervals in the global picture of Newton method dy-
namics for f .x/ D x.x 1/.x C 1/, and on the right is a magnification centered at x D
0:4472135951871958.
We have seen that Newton’s method gets really complicated to understand globally
when we switch from a quadratic to a cubic, as from f .x/ D .x 1/.x 2/ to f .x/ D
x.x 1/.x C 1/. This suggests the questions: Why? How? Is there a way to know ahead
of time when a system will be simple or complicated?
Answers will come from taking the advice of Jacques Hadamard who said, “The short-
est path between two truths in the real domain passes through the complex domain.” So
let’s look at Newton’s method applied to complex-valued functions of a complex variable.
There are many wonderful theorems at our disposal when we consider complex analytic
maps instead of mere real-valued differentiable maps. Let’s take advantage of them.
Before we investigate Newton’s method applied to complex-valued functions, however,
we first develop some concepts that we will need to understand any iterative dynamics.
1.2.3 Orbits, Examples, and Fixed Points

The general questions we consider in iterative dynamical systems concern describing and
predicting orbits. Let g be a function mapping its domain set domain.g/ into itself (the
domain set of a function g is the set of its possible inputs), which we take to be a subset
of the Riemann sphere C. (The Riemann sphere is the natural setting for this chapter and
so familiarity with the material in Appendix B starting on page 357 is crucial.) We denote
the nth iterate of g by gn . Thus gn .z/ D .g ı ı g/.z/ where g is applied n times, e.g.,
g3 .z/ D g.g.g.z///. In this chapter g3 .z/ does not denote the value g.z/ raised to the
third power, which would instead be denoted Œg.z/3 . We define g0 to be the identity map,
i.e., g0 .z/ D z. Furthermore, for any starting (seed) value z0 2 domain.g/, the sequence
of values zn D gn .z0 /, for n D 1; 2; 3; : : : , is called the orbit of z0 under the map g.
When trying to predict the behavior of the evolution of a seed value, we ask: Does the
orbit converge, fall into a repeating pattern, or show no signs of following any pattern?
What happens for different starting values z0 ? Do we get the same (or even similar) be-
havior if we choose starting values near z0 ? Keep these questions in mind as you consider
the following examples.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 8 — #26

i i
Example 1.9. Let f .x/ D e x for all x 2 R. Thus the iterates are f 0 .x/ D x; f 1 .x/ D
x ex
e x ; f 2 .x/ D e e ; f 3 .x/ D e e , and so on. Experiment with the Real Function Iterator
Applet to convince yourself that f n .x/ ! C1 no matter what x we start with. Additional
Exercise 1.157 asks for a formal proof.
Example 1.10. Let f .x/ D sin x where x 2 R is given in radians. Experiment with the
Real Function Iterator Applet to convince yourself that f n .x/ ! 0 for any real number x.
Additional Exercise 1.158 asks for a formal proof.
Example 1.11. Let f .x/ D cos x where x 2 R is given in radians. Experiment with the
Real Function Iterator Applet to convince yourself that f n .x/ ! 0:739085: : : : for any
real number x. Additional Exercise 1.159 asks for a formal proof.
Example 1.12. Let f .x/ D x 2 1 and x0 D 0:9. Use the Real Function Iterator Applet
to convince yourself that the tail end of the orbit xn D f n .0:9/ appears to oscillate back
and forth between 0 and 1.
Example 1.13. Let f .x/ D 4x.1 x/ and x0 D 0:2. Use the Real Function Iterator
Applet to see that the orbit xn D f n .0:2/ appears to have no pattern to it at all, even after
the first 25,000 orbit points are plotted. Zoom in on the orbit points to see how it appears
that they are dense in the interval Œ0; 1, that is, for every open interval .a; b/ that meets
Œ0; 1, there is some orbit point xn 2 .a; b/.
Though many types of behavior can be exhibited in orbits, we focus on one for the
moment. In Examples 1.9, 1.10, and 1.11, we see that there is a point that attracts the orbits
of many seeds. This motivates the following fundamental definition.
Definition 1.14 (Attracting Basin). Let w 2 C. For a complex-valued function g mapping
its domain set into itself, we define the basin of attraction of w (also called the attracting
basin) under the function g to be the set Ag .w/ of seed values whose orbits approach the
point w, i.e., Ag .w/ D fz 2 domain.g/ W gn .z/ ! wg.
The point w in this definition does not have to lie in domain.g/ (as in Example 1.9).
However, if w 2 domain.g/ and g is continuous, as in Examples 1.10 and 1.11, then The-
orem 1.15 shows that w must necessarily be a fixed point of g, i.e., g.w/ D w, whenever
Ag .w/ is non-empty.
Theorem 1.15. Let f W domain.f / ! domain.f / be a continuous map where domain.f /
is contained in C. Suppose a and x0 are in domain.f / and f n .x0 / ! a. Then f .a/ D a.
Proof. Since the sequence f n .x0 / ! a and f is continuous at a, we have f .a/ D

f .limn!1 f n .x0 // D limn!1 f .f n .x0 // D limn!1 f nC1 .x0 / D a.
Fixed points play a major role in dynamical systems and so we will pay special atten-
tion to them whenever they arise. In our previous examples the roots of f appear to be
fixed points of the corresponding Newton map F .z/, a fact we prove in Proposition 1.20.
Additional Exercise 1.161 will shed some light on the extent to which the converse holds.
We call the fixed points in Examples 1.10 and 1.11 attracting fixed points because the
orbits of seed values near the respective fixed points will converge to them. To be more
precise we give a formal definition, but first we give a reminder about a key relationship
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 9 — #27

i i
between the Euclidean metric on C and spherical metric on C, which is stated in Ap-
pendix Proposition B.7 on page 361. Namely, for points z; w 2 C; we have jzj > jwj if
and only if .z; 1/ < .w; 1/.
Definition 1.16 (Attracting Fixed Point). Let f be a map from its domain set C into
itself ( could be a subset of R).
(a) We call a (finite) fixed point a 2 C an attracting fixed point of f if there exists a
neighborhood U of a such that for any point z 2 \ U n fag, we have jf .z/ aj <
jz aj, i.e., the action of f is to move each point in \ U n fag closer to a.
(b) Suppose f .1/ D 1. We call 1 an attracting fixed point of f if there exists a neigh-
borhood U C of 1 such that for any point z 2 \ U n f1g, we have jf .z/j > jzj,
i.e., the action of f is to move each point in \ U n f1g closer to 1, as measured
by the spherical metric.
In Definition 1.16(a) we used the Euclidean metric to describe when the action of f
moves points closer to a, but we could have equivalently used the spherical metric by
writing .f .z/; a/ < .z; a/. It is important to become comfortable with understanding
when a particular metric used in a definition or result could equivalently be changed to
another. We often use the simplest metric at our disposal, assuming it is clear when another
metric could also be used. A good way to try to become comfortable with this is to consider
how sets appear when visualized in the flat plane C and in the sphere C. An exposition of
this is in Appendix B on page 357.
If a is an attracting fixed point of a continuous map f , then there exists a neighborhood
U of a such that U Af .a/. The proof of this does not require that f be (real or complex)
differentiable at a, but without a differentiability condition the proof is more technical.
Since we are interested only in specific differentiable functions in this chapter, we only
provide the following.
Theorem 1.17. Suppose R or C. Let f W ! be such that f .a/ D a and
jf 0 .a/j < 1. Then a is an attracting fixed point of f . Furthermore, there exists some " > 0
such that 4.a; "/ \ Af .a/, where 4.z0 ; r / D fz 2 C W jz z0 j < r g is the open
Euclidean disk with center z0 and radius r .
The proof applies to both cases R or C, where f 0 denotes, respectively, the
real or complex derivative.
Proof. Since jf 0 .a/j < 1 we may select ˇ such that jf 0 .a/j < ˇ < 1. Since, by definition,
f 0 .a/ D limz!a f .z/z af .a/ , there exists " > 0 such that for any z 2 n fag for which
jz aj < ", we have
ˇ ˇ ˇ ˇ
ˇ f .z/ a ˇ ˇ f .z/ f .a/ ˇ
ˇ ˇDˇ ˇ < ˇ:
ˇ z a ˇ ˇ z a ˇ
Hence for z 2 with jz aj < ", we know that jf .z/ aj ˇjz aj < ". This
says that for points z near a (within a distance of "), the function f moves z closer to
a by a factor of at least ˇ < 1. Hence a is an attracting fixed point by Definition 1.16.
If we iterate the map f at z, generating the orbit of z, we know that each application
of f takes each orbit point a step closer to a, by a factor of ˇ. Hence induction shows
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 10 — #28

i i
that jf n .z/ aj ˇ n jz aj ˇ n " ! 0 whenever z 2 with jz aj < ". Thus

4.a; "/ \ Af .a/.
Remark 1.18. The smaller the value of jf 0 .a/j is (and hence the smaller the value of ˇ
may be chosen) in the theorem, the faster the convergence is. In particular, if f .a/ D a and
f 0 .a/ D 0, then ˇ can be taken to be extremely small leading to very fast convergence,
and so such a fixed point a is called super attracting.
Example 1.19. Use the Real Function Iterator Applet to compare the rates of convergence
given by the following maps. For f .x/ D sin x consider the rate at which f n . 21 / converges
to 0. For g.x/ D x 2 consider the rate at which gn . 12 / converges to 0. For h.x/ D 21 x con-
sider the rate at which hn . 12 / converges to 0. Compare the absolute value of the derivative
at each function’s fixed point and note the correspondence with Remark 1.18. Try it out!
By Theorem 1.17, to check if a fixed point is attracting, you can calculate the abso-
lute value (or modulus) of the derivative at the fixed point. In Example 1.11, we see that
jf 0 .0:739085::::/j < 1, which proves that the fixed point is attracting. When considering
the Newton’s method dynamics, it appears in the cases we explored experimentally, that
the roots of f are attracting fixed points for the Newton map F . This is the case, and we
will prove it in Proposition 1.20 by showing that jF 0 j < 1 at each root of f . But before we
do, we give a word of caution.
We must be careful using Theorem 1.17 since it is not an if and only if statement.
Consider the fixed point a D 0 for the map f .x/ D sin x in Example 1.10. Here f 0 .0/ D
1, but this function as a real map, defined only for all real numbers x, does have a genuine
attracting fixed point at a D 0. However, if we consider the complex map g.z/ D sin z,
defined for all complex numbers z, the fixed point a D 0 is no longer attracting. Indeed, as
you are asked to prove in Additional Exercise 1.160, gn .˙i "/ ! 1 for any " > 0, which
you can illustrate using the Complex Function Iterator Applet. Try it out!
1.2.4 Complex Newton’s Method

We now investigate Newton’s method when we allow our variables and output values to
be complex. As we saw in the examples, our experimentation with the applets suggest that
the Newton map always has an attracting fixed point at each root of f . We can now state
and prove this fact in the real and complex cases.
Recall that a real-valued function of a real variable is said to be real analytic if it
possesses derivatives of all orders and agrees with its Taylor series in a neighborhood of
every point in its domain set.
Proposition 1.20 (Attracting Property of Newton’s Method). Given a non-constant real or
complex analytic function f with a root at ˛ 2 C, the point ˛ is an attracting fixed point
of the Newton map F and thus there exists r > 0 such that all points within a distance r of
˛ are in the attracting basin AF .˛/.
Put another way, the proposition states that for initial values z0 that are close enough
to ˛ (within a distance r ) the successive approximations F n .z0 / converge to ˛. That is,
starting from such a z0 and defining znC1 D zn ff0.z n/
.zn /
, we have zn ! ˛.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 11 — #29

i i
Proof. We consider the case that f .z/ is complex analytic. The same proof applies when
f is real analytic. From Theorem 1.17, it suffices to show that for F .z/ D z ff0.z/
.z/
we
00
have F .˛/ D ˛ and jF 0 .˛/j < 1. We note that F 0 .z/ D f.f.z/f .z/
0 .z//2 .
Since f .˛/ D 0, we apply Appendix A Lemma A.22 on page 354 to express f .z/ D
.z ˛/k h.z/ where h.˛/ ¤ 0 and k 2 N is the multiplicity of the root. If f 0 .˛/ D 0, the
quotient ff0.z/
.z/ appearing in the definition of F .z/ is not formally defined at ˛. However,
f .z/
f 0 .z/
has a removable singularity at ˛ because
f .z/ .z ˛/k h.z/ .z ˛/h.z/

D D ;
f 0 .z/ k.z ˛/k 1 h.z/ C .z ˛/k h0 .z/ kh.z/ C .z ˛/h0 .z/
which equals 0 for z D ˛: This lets us define F .˛/ D ˛ whether f 0 .˛/ D 0 or not.
Exercise 1.21 asks you to show that F 0 .z/, which has a removable singularity at ˛
when f 0 .˛/ D 0, gives F 0 .˛/ D k k 1 . Thus jF 0 .˛/j < 1, which completes the proof.
Exercise 1.21. Provide the details in the proof that F 0 .˛/ D k k 1 when f has a root of
order k. Also, note that ˛ is a super attracting fixed point of the Newton map F if and only
if k D 1, i.e., ˛ is a simple root of f . Try it out!
It is interesting to note that the Newton map of f .z/=f 0 .z/ (as opposed to the Newton
map of f .z/) always has super attracting fixed points at the roots of f regardless of the
order of the root of f . You are asked to show this in Additional Exercise 1.162.
Exploration 1.22 (Convergence Rates for Newton’s method). Let f .z/ and g.z/ be an-
alytic, with Newton maps Ff .z/ D z ff 0.z/
.z/ and Fg .z/ D z
g.z/
g 0 .z/ . Suppose f has
a root at ˛ of order k and g has a root at ˛ of order m, where k < m. The rate of
convergence of Ffn near ˛ is faster than the rate of convergence of Fgn near ˇ since
jFf0 .˛/j D k k 1 < mm 1 D jFg0 .˛/j (Remark 1.18 relates the rate of convergence to the
derivative at the attracting fixed point). Let’s explore this in the real variable case with
f .x/ D x k and g.x/ D x m . Use the Real Newton Method Applet to visualize how the or-
der of the root of f (respectively, g) influences the tangent lines used in Newton’s method.
Note the effect the order of the root has on the curvature of the graph near the root and how
this provides a visual way to understand the relative rates of convergence of Ffn and Fgn
near ˛. Try it out!
Remark 1.23. We see that the value r in Proposition 1.20 gives us a lower bound on how
close an initial guess z0 must be to a root ˛ for Newton’s method to be guaranteed to find
˛. Because of this, we call any such r a radius of convergence (for F and ˛) and call the
corresponding circle C.˛; r / D fz W jz ˛j D r g a circle of convergence. In practice, it is
useful to gauge r so that we can guarantee the success of Newton’s method. In Additional
Exercise 1.163 you are asked to show that there is no universal estimate for r that always
works since it very much depends on the particular map f . However, for certain classes
of maps f , useable estimates for r can be found and you can explore these in Additional
Exercise 1.164 and Small Projects 1.165 and 1.166.
Let us now consider some examples using complex functions.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 12 — #30

i i
Example 1.24. For constants ˛; ˇ 2 C, let f .z/ D .z ˛/.z ˇ/, with roots at ˛ and ˇ.
2 ˛ˇ
The Newton map for f is F .z/ D 2zz .˛Cˇ/ :
Exploration 1.25. Set ˛ D 0 and ˇ D 1 C i in Example 1.24 and iterate the Newton
map F with starting values z0 D 2, z0 D 3 2i , and z0 D i C 1, which you can
compute using the Complex Newton Method Applet. As expected, different starting values
for z0 find different roots of f .z/. Can you make a guess as to which seed values will find
which root? Try to determine if there are any seed values z0 for which Newton’s method
fails to find either root. Experiment with the Complex Newton Method Applet to test your
predictions. Try it out!
Exploration 1.26. Using f .z/ D z 3 1, determine the Newton map for f and then an-
alyze Newton’s method using different initial guesses in the Complex Newton Method Ap-
plet. Describe (as best as you can) which seed values will find which of the roots 1; e 2 i=3,
and e 2 i=3 . We are asking you to describe (as best as you can) the basins of attraction
AF .1/; AF .e 2 i=3 /; and AF .e 2 i=3 /. Try it out!
1.2.5 Global Picture of Complex Newton Method Dynamics

As with the real-valued maps, we wish to understand the global dynamics of Newton’s
method, i.e., how the orbits behave for all seed values. We employ the Graph basins
of attraction feature of the Complex Newton Method Applet, which uses different
colors to display which initial guesses will find which roots.
Exercise 1.27. Use the Complex Newton Method Applet to view the basins of attraction
for the Newton maps in Explorations 1.25 and 1.26. Try it out!
Looking at the picture of the two attracting basins corresponding to Exploration 1.25,
it appears that the boundary between them is a straight line; points on one side look closer
to the root of f on that side than to the other root. That is, this boundary appears to be
the perpendicular bisector (denoted by L in Figure 1.3) of the line segment from ˛ to ˇ.
This is true (pictures, however, can sometimes be misleading) and we can prove it using
the technique of global conjugation.
1.2.6 Global Conjugation

We learn in linear algebra that a change of basis can facilitate calculations and procedures,
and lead to better understanding. Similarity of matrices plays a key role (matrices A and
B are called similar when there is an invertible matrix P such that A D PBP 1 ). An
analog often used in dynamics is a type of change of coordinates provided by what we call
conjugation.
Definition 1.28. Let be a Möbius map, i.e., .z/ D azCb
czCd
where ad bc ¤ 0, noting it is
a bijection of C onto itself. When f and g are rational maps (quotients of two polynomials)
such that g D ı f ı 1 , we say f and g are globally conjugate by the map .
Often the point of conjugating a map f to a map g is that g is easier to work with than
f . And, as we will see, the information we usually want from f can be obtained from the
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 13 — #31

i i
simpler map g. Iterates are related as such, gn D . ı f ı 1 / ı . ı f ı 1 / ı ı

. ı f ı 1 / ı . ı f ı 1 / D ı f n ı 1 . Hence transfers information between
the iterates of f and the iterates of g. In particular, fixed points and their derivatives are
transferred in the following way.
Exercise 1.29. Suppose maps f and g are globally conjugate by the Möbius map , i.e.,
g D ı f ı 1 . Prove that for a 2 C, f .a/ D a if and only if g..a// D .a/.
Furthermore prove that if a; .a/ 2 C and f .a/ D a, then we have f 0 .a/ D g0 ..a//.
Try it out!
Additional Exercises 1.167–1.170 explore several examples of global conjugation.
Remark 1.30. There is also a very useful technique called local conjugation that can
simplify calculations. In fact, an important problem is determining for which maps of the
form f .z/ D a1 z C a2 z 2 C : : : , having a fixed point at 0, can there be found be an analytic
map defined near 0 such that ı f ı 1 is simply z 7! a1 z. This is called linearizing
the map f near 0. It can always be done when 0 < ja1 j ¤ 1, but only sometimes when
ja1 j D 1. Such results and their proofs can be found in the literature (e.g., [1, 3, 24]).
1.2.7 Newton Map of a Quadratic Polynomial

We now use global conjugation to simplify the analysis of the Newton map F .z/ D
z 2 ˛ˇ
2z .˛Cˇ/
in Example 1.24. We first choose a Möbius map that sends ˛ and ˇ to 0 and
1, respectively, and then analyze the simpler map obtained by conjugation. In particular,
the map .z/ D zz ˇ˛ conjugates F to g.z/ D ı F ı 1 .z/ D z 2 . We leave it to you to
n
show this, as well as to show that gn .z/ D z 2 .
Our goal is to show that if z 2 C is closer to ˛ than to ˇ, then F n .z/ iterates to
˛, i.e., jz ˛j < jz ˇj implies F n .z/ ! ˛. Let jz ˛j < jz ˇj; which implies
n n
j.z/j < 1. Since jgn ..z//j D j.z/2 j D j.z/j2 ! 0, we have gn ..z// ! 0. By
the conjugation property we have F n .z/ D 1 .gn ..z/// ! 1 .0/ D ˛. Thus we have
shown that the points z 2 C which are closer to ˛ than to ˇ are in AF .˛/.
Exercise 1.31. Show that the points z 2 C that are closer to ˇ than to ˛ are in AF .ˇ/. Try
it out!
We illustrate this conjugation in Figure 1.3, called a commutative diagram because the
maps ı F and g ı from the upper left to the bottom right are equal.
The points ˛ and ˇ are moved by to the points 0 and 1, and the line L [ f1g
in the top pictures is transformed to the unit circle C.0; 1/ in the bottom pictures. Using
this conjugation, we are able to analyze the relatively simple dynamics of g to get in-
formation about the dynamics of F , in particular, AF .˛/ D 1 .Ag .0// and AF .ˇ/ D
1 .Ag .1//.
Let’s return to the question of whether Newton’s method can fail in this example. Are
there initial values for which the Newton’s method orbit never finds a root of f ? In Ex-
ploration 1.5, we saw that Newton’s method fails when an initial value x0 is such that
f 0 .x0 / D 0. This, what we term an analytic obstruction of having a zero in the denomina-
tor, however, is overcome when we allow 1 to take its equal place with the other values in
the Riemann Sphere C. Even though in Example 1.24 we have f 0 . ˛Cˇ 2
/ D 0, the Newton
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 14 — #32

i i
L β L β
F(z)
α α
φ φ
g(z) = z2
0 0
C(0,1) C(0,1)
F IGURE 1.3. Commutative diagram for global conjugation of the Newton map F .z/ of f .z/ D
.z ˛/.z ˇ/.
map gives F . ˛Cˇ 2

/ D 1 (see Appendix Section B.4 on page 361 for a review of a dis-
cussion on functions defined at 1). Also, since F .1/ D 1, we see that an initial value
z0 D ˛Cˇ 2
will never find either root ˛ or ˇ because it generates the following sequence of
Newton approximations ˛Cˇ 2
; 1; 1; 1; : : : . The obstruction for the success of Newton’s
method starting with seed value z0 is not that F cannot be appropriately defined at z0 , but
that such a definition forces F to never iterate z0 to a root of f .z/ (since F n .z0 / D 1 for
all n 1). The seed ˛Cˇ 2
is not the only complex number that fails to find a root of f .z/
though. We can show, again with the help of conjugation, that the boundary line L that di-
vides AF .˛/ from AF .ˇ/ consists of exactly those points in C for which Newton’s method
fails to find either root. As in the real variable examples, we can understand this dynami-
cally since it divides the points that are pulled or attracted to ˛ from those that are attracted
to ˇ. We would think that something must fail to work out for points on this line. We have
not proven anything carefully yet. We have only looked at compelling computer-generated
evidence, which is always to be viewed with a healthy bit of skepticism.
However, a formal argument can be made. For a point z 2 L, we have j.z/j D
jz ˛j n
jz ˇj
D 1 and thus by the conjugation property we have jgn ..z//j D j.z/2 j D
n
j.z/j2 D 1 for all n 2 N. Hence for all n 2 N, we see that F n .z/ D 1 .gn ..z/// 2
1 .C.0; 1// D L [ f1g. In particular, F n .z/ converges to neither ˛ nor to ˇ.
Exercise 1.32. The analysis applies to any monic quadratic polynomial p.z/ with distinct
roots, but what if the leading coefficient is not 1? What if the quadratic polynomial p.z/
has a double root instead of two distinct roots? Analyze what happens in these situations.
Try it out!
1.2.8 Newton Map of a Cubic Polynomial

The behavior found in Exploration 1.26 with the cubic function f .z/ D z 3 1 is far more
complicated than the quadratic case. A starting point under Newton’s method will not
always find the root to which it is nearest. If it did, then the picture of the global dynamics
would look like Figure 1.4. However, the better picture that represents the dynamics is
Figure 1.5, which shows that the set of initial guesses in the complex plane is divided into
very intricate regions.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 15 — #33

i i
e 2pi/3
e –2pi/3
F IGURE 1.4. A reasonable (but false) guess for the description of the global dynamics of F .z/, the
Newton map for f .z/ D z 3 1.
F IGURE 1.5. A more accurate picture of the global dynamics of F .z/, the Newton map for f .z/ D
z 3 1 (with magnification on the right). Here the turquoise regions represent AF .1/, the blue regions
represent AF .e2 i=3 / and the red regions represent AF .e 2 i=3 /.
Let’s experiment with the zoom feature of the Complex Newton Method Applet to in-
vestigate this. When you zoom in on a point that is on the boundary of one colored region
(attracting basin), you always find tiny bulbs of the other two colors (attracting basins)
nearby. This happens no matter how much you zoom in! This shocking feature is why we
call such sets fractals (a set which when you zoom in reveals new features not seen from
the coarse larger scale picture). In this case, at a large scale there are extremely tiny bulbs
(smaller than a single pixel) that are not revealed in the picture unless one zooms in far
enough to see them.
So again we see that Newton’s method became very complicated when f .z/ changed
from a quadratic to a cubic map. Earlier we posed the questions: Why? How? Is there a
way to know when a system will necessarily be simple or complicated ahead of time? With
the depth of complex analysis knowledge to aid us we can give some good reasons why the
pictures, and hence the dynamics they represent, must be so complicated. We first remind
ourselves of some key concepts.
Definition 1.33. The boundary of a set E in C is @E D E \ C n E, which is the set of

points z 2 C that have the property that every open disk in the spherical metric 4 .z; r /
intersects both E and the complement of E no matter how small r > 0 is.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 16 — #34

i i
We can now describe the fractal features we observed in Figure 1.5 by saying @AF .1/ D
@AF .e 2 i=3 / D @AF .e 2 i=3 /: We see this phenomenon in Example 1.4 and Explo-
rations 1.6 and 1.25 as well when we consider the complex versions of these maps. It
turns out that, in general, all attracting basins of a Newton map F share the same set of
boundary points. Specifically, we have the following Common Boundary Condition. The
proof, however, uses Proposition 1.20 and the forthcoming Theorem 1.59 and so cannot be
presented until we develop a few more concepts.
Theorem 1.34 (Common Boundary Condition). Let f .z/ be an analytic function such that
its Newton map F .z/ D z ff0.z/
.z/
is a rational map. If w1 and w2 are roots of f .z/, then
@AF .w1 / D @AF .w2 /.
Let’s examine how Theorem 1.34 forces the dynamics in Figure 1.5 to be compli-
cated. By Proposition 1.20 there exists r > 0 such that 4.1; r / AF .1/; 4.e 2 i=3 ; r /
AF .e 2 i=3 /, and 4.e 2 i=3 ; r / AF .e 2 i=3 /: So, starting from Figure 1.6, we need to
consider how to color the rest of the points in AF .1/; AF .e 2 i=3 /, and AF .e 2 i=3 / with
turquoise, blue, and red, respectively, while being certain to make sure that the boundary
of each color matches the boundary of the other two colors. Reflecting on what this means
we see that this Common Boundary Condition forces the picture to be very complicated,
and also rules out having the global dynamics in Figure 1.4 since, for example, a point on
the negative real axis lies on the boundary of only two of the three color basins.
∆(e2 i/3, r)
∆(1, r)
∆(e-2 i/3, r)
F IGURE 1.6. Consequence of the Attracting Property of Newton’s Method (Proposition 1.20).
This Common Boundary Condition is also at the heart of what we call chaos in the
dynamics of Newton’s method. Consider a point z on the boundary of any basin and then
draw a tiny disk B around it. According to the Common Boundary Condition, it contains
all three colors. So if we wish to determine the orbit of z with a computer, what we will find
is that tiny errors, such as roundoff error, in the coordinates of z could lead to drastically
different results. By changing z even by the slightest amount, we can change the orbit of
z tremendously as z could slip into any of the red, turquoise, or blue regions. This is the
essence of what we call chaos. Although a more formal and more thorough understanding
of chaos will come later in the text, use our current understanding to explain why each
point on the line L that appeared in the analysis of the quadratic Newton method case has
this behavior. Also, explain why each point in one of the attracting basins (which has a tiny
disk of all the same color) is not such a point. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 17 — #35

i i
Remark 1.35. We mention here a remarkable fact about sets that share a complicated
boundary. The sets AF .1/; AF .e 2 i=3 /, and AF .e 2 i=3 / share the same boundary be-
cause each is broken up into an infinite number of pieces that are then arranged in the
complicated fractal pattern in Figure 1.5. However, it is true that three open sets (or even
n open sets for any fixed n 2 N) can all share the same boundary set and also have the
property that each is connected (an open set A is called connected when given any two
points z; w 2 A there exists a polygonal line in A which connects z to w). Such sets are
complicated indeed! You can read about the Lakes of Wada in [16, p. 143].
Let us now consider the question of whether there are any seed values for which the
Newton map for f .z/ D z 3 1 fails to find a root of f .z/. In the previous examples there
always existed such points; however, they were, in some sense, relatively few. You might
guess that each point in @AF .1/ D @AF .e 2 i=3 / D @AF .e 2 i=3 / is such a point, and
you would be right. We can use our intuition to imagine that such points fail to find any
root because they are being attracted by the three different roots with equal force. You are
asked to prove this in Additional Exercise 1.171. However, assuming that each point in
@AF .1/ D @AF .e 2 i=3 / D @AF .e 2 i=3 / fails to iterate under F .z/ to any of the roots of
f .z/ does not necessarily give the fate of all starting points. Is it true that all points in C
lie in either one of the attracting basins or on the common boundary of these sets? We ask
1. Is there a point not in @AF .1/ D @AF .e 2 i=3 / D @AF .e 2 i=3
/ that fails to find any
root of f .z/?
2. Can there be an open disk of such points?
When considering f .z/ D z 3 1 (or any of the maps f mentioned thus far in this
chapter) and its related Newton map F .z/, the answer to both questions is no, as you are
asked to explore in Additional Exercise 1.182. However, it is possible for the answer to the
second question, and hence the first, to be yes.
Exploration 1.36. Consider the map f .z/ D z.z 1/.z :909 :416i / and its related
Newton map F .z/. Using the Complex Newton Method Applet you can find regions of
seed values colored black that fail to find a root of f .z/. For example, there is one near
0:64 C 0:14i . Use the applet to select such a point and then iterate F .z/ to explore the
behavior. Experiment with many seed values chosen from the black regions. Zoom to see
whether or not it appears that the boundary of the black region matches the boundary of
the attracting basins for the roots of f .z/. Try it out!
You can investigate Newton’s method applied to other cubic polynomials in Additional
Exercise 1.172.
1.2.9 Newton’s Method Applied to a Family of Cubic Polynomials

In this section we apply Newton’s method to an arbitrary cubic polynomial. Let F denote
the collection of all polynomials p .z/ D z.z 1/.z / where is in D D f W Im
0; jj 1; j 1j 1g.
This class of polynomials F is representative of all polynomials with three distinct
roots in the following sense.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 18 — #36

i i
0 1
F IGURE 1.7. Region D of values corresponding to the maps p in F .
Proposition 1.37. For a cubic polynomial p.z/ with three distinct roots, there exists 2 D
such that the Newton map Fp is globally conjugate by a linear map T to the Newton map
Fp of p .z/, i.e., T ı Fp ı T 1 D Fp .
In order to prove it we require the next proposition, which applies to polynomials of
any degree.
Proposition 1.38. Let p.z/ be a polynomial and let T .z/ D az C b for a ¤ 0 where
a; b 2 C. Then for the polynomial q.z/ D p.T .z//, we have
1
T ı Fq ı T D Fp
where Fq and Fp are the Newton maps of q and p respectively.

We leave the proof of Proposition 1.38 to the reader. However, we say a few words
about its meaning and usefulness. The polynomial q (which is sometimes called the “rescal-
ing” of p) has the same degree as p. Furthermore, if p has roots at r1 ; : : : ; rd , then q will
have roots at T 1 .r1 /; : : : ; T 1 .rd /. So the proposition says that we can move the roots
of p by choosing T 1 appropriately and generating a new polynomial q. Studying the dy-
namics of Fq will be essentially the same as studying the dynamics of Fp since they are
globally conjugate to each other.
Example 1.39. Let p.z/ D .z C i=2/.z 1/.z C 1/. We illustrate Proposition 1.37 by
finding 2 D so that Fp is conjugate to Fp . A triangle is formed by the roots of p.z/
at i=2; 1, and 1. We construct a linear map T 1 to transform it into a triangle whose
longest side is the interval Œ0; 1 and whose third vertex (which will be our choice of ) is
in the upper half plane fIm z 0g. Then T 1 .1/ D 0 and T 1 . 1/ D 1, which determine
that T 1 .z/ D 21 z C 12 and D T 1 . i=2/ D 1=2 C i=4 which is in D.
Let’s verify that this choice of works. Because q.z/ D p.T .z// has roots at 0, 1, and
and q and p share the same roots, they agree up to a multiplicative constant, i.e., we
have q.z/ D cp .z/ for some constant c ¤ 0. This implies that Fq D Fp (why?). Hence,
by Proposition 1.38, we have that T ı Fp ı T 1 D T ı Fq ı T 1 D Fp as desired.
Use the Complex Newton Method Applet to compare the pictures of the attracting basins
for Fp and Fp noting the similarities because they are globally conjugate. Try to see the
effects of T when you compare the two pictures. Try it out!
Exercise 1.40. Find 2 D so that for p.z/ D .z 4/.z C i /.z C 4/ the map Fp is
globally conjugate to Fp . Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 19 — #37

i i
1.3. Iteration of an Analytic Function 19
Proposition 1.37 can be proven as in Example 1.39 and so we omit the details. We leave
it to the reader to investigate what can be said about the dynamics of Fp when p.z/ is a
cubic polynomial with a double or triple root. In addition, we invite the reader to consider
generalizations such as the following.
Small Project 1.41. Consider how Proposition 1.38 is used to move roots by a linear
transformation T 1 in the proof of Proposition 1.37. With this in mind, can you justify the
statement that all cubic polynomials whose roots form an equilateral triangle have globally
conjugate Newton maps (and hence essentially the same dynamics)? What about quartic
polynomials whose roots form a square? Does this generalize to higher degree polynomi-
als? Do the roots need to form a regular n-gon? What can you say about generalizing the
class F (and the set D) when considering polynomials of fixed higher degree?
Aside from the exceptional cases mentioned above, we can study the Newton’s method
dynamics of all cubic polynomials by studying the maps p where 2 D. We have cre-
ated the Cubic Polynomial Complex Newton Method Applet to help. It allows the user to
generate the pictures of the attracting basins for Newton’s method applied to any p , not
just 2 D. Also, it allows the user to investigate the parameter plane of values since
each such value will be colored according to the dynamics of Fp .
We present a note about the coloring of the parameter plane of values. You will see
in Section 1.4 that the orbit of points where Fp0 .z/ D 0 are critically important to under-
p .z/p00 .z/
standing the dynamics of Fp . Since Fp0 .z/ D .p
0
0
2 , we see that Fp .z/ D 0 only at
.z//
00
the roots of p .z/ or when p .z/ D 0. Since the roots of p .z/ are attracting fixed points
for Fp , their orbits are understood. However, since p00 .z/ D 0 for z D .1 C /=3 this, so-
called, free critical point is important. In the Cubic Polynomial Complex Newton Method
Applet, we track, for each , the critical orbit of z0 D .1 C /=3. When it is attracted to
one of the roots of p , we color in the parameter plane the corresponding color of the
attracting basin. To illustrate, click on the Show Critical Orbit checkbox to see the
critical orbit appear as white dots in the right picture (dynamical plane). Thus, if the critical
orbit converges to, say, the red root of p , then (marked by a +) is colored red in the left
picture (parameter plane). If this orbit is not attracted to any of the roots of p , we color
black. Try clicking in the parameter plane to a select different colored value, and then
observe the convergence of the critical orbit to a different colored root of p .
Exploration 1.42. Use the Cubic Polynomial Complex Newton Method Applet to investi-
gate the dynamics of Newton’s method applied to any p . Look for symmetries and exper-
iment with the dynamical behavior you find. Make conjectures, and see if you can prove
them. Try it out!
1.3 Iteration of an Analytic Function

In this section we study the dynamics of analytic maps that do not necessarily arise as New-
ton maps of polynomials. One goal is to understand which dynamical features of Newton’s
method extend to such a wider class of maps. We pay particular attention to the iteration
of polynomial maps of the form z 2 C c where c is a complex parameter since these are the
simplest maps that produce a rich variety of dynamical behaviors. (The iteration of Möbius
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 20 — #38

i i
maps is simpler to study because they have relatively simple dynamical behavior. This is
shown in Appendix 1.B.) Additional Exercise 1.169 asks you to use the global conjugation
technique of Section 1.2.6 to show that every quadratic map is globally conjugate, and in
some sense dynamically equivalent, to exactly one map of the form z 2 C c. Hence, in this
way the dynamics of maps z 2 C c represent the dynamics of all quadratic maps.
Sections 1.2.3 and 1.2.6 must be read before proceeding with this section, but not all
of Section 1.2 is necessary.
Though quadratic maps of the form z 2 C c are genuinely very simple and well under-
stood as functions, we will see that their dynamics can be complicated. We will look at the
dynamics of a map z 2 C c noting how the dynamics change as c changes, thus leading us
into bifurcation theory.
In Section 1.6, we consider the dynamics of more exotic transcendental entire complex
analytic maps such as ce z ; c sin z, and c cos z.
1.3.1 Classification of Fixed Points for Analytic Maps

In Section 1.2.3 we defined attracting fixed points. Now we extend our discussion to two
more types of fixed points, repelling and indifferent. We begin with an example.
p
Example 1.43. Let f .x/ D x 2 and g.x/ D x, restricted so that f; g W RC ! RC where
RC D fx 0g: Clearly 0 and 1 are fixed points of both f and g. We have f n .x/ ! 0
if 0 < x < 1; f n .x/ ! 1 if x > 1, and gn .x/ ! 1 for x 2 RC n f0g. Note that 0 is
an attracting fixed point for f (why?). If x is close to, but not equal to 1, the orbit f n .x/
moves away from 1. We then call 1 a repelling fixed point for f . The function g.x/ has
an attracting fixed point at 1 and a repelling fixed point at 0. We illustrate the dynamics in
Figure 1.8.
f f g g
0 1 0 1
F IGURE 1.8. Dynamics of f and g D f 1.
Because f and g are inverses of each other, it is reasonable that an attracting fixed
point for f must be a repelling fixed point for g, and vice versa. This becomes clear by
considering the following definition.
Definition 1.44 (Repelling Fixed Point). Let f be a map with domain set C, which
could be a subset of R.
(a) We call a (finite) fixed point a 2 C a repelling fixed point of f if there exists a
neighborhood U of a such that for z 2 \ U n fag, we have jf .z/ aj > jz aj,
i.e., the action of f is to move each point in \ U n fag farther from a.
(b) Suppose f .1/ D 1. We call 1 a repelling fixed point of f if there exists a neigh-
borhood U C of 1 such that for z 2 \ U n f1g, we have jf .z/j < jzj, i.e., the
action of f is to move each point in \ U n f1g farther from 1, as measured by the
spherical metric.
Theorem 1.45. Suppose R or C. Let f W ! be such that f .a/ D a and
jf 0 .a/j > 1. Then a is a repelling fixed point of f . Also, there exists " > 0 such that for
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 21 — #39

i i
each z 2 \ 4.a; "/ n fag the orbit f n .z/ eventually leaves 4.a; "/, i.e., there exists N
such that f N .z/ … 4.a; "/.
Remark 1.46. The proof is a modification of the proof of Theorem 1.17, and so we leave
the details to the reader. Additional Exercise 1.173 illustrates the fact that the theorem does
not preclude the possibility that the orbit of z, after leaving 4.a; "/, might reenter 4.a; "/.
Remark 1.47. Theorem 1.45 is not an if and only if result. Find a real-valued function
f W R ! R such that 0 is a repelling fixed point that has jf 0 .0/j D 1. Hint: Consider maps
of the form x 7! x ˙ x n .
We provided examples of real-valued maps which show that the converses of Theo-
rems 1.17 and 1.45 do not hold. However, the next theorem shows that such examples
cannot be found amongst the complex analytic maps.
Theorem 1.48. Let f .z/ be an analytic map on an open set C such that f .a/ D a.
Then,
(a) the point a is an attracting fixed point if and only if jf 0 .a/j < 1, and
(b) the point a is a repelling fixed point if and only if jf 0 .a/j > 1.
Theorems 1.17 and 1.45 provide two of the four implications. The proofs of the remain-
ing two, outlined in Additional Exercise 1.174, use special properties of complex analytic
maps.
Theorem 1.48 applies only to finite fixed points and so we ask if there is a correspond-
ing result when 1 is fixed. The following examples prepare for handling this.
Example 1.49. The map h.z/ D z C 1 on C fixes 1 and shifts a point in the complex
plane one unit to the right. For large and positive real z, we have jh.z/j > jzj, which
makes it appear that 1 is attracting. However, for large and negative real z, we see that
jh.z/j < jzj, which makes it appear that 1 is repelling. We conclude that the fixed point
at 1 is neither attracting nor repelling. The nth iterate hn .z/ D z C n and so for z in
C we have hn .z/ ! 1, always moving parallel to the x-axis. Thus Ah .1/ is the entire
Riemann Sphere C. This is somewhat surprising seeing that 1 is not an attracting fixed
point. We represent the dynamics on C and C in Figure 1.9.
F IGURE 1.9. Graphical representations of the dynamics of h.z/ D z C 1 on the plane C (left) and
on the Riemann sphere C (right).
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 22 — #40

i i
Example 1.50. The map g.z/ D z=2 on C takes z 2 C to a point with one half the
modulus, but with the same argument. The origin is an attracting fixed point and Ag .0/ D
C. Also, g.1/ D 1 and if z 2 C is near 1 then g.z/ moves away from 1, i.e., jg.z/j <
jzj. Thus 1 is a repelling fixed point. Note also that gn .z/ D z=2n . We represent the
dynamics on C and C in Figure 1.10.
F IGURE 1.10. Graphical representations of the dynamics of g.z/ D z=2 on the plane C (left) and
on the Riemann sphere C (right).
n
Example 1.51. For f .z/ D z 2 on C we have that f n .z/ D z 2 for all n D 1; 2; : : : .
n n
Thus, jzj < 1 implies jf .z/j D jzj2 < jzj and jf n .z/j D jz 2 j D jzj2 ! 0. Also,
n n
jzj > 1 implies jf .z/j D jzj2 > jzj and jf n .z/j D jz 2 j D jzj2 ! C1. Thus, we
conclude that 0 and 1 are attracting fixed points with attracting basins Af .0/ D 4.0; 1/
and Af .1/ D C n 4.0; 1/. We represent the dynamics in Figure 1.11, noting that the
angle doubling property of the map is not represented. If z0 D r e i in polar form we have
that f .z0 / D r 2e i 2 . So, starting with seed z0 D 0:999e i =100; the orbit zn will converge
towards 0, doubling the angle at each step. Observe this in the Complex Function Iterator
Applet, using the Polar seed form and the Polar computation mode.
F IGURE 1.11. Graphical representation of the dynamics of f .z/ D z 2 .
The last three examples were chosen to illustrate fixed points at 1. We were able to
calculate a formula for f n . This is rare. However, having a formula for the iterates f n is
unnecessary because we can analyze the dynamics without appealing to them.
In Examples 1.49, 1.50, and 1.51 we saw that 1 is a fixed point that we were able
to classify by examining the dynamics near 1. However, we wonder (in light of Theo-
rem 1.48) if we can use the derivative to classify fixed points at 1. It turns out that we can,
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 23 — #41

i i
but we must be careful. We saw that jh0 .1/j D 1 corresponded to a fixed point that was
neither attracting nor repelling, jg0 .1/j D 1=2 corresponded to a repelling fixed point,
and f 0 .1/ D 1 corresponded to an attracting fixed point. The derivative at 1 does not
play the same role as it does in Theorem 1.17 for finite points. However, as we now see, it
is not far off.
Definition 1.52 (Multiplier of a Fixed Point). When f is analytic at 1 with f .1/ D 1,
we define the multiplier at 1 to be 1=f 0 .1/ D limz!1 1=f 0 .z/. When a is a finite
fixed point in C, we define the multiplier at a to be f 0 .a/.
We now classify a fixed point of an analytic map, whether finite or not, based on its
multiplier.
Definition 1.53 (Classification of Fixed Points). Suppose f W ! is analytic where
is an open subset of C and a is a fixed point with multiplier . Then a is called
(a) super attracting if D 0
(b) attracting if 0 < jj < 1
(c) repelling if jj > 1
(d) indifferent if jj D 1.
You can check that this matches what we found in Examples 1.49, 1.50, and 1.51. The
motivation for the definition of the multiplier when f has a fixed point at 1 is that by
globally conjugating f by .z/ D 1=z we obtain the map k.z/ D 1=f .1=z/, which has
a fixed point at 0. We defined the multiplier of f at 1 to be the multiplier of k at 0,
which is calculated in Appendix B Lemma B.20 on page 363 to be k 0 .0/ D 1=f 0 .1/. In
Additional Exercise 1.175, you are asked to prove that the multiplier correctly corresponds
to the notions of attracting and repelling fixed points as given in Definitions 1.16 and 1.44.
Remark 1.54. Indifferent fixed points can exhibit both a partial attracting nature and a
partial repelling nature. For example, we saw in Section 1.2.3 that the indifferent fixed
point at the origin for the complex map g.z/ D sin z is attracting for real-valued seeds,
but repelling for purely imaginary seeds. Likewise, the indifferent fixed point at 1 for
the map h.z/ D z C 1 attracts large positive real z, but repels (at least initially) large
negative real z.pHowever, sometimes an indifferent fixed point, like the origin under the
map f .z/ D e 2 i z, neither attracts nor repels. Indifferent fixed points of analytic maps
can exhibit many types of dynamical behavior and their study can be complicated. It is
complicated enough that we will not say more about them here, but refer the interested
reader to [1, 3, 24].
1.3.2 A Closer Look at the Dynamics of f .z/ D z2

n n
Let us return to the dynamics of f .z/ D z 2 . If jzj D 1; then jf n .z/j D jz 2 j D jzj2 D
n
12 D 1, i.e., if a point is on the unit circle C.0; 1/ D fz W jzj D 1g, then so is its entire orbit
ff n .z/g1nD1 . From this and from our previous work in Example 1.51 you might conclude
that its dynamics has been solved. As we will see, this is not at all close to the truth!
If we know the orbit of one seed value, will nearby seed values have similar orbits? If
z lies within the unit disk, then it and points near it in the unit disk have the same behavior,
namely, each orbit converges to 0. If z lies outside the closed unit disk, then it and points
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 24 — #42

i i
near it outside of the closed unit disk have the same behavior, namely, each orbit converges
to 1. However, the story is different for seed values on the unit circle. For any seed on
C.0; 1/ we can find other seed values arbitrarily close to it that have drastically different
orbits (namely that limit to either 1 or 0).
Let’s examine this using the computer. For the map f .z/ D z 2 use the Complex Func-
tion Iterator Applet to iterate the seed value z0 D 1 C i as well as seed values close to z0 .
You can see that z0 and the nearby seed values have the same behavior. Repeat by iterating
z0 for the values 0:4 C 0:5i; 1; i; and 0:6 C 0:8i , testing nearby seed values that you
choose yourself.
Remark 1.55 (A word of caution about using computers). If we are trying to calculate
the orbit under the map f .z/ D z 2 of a point on C.0; 1/, we may have problems getting
our computer to provide accurate results. Let’s test this on the Complex Function Iterator
Applet. For the map f .z/ D z 2 try iterating seed value z0 D 0:6 C 0:8i (which lies on
C.0; 1/) by entering the seed value in Euclidean form and using the Euclidean
computation mode. Iterating 20 times, you will see what you would expect. However,
after iterating about 55 times you will see the orbit move inside the unit disk where it
will then iterate to 0. Why? If the computer truncates, rounds, or approximates any of the
values in the orbit (as computers often do), these small errors will likely push the orbit
outside or inside the unit circle, causing the computer mistakenly to calculate the orbit as
tending to either 1 or 0. When using a computer it is important to know if the computer
will approximate values it uses, and whether the approximations will be significant or not
in the end result.
As an illustration of the subtle issues that can wreak havoc on your computations, we
show the problems that can arise from the fact that z 2 and z z are not always equivalent.
Of course the expressions z 2 and z z are mathematically equivalent, but they are not
computationally equivalent. The former is evaluated as e 2Log z , where Log z is the princi-
pal logarithm, and the latter is evaluated with complex multiplication. Each incorporates
different rounding errors at times. The result of this very subtle difference is quite evident
when iterating the seed 0:6 C 0:8i in Euclidean computation mode under these
maps. Try it out!
So we see that it is important to identify those seed values where approximations (or
errors introduced by using approximations) lead to significant errors in calculations of orbit
values. Such seed values are said to be in the chaotic set. The chaotic set is called the Julia
set, in honor of the mathematician Gaston Julia who in 1918, at the age of 25, published
his 199-page masterpiece [17] titled “Mémoire sur l’itération des fonctions rationnelles”
describing the iteration of complex rational functions.
Remark 1.56. The Julia set is not tied only to errors or approximations that a computer
might introduce. We need to know whether a tiny error in the seed value (no matter what
caused it) could produce a significant error in an orbit value. For example, the seed z0 D i
would be in the Julia set of f .z/ D z 2 even though its orbit i; 1; 1; 1; 1; : : : is computed
exactly by a computer. However, we still say it is a chaotic seed value since given any
allowable error in the seed value, even one as small as 10 631 , we can find another seed z00
close to z0 D i (i.e., within the error) such that z00 has a drastically different orbit from the
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 25 — #43

i i
orbit of z0 . Thus, even if a computer wouldn’t introduce an error in its orbit calculation for
a seed z0 , the seed value might still be in the Julia set.
The Julia set for f .z/ D z 2 is C.0; 1/ because the dynamics there are chaotic, i.e., for
any z0 2 C.0; 1/ there is a point z00 arbitrarily close by that has a drastically different orbit.
A computer often fails to be accurate since we would need it to store such values and each
point in its orbit with an infinite degree of accuracy. To put it another way, approximation
of the starting value (or some future iterate value) ruins our confidence in the calculations
of the long term behavior. This is called sensitive dependence on initial conditions, and it
is the defining feature in what is called chaos.
We call the complement C n C.0; 1/ the stable set for f .z/ D z 2 because orbits are
stable there, i.e., for any z0 2 C n C.0; 1/ its orbit zn will behave like the orbit zn0 of any
seed value z00 chosen sufficiently close to z0 . The stable set is commonly called the Fatou
set because of Pierre Fatou’s role in developing the theory of complex function iteration
in 1917 (see [12, 13, 14]). In honor of their pioneering work in the field, the dynamics of
complex analytic functions is often called the Fatou-Julia theory.
Exercise 1.57. Use your understanding of the meanings of the Fatou and Julia sets to
determine these sets for the maps h.z/ D z C 1; g.z/ D z=2, and k.z/ D 3z. Try it out!
Exercise 1.58. Try to write a precise mathematical definition of the Fatou set and the Julia
set that applies to any rational function g.z/ defined on C. Try to formulate what it means
for orbits to be drastically or significantly different. Try it out!
Notation: For a rational function g.z/, we employ the notation F .g/ for the Fatou set of g
and J.g/ for the Julia set of g. Formal definitions along with a more advanced development
can be found in the chapter Appendix 1.A on page 78. These should be read and compared
with your own definitions, but it is not necessary to know these formal definitions well to
continue on in the text; an intuitive understanding is enough.
We showed that J.f / D C.0; 1/ for the map f .z/ D z 2 by noting that for z0 2 C.0; 1/
there is a point arbitrarily close to z0 whose orbit tends to 1 and is thus drastically different
from the orbit of z0 which must remain on C.0; 1/. It is also true that f is chaotic on
C.0; 1/ if we restrict ourselves to using only seed values from C.0; 1/. To see this, consider
z0 D e i written in polar form. For any nearby point z00 D e i ˛ we must, because f doubles
angles, have jf .z0 / f .z00 /j > jz0 z00 j. Indeed, denoting the arclength distance between
z0 and z00 along the unit circle by ˇ D j ˛j, we see that the distance between f .z0 / and
f .z00 / along the unit circle is 2ˇ. By induction, we see that the distance between f n .z0 /
and f n .z00 / along the unit circle is 2n ˇ as long as 2n ˇ < . So no matter how close z0 and
z00 start out, i.e., no matter how small ˇ is, corresponding orbit points will eventually be far
apart. As soon as 2n ˇ > =3 we have jf n .z0 / f n .z00 /j > 1. Thus arbitrarily close seed
values on C.0; 1/ do not have corresponding orbit values that forever stay close. Use the
Complex Function Iterator Applet, using both Polar computation mode and Polar
seed form, to see the sensitive dependence. Simultaneously iterate (one step at a time)
Seed 1 z0 D e 2:18i and Seed 2 z00 D e 2:19i .
We close this section by reflecting on some commonalities between the Newton map
dynamics in Section 1.2.4 and the dynamics of f .z/ D z 2 . As we saw in each Newton’s
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 26 — #44

i i
method example and for the f .z/ D z 2 dynamics, any two basins of attraction of attracting
fixed points share a common boundary. For f .z/ D z 2 , the basins Af .0/ and Af .1/ share
a common boundary that is the Julia set J.f / D C.0; 1/ where the attractive pull of each
attracting fixed point is balanced by the other. If you read Section 1.2, you should now go
back and consider each complex Newton’s method example to convince yourself that the
Julia set of the Newton map is exactly the common boundary of any attracting basin. This
is no coincidence and we state the precise result as follows.
Theorem 1.59. Let f .z/ be a rational map. If w is an attracting fixed point of f .z/, then
@Af .w/ D J.f /.
Remark 1.60. Theorem 1.59 together with Proposition 1.20 imply Theorem 1.34.
Remark 1.61. If the fixed point is not attracting in Theorem 1.59, then the conclusion
might not follow. We saw this in Example 1.49 where Ah .1/ D C, which has empty
boundary, but J.h/ D f1g, which we leave the reader to show.
An incomplete sketch of the proof of Theorem 1.59. Showing the full details of the proof
requires a precise definition of the Julia set and complex analysis beyond the level of this
text. We provide a sketch of some of the arguments. By Additional Exercise 1.171, Af .w/
is an open set. Thus each point in Af .w/ has a neighborhood of points whose orbits act in
the same way, showing, informally, Af .w/ F .f /. Likewise, a point in @Af .w/ contains
points arbitrarily close to it that iterate to w and points arbitrarily close to it whose orbit
points stay far away from w. Such a point must be in the Julia set, i.e., @Af .w/ J.f /.
What remains is to show that C n Af .w/ is in the Fatou set. By showing f .C n Af .w//
C n Af .w/, we may use Montel’s Theorem (Appendix 1.A Theorem 1.210) to conclude
C n Af .w/ F .f /. The details of the proof can be found in [3, p. 58].
1.3.3 Dynamics of Maps of the Form fc .z/ D z2 C c

We have investigated the dynamics of the map f .z/ D z 2 , though there is still more to
investigate since a finer analysis of the dynamics on C.0; 1/ reveals some very interesting
behavior. This is pursued in Appendix 1.A Remark 1.214 on page 79 and Additional Ex-
ercises 1.173 and 1.208. Now we investigate what happens when we change the map by
adding a constant.
A justification for studying this (although the only justification a mathematician re-
quires is that the problem be interesting) is that we want to know what happens when a
system we are studying is slightly altered, or has some error causing us to believe that we
cannot be certain that our mathematical model is exactly correct as opposed to being a
good approximation. In the dynamics of f .z/ D z 2 , we identified which seed values z0
were stable (i.e., in the Fatou set) or chaotic (i.e., in the Julia set) by studying the effects
of allowing arbitrarily small errors in the seed value. Now we investigate what type of sta-
bility may or may not be present when we allow for an error or perturbation in the map we
are iterating.
We will consider the dynamics of maps of the form fc .z/ D z 2 C c where c is a
complex parameter. Specifically, we fix c D c0 and study the dynamics of fc0 , then vary
c to see if the dynamics of the maps fc have similar behavior. Just as we saw the seed
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 27 — #45

i i
values z for a fixed map split into the stable Fatou set and the chaotic Julia set, we will also
see the set of parameters c split into stable parameters and what we will call bifurcation
parameters.
We begin by noting a common aspect of the dynamics of fc for all values of c. Each
map fc has a super attracting fixed point at 1 (verify by checking the multiplier of the fixed
point at 1). Thus, if any orbit point zn D fcn .z0 / of a seed value z0 should be large enough
(by this we mean that jzn j is large enough), then the orbit will tend to 1. A calculation
in Additional Exercise 1.191 shows that for the maps fc , we have jzk j > maxf2; jcjg for
some k 2 N if and only if zn ! 1.
We now consider a small perturbation of the map f0 .z/ D z 2 . If we let c be small (by
which we mean jcj is small), then we might expect the dynamics of fc to be similar to
those of f0 .z/ D z 2 . This turns out to be true in many respects.
Exploration 1.62. Fix c D 0:1 and use the Complex Function Iterator Applet to study
the dynamics of f0:1 .z/ D z 2 C 0:1. Try many different seed values including z0 D
0; 1 C i; 2 0:5i; : : : and record the types of behavior you are able to find. Try it out!
Similarly to f0 , it seems that f0:1 has only two types of long term behavior. Iterates of
f0:1 seem to either approach 1 or the attracting fixed point p 0:1127. Our intuition tells
us that this cannot be the whole story. If there are two attracting fixed points, then we expect
there to be tension between attracting basins, and that points on their boundaries will not be
attracted to either fixed point. The map f0 .z/ D z 2 has two attracting basins each having
the chaotic set J.f0 / D C.0; 1/ as its boundary. This also occurs with the map f0:1 , except
that the chaotic set J.f0:1 / is not a circle (though some advanced mathematics can show
that it is a simple closed curve, see [3, p. 126]). Let’s use the Global Complex Iteration
Applet for Polynomials to see a picture of J.f0:1 /. The applet will color each seed in the
basin of attraction of 1 for the map f0:1 based on how many iterates it takes for the orbit to
become strictly larger than maxf2; j0:1jg D 2, and it will color the remaining points black.
By Theorem 1.59, the basin Af0:1 .1/ has boundary set equal to J.f0:1 /. Experiment with
iterating seed values (two at a time) near J.f0:1 / to see the sensitive dependence on initial
conditions. Try it out!
The set of points colored in black in the applet has a name that we present here for any
polynomial.
Definition 1.63. For a polynomial g.z/, we define the filled-in Julia set K.g/ to be the set
of points which do not iterate to 1, i.e.,
K.g/ D fz 2 C W fgn .z/g1

nD0 is bounded in Cg D C n Ag .1/:
Remark 1.64. For a polynomial g.z/ of degree greater than or equal to two (which has a
super attracting fixed point at 1), it is true that @Ag .1/ D @K.g/, and so by Theorem 1.59
we have that J.g/; @Ag .1/; and @K.g/ are identical sets. You are asked to prove this in
Additional Exercise 1.176.
Returning to the dynamics of the map f0:1 , we ask if we could have predicted that there
would be a finite attracting fixed point. The answer is yes, and here is how. We find the two
finite fixed points of f0:1 bypsolving f0:1 .z/ D z p (why?). We then show that the two
finite fixed points p D .1 0:6/=2 and q D .1 C 0:6/=2 are attracting and repelling,
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 28 — #46

i i
0 0
respectively, since jf0:1 .p/j < 1 and jf0:1 .q/j > 1.
One way to relate the dynamics of f0 and f0:1 is to say that when c moves from
c D 0 to c D 0:1, the super attracting fixed point at 0 with multiplier 0 D 0 becomes
0
an attracting fixed point at p 0:1127 with multiplier p D f0:1 .p/, the circle J.f0 /
becomes a distorted circle J.f0:1 /, and the super attracting fixed point at 1 persists, i.e.,
remains a super attracting fixed point for f0:1 . Thus the small change in the c parameter
led to only a small change in the dynamics.
Let us explore other c values to decide which c values have dynamics similar to f0 and
which do not.
Exercise 1.65. Let c D 0:2 C 0:2i and use the Global Complex Iteration Applet for Poly-
nomials to study the dynamics of f0:2C0:2i .z/ D z 2 C 0:2 C 0:2i and to see the global
picture of the attracting basins and the Julia set. Calculate by hand the exact value of the
attracting fixed point of this map. You can test your calculation by using the applet to iterate
a nearby seed. Try it out!
Now let us see what happens if we move c far from 0.
Exercise 1.66. Let c D 1 and use the Global Complex Iteration Applet for Polynomials
to study the dynamics of f 1 .z/ D z 2 1. Are they similar to those of f0 ? Try it out!
Exercise 1.67. By generalizing the calculations for c D 0:1, mathematically describe the
set K1 of all c values such that fc has a finite attracting fixed point in C. Start by solving
an equation to find the fixed points, and then consider what conditions need to be met to
make one of them attracting. For all c values, fc has fixed points. Our goal is to determine
what c values have a fixed point that is attracting. You can see the picture of K1 , which is
called a cardioid (a heart shaped region) in Figure 1.13 on page 32. Try it out!
By examining the attracting basins and Julia sets for maps fc where c 2 K1 (using
the Global Complex Iteration Applet for Polynomials) you can see that each function has
dynamics similar to the dynamics of f0 . Thus we have seen our first example of stability in
the parameter, i.e., for any c in this set of parameters, the dynamics does not fundamentally
change when you change the parameter slightly. Another way of saying this is that K1 is
an open set of parameter values.
We know that for c 2 K1 there is an attracting fixed point pc of the map fc . Let’s call
.c/ the multiplier at pc and thus, using Theorem 1.48, which says that the multiplier of
a (super) attracting fixed point must have modulus strictly less than 1, we may regard as
a map from K1 into 4.0; 1/. If we follow the calculations in Exercise 1.67 carefully, we
see that this multiplier map W K1 ! 4.0; 1/ is one-to-one, continuous, and onto (onto
means that for every 0 2 4.0; 1/ there is a parameter c0 2 K1 such that .c0 / D 0 ). In
fact, this map can be extended to be defined and continuous on all of K1 . (We say a map
g defined on its domain set D can be extended to a larger set DQ D, if there exists an
extension map gQ on DQ such that gQ D g on D. By a slight abuse of notation we use g to
denote the extension, which means that we assume g itself is defined on the larger set D.) Q
However, for c 2 @K1 we have only .c/ 2 C.0; 1/ and the corresponding fixed point pc
is indifferent.
Let us explicitly compute the multiplier map and consider its inverse. By a direct calcu-
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 29 — #47

i i

p p
lation, we see that pc D .1 1 4c/=2 and qc D .1 C 1 4c/=2 are the fixed points
of fc . Note that pc will be attracting for an appropriate choice of c, but qc will always be
repelling
p (why?). Hence, the multiplier map is given by D .c/ D fc0 .pc / D 2pc D
1 1 4c, when c 2 K1 . The inverse of the multiplier map, as you can readily compute,
2
is c D c./ D 2 4 , which gives the c value for a fixed point with multiplier . Since c
values in K1 correspond to (super) attracting fixed points, the set K1 is the image of the set
of all 2 4.0; 1/ under the map c./. This produces the set K1 illustrated in Figure 1.13
which you should take a moment to verify using the ComplexTool applet.
Exercise 1.68.
(a) Find the c value such that fc has an attracting fixed point pc with multiplier D
0:7e i=4 , and then use an applet to illustrate the attractiveness of pc by iterating points
near it.
(b) Find the c value such that fc has an indifferent fixed point pc with multiplier D
e i=3 , and then use an applet to illustrate the behavior of orbits of points near pc .
Additional Exercise 1.177 will assist in gaining a better understanding of the role of the
multiplier in the dynamics near a fixed point.
1.3.4 Cycles of fc .z/ D z2 C c

As we saw, the orbit of 0 under the map f 1 .z/ D z 2 1 is 1; 0; 1; 0; 1; 0; : : :. We
summarize this by saying that f 1 has a 2-cycle f0; 1g. Also, as we saw (for example,
by iterating the seed z0 D 0:2 0:3i ), the 2-cycle seems to be attracting (actually, we see
later that it can be called super attracting).
Exercise 1.69. Modify Definition 1.16 of an attracting fixed point to produce your own
definition of an attracting 2-cycle. Try it out!
Exercise 1.70. Modify Definition 1.14 of an attracting basin of a point to produce your
own definition of an attracting basin for a 2-cycle. Try it out!
Using an applet to produce the orbit of several seed values, we seem to have only two
types of long term behavior for the map f 1 . Iterates of f 1 seem to either approach 1
or approach the 2-cycle f0; 1g. Intuition suggests that this cannot be the whole story.
Tension between attracting basins probably leads to points which are neither attracted to
1 nor to the 2-cycle f0; 1g. Use the Global Complex Iteration Applet for Polynomials to
see the basins, and the corresponding Julia set. Try it out!
Example 1.71. Let c D 0:9 0:1i and use one of the applets to study the dynamics of
the map f 0:9 0:1i .z/ D z 2 C . 0:9 0:1i / to verify that it has an attracting 2-cycle. Try
it out!
Example 1.72. Let c D 0:13 C 0:73i and use one of the applets to study the dynamics
of the map f 0:13C0:73i .z/ D z 2 C . 0:13 C 0:73i / to verify that it has what we could
call an attracting 3-cycle. Try it out!
We would like to calculate which c parameters will lead to fc .z/ D z 2 C c having an
attracting 2-cycle, 3-cycle, 4-cycle, and so on. We will be able to make some progress, but
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 30 — #48

i i
it will be limited because our methods require us to find the roots of polynomials of high
degree. We begin by defining cycles and then we see how to classify them as attracting,
repelling, or indifferent by using the derivative (or more precisely the multiplier).
1.3.5 p-Cycles and their Classification

Definition 1.73 (Cycles). A point w 2 C is called periodic with period p for the map f
if f p .w/ D w and w; f .w/; : : : ; f p 1 .w/ are distinct points. In such a case we call the
set fw; f .w/; : : : ; f p 1 .w/g a p-cycle for the map f .
Periodic points correspond to fixed points of higher iterates f p of the map f . For
example, the periodic point of period p D 2 at w D 0 for the map f 1 is a fixed point of
the second iterate f 21 . We can then use this to classify a cycle as attracting, repelling, or
indifferent based on the multiplier of the corresponding iterate.
Definition 1.74 (Multiplier for Cycles). Suppose fw0; : : : ; wp 1 g forms a p-cycle for the
map f . We define the multiplier of this cycle (also called the multiplier of each point
w0 ; : : : ; wp 1 of period p) to be the multiplier of the map f p at its fixed point w0 . Then
the p-cycle fw0; : : : ; wp 1 g of the map f is called
(a) super attracting if D 0
(b) attracting if 0 < jj < 1
(c) repelling if jj > 1
(d) indifferent if jj D 1.
Example 1.75. The 2-cycle f0; 1g for f 1 is super attracting since D .f 21 /0 .0/ D
f 0 1 .0/ f 0 1 .f 1 .0// D f 0 1 .0/ f 0 1 . 1/ D 0. Instead of using 0 we could have used
the other point in the 2-cycle to calculate D .f 21 /0 . 1/ D f 0 1 . 1/ f 0 1 .f 1 . 1// D
f 0 1 . 1/ f 0 1 .0/ D 0.
Since the multiplier of a p-cycle is defined via the derivative of f p , the p-fold
composition of f with itself, it is important to understand the use of the chain rule in the
classification of p-cycles. Let’s examine it more closely.
Chain Rule in C: If f and g are analytic functions at finite points z0 and z1 , respectively,
f g
and if z0 7! z1 7! z2 , then .g ı f /0 .z0 / D g0 .f .z0 //f 0 .z0 / D g0 .z1 /f 0 .z0 /. In other
words, to compute the derivative of the composite function, we multiply the derivatives of
each function evaluated at the appropriate point along the way.
Suppose that the set of finite points fw0 ; : : : ; wp 1 g forms a p-cycle for the analytic
map f as pictured in Figure 1.12. We have f p .w0 / D w0, and so by the chain rule we
compute
D .f p /0 .w0 / D f 0 .w0 /f 0 .w1 / : : : f 0 .wp 1/ D .f p /0 .wj /; j D 0; : : : ; p 1; (1.1)
which shows, among other things, that the multiplier in Definition 1.74 is well defined
since the derivative of f p is the same at any point in the cycle.
We now wish to understand the relationship between the classification of the cycle as
attracting, repelling, or indifferent (determined by ) and the dynamics of the map f near
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 31 — #49

i i
z1
z0
f f
w1 z2
w 0 = w5 w2
z5
f f5 f
z4 z3
w4
f w3
F IGURE 1.12. Illustration of a 5-cycle fw0 ; w1 ; : : : ; w4 g with the partial orbit of a point z0 chosen
near w0 .
the cycle. By the continuity of the map f for any seed z0 sufficiently close to w0, the orbit
points z1 ; : : : ; zp 1 ; zp will be close to the points w1 ; : : : ; wp 1 ; w0. Supposing that jj <
1, the map f p has an attracting fixed point at w0 . Thus, if we choose a seed z0 sufficiently
close to w0 , we have that f p .z0 / is closer to w0 than z0 is (i.e., jf p .z0 / w0 j < jz0 w0 j)
by Definition 1.16 applied to the fixed point w0 of f p . This argument works for any of
w1 ; : : : ; wp 1, and so one way to describe such an attracting cycle is to say that when you
apply the map f a total of p times, points near a wk will move around the cycle only to
return closer to wk . See Figure 1.12 where the orbit of z0 exhibits this behavior. In a similar
way we justify the classification of repelling cycles.
The calculations require only minor modifications when one of the points wk in the
cycle is 1. In the spherical metric, where 1 is no more special than any other point in C,
we see that the dynamic behavior (attraction or repulsion) of the cycle behaves in the same
way as cycles in the finite plane C.
1.3.6 Attracting Cycles for the Maps fc .z/ D z2 C c

Let us return to investigating the dynamics of the maps fc . We have seen two examples
of maps fc .z/ D z 2 C c with attracting 2-cycles. Let us now determine the set K2 of
all c values such that fc has an attracting 2-cycle. As 2-cycles generally exist, the issue
here is to determine when such a cycle will be attracting. Any point in a 2-cycle must be
a fixed point of fc2 , and not be a fixed point of fc . Thus we wish to solve the equation A,
.z 2 Cc/2 Cc D z, and exclude the solutions of equation B, z 2 Cc D z. Since each solution
to B, rewritten as z 2 C c z D 0, is a solution to A, rewritten as .z 2 C c/2 C c z D 0, we
have that z 2 C c z must divide .z 2 C c/2 C c z. After long division, we rewrite A as
.z 2 Cz C1Cc/.z 2 Cc z/ D 0. Thus a 2-cycle fu; vg must be such that both u and v solve
z 2 CzC1Cc D 0, i.e., .z u/.z v/ D z 2 CzC1Cc, which implies uv D 1Cc. According
to (1.1) the multiplier of the 2-cycle fu; vg is D fc0 .u/fc0 .v/ D 4uv D 4.1 C c/. Hence
the 2-cycle will be attracting exactly when 4j1 C cj D jj < 1, i.e., when jc . 1/j < 1=4.
Hence K2 D 4. 1; 1=4/, which is the disk pictured in Figure 1.13.
Exploration 1.76. Test c values in K2 by using the Global Complex Iteration Applet for
Polynomials to see that you get an attracting 2-cycle. Which seed values in the picture
produced by the applet seem to find the attracting 2-cycle? Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 32 — #50

i i
K2 K1
–1.25 – 0.75 0.25
–1
F IGURE 1.13. Parameter space of c values showing the cardioid K1 with cusp at c D 1=4 and the
disk K2 D 4. 1; 1=4/. The boundaries of K1 and K2 meet at c D 3=4.
In Additional Exercise 1.178 you are asked to investigate the relationship between the
multiplier and the convergence towards the 2-cycle. In Additional Exercise 1.179 you are
asked to investigate a multiplier map defined on K2 and its inverse.
Let us now try to determine the set K3 of c parameters that lead to attracting 3-cycles
for fc .z/ D z 2 C c . A point in a 3-cycle must be a fixed point of fc3 , and not be a fixed
point of fc , and so must satisfy (when substituted for z) the eighth degree polynomial
equation Œ.z 2 C c/2 C c2 C c z D 0, but not be a root of z 2 C c z D 0. After long
division we are left with a degree six polynomial to solve if we are to find the points of
the 3-cycle. Because we cannot, in general, solve a polynomial of degree five or greater,
we will have considerably more difficulty determining the attracting 3-cycles than we did
the 2-cycles. However, we were able to locate the c values that correspond to attracting
2-cycles without having to solve for the points of the cycle, although we could have by
applying the quadratic formula. We wonder if it is possible to use similar techniques, or
devise new ones, to describe as much as we can of this set K3 .
Large Project 1.77. Is it possible to determine the set K3 precisely, as was done for K1
and K2 ? Though we might not be able to solve the degree six polynomial equation, perhaps
we could use other root solving techniques, such as Newton’s method, to approximate roots
closely enough to be useful. The roots could be tested with the Global Complex Iteration
Applet for Polynomials to see their role in the dynamics of fc . Related questions of interest
are: Can you show that K3 is an open set? It turns out that K3 has more than one connected
component (though K1 and K2 are connected). Can you determine how many connected
components K3 has and what relationship they have to each other? Can you find the c
parameters which lead to super attracting 3-cycles? The roots of the degree six polynomial
must include three points which make up the attracting 3-cycle, but what do the other
three points represent? What about finding the c parameters that yield attracting 4, 5, 6, . . .
cycles?
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 33 — #51

i i
We present a definition and theorem that may aid in understanding the project because
it shows a relationship between multiplier maps and components of K3 . (To recall the
definition of a component, see Appendix B.3 on page 359.)
Definition 1.78 (Hyperbolic Components). For each n 2 N, we define the set Kn to
be the set of parameters c such that fc has an attracting n-cycle. We call any connected
component W of some Kn a hyperbolic component of Kn .
Although K1 and K2 are connected sets, some Kn are disconnected. However, each
connected component of Kn has the following nice property corresponding to its multiplier
map. The proof, which is beyond the scope of this text, may be found in [3, p. 134].
Theorem 1.79 (Multiplier Map Theorem). Let W be a hyperbolic component of some Kn .
Let W W ! 4.0; 1/ be the multiplier map that takes c 2 W to the multiplier .c/ of
the associated attracting n-cycle. Then the map is one-to-one, analytic, and onto, i.e.,
it maps W conformally onto 4.0; 1/. Furthermore, extends to a one-to-one continuous
map of W onto 4.0; 1/.
Definition 1.80. For a hyperbolic component W of Kn , we call the unique c 2 W the
center of W if .c/ D 0, i.e., it is the unique c in W for which fc has a super attracting
n-cycle.
Example 1.81. We have already observed that the center of K1 is c D 0 and the center of
K2 is c D 1.
We can prove the theorem for K1 and K2 by studying the explicit form of the mul-
tiplier map, which you are asked to do in Additional Exercise 1.179. For the other sets
Kn , however, it is not so easy. The proof of Theorem 1.79 uses properties of the multiplier
map without explicitly constructing it. One important application is that it proves that each
hyperbolic component W of Kn is open since it is the conformal image of an open set,
namely W D 1 .4.0; 1//. Hence, each Kn is also open. Recall, that a map is called
conformal when it is both one-to-one and analytic.
As we have seen, if we move the parameter c within the cardioid K1 , the dynamics
of fc do not change much. Similarly, if we move the parameter c within the disk K2 , the
dynamics of fc do not change much. Thus we call K1 and K2 stable regions of parameter
space. It is also true that each Kn is a stable region of parameter space, which, given its
definition, is another way of saying that Kn is an open set. This follows from Theorem 1.79,
but also seems reasonable without appealing to the theorem. If you slightly alter a function
with an attracting cycle (keeping in mind the strict inequality condition on the multiplier),
then it seems reasonable the new function will have an attracting cycle (i.e., will have a
strict inequality condition on the multiplier) in roughly the same place. The details can be
shown by working through Additional Exercise 1.180. This is exactly what we see when
we move the c parameter by small amounts within the cardioid K1 or within the disk K2 .
We cannot, however, expect this to happen with an indifferent fixed point or cycle since
altering the multiplier of an indifferent fixed point could make the modulus be strictly less
than one (attracting) or strictly greater than one (repelling), instead of remaining exactly
equal to one.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 34 — #52

i i
As opposed to the stable parameters found in the sets Kn , we call a parameter c unsta-
ble if there are parameters arbitrarily close to it for which the maps fc have fundamentally
different dynamics. For example, c D 3=4 is an unstable parameter.
Exploration 1.82. Give several reasons why c D 3=4 is an unstable parameter. Use
the Global Complex Iteration Applet for Polynomials to explore the dynamics when c is
close to c D 3=4, paying special attention to the dynamics for c1 D 0:75 C :05; c2 D
0:75 :05, and c3 D 0:75 C :05i . Try it out!
Now that you have provided your own reasons, we illustrate the unstable nature of the
parameter c D 3=4 by describing three particular ways in which the dynamics changes
there. We note how the type of attracting cycle, the Julia set, and the orbit of the origin all
undergo fundamental changes. We call c D 3=4 a bifurcation point, since it is the param-
eter on the boundary of two regions in parameter space where the dynamics undergoes a
fundamental change.
1. Attracting Cycle: We don’t have to know the dynamics exactly at the point c D 3=4 to
show it is unstable. It is enough to know that there are parameters arbitrarily close to and
less than c D 3=4 (in K2 ) that give rise to an attracting 2-cycle (but no attracting fixed
point) and there are parameters arbitrarily close to and greater than c D 3=4 (in K1 ) that
give rise to an attracting fixed point (but no attracting 2-cycle). Verify that as c decreases
to and then past 3=4 the attracting fixed point becomes indifferent and then splits into an
attracting 2-cycle.
Remark 1.83. A different way to look at this is to instead say that as c decreases to 3=4
the attracting fixed point merges with a repelling 2-cycle to form an indifferent fixed point
exactly at c D 3=4. Then as c decreases further, the 2-cycle re-emerges as an attracting 2-
cycle and the fixed point becomes repelling. Thus, the fixed point and the 2-cycle exchange
“polarity” in this transition. Additional Exercise 1.181 asks you to provide the details.
2. Julia Set: We can also see the result of this fundamental change in dynamics by looking
at the Julia sets J.fc / as we vary c. If we start at c D 0 and then decrease c, we see the
Julia sets J.fc / change from a circle to a distorted circle. Decreasing c further towards c D
3=4, we see that the distorted circle J.fc / begins to have infinitely many bulbs partially
forming as the distorted circle J.fc / starts pinching in. Exactly at c D 3=4 the pinching
in for each of the infinitely many distinct bulbs simultaneously becomes complete. Thus we
have gone from having one bounded component of F .fc / for c 2 . 3=4; 1=4/ consisting
of one attracting basin of an attracting fixed point to the situation for c 2 . 5=4; 3=4/
where there exist infinitely many bounded components of F .fc / consisting of the attracting
basin of an attracting 2-cycle.
3. Orbit of the origin: We notice another change in the dynamics as the parameter c moves
from K1 to K2 . The orbit of the origin changes from being attracted to an attracting fixed
point to being attracted to an attracting 2-cycle. In each of the examples we have considered
where the map fc .z/ D z 2 Cc had an attracting cycle, the origin was absorbed into it in the
sense that the tail of the orbit (formally ffcn .0/ W n > N g for large N ) is nearly identical
to the attracting cycle. Another way to say this is that the origin was attracted to the cycle.
This is true in general (as we shall soon note in Remark 1.94), and the key fact is that z D 0
is a critical point, i.e., fc0 .0/ D 0.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 35 — #53

i i
1.4. Critical Points and Critical Orbits 35
Exercise 1.84. Verify that c D 1=4, the cusp of the cardioid K1 , is an unstable parameter.
Use the Global Complex Iteration Applet for Polynomials to investigate this with respect
to items 1–3 above. Try it out!
Exploration 1.85. What do you think happens to the attracting fixed point and repelling
2-cycle when the parameter c decreases to 3=4, but then makes a right turn and starts
heading towards 0:75 C 0:05i ? Investigate this with respect to item 1 (and Remark 1.83)
using the Global Complex Iteration Applet for Polynomials.
1.4 Critical Points and Critical Orbits

In this section, we address some natural questions you may have been asking as we looked
at the dynamics of fc .z/ D z 2 C c.
1. Does every such map have an attracting cycle (other than 1)?
2. How many attracting cycles can fc have?
3. In the examples we looked at, the orbit of the origin is always attracted to the attract-
ing cycle. Is this true in general?
In this section we see that the key to answering many of these questions involves the notion
of critical points, and investigating how the orbit of such points largely determines many
key dynamical features. We begin with a definition.
When the power series of an analytic map f .z/ at z0 2 C has the form f .z/ D
f .z0 / C ak .z z0 /k C akC1 .z z0 /kC1 C : : : where ak ¤ 0, we say that z0 maps to
f .z0 / with degree vf .z0 / D k (we also call vf .z0 / the multiplicity or valency).
Remark 1.86. The condition that z0 and f .z0 / are in C will always be met in the exam-
ples we look at, and so we leave it to the interested reader to make modifications to this
definition when z0 or f .z0 / is infinity. However, when the term critical point is applied to
a point z0 in the theorems below, we include the possibilities that z0 or f .z0 / is infinity.
Definition 1.87. We call z0 a critical point of f if vf .z0 / > 1.
If z0 and f .z0 / are in C, then z0 is a critical point exactly when f 0 .z0 / D 0 (as in
Calculus I). Also, as described in Appendix Section A.6.1 on page 353, since at z0 the map
f is locally a vf .z0 /-to-one mapping, we see that z0 is a critical point exactly when f is
not locally one-to-one.
Definition 1.88. Let f be a rational or entire map. If w is a (super) attracting fixed point of
f , then we define the immediate basin of attraction Af .w/ to be the connected component
of Af .w/ that contains w. We point out (but leave it to the interested reader to show) that
Af .w/ is the component of the Fatou set F .f / that contains w.
Theorem 1.89. Let w be a (super) attracting fixed point of a non-Möbius rational map f .
Then there exists a critical point z0 2 Af .w/ and hence f n .z0 / ! w.
The proof of Theorem 1.89 is beyond the scope of this text, but it appears as Theorem
7.5.1 in [1]. We note, however, that we have seen this many times now.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 36 — #54

i i
Example 1.90. For c in K1 (see Figure 1.13), we know that fc .z/ D z 2 C c has a finite
(super) attracting fixed point pc . Since the origin is the only critical point (other than 1,
which is fixed) we must then have by Theorem 1.89 that 0 2 Afc .pc / and fcn .0/ ! pc .
This is exactly what we observed (without proof) using the applets in many examples.
To present the corresponding result for attracting cycles we will need the following
definition.
Definition 1.91. Let f be a rational or entire map which is non-Möbius. For a (super)
attracting p-cycle w0 ; : : : ; wp 1 of f we define the immediate basin of this cycle to be the
p 1
union of components [j D0 Fj where Fj is the component of the Fatou set F .f / containing
wj .
Exercise 1.92. The map f 1 .z/ D z 2 1 has a super attracting 2-cycle f0; 1g and so
its immediate basin consists of two components of F .f 1 /, one of which contains 0 and
the other which contains 1. Use the Global Complex Iteration Applet for Polynomials
to investigate the picture shown in Figure 1.14. What are the dynamic properties of the
other black bulbs, i.e., components of F .f 1 /? Try to find a pattern to the bulbs so you can
experimentally approximate points z0 such that f 31 .z0 / D 0. How many such points are
there? Do the same, but changing 3 to 5. Can you generalize this? Try it out!
F–1 F0
F IGURE 1.14. The immediate attracting basin of the 2-cycle f 1; 0g for f 1 .z/ D z2 1 consists
of the Fatou components F 1 and F0 containing the cycle points.
Theorem 1.93. Let f be a rational map which is non-Möbius. Then the immediate basin
of each (super) attracting cycle contains a critical point of the map f .
The proof of this result uses the fixed point version given in Theorem 1.89, the chain
rule, and other results that are more than we want to take on at this point. The details can
be found in [1] or [3]. Theorem 1.93 leads to the following remark.
Remark 1.94. Since all maps of the form fc .z/ D z 2 C c have only one finite critical
point (namely z D 0), Theorem 1.93 implies that each such map can have at most one
finite (super) attracting cycle and any such cycle must absorb the orbit of the critical point
at the origin. Thus we have answered Question 2 and answered Question 3 if fc has an
attracting cycle.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 37 — #55

i i
Definition 1.95. Both because 0 is a critical point of the map fc and because its orbit is
important, we call ffcn .0/g1
nD1 the critical orbit of the map fc .
We have seen that the critical orbit ffcn .0/g1nD1 plays a special role in understanding
the dynamics of the maps fc .z/ D z 2 C c . For some c the critical orbit is attracted to
an attracting cycle (e.g., c D 0; 1; 0:278 C 0:534i ) and for other values it is attracted to
the super attracting fixed point at 1 (e.g., c D 0:3; 4 C i; 2 0:3i ). You might wonder
then, is it true that for every c value the critical orbit becomes attracted to some (super)
attracting cycle? We can use intuition to say that this is probably not true. If we consider
the parameter plane, also called the c-plane, of all c parameters for the maps fc , we can
think of a tension created by the c values whose critical orbits are attracted to 1 and the
c values whose critical orbits are not. There may be c values where the pull of the critical
orbit towards 1 and the pull of the critical orbit to stay bounded is balanced. Informal
reasoning can be the basis for a good guess, but it is far from a proof. We take the easier
route and settle the question by looking at the following example.
Example 1.96. Show formally with paper and pencil (or informally with an applet) that
for c D i; 1=4; 5=4; 2, and 3=4, the critical orbit under fc .z/ D z 2 C c is neither
attracted to 1 nor attracted to a finite attracting cycle. Thus we have answered Question
1. Try it out!
For c D i , you noticed that the critical orbit i 7! i 1 7! i 7! i 1 7! i 7! : : :
became cyclic, but the cycle did not include 0. Orbits like this have an important role in
dynamics and so we give them a special name, as well as the c values which lead to them.
Definition 1.97. We call z0 a pre-periodic point (or eventually periodic, but not periodic)
under the map f if it is not periodic, but some point on the orbit of z0 is periodic.
Definition 1.98. We call a parameter c a Misiurewicz point if 0 is pre-periodic under fc .
Thus we see that c D i and c D 2 are Misiurewicz points, but c D 0 is not since
f0 .z/ D z 2 has a critical orbit 0 7! 0 7! : : : that is periodic, and so not pre-periodic.
However, f0 .z/ D z 2 does have non-critical pre-periodic points, and in Additional Exer-
cise 1.183 you are asked to find them.
If c is a Misiurewicz point, then J.fc / D K.fc / (see [3, p. 133]), in which case we
call J.fc / a dendrite. Figure 1.15 shows an example.
i–1 i
–i
F IGURE 1.15. The dendrite J.fi / along with the critical orbit i; i 1; i; i 1; i; : : :
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 38 — #56

i i
Remark 1.99. To correctly understand pre-periodic points and Misiurewicz points we

note the important distinction between orbits that are pre-periodic (which we informally
described above as those which became cyclic) and those orbits that are attracted to a
cycle. The difference is the same as the difference between a sequence approaching a value
and a sequence eventually being a value. For example, the sequence 1; 1=2; 1=3; 1=4; : : :
approaches 0, but never becomes 0. However, the sequence 2; 1; 1=2; 0; 0; 0; : : : eventually
becomes and stays at 0. When using technology such as our applets it can be difficult, if
not impossible, to distinguish between the two concepts. For example, set c D 0:9 and
look at the numerical values of the first 100 points of the orbit of z0 D 0 under fc . Looking
at the 75th and higher orbit values, we see that the points bounce back and forth between
what appears to be the same two values. This might lead you to incorrectly conclude that
z0 D 0 is pre-periodic. The problem is that the true orbit in this case never exactly bounces
back and forth between the exact same values. In fact, it is true (which we leave to the
reader to show) that z1 < z3 < z5 < : : : and z0 > z2 > z4 > : : : with strict inequalities.
However, the odd terms approach 0:8872983346207418 : : : and the even terms approach
0:1127016653792581 : : : . Since the applet truncates the data for each zn , it appears that
the odd sequence and even sequence eventually become constant. When trying to see if a
point is pre-periodic, technology may mislead you. Careful analysis and proof need to be
used.
Let us return to the question, does the critical orbit for every map fc get attracted to an
attracting cycle? We saw that for c D i; 1=4; 5=4; 2, and 3=4 this is not so. However,
in each case the critical orbit either became cyclic or was attracted to a cycle, though not
necessarily an attracting cycle. As mathematicians we naturally wonder, is this always so?
Is it possible for a c value to be such that the critical orbit neither becomes cyclic nor is
attracted to any cycle? The answer is yes. We do not have the space to explain the deep
mathematics to justify this. Nevertheless, asking good questions like this, whether or not
we can answer them, is an important part of contributing to mathematics.
So it is not true that every c parameter has a critical orbit that either becomes cyclic
or is attracted to a cycle. What is true, however, is that every c parameter has a critical
orbit that is either attracted to 1 or remains bounded. It is this dichotomy that leads us to
consider one of the most beautiful objects in all of mathematics, the Mandelbrot set.
Definition 1.100. The Mandelbrot set is M D fc 2 C W fcn .0/ ¹ 1g.
We have already encountered some important aspects of M , namely, it contains both
K1 and K2 in Figure 1.13. We also know that it contains Kn for every n because for fc
to have an attracting n-cycle, the critical orbit must be attracted to it by Theorem 1.93,
and thus not be attracted to 1. The calculation of the sets Kn is an arduous task and a
complete description of all such sets has for years stumped mathematicians. Let us there-
fore use the computer to draw M for us and experimentally investigate the sets Kn and M
(keeping in mind that limitations discussed in Remarks 1.55 and 1.99 force us to moderate
the confidence that we can place in such pictures). We begin by using the following applet
to construct a picture of M .
The Mandelbrot Set Builder Applet will color each selected point c in the parameter
plane red if c … M or black if c 2 M . Thus, if the the critical orbit under fc approaches 1,
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 39 — #57

i i
then c is colored red, otherwise it is colored black. We cannot compute the infinite number
of points in the critical orbit, so the applet computes only the number of iterates allowed
in the Maximum Iterations input box. Setting this to 100 will produce nice results
(we encourage you, however, to vary the value to see the effect it has on the picture). The
applet will color a selected c value red if and only if one of the calculated critical orbit
points lands outside of 4.0; 2/. This is justified by the following lemma, which you are
asked to prove in Additional Exercise 1.191.
Lemma 1.101. If the critical orbit fcn .0/ ever escapes 4.0; 2/, then it converges to 1.
Exploration 1.102. Experiment with the Mandelbrot Set Builder Applet to get a feel for
the mathematics that defines the Mandelbrot set. Try it out!
A feature of the Mandelbrot set which stands out is that it is symmetric about the x-
axis. Another feature you may have observed is that M has no holes. You are asked to
prove these results in Additional Exercises 1.184 and 1.185. The Mandelbrot Set Builder
Applet is a tool for visualizing M and can lead us to pursue some of its interesting features,
but there are some important properties of M it cannot help with, such as whether M is a
closed set or not. To do this we need more formal mathematics.
Lemma 1.103. The Mandelbrot set M is closed.
Proof. Suppose ck ! c where ck 2 M . We will show that c 2 M , thus proving that M

is closed. By observing that for fc the terms of the critical orbit are 0; c; c 2 Cc; .c 2 Cc/2 C
c; : : : , we see that the nth term can be written Qn .c/ for a polynomial Qn . Fix n 2 N and
note that Qn is continuous. Since jQn .ck /j 2 for each k 2 N by Lemma 1.101, we have
jQn .c /j 2 since jQn .ck /j ! jQn .c /j by the continuity of Qn . Since this holds for
every n, we have shown that the critical orbit of fc is contained within 4.0; 2/, and thus
c 2 M .
We close this section by presenting without proof two interesting facts about M . First,
the set of Misiurewicz points, each of which is in M (why?), is dense in the boundary
@M . This means that given an open set U that contains a point in @M , the set U must also
contain a Misiurewicz point (see [3, p. 133]). Second, @M is contained in the closure of
the centers of the hyperbolic components defined in Definition 1.80 (see [2, p. 100]). Thus,
an open set that contains a point in @M also contains the center of a (small) hyperbolic
component. Because a center of a hyperbolic component is in the interior of M and is
separated away from any other center, we see that @M must be quite complicated. We
encourage you to take a moment to reflect and fully digest the previous statement. Start
by explaining how the set K1 [ K2 in Figure 1.13 fails to have this property, but that by
attaching many tiny bulbs we can generate a set with this property.
Remark 1.104. Although a lot is known about M , one important question that has stumped
mathematicians thus far, is whether or not M contains an open set which does not meet any
Kn , that is, an open set of c values for which no fc has an attracting cycle other than 1.
The conjecture that asserts that this cannot happen is the density of hyperbolicity conjecture
and remains the focus of research.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 40 — #58

i i
1.5 Exploring the Mandelbrot Set

The Mandelbrot set has been called one of the most beautiful objects in mathematics. Our
red and black picture created by the Mandelbrot Set Builder Applet does not do it justice.
The set M has many tiny intricate hairs and bulbs that are hard to see with this bichromatic
picture, and the complexity of the picture cries out for a way to zoom in on them.
Initially it was conjectured that the Mandelbrot set is disconnected because low res-
olution pictures did not show the fine details and thin filaments that connect all the parts
of M . However, M is connected (see [1, p. 239]). We cannot prove this here, but a more
sophisticated applet will give us pictures that hint at this.
From now on we use the Parameter Plane and Julia Set Applet, which colors points in
M black, but colors the parameters c … M a different shade based on how many iterations
it takes for the critical orbit to escape 4.0; 2/. This gives a better feel for the immense
detail of the Mandelbrot set M (see Figure 1.16). The applet also shows, for each selected
c, a picture of the corresponding Julia set.
F IGURE 1.16. The Mandelbrot set.
Exploration 1.105. Experiment and play with the Parameter Plane and Julia Set Applet,
zooming in on parts of M and the Julia sets. Look at not just the geometry of the pictures,
but also at the dynamics. What do the bulbs and hairs mean dynamically? Is there a rela-
tionship between the geometry and the dynamics? Look for patterns and discover new fine
details in the hairs and bulbs. Describe what you find and make a list of observations, ques-
tions, and conjectures. You have the tools to explore an infinitely complex world. There are
millions of features of M and the related Julia sets to be found by using this applet. In fact,
after exploring for a sufficient amount of time, zooming in on the fine details of the sets,
you will likely find a picture that no other human has ever seen before! Try it out!
We have quite a menagerie of pictures to see and investigate, including Rabbits, Drag-
ons, and Elephants. We even have Star Clusters, Galaxies, and baby Mandelbrot sets (see
Figure 1.17).
In the next section we investigate star cluster sets more carefully so that we can better
understand what we are seeing (or not seeing) in such pictures.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 41 — #59

i i
1.5. Exploring the Mandelbrot Set 41
F IGURE 1.17. From top left to bottom right we see a Rabbit (Julia set when c D 0:12 C 0:75i ),
a Dragon (Julia set when c D 0:36 C 0:1i ), Elephants (zoom in on the Mandelbrot set centered at
c D 0:77 C :173i ), a Star Cluster (Julia set when c D 0:4387 C 0:784i ), a Galaxy (zoom in
on the Mandelbrot set centered at c D 0:75623053 C 0:06418323i), and a baby Mandelbrot set
(zoom in on the Mandelbrot set centered at c D 1:625 and adjust the Parameter plane max
iterations to 150 for a better resolution).
1.5.1 Cantor Dust Sets

We should always be careful when using technology to represent mathematics. What we
see is not always an accurate representation of what we are trying to see. For example,
using c D 0:21 C 0:64i in the Parameter Plane and Julia Set Applet, it might appear
that J.fc / is empty and that all points iterate to 1 under fc . If you display the picture
in black and white via the Dynamic plane black/white plot checkbox on the
Settings tab, you will see an all white screen, which taken at face value would mean
J.fc / is empty. That, however, is far from the truth and in Additional Exercise 1.186 you
are asked to show that there are infinitely many points that do not iterate to 1. However
J.fc /, especially for having so many points, is small in the sense that it does not show up
on the computer screen very well. To see the points better we view the color picture and
then zoom in far on the non-red parts (e.g., center each zoom in the middle of the largest
star cluster) to eventually see regions of black, which represent points which do not iterate
to 1. However, even the regions of black are not what they seem to be. Because we use
only a finite number of Dynamic plane max iterations some of the black points
would ultimately iterate to 1 if we increased the number of iterations. However, others
will not iterate to 1 and are correctly colored black.
Exercise 1.106. Adjust the Dynamic plane max iterations value to make J.fc /,
using c D 0:21 C 0:64i , harder to see and more accurate, or easier to see and less accu-
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 42 — #60

i i
rate. Also, use the Dynamic plane black/white plot feature to see how it can
be used to give you a better picture of J.fc /, keeping in mind that what we see on the
computer screen is only an approximation. Try it out!
The Star Cluster set J.f0:21C0:64i / and other hard-to-see though infinite sets of points
are called Cantor dust sets. It is hard to see them because each point in the set is discon-
nected from any other point. That is, for two points z0 and w0 in the set, there is a simple
closed curve that never meets the Cantor dust set, but winds around z0 without winding
around w0. Such a set is called totally disconnected because the only connected subsets
are single points. In Section 1.5.2 we explore one method to prove that certain Julia sets
have this property. We first use our applet to illustrate many examples of this phenomenon.
Exploration 1.107. Experiment with the Parameter Plane and Julia Set Applet to find
examples of Cantor dust Julia sets J.fc /. Can you make a conjecture about which c values
correspond to them? Try it out!
1.5.2 Connectedness Locus

One of the properties of M that you may have observed in your experimentation is that
for each c 2 M , the Julia set J.fc / is connected, and for each c … M , the Julia set
J.fc / is disconnected. (Your informal understanding of connectedness will suffice in this
text; however, the following is a formal definition. A compact set E in C is disconnected
if there exists a simple closed curve in C n E which winds around some points of E,
but not all points of E.) Move the c value around the parameter plane in the Parameter
Plane and Julia Set Applet to observe this. This is not a coincidence, and it is because of
this that we call M the connectedness locus for the family of maps ffc W c 2 Cg, i.e.,
M D fc 2 C W J.fc / is connectedg. This is proved by using the following theorem and
noting that fc has its sole finite critical point at the origin.
Theorem 1.108. Let f be a polynomial of degree greater than or equal to two. Then every
finite critical point of f has a bounded orbit if and only if J.f / and K.f / are connected.
The proof is beyond the scope of this text (see [1, p. 202] for details), but for the maps
fc we will be able to illustrate the reasoning behind the following related dichotomy.
Theorem 1.109. For the map fc , either
(a) ffcn .0/g1
nD1 is bounded in C and J.fc / is connected, or
(b) fcn .0/ ! 1 and J.fc / is totally disconnected.
We illustrate the method of proof, though some details requiring advanced methods are
left to be found in [3, p. 67]. We first make an important definition.
Definition 1.110. A compact topological disk (ctd) is a compact set (i.e., a closed and
bounded set) in C whose boundary is a simple closed smooth path.
Its importance comes from the following lemma and from noting that any ctd is con-
nected.
Lemma 1.111. If E is a ctd and c … @E, then fc 1 .E/ consists of
(a) one ctd containing 0, if c 2 Int.E/,
(b) two ctd’s, neither containing 0, if c … E.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 43 — #61

i i
We do not prove this, but with the help of our applet we will illustrate it.
Sketch of proof of Theorem 1.109. Choose r > maxf2; jcjg such that jfcn .0/j ¤ r for
all n D 1; 2; : : : . Define E0 D .0; r / and En D fz 2 C W fcn .z/ 2 E0 g for each
n D 1; 2; : : : . Thus En D fc 1 .En 1 / for each n D 1; 2; : : : and c … @En for each
n D 0; 1; 2; : : : . From Additional Exercise 1.191, we know that jzj > r implies jfc .z/j >
jzj > r and fcn .z/ ! 1. With this information it can be shown that E0 E1 E2 : : :
and K.fc / D \1 nD0 En . We now investigate cases (i) and (ii) of the theorem by analyzing
the sets En .
Suppose ffcn .0/g1 nD1 is bounded in C, i.e., 0 2 K.fc /. Since c D fc .0/, we have
c 2 K.fc /, and so c 2 En for all n D 0; 1; 2; : : : . Since c … @En, we may then also
conclude 0 2 Int.En / for all n D 0; 1; 2; : : : . Applying Lemma 1.111 inductively we may
then show that each En is connected as it consists of a single ctd (see Figure 1.18 (left)).
Though we omit the details, it follows from a more advanced topological argument that
K.fc / D \1 nD0 En is connected, which by another advanced topological argument implies
J.fc / D @K.fc / is connected.
Suppose fcn .0/ ! 1, i.e., 0 … K.fc /. By the choice of r we know c 2 Int.E0 /.
However, since 0 does not lie in every En (else 0 would lie in K.fc /), there is a positive
integer m such that c 2 Int.Em 1 / but c … Em . Thus by Lemma 1.111 we see that Em
consists of one ctd, but EmC1 D fc 1 .Em / consists of two ctd’s (see Figure 1.18 (right)
where m D 2). Applying Lemma 1.111 to each ctd in EmC1 , we see that EmC2 consists
of four ctd’s. Proceeding inductively we see that EmCk consists of 2k ctd’s. Although we
do not have the tools here to prove it, the size of the ctd’s that make up EmCk shrinks to
zero as k ! 1, and so we conclude that any two distinct points z and w in K.fc / must
lie in distinct ctd’s of EmCk when k is sufficiently large. Hence the boundary of such a
ctd containing z, which lies in Afc .1/ (why?), is a curve which encloses z but not w.
c
E0
E2 E2 E3
E1
E1 0 0 c
E0
F IGURE 1.18. The sets E0 ; : : : ; E5 drawn using the Parameter Plane and Julia Set Applet with
Dynamic plane escape radius r D 2, Dynamic plane max iterations = 5, and
Color sample rate = 7. In both pictures E0 D .0; r/. The left picture shows c D i with each
En connected. The right picture shows c D 0:9 with each E0; E1 , and E2 connected, but subsequent
En disconnected.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 44 — #62

i i
We conclude that K.fc / is totally disconnected, and thus so is J.fc / since by utilizing
Remark 1.64 we can show that J.fc / D @K.fc / D K.fc /.
So, omitting a few details, we were able to understand how Cantor dust sets naturally
arise when considering J.fc /. Another topological property which you may have observed
in your explorations is that K.f / has no holes. This we have all the tools to prove, where
we define a hole as a bounded domain U in C n K.f / such that @U K.f /. (The
advanced reader will recognize that the absence of holes in K.f / is equivalent to the set
C n K.f / being connected.) We prove this by appealing directly to the definition of K.f /
and applying the maximum modulus theorem.
Lemma 1.112. Let f be a polynomial of degree greater than or equal to two. Then K.f /
has no holes.
Proof. Suppose the set U is a hole in K.f /. Let R > 0 be such that jzj > R implies
jf .z/j > jzj > R (such an R must exist because f is a polynomial of degree greater
than or equal to two). From this, we can show f n .z/ ! 1 whenever jzj > R. Thus
jf n .z/j R for all n 2 N for any z with a bounded orbit. This means, by the definition of
K.f /, that we have K.f / D fz 2 C W jf n .z/j R for all n 2 Ng. Since @U K.f /,
we see that for any n 2 N, the polynomial f n is bounded by R on @U . By a version
of the maximum modulus theorem (given as Appendix Corollary A.19 on page 352), we
conclude that f n is bounded by R on U as well. Since this works for all n 2 N, we see
that U K.f /, contrary to our assumption that U is a hole in K.f / (that does not meet
K.f /). This contradiction proves the result.
1.5.3 Self-similarity and Symmetry

One of the properties you may have noticed in dynamical plane pictures for the maps fc is
that they are symmetric about the origin. That is, each set Afc .1/; J.fc /; and K.fc / has
the property that it contains z if and only if it contains z. You are asked to prove this in
Additional Exercise 1.187.
Another property you may have noticed in each set J.fc / is that small parts of J.fc /
look like other larger parts of J.fc /. We call this property self-similarity and note that it is
a property of all Julia sets.
Exploration 1.113. Set c D 0:112C0:74667i in the Parameter Plane and Julia Set Applet
and in the dynamic plane (z-plane) window set the x range to be 0.22026 to 0.28746 and
set the y range to be 0.4648 to 0.532, and then hit the Update button. Notice how similar
the picture looks when we zoom in by setting the x range to be 0.2277184 to 0.2475200
and the y range to be 0.50166592 to 0.52146752. Continue zooming in on a point of
J.fc / to see how at all scales (i.e., depth of zoom) the picture, after rotating, looks like the
first picture. You can click on the thumbnails at the bottom to see any previously created
pictures. Repeat this with other c values to get a sense that this is a general property of
Julia sets. Try it out!
How about the Mandelbrot set M ? Is it self-similar too? The answer is a definite . . . sort
of. There are places in M that have small pieces that look like larger pieces, but it is not
the case, as in the Star Cluster sets above, that the whole set M looks like a bunch of small
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 45 — #63

i i
copies of one piece of itself. The Mandelbrot set is sometimes called quasi-self-similar for
this reason.
Exploration 1.114. Set c D 0:41491386 C 0:60134804i in the Parameter Plane and
Julia Set Applet and zoom in on the Mandelbrot set, centering the zoom on c. What you
see are pictures that appear to look the same as you zoom in. If you zoom in on the tips of
any of the antennae of the Mandelbrot set you will see a type of self-similarity. Now see
what you find when you repeatedly zoom in on a point in the middle of such an antenna.
Try it out!
Zoom in on the point c D 1:2418406 0:32366967i to find what we call a baby Man-
delbrot set. Set the Parameter plane max iterations to a higher value (such as
200 or 300) to see this picture better. You can also change the Color sample rate to
adjust the color scheme to create a nicer picture. What you have found is not an exact copy
of the full Mandelbrot set, as you can tell by the long antennae coming out of it, but it has
the unmistakeable look of M . It turns out, due to a deep result, that these baby Mandelbrot
sets are dense in the boundary of M , that is, given any neighborhood of a point in @M , no
matter how small, there exists a baby Mandelbrot set in it. To see it we may have to zoom
in pretty far, but it’s there. This phenomenon, and more, is discussed in Section 1.7.
Exploration 1.115. Try zooming in on an arbitrary point in the @M , and then see if you
can find a baby Mandelbrot set. You might need to adjust the Parameter plane max
iterations to give you a better view. Try it out!
1.5.4 Bulbs in M
Let’s now try to classify systematically some of what we see in M , in terms of geometry
and dynamics. Each point c in the parameter plane (also called the c-plane) corresponds
to a function fc whose Julia set J.fc /, Fatou set F .fc /, and dynamics are viewed in the
dynamic plane (also called the z-plane). We will see that M , and the parts that make it
up, provide us with a sort of dictionary or index of the types of dynamics we find for
fc .z/ D z 2 C c.
We have already seen that the cardioid K1 in Figure 1.16 consists of c values corre-
sponding to maps with attracting fixed points. Off it there are infinitely many bulbs attached
that we would like to understand both by their geometrical properties as sets in the c-plane
and by the corresponding dynamic properties in the z-plane.
The most prominent bulb off K1 is the disk K2 D 4. 1; 1=4/ representing the pa-
rameters corresponding to maps with attracting 2-cycles. The next largest bulb off K1 is
near the top. Select c D 0:12 C 0:77i from it and consider the critical orbit in the dy-
namic plane to see that there exists an attracting 3-cycle. Use the Parameter Plane and
Julia Set Applet to observe this by iterating the critical orbit one iterate at a time (by check-
ing the Show critical orbit box, then checking the Iterate orbits box, and
then hitting the + button that appears next to the Iterate orbits box). For this c,
the Julia set J.fc / in the dynamic plane is pinched in such a way that three bulbs meet
at every pinch point. Let’s call the pinch point that corresponds to the immediate basin of
this attracting 3-cycle the main pinch point (which in this case is near 0:282 C 0:492i ).
Then by iterating the origin one step at a time you can see that each iterate makes roughly
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 46 — #64

i i
a 1=3 counterclockwise rotation around it. You can use the Connect orbit points
checkbox to help you track the path of the orbit. Trying other c values from the bulb in the
c-plane we find that we always get this same type of behavior. For this reason we call the
bulb in M the 1=3 bulb.
Exercise 1.116. Use the Parameter Plane and Julia Set Applet to investigate the dynam-
ics for c D 0:513 C 0:5693i . Decide what fraction p=q would best describe the bulb
containing this point. Try it out!
In the dynamic plane we see an attracting 5-cycle, which cycles around a main pinch
point at which five bulbs meet. We see by iterating the origin one step at a time that the
iterates make a 2/5 rotation about the main pinch point in each step. For this reason we call
the bulb of M that contains c D 0:513 C 0:5693i the 2/5 bulb. You should experiment to
see this behavior for other c values within this 2/5 bulb. Try it out!
Definition 1.117. Let B be a hyperbolic component of Kq whose boundary meets the
boundary of the main cardioid K1 . We call B the p=q bulb, and denote it Bp=q , if for each
c 2 B the map fc has an attracting q cycle and each step in the critical orbit in the dynamic
plane makes roughly a p=q rotation about the main pinch point.
From what follows it will become clear that for a rational number p=q, 0 p=q 1,
there is only one p=q bulb and so we are justified in calling Bp=q the p=q bulb, as opposed
to a p=q bulb.
Exercise 1.118. Experiment and prove a relationship between the p=q bulb and its conju-
gate bulb, that is, the bulb reflected over the x-axis. Find p 0 and q 0 if the conjugate of the
p=q bulb is the p 0 =q 0 bulb. Try it out!
Remark 1.119. Theorem 1.79 shows that the multiplier map p=q maps Bp=q conformally
onto 4.0; 1/. The map p=q extends to be a one-to-one continuous map of Bp=q onto
4.0; 1/.
It turns out that Bp=q meets the boundary of the main cardioid @K1 in a single point,
which we denote cp=q , and call the root of the p=q bulb. For example, c1=2 D 3=4 is the
root of the 1/2 bulb B1=2 D K2 . In Additional Exercise 1.188 you are asked to investigate
the multiplier maps evaluated at the root. Note that since the root cp=q of the p=q bulb lies
on the boundary of two hyperbolic components, namely K1 and Bp=q , we see that there
are two multiplier maps (namely, W K1 ! 4.0; 1/ and p=q W Bp=q ! 4.0; 1/) that are
defined at each cp=q . The root points play a special role and so we wish to understand them
further. In particular, we wish to understand the following dynamic property.
Exercise 1.120. The root cp=q of the p=q bulb is the c parameter for which fc has an
indifferent fixed point with multiplier e 2 ip=q , i.e., cp=q D c.e 2 ip=q / where c./ is the
inverse of the multiplier map W K1 ! 4.0; 1/. Since cp=q is a bifurcation parameter,
you should describe three dynamical changes (as done in Exercise 1.84) that occur as the
parameter c moves from the main cardioid K1 into Bp=q passing through the root cp=q . Try
it out!
Exploration 1.121. Explore and label other bulbs in the same way as in Exercise 1.116.
We recommend that you print a large image of M and label each bulb as you go. Try it
out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 47 — #65

i i
While doing Exploration 1.121 you may have seen a pattern that shows how you can
quickly compute p=q for the largest bulb between two bulbs. We explain the pattern, but
leave the detailed proofs to be found in [7]. Looking at M we see B1=2 and B1=3 each
attached to the main cardioid, with infinitely many p=q bulbs in between decorating the
boundary of the main cardioid. The largest, as we have seen, is B2=5 . We can correctly “do
the math” in this situation by using Farey addition to compute
1 1 2
˚ D :
2 3 5
In Farey addition, we add fractions in the unusual way of adding the numerators and adding
the denominators. We call the two addends the Farey parents (e.g., 21 and 13 ) and the result-
ing fraction the Farey child (e.g., 52 ). Correspondingly, we call the bulbs B1=2 and B1=3 the
Farey parents of the Farey child bulb B2=5 .
Example 1.122. The Farey child of B2=5 and B1=3, that is, the largest bulb between them,
is B3=8 since
2 1 3
˚ D :
5 3 8
A quick check with the Parameter Plane and Julia Set Applet will confirm this. Remember
to use the Connect orbit points checkbox to help you track the path of the orbit.
Try it out!
Remark 1.123. We must be careful with our use of Farey addition of bulbs. The main
rule we must be sure to adhere to is: Two bulbs can only be Farey parents if all the bulbs
between them are smaller than they are. For example, the largest bulb between the 4=11
bulb (the one containing c D 0:292 C 0:633i ) and the 2=5 bulb is the 3=8 bulb. Here
Farey addition does not work. Because the 3=8 bulb is larger than the 4=11 bulb (which
you should check on the applet), we know that Farey addition is not applicable.
There is another issue to be dealt with if Farey addition is to help us compute the bulb
fraction for all p=q bulbs. Between B1=3 and the cusp of the cardioid K1 there is a largest
bulb, but how can we use Farey addition to determine it? The key is to treat the cusp like
a Farey parent which is larger than all other bulbs. Additional Exercise 1.190 asks you to
determine experimentally what Farey fraction should be used to represent the cusp.
1.5.5 Sub-bulbs of M
Just as the main cardioid K1 of M has many p=q bulbs attached to it, each p=q bulb
has many sub-bulbs attached. Let’s investigate them. Use the Parameter Plane and Julia
Set Applet to view the change in dynamics as we let c vary from c1 D 0:16097811 C
0:80545706i within the 1=3 bulb to c2 D 0:17462417 C 0:8296561i in an attached sub-
bulb. We see that the attracting 3-cycle becomes an attracting 15-cycle. The picture of the
Julia set with the attracting 15-cycle has features common to Julia sets with an attracting
3-cycle and Julia sets with an attracting 5-cycle. We see that three bulbs of K.fc1 / meet at
each pinch point. In changing c to c2 , we see that the pinch points from K.fc1 / persist, but
also, within each bulb of K.fc1 /, new pinching occurs, such that at each of these newly
formed pinch points five bulbs of K.fc2 / meet. See Figure 1.19 and use the zooming
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 48 — #66

i i
F IGURE 1.19. Left shows J.fc1 / and right shows J.fc2 /, slightly magnified.
features in the Parameter Plane and Julia Set Applet to see the pattern repeat at all scales
(remembering to adjust the Dynamic plane max iterations value as needed).
Small Project 1.124. Investigate the behavior seen above by entering into a variety of sub-
bulbs attached to a variety of p=q bulbs. Can you discern a pattern? If shown an example
of a Julia set such as the one in Figure 1.20, can you identify which sub-bulb the c value
came from?
F IGURE 1.20. How can you tell where to find the c parameter for this picture of J.fc /?
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 49 — #67

i i
1.5.6 Limbs and Antennae in M

The Mandelbrot set M is a connected set. We can formally describe this by saying that
a simple closed curve in the complement of M must either wind around no point of M
or must wind around all of M . That M is connected is not easy to prove and we shall
not attempt to do it, but we will use it to describe some other important aspects of M . In
particular, we make use of the fact that we can naturally disconnect M into two connected
pieces by removing a root cp=q of any p=q bulb (which we also will not prove here).
Definition 1.125. The set M n fcp=q g consists of two connected sets, one that contains the
cardioid K1 and the other containing Bp=q , which we call the p=q limb.
On each p=q limb there is the main p=q bulb with infinitely many sub-bulbs attached.
Each of these sub-bulbs has an infinite number of tinier sub-sub-bulbs, and so on. However,
the p=q limb contains more than just these bulbs. It also contains what we informally call
antennae made out of thin filaments that reach out from the p=q bulb. Looking straight
above the 1/3 bulb there is a junction where three equally spaced filaments, which we call
spokes, meet (see Figure 1.21). The main or principal spoke is the one that attaches to the
1/3 bulb, and the shortest spoke is a 1/3 rotation about the junction point from the main
spoke. This relationship between the short spoke and the p=q value occurs frequently, but
not always (e.g., examine B1=5, which contains the point c D 0:39 C 0:33i , using the
Parameter Plane and Julia Set Applet). This generalization has been made into a formal
theorem, but to do so, the notion of shortest had to be changed slightly (see [7]). We also
note that if you zoom in on other junctions found on the antennae on the 1/3 limb, you will
often, but not always, see three spokes meeting at the junction. This seeming coincidence
cries out for further exploration. First we make a definition.
Junction point
Principal Shortest
spoke spoke
B1/3
F IGURE 1.21. Illustration of main junction point of 1/3 limb of M .
Definition 1.126. When q spokes meet at a junction point, we will call it a junction point
of order q.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 50 — #68

i i
Exploration 1.127. Investigate other p=q limbs and their principal spokes to get an idea
of how often the spoke that is a p=q rotation from the main spoke is the shortest. Is there a
pattern for the longest spoke? Try it out!
Exploration 1.128. Investigate other p=q limbs and the junction points within them to
get an idea of how often the junctions there are of order q. What is the order of the other
junctions? Is there a pattern to the types of junction orders on a p=q limb? Try it out!
In the course of exploring the p=q limbs you may have seen many baby Mandelbrot sets
like the one in Figure 1.22. To see them clearly you will want to adjust the Parameter
plane max iterations value to 300, 400, 500, or even 1,000 or more depending
on the size of the baby Mandelbrot set that you want to see. Also, adjusting the Color
sample rate can help. We mentioned before that baby Mandelbrot sets are everywhere
(they are dense in @M ); however, even though they are all seeming copies of the original
M set, they are not all the same. They have different antennae sprouting from them. If we
pay attention to these antennae and to how many meet at the junction points, can we get
an idea of how the baby Mandelbrot set is related to the p=q limb it is in? For example,
in Figure 1.22 we see that the junction points near the tips of the antennae are all of order
three. Furthermore, we see that along these filaments, closer to the baby Mandelbrot set, we
have junction points of order five. Can this information be a clue to help you find where this
baby Mandelbrot set is? Imagine playing a game where your friend shows you a picture of
a baby Mandelbrot set. Can you win the game by locating it? These are hard questions, but
maybe by investigating them you will find some partial answers, connections, or maybe
pose some interesting questions of your own.
F IGURE 1.22. Which p=q limb contains this baby Mandelbrot set? Off of which sub-bulb of Bp=q
is it?
In your investigation of the antennae, bulbs, and limbs of M you may have also no-
ticed that some of the parameter plane pictures, specifically enlarged areas of the tips of
the antennae of M , look very much like some of the dynamic plane pictures, specifically
enlarged areas of the tips of certain Julia sets. The pictures can look so much alike that it
can be confusing which is which (see Figure 1.23). In [20] it was shown that zooming in
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 51 — #69

i i
F IGURE 1.23. Which is part of the Mandelbrot set and which is part of a Julia set?
the parameter plane near a Misiurewicz point c (see Definition 1.98) shows a portion of
the Mandelbrot set that is a rotation of an enlargement of the Julia set of fc in the dynamic
plane near c. This is curious given that the parameter plane and the dynamic plane are, on
the face of things, different animals. (Most mathematical discoveries are not accompanied
by a shout of ‘Eureka!’, but rather a quietly spoken ‘Huh, now that’s curious.’) Why do
parts of the Mandelbrot set seem identical to parts of certain Julia sets? The proof is too
complicated for these pages, but we can see examples. Use the Parameter Plane and Julia
Set Applet to zoom in very far on the point c D 0:4711819 C 0:3541484i in both the
parameter plane and the dynamic plane (in the dynamic plane, c is the the first point in the
critical orbit). Then compare the portion of M near c to the portion of J.fc / near c. Up to
a rotation they seem identical.
1.5.7 Fixed points for fc .z/ D z2 C c

In this section we look at the dynamics near the fixed points of fc .z/ D z 2 C c.
Example 1.129. Using the Parameter Plane and Julia Set Applet set c D 0:513 C
0:5693i in the parameter plane and choose seed z0 D 0:4034 C 0:2896i (near the main
pinch point) in the dynamic plane. Zoom in close to the fixed point (use the Plot fixed
points box to color the fixed points purple) and then iterate the seed value one step at
a time and observe the behavior. Iterate several other seeds near the pinch point to see the
same behavior. We see that the main pinch point is a repelling fixed point whose multiplier
has argument very close to .2=5/2 .
Exercise 1.130. Using paper and pencil (along with your calculator or computer) solve
for the fixed points of fc where c D 0:513 C 0:5693i as in Example 1.129. Compute
their multipliers and verify that the pinch point’s multiplier has argument very close to
.2=5/2 . Use the Complex Function Iterator Applet or Global Complex Iteration Applet
for Polynomials to iterate the map z 7! z for seed values near the repelling fixed point
at the origin. Compare the dynamics of z 7! z near the origin and z 7! fc .z/ near the
pinch point. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 52 — #70

i i
Exploration 1.131. Investigate other c values in the 2/5 limb to see if the repelling fixed
point has a multiplier with argument very close to .2=5/2 . You can do this experimentally
by checking the behavior of orbits that start near the repelling fixed point (use the Plot
fixed points feature in the Parameter Plane and Julia Set Applet to locate the fixed
points). Relate the argument of the multiplier to the dynamics near the repelling fixed point
and note how the argument corresponds to the spiralling we see in K.fc / near the repelling
fixed point. How would this behavior change in other p=q limbs? Try it out!
We saw the role of the fixed point of fc that also serves as a main pinch point in the
set K.fc /. What can you observe about the other fixed point? How does it differ from the
main pinch point? Does it have a special dynamic role? Can you prove anything or make a
conjecture about it?
1.5.8 Concluding Remarks about the Family fc .z/ D z2 C c

Though we have seen and discussed many aspects of the Mandelbrot set there are many
more things to know. We encourage you to delve deeper into the topics we discussed and
pursue your own line of questions. There are many fascinating things left to discover in
this infinite playground of mathematics. Go enjoy it!
We end this section with a remark about one of the goals in the introduction to this
chapter, to introduce, investigate, and understand what we can about chaotic dynamical
systems. It is easy to believe that systems involving the weather, the stock market, or the
motion of the bodies in the Milky Way galaxy are chaotic. There are many uncontrollable
variables in these systems, and changing one, even just a little, may affect the other vari-
ables. What we see with this family of maps fc .z/ D z 2 C c is also chaos though it exists
without unknown variables or random processes. This simple system of one complex vari-
able and one complex parameter produces chaotic behavior of an unimaginably wide and
rich sort. This gives us a small taste of what is truly a global phenomenon; chaos is all
around us, even in what seems like the simplest of systems. For a non-technical look at
some history of chaos, chaos theory, and the chaoticians who investigate it, we recommend
Chaos by James Gleick [15].
1.5.9 Other Uni-critical Families of Polynomials

In this section we investigate the dynamics of polynomials Pc .z/ D z d C c where d D
2; 3; 4; : : : (we have already studied d D 2). Since these maps have only one finite critical
point (at the origin), we can analyze the parameter space of these maps in the same way
as for the maps fc .z/ D z 2 C c. We define the d th degree Mandelbrot set to be Md D
fc 2 C W Pcn .0/ ¹ 1g. As we were told for the Mandelbrot set, the condition that the
critical orbit fPcn .0/g tends to 1 is equivalent to the condition that some point in this orbit
escapes 4.0; 2/. Steps to prove this, and more, are given in Additional Exercise 1.191.
You are encouraged to explore the dynamics of such maps using the Parameter Plane
and Julia Set Applet. Where z 2 C c appears, use the drop down menu to select z d C c, and
then enter an integer d > 2. Additional Exercise 1.192 asks you to prove some symmetries
exhibited in the parameter plane and the dynamic plane. Additional Exercise 1.193 asks
you to show that Md is the connectedness locus for this new family.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 53 — #71

i i
1.6. Transcendental Dynamics 53
Large Project 1.132. Investigate these families of maps using the Parameter Plane and
Julia Set Applet in the same way as as we did for fc .z/ D z 2 C c. Try to find patterns and
relationships in the bulbs and antennae. Develop your own questions, make conjectures,
and describe Md .
1.6 Transcendental Dynamics

In this section we investigate the dynamics of three different families of maps: Ec .z/ D
ce z ; Sc .z/ D c sin z; and Cc .z/ D c cos z, where c 2 C n f0g is a parameter. We first
consider their dynamic properties for fixed c, then we study the parameter plane, that is,
we investigate what changes occur in the dynamics when c varies. The functions have one
striking difference from the maps Pc .z/ D z d C c studied above. They are transcendental
entire maps, i.e., are analytic on all of C (entire), but are not polynomials. Hence, they not
only fail to have an attracting fixed point at 1, they fail to even be defined at 1. These
maps have an essential singularity at 1 which means, among other things, that they cannot
be defined at 1 in any continuous way (Appendix Example B.14 on page 362 illustrates
this). However, it turns out that 1 and its basin play a central role. Instead of the basin of
1 being in the Fatou set as we have for polynomials, we have the opposite.
Proposition 1.133. For the maps Ec .z/ D ce z ; Sc .z/ D c sin z; and Cc .z/ D c cos z,
where c 2 C n f0g, the Julia set is the closure of the attracting basin of 1, e.g., J.Ec / D
AEc .1/.
This proposition asserts a striking difference between polynomial dynamics and the
dynamics of these transcendental maps. Though the proof of this is beyond the scope of
this text (see [6, 9] as general references on the dynamics of Ec ), we use it, in particular, to
program the computer to illustrate their Julia sets. To do so, we must first understand how,
or rather in what direction, a point z0 can iterate to 1 under the maps. We first investigate
Ec .z/ D ce z .
Remark 1.134. For those who are familiar with Picard’s theorem (see [1, p. 242]), we give
some idea of why the Julia set of certain transcendental entire functions is related to the
closure of the points which iterate to 1. Picard’s Theorem says that given an entire map
g with an essential singularity at 1 and given any neighborhood U of 1 (no matter how
small), g will map U n f1g infinitely often onto the entire complex plane C except for at
most one point. This means that, with the exception of at most one point, for each w 2 C
there are infinitely many points z 2 C such that g.z/ D w (the map e z illustrates this with
an exceptional value at w D 0). Hence, there is a high degree of sensitive dependence as
some points near 1 must have very different orbits even in the first application of the map
(let alone after repeated iteration of the map).
1.6.1 Exponential Dynamics

In this section we investigate the dynamics of Ec .z/ D ce z . Because Ec .z/ D ce z D
ce x e iy where x D Re z and y D Im z, we have jEc .z/j D jcje x . Hence Ec .z/ is large,
and thus close to 1, when x D Re z is large. If Re z > 50, then jEc .z/j > jcje 50 is
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 54 — #72

i i
extremely large for any c that is not so small that a computer would recognize it as 0. We
use this to justify the algorithm implemented in the Parameter Plane and Julia Set Applet
for drawing J.Ec /, which you can access by selecting ce z from the function menu. A point
in the dynamic plane is colored based on how many iterates it takes to escape, by which we
mean have its real part become greater than 50. Thus the colored points represent AEc .1/,
which by Proposition 1.133 must look the same as the Julia set. Points colored black do not
escape, at least not after iterating the number of times set in the Dynamic plane max
iterations box, and so these points represent the Fatou set. Since, as you are asked to
prove in Additional Exercise 1.194, Ecn .z0 / ! 1 if and only if Re Ecn .z0 / ! C1, we
say that points that iterate to 1, do so in the direction of the positive real axis.
Remark 1.135. A point z may be identified by the algorithm as being in the Julia set when
it is not. However, we can show that if that occurs, there would have to be a point very close
to z that does lie in the Julia set. Hence, for a visual approximation of the Julia set, this
technical issue does not pose a serious concern. Large Project 1.195, however, asks you to
examine this matter in depth.
Example 1.136. We show that for c D 0:2 the Julia set J.E0:2 / is what we will call a
Cantor bouquet. Our proof will not use the applet, nor its algorithm, but will employ only
the notion of sensitive dependence on initial conditions. However, we find it useful to first
view the Julia set in Figure 1.24 as drawn by the Parameter Plane and Julia Set Applet.
Because points in black do not escape and other points iterate to 1 (in the direction
of the positive real axis), we see that the colored set depicts J.E0:2 /. It appears from
the picture that J.E0:2 / contains large open sets, but this is an artifact of our algorithm’s
inability to iterate infinitely many times. Were we able to set the Dynamic plane max
iterations equal to infinity we would not find any open sets in J.E0:2 /. We explain
why in a moment, but first we use the zooming feature to see that what appears to be tiny
fingers in J.E0:2 / are actually made of tinier fingers, which are themselves made of even
tinier fingers and so on. Try it out! We see self similarity again. We now explain why the
picture of J.E0:2 / shows fingers inside of fingers.
F IGURE 1.24. A portion of the Cantor bouquet Julia set of E0:2 shown for 0 x 10 and
10 y 10.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 55 — #73

i i
We begin by showing that E0:2 has a real attracting fixed point. By the intermediate
value theorem, E0:2 has a fixed point p for some real value 0 < p < 1 since E0:2.x/ x
0
is positive for x D 0 and negative for x D 1. Also, p is attracting since jE0:2 .p/j < 1.
We now show that the half plane H D fRe z < 1g is contained in the attracting basin
0 0
AE0:2 .p/. Set D jE0:2 .1/j D 0:2e and note that jE0:2 .z/j < < 1 for z 2 H . Thus,
Rz 0
for z 2 H , we have jE0:2 .z/ pj D jE0:2 .z/ E0:2 .p/j D j p E0:2 .s/ dsj jz pj.
Hence the action of E0:2 is to move points in H closer to p by a factor of at least . Using
induction, along with the fact that E0:2 .H / H (verify), it can be shown that for z 2 H
n
we have E0:2 .z/ ! p, i.e., H AE0:2 .p/.
Exercise 1.137. Use the Parameter Plane and Julia Set Applet with E0:2 .z/ D 0:2e z to
see this contraction. In the dynamic plane window with 2 Re z 2 and 2 Im z 2
(which will be all black), pick two points and iterate the map one step at a time to see that
after each step the points are closer together, as they iterate toward p. Try it out!
The technique that shows how a bound on the derivative can be used to show contrac-
tion of a map is general, and very useful. We interrupt our proof of the Cantor bouquet
result to state a lemma for future use, which you are asked to prove in Additional Exer-
cise 1.196. First, recall that a set D is called convex if for any points z; w 2 D, the line
segment connecting z and w is a subset of D.
Lemma 1.138 (Contraction Lemma). Let f be analytic and map the convex domain D
into itself. Suppose that jf 0 .z/j < < 1 for z 2 D. Then f contracts distances by a
factor of at least on D, i.e., jf .z/ f .w/j < jz wj for z; w 2 D. Furthermore,
jf n .z/ f n .w/j < n jz wj for z; w 2 D. Hence, if a 2 D is a fixed point of f , then
D Af .a/.
Returning to the proof that J.E0:2 / is a Cantor bouquet we now consider E0:21 .H / D
fz 2 C W E0:2 .z/ 2 H g. Use the transformation properties of the exponential map to
justify the picture in Figure 1.25 showing that C n E0:21 .H / consists of infinitely many
components, which we call fingers. Label them Ck for k 2 Z, each Ck being a translate by
2 i k of C0 . We also note that H E0:21 .H /.
C1
C0
C–1
F IGURE 1.25. In this window of the complex plane, the set E0:21.H / is colored in black with
components of its complement in red for 2 x 18 and 10 y 10. For reference, the unit
circle is shown in white.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 56 — #74

i i
Each Ck is mapped conformally (a conformal map is one-to-one and analytic) onto the
half plane fRe z 1g, and thus each Ck contains a preimage under E0:2 of each Cj for all
j 2 Z. The preimages, shown in red in Figure 1.26, are sub-fingers of Ck with what we
call gaps (in black) in between. The gaps, as in Figure 1.25, extend all the way to 1 in the
direction of the positive x-axis.
F IGURE 1.26. The left image shows the tip of the finger C0 and the right image shows a portion of
E0:22 .H / D E0:21 .E0:21.H // for 1 x 6 and 2 y 2: The red sub-fingers do not meet the
(white) boundary of C0 .
We continue to take inverse images of H to see that, for n 2 N, the set E0:2n .H / has
a complement C n E0:2n .H / consisting of fingers inside of fingers inside of fingers, and so
on. More precisely, we make the following definition.
Definition 1.139. We call each component of C n E0:2n .H / a stage n finger.
Figure 1.25 depicts stage 1 fingers in red and the right picture in Figure 1.26 depicts
stage 2 fingers in red. The gaps (in black) are then portions of E0:22 .H / that separate the
fingers from each other. We encourage you to investigate stage n fingers more closely
using the Parameter Plane and Julia Set Applet by setting both the Dynamic plane
max iterations and the Dynamic plane min iterations to n, while setting
the Escape criterion to Re z > 1. Each stage n finger is contained in a stage n 1
finger for n 2. The boundary of a stage n finger for n 2 is mapped by E0:2 onto
n
the boundary of a stage n 1 finger and thus is mapped by E0:2 onto the vertical line
fRe z D 1g.
It will be important to understand how thick fingers can be and so we make the follow-
ing definition.
Definition 1.140. Let F be a stage n finger. We define the thickness of F to be the value
of t such that F cannot contain an open Euclidean disk with diameter strictly larger than
t, but for any value s < t, the set F contains an open Euclidean disk with diameter s, i.e.,
t D supf2r > 0 W .z; r / F for some z 2 F g.
Exercise 1.141. Prove that the thickness of a stage 1 finger Ck is by using the mapping
properties of the exponential map. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 57 — #75

i i
The set Fn D CnE0:2n .H / is the union of all stage n fingers and because a stage n finger
is contained in a stage n 1 finger we see that F1 F2 and F1 \F2 \ \Fn D Fn .
Thus we regard \1 nD1 Fn as the union of the stage infinity fingers. They, together with the
point at 1, comprise the Julia set, a proposition we state as follows.
Proposition 1.142. The Julia set J.E0:2 / D \1
nD1 Fn [ f1g D C n AE0:2 .p/.
Proof. Let J D \1 nD1 Fn [ f1g. We then have, whose proofs we leave to the exercises:
(a) any two fingers are separated by an infinitely long gap (of black) points in AE0:2 .p/
(Additional Exercise 1.197), and (b) the thickness of the stage n fingers shrinks to zero as
n goes to infinity (Additional Exercise 1.198). Facts (a) and (b) show that J cannot contain
any open set (Additional Exercise 1.199). Hence for z 2 J there are points arbitrarily
close to it that lie in [1 n
nD1 E0:2 .H /, and thus iterate to p. Since the orbit of every z 2 J
lies forever in fRe z 1g by the definition of J (and thus differs from an orbit which
converges to p), we get that z is sensitive to initial conditions. So z 2 J.E0:2 /. Likewise,
for a point not in J , it and nearby points which also lie in the open set AE0:2 .p/ have
the same dynamic behavior (they all iterate towards p), placing the point in the Fatou set
F .E0:2 /. Hence J D J.E0:2 /.
Exploration 1.143. Use the Parameter Plane and Julia Set Applet to track the orbit of
nearby points in J.E0:2 / to see the sensitivity to initial conditions. It is a bit difficult to
pick points that stay in the viewing window for more than a few steps, but by looking at
the data in the Orbits 1 and 2 tab at the bottom, you can see the orbits separate from
each other. You can also zoom in on the picture so that you can choose your initial seed
values close to each other, and then zoom out to see their orbits. Try it out!
1.6.2 Hairs and Endpoints

Thinking again about this construction of J.E0:2 / we now describe this Cantor bouquet
by describing the hairs (previously called the stage infinity fingers) that make up the set.
This section is not crucial to the rest of the text so we will not provide all of the details to
prove everything claimed, but we will discuss some fascinating aspects of what we saw in
Example 1.136.
Pick a stage 1 finger Ck1 , and from the infinite number of sub-fingers within Ck1 pick
a sub-finger Ck1 k2 . From the infinite number of sub-sub-fingers within that, pick a sub-
sub-finger Ck1 k2 k3 . Continuing so that each Ck1 k2 :::kn is a stage n finger in Ck1 k2:::kn 1 ,
we see that D \1 nD1 Ck1 k2 :::kn must, by (b) be infinitely thin (i.e., contain no open disk
of any positive radius), but also stretch to 1 in the positive x-direction. We call a hair,
and note that there are infinitely many of them in J.E0:2 / corresponding to the different
ways one can choose k1 ; k2 ; : : : . Each hair is a curve, an image of a continuous map
h W Œ0; 1/ ! C such that h .t/ ! 1 as t ! 1. We call h .0/ the endpoint of the
hair . There are a few important details necessary to prove all this. One is showing that
there is something left in the set D \1 1
nD1 Ck1 k2 :::kn , or even in the larger set in \nD1 Fn .
Perhaps, the left tips of the stage n fingers move farther and farther to the right so that
the set \1nD1 Fn is empty. These matters can be pursued in [6]. Additional Exercise 1.200,
however, gives another way to see that there are infinitely many hairs (shown to have finite
endpoints) in J.E0:2 /.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 58 — #76

i i
As in the construction, each hair is separated from another by an infinite (though pos-
sibly thin) gap that stretches out to 1 in the direction of the positive x-axis. Thus in C we
would say these hairs are all separated from each other. However, by including 1, where
all these hairs meet, we produce a set J.E0:2 / that is a connected subset of the sphere C.
We conclude this section with remarks about two fascinating properties of the Cantor
bouquet J.E0:2 /. To discuss the first properly we need a more general definition of what it
means for a subset of C to be connected.
Definition 1.144. A set C is connected in C if there do not exist open subsets U and V of
C such that (i) U \ C ¤ ;, (ii) V \ C ¤ ;, (iii) C U [ V , and (iv) U \ V \ C D ;.
A more intuitive definition is: A set C C is connected in C if we cannot break up
(disconnect) C into a disjoint union of non-empty sets A and B such that no sequence
from A converges to a point in B, and vice versa.
We can now present (without proof) two startling properties of the Cantor bouquet
J.E0:2 /.
Property 1. Let E denote the set of endpoints of all the hairs in J.E0:2 /. Then E D
E [ f1g is connected in C. Informally, this says that the endpoints together with 1 are
somehow bunched so tightly that it is impossible to disconnect the set. However, and this
is the strange part, removing 1 from E creates a totally disconnected set E (i.e., the
only connected subsets of E are those that contain only a single point). We can show this
easily by noting that between any two hairs there is a gap in F .E0:2 /, thus showing that
the endpoints cannot lie in the same connected subset of E. The gap extends to 1 and
so cannot be used to create a disconnection of E . It remains to be shown that no other
method of disconnecting E works either. This is too complicated for the present text, but
the details are in [21].
Property 2. The Hausdorff dimension of a set in C is a number that relates to the size of
the set. It is defined (see [11] for further details) so that it shares many of the properties of
our usual notion of dimension and so we refer to it as a dimension though it need not be an
integer. The Hausdorff dimension of a set measures, in a sense, how much space the set fills
up. The Hausdorff dimension of any smooth curve, in particular any hair in J.E0:2 / (which
is shown to be smooth in [5]), is 1, whether the endpoint is included or not. Let H denote the
union of all open hairs, that is, of hairs with endpoints removed. Thus H D J.E0:2 / n E .
The hairs are separated by gaps that prevent them from accumulating too much and filling
up a lot of space. So it seems plausible that H has Hausdorff dimension 1, which it does.
What is astounding is that the set of endpoints E has dimension 2. Thus the endpoints seem
to bunch up so that the set of all of them fills up two dimensions of space. Since for each
infinitely long hair there is only one endpoint, it seems impossible that the dimension of
the set of endpoints could be larger than the dimension of the set of open hairs. But this is
what happens. Details are in [18, 19].
1.6.3 Critical Orbits, Exploding Julia Sets, Parameter Space for Ec

As we have seen, critical points dictate much about the dynamics of a rational map, espe-
cially for the polynomial fc .z/ D z 2 C c. The exponential map Ec .z/ D ce z , however,
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 59 — #77

i i
has no critical points (verify), but has 0 as what we call an asymptotic value and this will
play a similar role.
Definition 1.145. An asymptotic value for an entire function f is a value a 2 C such that
there exists a curve W Œ0; C1/ ! C tending to 1 (i.e., .t/ ! 1 as t ! C1) such
that f . .t// ! a as t ! C1.
For Ec we may choose the curve to be the negative real axis, along which the values
Ec .z/ ! 0 since Re z ! 1. The following theorem, whose proof can be found in [6],
parallels Theorems 1.89 and 1.93 for rational maps.
Theorem 1.146. Let f be a transcendental entire map. Then the immediate basin of each
(super) attracting cycle of f contains a critical point of f or an asymptotic value of f .
From this we see that the orbit fEcn .0/g1
nD1 plays a special role and so we call it the
critical orbit.
Fact 1. An attracting cycle of Ec attracts the critical orbit fEcn .0/g1
nD1 .
Fact 1 follows from Theorem 1.146. The next fact has a proof beyond the scope of this
text, but we shall nonetheless make frequent use of it.
Fact 2. If Ecn .0/ ! 1, then J.Ec / D C.
Fact 2 is different from anything we have seen. The super attracting fixed point at 1
for any polynomial implies that we could never have an empty Fatou set.
Fact 2 also leads us to discover the fascinating exploding Julia sets we can see when
we vary c in the map Ec . Example 1.136 showed that J.E0:2 / is a Cantor bouquet of
hairs and F .E0:2 / is the attracting basin of a finite attracting fixed point p 0:26. If we
n
increase c from c D 0:2 to c D 0:5 we see that E0:5 .0/ ! C1 and so, by Fact 2, we have

J.E0:5 / D C. Hence some real value c between c D 0:2 and c D 0:5 is a bifurcation
point where the dynamics drastically change. What we see is that as c grows from being
smaller than c to larger than c the Julia set goes from being confined to fRe z 1g to
exploding to be all of C. You are asked to determine c and show this exploding behavior
in Additional Exercise 1.201.
So we get different dynamic behavior for Ec depending on whether the critical orbit
escapes to 1. We use this dichotomy to color the parameter plane based on how many
iterations it takes for the critical orbit to escape (that is, have real part become greater than
50), leaving c colored black if the critical orbit does not escape. Investigate the parameter
space using the Parameter Plane and Julia Set Applet. Set the window in the parameter
plane to show Œ 3:2; 3:2 Œ 3; 3 to find a cardioid shaped region, with many bulbs
attached. You may want to set the Parameter plane max iterations value to
40, 60, 80, or 100 to see the cardioid in more detail. Because of the computing power
needed to compute the complex exponential iterates, each picture may take a while for the
applet to complete.
Exploration 1.147. Experiment with the applet to see if you can guess the dynamical
significance of the cardioid and some of the bulbs. Try it out!
Let’s investigate the exponential parameter plane as we did for the family fc .z/ D
2
z C c by calculating the set L1 of c values for which Ec has an attracting fixed point w.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 60 — #78

i i
For w to be an attracting fixed point we require Ec .w/ D w and jEc0 .w/j < 1: The first
condition yields ce w D w which used with the second gives jwj D jce w j < 1: Notice w is
both the fixed point and the multiplier. Thus we are looking for c such that c D we w for
jwj < 1. By moving w around the unit circle, the c values trace out the cardioid-like set in
the parameter plane pictured in Figure 1.27. Use the ComplexTool applet to draw this.
–1 –2
–1
F IGURE 1.27. This cardioid-like shape is the boundary of L1 , the set of c parameters that give rise
to an attracting fixed point for the map Ec .z/ D cez .
This cardioid-like L1 has similarities with the cardioid K1 in the Mandelbrot set. For
example, we can find p=q bulbs attached to its boundary. The p=q bulbs might not be
located where you first expect. The 1=3 bulb in the Mandelbrot set M is at the top of the
cardioid K1 whereas the top of L1 has the 2=5 bulb attached.
Exploration 1.148. Use the Parameter Plane and Julia Set Applet to investigate the new
locations of the p=q bulbs. One way to determine their locations is to experiment long
enough to find a pattern (can you use Farey addition here too?). Another way is to use
the multiplier map, or its inverse, which you are asked to determine in Additional Exer-
cise 1.202. Be sure the settings for the applet have values that produce good pictures. You
may want to check the values for the escape criterion and max and min iterations which
you adjusted earlier, or reload the applet to reset them. Try it out!
Exploration 1.149. Where is the 1=2 bulb located? Does it have a sub-bulb that corre-
sponds to attracting 4-cycles? Try it out!
Exercise 1.150. Whether or not you think you can solve it, try to find an interesting math-
ematical question to ask about the dynamics of exponentials. Try to draw parallels with the
family fc .z/ D z 2 C c, or show where parallels do not hold. Try it out!
1.6.4 Trigonometric Dynamics

Recall that sin z D 2i .e iz e iz / and cos z D 12 .e iz C e iz /, and set Sc .z/ D c sin z
and Cc .z/ D c cos z. Additional Exercise 1.203 asks you to show that orbits under the
maps Sc .z/ and Cc .z/ escape to 1 in the direction of the positive or negative imaginary
axis. Thus, using Proposition 1.133 to note J.Sc / D ASc .1/ and J.Cc / D ACc .1/, the
algorithm in the Parameter Plane and Julia Set Applet for creating the Julia sets for the
exponential maps Ec .z/ D ce z can be modified. An orbit under either Sc or Cc is deemed
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 61 — #79

i i
1.7. The Mandelbrot Set is Universal 61
to escape to 1 if any point in the orbit has imaginary part larger than 50 in absolute value.
Additional Exercise 1.204 shows that neither Sc nor Cc has a finite asymptotic value, but
both have infinitely many critical points. This may appear to present a difficulty since a
map having more than one critical orbit would pose a challenge to coloring and analyzing
the parameter space as we did for z d C c and ce z . However, as we ask you to verify in
Additional Exercise 1.205, the maps Sc and Cc have only one critical orbit. The critical
orbit for Sc is fScn .c/g1 n 1
nD1 and the critical orbit for Cc is fCc .c/gnD1 .
As in the exponential case we have key facts that will aid our understanding of the
dynamics of these trigonometric functions.
Fact 1. Any attracting cycle of Sc attracts its critical orbit fScn .c/g1
nD1 .
Fact 2. Any attracting cycle of Cc attracts its critical orbit fCcn .c/g1
nD1 .
Fact 3. If Scn .c/ ! 1, then J.Sc / D C.
Fact 4. If Ccn .c/ ! 1, then J.Cc / D C.
Because the orbit fScn .c/g1

nD1 is critical to the dynamics of the map Sc , we can color
the parameter plane as we did for the family of maps Ec . The Parameter Plane and Julia
Set Applet will color c based on how many iterations it takes for the critical orbit to escape,
i.e., have imaginary part larger than 50 in absolute value. It is left black if the critical orbit
does not escape. The applet for Cc works similarly. Use it to experiment and investigate
the dynamics of Sc and Cc . If you look closely in the parameter planes, you will find a
familiar friend. Try it out!
1.7 The Mandelbrot Set is Universal

You may have noticed that the parameter planes for the maps fc .z/ D z 2 C c, Cc .z/ D
c cos z, and Sc .z/ D c sin z drawn by the Parameter Plane and Julia Set Applet show
black regions that contain sets resembling the Mandelbrot set M . See Figure 1.28.
The appearance of copies of the Mandelbrot set in these diverse settings is unexpected.
Why should the Mandelbrot set, a picture of the fundamental dichotomy (whether the crit-
ical orbit escapes or not) in the parameter space of quadratic maps fc .z/ D z 2 C c, have
anything to do with the parameter space of the transcendental maps Cc .z/ D c cos z and
Sc .z/ D c sin z? As it turns out copies of the Mandelbrot set appear in so many parameter
planes that it is a fundamental mathematical object, like or e, that arises in many unex-
pected places. As baby Mandelbrot sets are dense in the boundary of the full Mandelbrot
set, so are they dense in many parameter spaces. The main reason, which we state only
loosely, is that often iterates of maps have dynamic behavior like the dynamic behavior of
iterates of a quadratic polynomial when the maps are considered on small domains. Thus,
as a parameter is changed, we see the same changing behavior in the parameter plane as
we do in the family fc .z/ D z 2 C c corresponding to the original Mandelbrot set. In [22]
you can find the details showing that small Mandelbrot sets are dense in the bifurcation
locus for what are called holomorphic families of rational maps.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 62 — #80

i i
F IGURE 1.28. The original Mandelbrot set M (upper left), a copy of M containing c D 3:2 in the
parameter space for c cos z (upper right), a copy of M containing c D 2:17 C 1:3i in the parameter
space for c sin z (lower left), and a copy of M containing D 0:906 C 0:423i in the parameter space
for Newton’s method applied to the map p .z/ D z.z 1/.z /.
1.8 Concluding Remarks and New Directions

We investigated the chaos that arises in a number of complex dynamical systems. We saw
how Newton’s method can (and often must) produce chaotic behavior, giving beautiful
fractal images. To investigate chaos more generally, we looked at iteration of analytic maps
that were not generated as the Newton map of a polynomial. We found that even in the
seemingly simple class fc .z/ D z 2 C c there was a richness and complexity we could not
have imagined. Similar features were present in the dynamics of the entire maps Ec .z/ D
ce z ; Sc .z/ D c sin z, and Cc .z/ D c cos z, though some new phenomena also appeared.
Throughout, an important part of our investigations was to track, if we could, how the
behavior of the systems changed when the maps were changed. We did this by varying
the parameter c. In some sense, however, these were really mild perturbations of the maps
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 63 — #81

i i
1.8. Concluding Remarks and New Directions 63
since many of the salient features of the maps (such as the super attracting fixed point at 1
for fc .z/ D z 2 C c or the essential singularity at 1 for the transcendental maps) persisted
no matter how the c values were changed.
1.8.1 Perturbation with a Pole

There are many interesting ways to perturb a map that produce beautiful mathematics. The
maps Fc .z/ D z d C c=z m with d; m 2 N are, for c 2 C n f0g, perturbations of the
dynamically well understood map z 7! z d . Unlike the fc .z/ D z 2 C c perturbations of
z 7! z 2 , they add a new dimension to the analysis due to the pole at the origin. For c small,
Fc behaves like z 7! z d , but only as long as z is sufficiently far from the origin. Near the
origin the pole changes the dynamics considerably from z 7! z d . These systems lead to
fascinating mathematics that can sometimes be represented by pictures such as Figure 1.29
showing features not seen before in this chapter.
F IGURE 1.29. A Sierpinski curve Julia set for F .z/ D z 3 C 0:13i=z 3. The regions in shades of
orange are in AF .1/ and the points in yellow represent J.F /.
Large Project 1.151. Investigate the dynamics of the maps Fc .z/ D z d C c=z m . You
can use the Parameter Plane and Julia Set Applet and your analytic paper and pencil tech-
niques. We have not described how the parameter plane is being drawn. It is up to you to
determine this and to figure out what you can on your own. Of all the maps mentioned in
this chapter, these are the newest and least studied.
1.8.2 Random Dynamics

In the examples of perturbed maps throughout this chapter, though the perturbation could
be mild by changing a simple parameter or more severe by adding a pole as in Fc , there
was one fundamental assumption made regarding the perturbed system – once the map was
perturbed, it was fixed and iterated again and again to create the dynamics. Another way to
perturb a system is to allow the map to change at every step in the orbit.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 64 — #82

i i
Suppose we start with two maps f and g. The usual iteration dynamics considers the
orbit generated by f n .z0 / or by gn .z0 /. It is natural to ask what happens if at each stage of
the orbit either map f or g can be applied. In some cases the dynamics can be much more
fun. We study the behavior of orbits hi1 .z0 /; hi2 .hi1 .z0 //; hi3 .hi2 .hi1 .z0 ///; : : : where the
maps hi are chosen to be either f or g. We use the term random dynamics when the choices
of maps are made at random. Such systems are connected to iterated function systems and
their attractor sets (see [8]), such as the van Koch curve and Sierpinski triangle.
Instead of investigating such attractor sets, we look in another direction. We investigate
what we call a random basin of attraction as follows. Fix a point z0 2 C and randomly
select h1 to be either f or g, each with probability 1/2. Then set z1 D h1 .z0 /. Randomly
select h2 to be either f or g and set z2 D h2 .z1 / D h2 .h1 .z0 //. Continue to produce a
random orbit z0 ; z1 ; z2 ; : : : , and consider the probability that it converges to infinity.
Example 1.152 (Devil’s Staircase). Let f .x/ D 3x and g.x/ D 3x 2 be defined on
the real line R. For x0 2 R generate a random orbit x0 ; x1; x2; : : : . Let P .x0 / denote the
probability that it converges to C1, noting that we distinguish between convergence to
C1 and 1. It is plausible that if x0 > 1, then P .x0 / D 1 since no matter how we
choose f and g at each step, we have the orbit points grow larger and positive. Similarly,
if x0 < 0, then P .x0 / D 0 since both f and g force the orbit points to grow larger and
negative. What happens for 0 x0 1, however, is more interesting. We show the graph
of P , called the Devil’s Staircase, in Figure 1.30.
1
F IGURE 1.30. The Devil’s Staircase is the graph of the probability P.x/ that a random orbit gen-
erated by the maps f .x/ D 3x and g.x/ D 3x 2 tends to C1.
An interesting property of this graph is that between any two steps (i.e., intervals where
P 0 .x/ D 0), we have infinitely many steps. Also, the sum of the lengths of the steps
between 0 and 1 is exactly 1, so we say that this graph is almost always flat, i.e., almost
always has P 0 .x/ D 0. However, P .x/ is both continuous and increasing from y D 0 to
y D 1 as x goes from 0 to 1.
Exercise 1.153. Properties of the Devil’s Staircase.
(a) Show that P .x/ D 0 for x < 0 and P .x/ D 1 for x > 1.
(b) Show that P .0/ D 0 and P .1/ D 1.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 65 — #83

i i
1.8. Concluding Remarks and New Directions 65
(c) Show that P .x/ is increasing (not strictly), i.e., x < y implies P .x/ P .y/.
(d) Show that P .x/ D 1=2 and P 0 .x/ D 0 on the interval .1=3; 2=3/.
(e) Show that P .x/ D 1=4 and P 0 .x/ D 0 on the interval .1=9; 2=9/.
(f) Find where P .x/ D 3=4.
(g) In parts (d)-(f), you located the largest step (corresponding to .1=3; 2=3/), and the two
next largest steps. Describe the four next largest steps and the pattern relating all steps.
(h) Show that the sum of the lengths of the steps is 1.
(i) Show that P .x/ is continuous.
Example 1.154 (Devil’s Colosseum). The Devil’s Colosseum is a higher dimensional ana-
log of the Devil’s Staircase, and is also understood through random dynamics. Instead of
functions defined on the real line, we let g1 .z/ D z 2 =4 and g2 .z/ D z 2 1 be complex-
valued maps, and then use their second iterates to define f .z/ D g12 .z/ and g.z/ D g22 .z/.
We now investigate the random basin of attraction for 1 as follows. Fix a point z0 2 C.
For the random orbit z0 ; z1 ; z2 ; : : : , let P .z0 / be the probability it tends to 1. For z0 large
(jz0 j > 10 will do), we have P .z0 / D 1 since no matter what choices of f or g we make
at each step, the orbit of z0 will go to 1. Though it is not as simple, we can also show that
P .z0 / D 0 for z0 near zero. In between, however, we get interesting behavior. The function
P .z/ has a graph illustrated in Figure 1.31 called the Devil’s Colosseum and has properties
similar to those of the Devil’s Staircase. To climb out from the bottom, you would walk a
path like the Devil’s Staircase.
F IGURE 1.31. The left picture shows the Devil’s Colosseum, the graph of the probability P.z/ that
a random orbit tends to 1. The picture on the right is the inverted Devil’s Colosseum, sometimes
called a fractal wedding cake. The first pictures of these objects were created and studied by Hiroki
Sumi (see [25]).
We can describe the dynamics that underlie the Devil’s Colosseum in terms of stability
of orbits, and in doing so introduce the following extension of the notions of the Fatou and
Julia sets. Let F .hf; gi/ denote the Fatou set of the random system, defined to be the set of
points z0 2 C such that every randomly generated sequence of maps hin produces a stable
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 66 — #84

i i
orbit in the sense that nearby seed values have similar orbits when the sequence hin is used.
We define the Julia set of the random system to be J.hf; gi/ D C n F .hf; gi/: We can then
describe the oddity of the Devil’s Colosseum. The function P is continuous on C, yet it
is constant on each component of the Fatou set even though the Julia set contains no open
set (see [25]). It is these components of the Fatou set that make up the various horizontal
levels, or steps, in Figure 1.31. The Julia set is the set on which P .z/ varies (since P .z/
does not vary, i.e., it is flat, on F .hf; gi/). The Julia set is pictured in Figure 1.32.
F IGURE 1.32. The Julia set J.hf; gi/ in C is the set on which P.z/ varies. This picture was drawn
via the computer application Julia 2.0 (see [23]).
Remark 1.155. The random dynamics in the Devil’s Colosseum produces something that
cannot happen with usual iteration dynamics. Given a polynomial f .z/, define P .z0 / to be
the probability of an orbit with seed z0 tending to 1. Since there is only one map to choose,
there is no randomness involved, and P .z0 / must be either 0 or 1. We have P .z/ D 1 for
z 2 Af .1/ and P .z/ D 0 for z 2 K.f /. The function P is then discontinuous at every
point z 2 @Af .1/ D @K.f / D J.f /. Thus we see that the random dynamics exhibited
by the Devil’s Colosseum, where P .z/ is continuous on all of C, is a change from iteration
dynamics.
Exercise 1.156. You are encouraged to investigate random dynamical systems visually by
generating your own pictures similar to the Devil’s Colosseum. Sample code is provided
in Appendix 1.C.
There is a large world of chaotic phenomena to explore. Whether you stay within the
confines of the topics of this book or venture into new areas, there is an infinite amount of
mathematics to discover. We encourage you to explore and make your own contributions
to mathematics.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 67 — #85

i i
1.9. Additional Exercises 67
1.9 Additional Exercises

Orbits, Examples, and Fixed Points
Exercise 1.157. Prove f n .x/ ! C1 for all x 2 R when f .x/ D e x . The advanced
reader can show that the convergence is uniform, i.e., given M > 0, there exists an integer
N > 0 such that n > N implies f n .x/ > M for all x 2 R.
Exercise 1.158. For f .x/ D sin x where x 2 R is given in radians prove that f n .x/ ! 0
for all x 2 R. Hint: Apply the mean value theorem to show that j sin xj < jxj for all
x 2 R n f0g. The advanced reader can show that the convergence is uniform, i.e., given
" > 0, there exists an integer N > 0 such that n > N implies jf n .x/ 0j < " for all
x 2 R.
Exercise 1.159. For f .x/ D cos x where x 2 R is given in radians prove that f n .x/ !
0:739085:::: for all x 2 R where x D 0:739085:::: is the number such that cos x D x .
Hint: Apply the mean value theorem to show that j cos x x j < jx x j for all x 2
R n fx g. The advanced reader can show that the convergence is uniform, i.e., given " > 0,
there exists an integer N > 0 such that n > N implies jf n .x/ x j < " for all x 2 R.
Exercise 1.160. For g.z/ D sin z, show that gn .˙i "/ ! 1 for any " > 0. Hint: Show
that g.iy/ is purely imaginary and jg.iy/j > jyj for each real y ¤ 0.
Complex Newton’s Method

Exercise 1.161. Let F be the Newton map of a polynomial f of degree two or more.
(a) Suppose F n .z0 / ! a for some z0 2 C and some finite point a 2 C. Prove f .a/ D 0.
(b) Show that F .1/ D 1 and so it is necessary that part (a) consider only finite points.
Exercise 1.162. Prove that the Newton map of f .z/=f 0 .z/ has super attracting fixed points
at the roots of f regardless of the orders of the roots of f . Hint: Given a root ˛ of f of
order k, determine the order of the root of f .z/=f 0 .z/.
Exercise 1.163. Construct examples of analytic maps f to justify the statement that there
is no universal r > 0 such that 4.˛; r / AF .˛/ for all analytic maps f with a root at
˛, where F denotes the Newton map of f .
Exercise 1.164. Let p.z/ D .z ˛/k .z ˛1 / : : : .z ˛s / be a polynomial of degree n D
k C s, where the ˛j need not be distinct. Follow the steps below to show that r D d.2k 2n 1
1/
is a radius of convergence for the Newton map F at ˛, where d D minj D1;:::;s j˛ ˛j j.

(a) For the Newton map F .z/ D z pp.z/ 0 .z/ , we wish to show that jz ˛j < r implies
jF .z/ ˛j < jz ˛j. This says that the action of F is to move points in the disk
.˛; r / closer to ˛. Show that it implies that F n .z/ ! ˛ as n ! 1 when jz ˛j < r .
ˇ ˇ
ˇ ˇ
(b) Show that jF .z/ ˛j < jz ˛j is equivalent to jFjz.z/˛j˛j D ˇ1 .z p.z/ ˛/p .z/ ˇ
0 < 1.
ˇ ˇ ˇ ˇ ˇ ˇ
(c) Show that ˇ1 w1 ˇ D ˇ ww 1 ˇ D ˇ w 1ˇ
w 0
< 1 holds if and only if Re w > 1=2.
h P i
˛/p0 .z/
(d) Justify the following. The term w D .z p.z/ D .z ˛/ z k ˛ C js D1 z 1˛ D
j
P P
k C js D1 zz ˛˛ . Thus Re w > 12 exactly when Re js D1 zz ˛˛ > 12 k, which holds,
j
P ˇ ˇ j
ˇ ˇ
in particular, when js D1 ˇ zz ˛˛ ˇ < k 12 .
j
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 68 — #86

i i
(e) Show that if jz ˛j < r , then jz ˛j j > d r , where d D minj j˛ ˛j j. Hence

Ps ˇˇ z ˛ ˇˇ r
j D1 ˇ z ˛j ˇ < s d r D .n k/ d r r .
(f) Combine the above to reach the desired conclusion. Also, note that for z 2 .˛; r / we
have jF .z/ ˛j < jz ˛j, a stronger statement than saying that .˛; r / Af .˛/. For
this reason, the value r D d.2k
2n 1
1/
may be called a radius of contraction for Newton’s
method at ˛.
Small Project 1.165. Fix an integer n 2 and let f .z/ D z n 1. Express the value for r
found in Exercise 1.164 in terms of only n. See if you can improve on it for the radius of
contraction by investigating the proof applied to the maps f .z/ D z n 1. Hint: You may
wish to calculate an r > 0 such that jF 0 .z/j < 1 on 4.˛; r / for each root ˛ of f , and
then apply the Contraction Lemma 1.138.
Small Project 1.166. This is an open-ended project to investigate whether the value for
r found in Exercise 1.164 can be improved if we know something about the geometry
of the roots of p. See if you can improve on the value for the radius of contraction by
investigating the proof in Exercise 1.164 for polynomials f of degree 3 with distinct roots
˛p ; ˇp ; p . If the roots all lie on a line can you get more from the proof? What if the roots
form an equilateral triangle? Are there geometric configurations of the roots that allow for
better results than what is given in Exercise 1.164? Can you generalize to higher degree
polynomials? What if you allow for multiple roots of f ? If a paper and pencil result is
too hard, provide a conjecture based on examples you consider with the Complex Newton
Method Applet. Whether or not you can answer them, give some related questions that one
might consider. Remember, asking questions is important, even if you can’t answer them.
Hint: You may wish to calculate an r > 0 such that jF 0 .z/j < 1 on 4.˛; r / for each
root ˛ of f , and then apply the Contraction Lemma 1.138.
Global Conjugation
Exercise 1.167. Prove that if ab ¤ 0, then the maps z 7! az and z 7! bz are globally
conjugate if and only if a D b or a D 1=b.
Exercise 1.168. Prove that if ab ¤ 0, then the maps z 7! z C a and z 7! z C b are
globally conjugate.
Exercise 1.169. Prove that a quadratic map f .z/ D az 2 C bz C d is globally conjugate
to one and only one map of the form z 7! z 2 C c.
Exercise 1.170. Prove that a rational map R is globally conjugate to a polynomial if and
only if there exists w 2 C with R 1 .fwg/ D fwg.
Newton Map of a Cubic Polynomial
Exercise 1.171. Prove that the attracting basin of an attracting fixed point of a rational or
entire map is an open set in C, and so does not contain any of its boundary points. Hint:
Use Theorem 1.17 together with the fact that if f is a continuous map, then f 1 .U / is
open whenever U is open.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 69 — #87

i i
Exercise 1.172. Use the Complex Newton Method Applet to investigate the behavior of
Newton’s method applied to the maps f1 .z/ D z.z 1/.z 0:908 0:423i / and f2 .z/ D
z.z 1/.z 0:913 0:424i /. Describe the behavior you see, especially for the black seed
values where Newton’s method fails.
Iteration of an Analytic Function
Exercise 1.173.
(a) Show that for the map f .z/ D z 2 there is a point z0 2 C.0; 1/ whose orbit is dense on
C.0; 1/, that is, ff n .z0 / W n 2 Ng D C.0; 1/. Hint: On C.0; 1/, we have f .e i 2 / D
e i 4 , so the angle 2 Œ0; 1/ is doubled (mod 1). Consider 0 2 Œ0; 1/ given in bi-
nary form as 0 D 0:0 1 00 01 10 11 000 001 010 011 100 101 110 111 0000 : : : where
spaces have been included in the binary expansion to illustrate the pattern.
(b) Use (a) to justify the second claim in Remark 1.46.
(c) Because of computer limitations, any point you choose on C.0; 1/ will have a com-
puted orbit that eventually becomes periodic, although it may take a large number of
orbit points to see this. Justify this statement.
Exercise 1.174. Complete the proof of Theorem 1.48.
(a) Use Schwarz’s Lemma (Appendix Theorem A.21 on page 353) to show that jf 0 .a/j <
1 for an attracting fixed point a 2 C of an analytic map f . Hint: Consider a small disk
4.a; r / where f is attracting and investigate the map g W 4.0; 1/ ! 4.0; 1/ defined
by g.z/ D f .r zCa/
r
a
.
(b) Suppose a 2 C is a repelling fixed point of an analytic map f . Use the fact that f
is locally one-to-one with a locally defined inverse to show jf 0 .a/j > 1. Hint: Since
jf 0 .a/j 1 (else a would be attracting), we have that f is locally one-to-one (see
Appendix Section A.6.1 on page 354 regarding the open mapping theorem and the
Inverse Function Theorem A.24). Letting h be f 1 defined on a small disk 4.a; r /,
show that a is an attracting fixed point for h (since it was repelling for f ) and then
apply part (a) to h. Use this to argue for the conclusion.
Exercise 1.175. Let f be a map that is analytic at 1 such that f .1/ D 1. Let D
1=f 0 .1/. Prove that if jj < 1, then 1 is an attracting fixed point according to Defini-
tion 1.16. Also, prove that if jj > 1, then 1 is a repelling fixed point according to Defini-
tion 1.44. Hint: Noting that the map k.z/ D 1=f .1=z/ has a fixed point at the origin with
multiplier (see Appendix Lemma B.20 on page 363), study how the map .z/ D 1=z
transfers information about k at the origin to information about f at 1.
Exercise 1.176. Let g.z/ be a polynomial of degree greater than or equal to two. Prove
that @Ag .1/ D @K.g/, and by Theorem 1.59 we have J.g/ D @Ag .1/ D @K.g/.
Exercise 1.177.
(a) For c D 0:467 C 0:513i determine the attracting fixed point pc of fc .
(b) Write the multiplier for pc in polar form.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 70 — #88

i i
(c) Use the Global Complex Iteration Applet for Polynomials to iterate a seed z0 near
pc one step at a time (use the zoom feature to get a better look) to see how the orbit
approaches pc . The way this orbit approaches pc is related to the multiplier . Describe
the connection.
(d) Use the Complex Function Iterator Applet to iterate the map z 7! z for seed values
near the attracting fixed point at the origin. Describe the convergence.
(e) Compare (c) and (d), explaining in as much mathematical detail as you can what you
find. Hint: Consider the Taylor series of fc expanded around the point pc .
(f) Choose various c in K1 and study the convergence of orbits to the attracting fixed point
using the applet. Use the picture from the applet to approximate the argument of the
multiplier. Note: Some c values will be easy to work with, while others are not as easy.
Exercise 1.178.
(a) For c D 0:92568 C 0:22512i determine the attracting 2-cycle fuc ; vc g of fc .
(b) Write the multiplier for fuc ; vc g in polar form.
(c) Use the Global Complex Iteration Applet for Polynomials to iterate a seed z0 near one
of the cycle points uc one step at a time (use the zoom feature to get a better look)
to see how every other orbit point approaches uc . The way this orbit approaches uc is
related to the multiplier . Describe the connection.
(d) Use the Complex Function Iterator Applet to iterate the map z 7! z for various seed
values near the attracting fixed point at the origin. Describe the convergence.
(e) Compare (c) and (d), explaining in as much mathematical detail as you can what you
find. Hint: Consider the Taylor series of the iterate fc2 expanded around the point uc .
(f) Choose any c in K2 and study the convergence of orbits to the attracting 2-cycle us-
ing the applet. Use the picture from the applet to approximate the argument of the
multiplier. Note: Some c values will be easy to work with, while others are not as easy.
Exercise 1.179. Determine the form of the multiplier map on K2 . The function W K2 !
4.0; 1/ maps c 2 K2 to the multiplier of the attracting 2-cycle of fc . Show that it can be
extended to a map W K2 ! 4.0; 1/ that is continuous, one-to-one, and onto.
Exercise 1.180. Without appealing to Theorem 1.79, prove that if fc0 has an attracting
n-cycle, then for all c close to c0 , the map fc has an attracting n-cycle. Conclude that Kn
is an open set. Hint: You may use Rouché’s Theorem (see [1, p. 294]) to prove the general
result that if the coefficients of a polynomial P are sufficiently close to the corresponding
coefficients of a polynomial Q of the same degree, then the roots of P and the roots of Q
are close. Apply this to the polynomials P1 .z/ D fcn0 .z/ z and Q1 .z/ D fcn .z/ z.
Exercise 1.181. Mathematically justify Remark 1.83 on page 34.
Critical Points and Critical Orbits

Exercise 1.182. Let f .z/ D z 3 1 and let F denote the corresponding Newton map. Show
that F cannot have an attracting cycle of any order other than the fixed points at the roots
of f . Hint: Check the behavior of each critical point of F and consider Theorem 1.89.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 71 — #89

i i
Note: More can be said. It is true that Newton’s method fails to find a root of f only
for seed values from @AF .1/ D @AF .e 2 i=3 / D @AF .e 2 i=3 /. In fact, C D AF .1/ [
AF .e 2 i=3 / [ AF .e 2 i=3 / [ @AF .1/: The method of proof is along the same lines but
uses a more powerful set of results. The results imply that if Newton’s method fails for all
seed values in an open set of points, then there is a critical point of F whose orbit does
not converge to any of the roots of f . The key ingredients to this proof are Sullivan’s No
Wandering Domains Theorem, the classification of forward invariant components of Fatou
sets, and the role of critical points in parabolic domains, Siegel disks, and Herman rings.
The results can be found in [1].
Exercise 1.183. Show that the periodic points for the map f .z/ D z 2 are of the form
e i 2p=q where p; q 2 N with p=q in lowest terms and q odd. Show that the pre-periodic
points have the same form, but with q even. Hint: Show that if p=q D 2kn 1 for k 1; n
1, then e i 2p=q has a period of k or some divisor of k. Then use the fact that every rational
number p=q with q odd can be expressed in this form. This can be proved by using the
fact that the element 2 in the multiplicative group consisting of those elements of Zq that
are relatively prime to q has a finite multiplicative order which we call k.
Exercise 1.184. One feature of the Mandelbrot set that stands out is that it is symmetric
about the x-axis. Prove this by showing that c 2 M if and only if cN 2 M . Hint: Compare
the critical orbits ffcn .0/g and ffcNn .0/g.
Exercise 1.185. Prove that the Mandelbrot set M has no holes (i.e., its complement C nM
is connected) by appealing directly to its definition. Hint: Suppose that U is a bounded
domain of C n M such that @U M . Apply Appendix Corollary A.19 on page 352 (a
version of the maximum modulus theorem) to the maps Qn on U , where Qn are as in the
proof of Lemma 1.103.
Exploring the Mandelbrot Set

Exercise 1.186. Let P be a polynomial of degree n 2. Show that the filled-in Julia set
K.P / has infinitely many points as follows.
(a) Show that P .z/ has a fixed point z0 by applying the fundamental theorem of algebra.
(b) Show that if z0 is such that P 1 .fz0 g/ D fz0 g, then P .z/ D z0 C a.z z0 /n for some
a ¤ 0 from which it can be shown that K.P / D .z0 ; r / where r D jaj1=.1 n/ .
(c) Show that if P 1 .fz0 g/ ¤ fz0 g, then there exists an infinite sequence of distinct points
z n such that 7! z n 7! z nC1 7! 7! z 2 7! z 1 7! z0 . Hence, fz n g
K.P /:
It is true that J.P / contains uncountably many points, but to show this requires more effort
(see [1, p. 95]).
Exercise 1.187. Prove that for c 2 C, we have z 2 Afc .1/ if and only if z 2 Afc .1/:
Do the same for J.fc / and K.fc /.
Exercise 1.188. Since the root cp=q of the p=q bulb in the Mandelbrot set lies on the
boundary of two hyperbolic components, namely K1 and Bp=q , Theorem 1.79 implies
that there are two multiplier maps that are defined at cp=q . The map W K1 ! 4.0; 1/
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 72 — #90

i i
can be extended to be defined at cp=q , as can p=q W Bp=q ! 4.0; 1/. Does .cp=q / D
p=q .cp=q /? What does it mean if they are the same? different?
Exercise 1.189. Find a c value in B5=32 . Formal proof is not required.
Exercise 1.190.
(a) Experiment with the Parameter Plane and Julia Set Applet to determine what Farey
fraction should be used to represent the cusp of the cardioid K1 in the Farey procedure
for finding child bulbs. We treat the cusp as a Farey parent which is larger than all other
bulbs.
(b) It is true (though not easy to prove) that starting with Farey parents B1=2 and the cusp
one can compute through Farey addition all of the p=q bulbs attached to the upper
half of K1 . Illustrate this by producing a Farey family tree that contains the lineage of
B5=32 up to the ancestors B1=2 and the cusp of K1 .
(c) Use Exercise 1.118 and part (b) to show that all of the p=q bulbs attached to K1 can
be identified through Farey addition.
Other Uni-critical Families of Polynomials

Exercise 1.191. Let Pc .z/ D z d C c for d D 2; 3; 4; : : : .
(a) Show that if jzj jcj and jzj > 2, then there exists " > 0 such that jPc .z/j >
jzj.1 C "/.
(b) Use induction to show that if jzj jcj and jzj > 2, then Pcn .z/ ! 1.
(c) Apply (b) to prove Pcn .0/ ! 1 as n ! 1 if and only if jPck .0/j > 2 for some
k 2 N. In particular, Md 4.0; 2/.
(d) Prove that the filled-in Julia set of Pc is contained in 4.0; 2/ when jcj 2. Show by
example that this does not hold when jcj > 2, i.e., it is not the case that for all c we
have Pcn .z/ ! 1 whenever jzj > 2.
Exercise 1.192.
(a) Use the Parameter Plane and Julia Set Applet to explore the parameter plane for the
family Pc .z/ D z d C c, and then identify and explain the symmetry you see there.
(b) Use the Parameter Plane and Julia Set Applet to explore the Julia sets of maps in the
family Pc .z/ D z d C c, and then identify and explain the symmetry you see there.
Exercise 1.193. For an integer d > 2, prove Md is the connectedness locus for the family
of maps Pc .z/ D z d C c, i.e., show that Md D fc 2 C W J.Pc / is connectedg.
Transcendental Dynamics
Exercise 1.194. Prove that Ecn .z0 / ! 1 if and only if Re Ecn .z0 / ! C1, where
Ec .z/ D ce z and c 2 C n f0g.
Large Project 1.195. This large project, to determine the accuracy of the exponential Julia
set applet pictures, is open-ended and should be attempted only after Section 1.6.1 has been
covered.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 73 — #91

i i
As mentioned in Remark 1.135 on page 54, the algorithm used to illustrate the expo-
nential Julia set J.E0:2 / can color some seed values z0 non-black (which it does when
20
at least one of the orbit points E0:2 .z0 /; : : : ; E0:2 .z0 / lands in the set fRe z > 50g), that
are actually not in J.E0:2 /. If z0 maps to E0:2 .z0 / D 51 C i , it will not be left black.
2
However, we see that E0:2 .z0 / D 0:2e 51 2 H D fz W Re z < 1g AE0:2 .p/, which
implies z0 2 AE0:2 .p/ D F .E0:2 /. The goal of the project is to estimate how close points
like z0 are to points in J.E0:2 /. This is, in a sense, a measure of the error in the algorithm.
One method to use is to consider an open rectangle R centered at z0 and its expanded
image E0:2 .R/. If it meets a point that has an orbit that truly tends to 1, then there is
a corresponding point in R that lies in J.E0:2 /. So you must consider how much R gets
expanded, and how big the expansion needs to be before you know that the expanded
image E0:2 .R/ contains a point with an orbit that truly tends to 1. Hint: Consider the
open interval .q; C1/ and its 2 i k-translates where q is the repelling fixed point for
E0:2 .
Also, you should consider what estimates you can get if w0 is such that Re E0:2 .w0 / <
2
50 but E0:2 .w0 / D 51 C i . Thus w0 takes two steps before the algorithm gives it a non-
black color, as opposed to z0 that used only one step. The point w0 is not in J.E0:2 /, but
there are points nearby which are in J.E0:2 /. Can you estimate how near these points are?
Some related questions are: Is there a relationship between the size of the viewing
window and the accuracy of the pictures? Can you quantify this, in general or in specific
cases? How does the relationship depend on the value 50 that was chosen to determine
the escape criterion or on the value 20 that determined how many orbit points might be
checked? If you set the Dynamic plane min iterations to 2 (which means that
a seed value is iterated twice before the escape condition is tested, i.e., a seed z0 is given a
non-black color only if at least one of the points z2 ; : : : ; z20 has real part greater than 50),
do we get a more accurate or less accurate picture?
This is a technical project. Some of these questions are addressed in [10].
Exercise 1.196.
(a) Prove the Contraction Lemma 1.138 stated on page 55. Hint: Consider the argument
preceding the statement of the lemma.
(b) Show that if f is a one-to-one analytic function such that jf 0 .z/j > > 1 for all
z in some domain D and f .D/ is convex, then f is expanding on D, i.e., we have
jf .z/ f .w/j jz wj for all z; w 2 D. Hint: Consider f 1 , which by the Inverse
Function Theorem A.24 on page 354, is analytic.
Exercise 1.197. Prove that two fingers (of any stage) constructed in Example 1.136 are
separated by an infinitely long gap of black points in AE0:2 .p/. Hint: The horizontal lines
Kn D fz 2 C W Im z D .2n C 1/ g; n 2 N lie in AE0:2 .p/ and separate the stage 1
fingers. Show that the inverse image of these lines under E0:2 separates the stage 2 fingers,
and then proceed inductively.
Exercise 1.198. Follow the steps (a)-(d) to prove that the thickness of a stage n finger is
no greater than = n 1 , where D ln 5 > 1.
(a) Prove that a stage n finger is contained in the right half plane fRe z g by showing
Re E0:2 .z/ jE0:2.z/j < 1 whenever Re z < .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 74 — #92

i i
(b) Suppose E0:2 .4.z0 ; r // fRe z > g for some r < . Show that E0:2.4.z0 ; r //
4.E0:2 .z0 /; r / by filling in the details of the following argument. Define a branch
L of the inverse of E0:2 on the convex set fRe z > g such that L.E0:2 .z0 // D z0
(given by L.z/ D Log z C ln 5 C 2 i n for some n 2 Z). For jz z0 j D r , we
then have r D jz z0 j D jL.E0:2 .z// L.E0:2 .z0 //j .1= /jE0:2 .z/ E0:2 .z0 /j
since jL0 .z/j D j1=zj < 1= when Re z > . Because E0:2 is univalent on 4.z0 ; r /,
it follows that E0:2 maps the circle jz z0 j D r to a simple closed curve C that
encloses 4.E0:2 .z0 /; r /. (At this point one could complete the proof by applying the
Argument Principle A.20 on page 352 to see that the number of zeros of E0:2.z/
E0:2 .a/ is equal to one for all values a in the interior of C .) Applying the Maximum
Modulus Theorem A.18 on page 352 to L.z/ z0 on E0:2 .4.z0 ; r //, we see that
L.4.E0:2 .z0 /; r // 4.z0 ; r /, and the result follows.
(c) Use parts (a) and (b) to prove that if 4.z0 ; r / is contained in a stage n finger for n 2,
then 4.E0:2 .z0 /; r / is contained in a stage .n 1/ finger.
(d) Use Exercise 1.141 and induction to complete the proof.

Exercise 1.199. Prove that the set J in Proposition 1.142 cannot contain an open set.
Exercise 1.200. Follow the steps below to construct infinitely many (though not all) of the
hairs in J.E0:2 /. Each hair is the image of a continuous map h W Œ0; 1/ ! C such that
h.t/ ! 1 as t ! 1.
(a) Prove that there exists a repelling fixed point q 2 R such that for any x > q we have
n
E0:2 .x/ > x and therefore E0:2 .x/ ! C1. Use this to conclude that the interval
h0 D Œq; C1/ is a hair in J.E0:2 /. We call it a straight hair since it extends to 1 in a
straight line.
(b) Use the 2 i periodicity of E0:2 to show that for each k 2 Z the set hk , defined to
be the k2 i translate of h0 , is also a hair in J.E0:2 /. We call each hk a stage 1 hair,
which could be called the main hair in the corresponding stage 1 finger.
(c) Argue that inside of the stage 1 finger C0 , there exist infinitely many hairs h0;k in
J.E0:2 / such that E0:2.h0;k / D hk . No h0;k is a straight hair, except for h0;0 which
equals h0 D Œq; C1/.
(d) Use the 2 i periodicity of E0:2 to show that for each j; k 2 Z the set hj;k , defined to
be the j 2 i translate of h0;k , is also a hair in J.E0:2 /. We call each hj;k a stage 2 hair.
(e) Repeat the arguments to show that there exist infinitely many stages of hairs in J.E0:2 /.
(f) If we let Hn denote all the stage n hairs, then [1

nD1 Hn [f1g is not quite all of J.E0:2 /.
However, we can show that J.E0:2 / D [1 nD1 n [ f1g by showing that each point in
H
the hairs described in Section 1.6.2 is a limit of points from the collection of stage n
hairs.
Exercise 1.201. Determine the exact real bifurcation value c between c D 0:2 and c D
0:5 for the family of maps Ec . Show that for c < c we have fRe z < 1g F .Ec /, but for
c > c we have F .Ec / D ;.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 75 — #93

i i
Exercise 1.202. For each c 2 L1 , the map Ec .z/ D ce z has an attracting fixed point with
multiplier .c/ (see Figure 1.27 on page 60). Thus we have a multiplier map W L1 !
4.0; 1/.
(a) Find the inverse of the multiplier map.
(b) Use the inverse of the multiplier map to find c where the 7=13 bulb attaches to the
cardioid L1 . Hint: What kind of fixed point will there be at this c value?
Exercise 1.203. Show that orbits under the maps Sc .z/ D c sin z and Cc .z/ D c cos z
escape to 1 in the direction of the positive or negative imaginary axis:
(a) Show that Scn .z/ ! 1 if and only if j Im Scn .z/j ! C1.
(b) Show that Ccn .z/ ! 1 if and only if j Im Ccn .z/j ! C1.
Exercise 1.204. Show that neither Sc .z/ D c sin z nor Cc .z/ D c cos z has a finite asymp-
totic value:
(a) Examine j sin.x C iy/j to show that if Sc had a finite asymptotic value, then the curve
(along which Sc has a finite asymptotic value) would be vertically bounded, that is,
lie in a horizontal strip fj Im zj M g.
(b) Show that sin z maps the imaginary axis into itself.
(c) Show that sin z maps the vertical line f=2 C iy W y 2 Rg into Œ1; C1/.
(d) Use the fact that sin z is 2 periodic along with parts (a), (b), and (c) to show that sin z
cannot have a finite asymptotic value.
(e) Show that Sc cannot have a finite asymptotic value since sin z does not.
(f) Show that Cc cannot have a finite asymptotic value since Cc .z/ D Sc .z C =2/.
Exercise 1.205.
(a) Explain in what sense fScn .c/g is the only critical orbit of Sc .
(b) Explain in what sense fCcn .c/g is the only critical orbit of Cc .
Definitions and Properties of the Julia and Fatou Sets

Exercise 1.206. Let an and bn for n 0 be sequences of positive real numbers such that
1 D a0 > b0 > a1 > b1 > : : : and an ! 0. Note bn ! 0 also.
(a) Construct a strictly increasing function g W Œ0; C1/ ! Œ0; C1/ such that g is dif-
ferentiable on .0; C1/ and g.an / D an and g.bn / D bn with g0 .an / D 0 and
g0 .bn / D 2. Construct g so that it does not fix any points other than the an , the bn ,
and 0. Hence each an is a super attracting fixed point of g and each bn is a repelling
fixed point of g. Sketch a graph of g to convince yourself that this can be done. It is
not necessary that you produce a formula for g.
(b) Define a function f W C ! C by f .r e i / D g.r /e i . Consider the dynamics of points
both on and near the circles jzj D an and jzj D bn to see that each “repelling circle”
jzj D bn , which is fixed by f , must lie in J.f /.
(c) Show that 0 2 F .f /. Hint: Use the fact that f ..0; an // .0; an / for each n 0.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 76 — #94

i i
(d) Show that (b) and (c), together with the fact that bn ! 0, implies that 0 is not in the
interior of F .f /.
Exercise 1.207. Use the definitions of Fatou and Julia set to prove the statements in Propo-
sition 1.213 parts (a)-(d). Hint: For the complete invariance statements, use the Open Map-
ping Theorem A.23 on page 354 to show that when it is non-constant, f is an open map,
i.e., if U is an open set in the domain of f , then the image f .U / is also an open set in C.
For (d), use Montel’s Theorem 1.210 and part (c).
Exercise 1.208. Show that the set of repelling cycles for f .z/ D z 2 is dense in J.f / D
C.0; 1/ by explicitly solving for the set of p-periodic points for each p 2 N.
1.10 Bibliography
[1] Alan F. Beardon, Iterations of Rational Functions. Springer-Verlag, New York, 1991.
[2] Bodil Branner. The Mandelbrot set. In Chaos and Fractals (Providence, RI, 1988),
volume 39 of Proc. Sympos. Appl. Math., pages 75–105. Amer. Math. Soc., Provi-
dence, RI, 1989.
[3] Lennart Carleson and Theodore W. Gamelin, Complex Dynamics. Springer-Verlag,
New York, 1993.
[4] Ruel V. Churchill and James Ward Brown, Complex Variables and Applications.
McGraw-Hill Book Co., New York, eighth edition, 2009.
[5] M. Viana da Silva, The differentiability of the hairs of exp.Z/. Proc. Amer. Math.
Soc., 103(4): 1179–1184, 1988.
[6] Robert L. Devaney. Chapter 4—Complex Exponential Dynamics. In F. Takens,
Henk Broer, and B. Hasselblatt, editors, Handbook of Dynamical Systems, volume 3
of Handbook of Dynamical Systems, pp. 125–223. Elsevier Science, 2010.
[7] Robert L. Devaney and Mónica Moreno Rocha, Geometry of the antennas in the
Mandelbrot set. Fractals, 10(1): 39–46, 2002.
[8] Robert L. Devaney, A First Course in Chaotic Dynamical Systems. Addison-Wesley
Studies in Nonlinearity. Addison-Wesley Publishing Company Advanced Book Pro-
gram, Reading, MA, 1992.
[9] Robert L. Devaney, A survey of exponential dynamics. In Proceedings of the Sixth
International Conference on Difference Equations, pp. 105–122, CRC. Boca Raton,
FL, 2004.
[10] Marilyn B. Durkin, The accuracy of computer algorithms in dynamical systems.
Internat. J. Bifur. Chaos Appl. Sci. Engrg., 1(3): 625–639, 1991.
[11] Kenneth Falconer, Fractal Geometry—Mathematical Foundations and Applications.
John Wiley & Sons, Chichester, 1990.
[12] Pierre Fatou, Sur les équations fonctionnelles. Bull. Soc. Math. France, 47: 161–271,
1919.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 77 — #95

i i
1.10. Bibliography 77
1920.
1920.
[15] James Gleick, Chaos. Penguin Books, New York, 1987.
[16] John G. Hocking and Gail S. Young, Topology. Dover Publications Inc., New York,
second edition, 1988.
[17] Gaston Julia, Mémoire sur l’itération des fonctions rationnelles. Journal de Math.
Pure et Appl., pp. 47–245, 1918.
[18] Bogusława Karpińska, Area and Hausdorff dimension of the set of accessible points
of the Julia sets of e z and sin z. Fund. Math., 159(3): 269–287, 1999.
[19] Bogusława Karpińska, Hausdorff dimension of the hairs without endpoints for
exp z. C. R. Acad. Sci. Paris Sér. I Math., 328(11): 1039–1044, 1999.
[20] Tan Lei, Ressemblance eintre l’ensemble de Mandelbrot set l’ensemble de Julia au
voisinage d’un point de Misiurewicz. In A. Douady and J. H. Hubbard, editors,
Étude dynamique des polynômes complexes. Partie II, volume 85 of Publications
Mathématiques d’Orsay [Mathematical Publications of Orsay], pp. 139–152. Uni-
versité de Paris-Sud, Département de Mathématiques, Orsay, 1985.
[21] John C. Mayer, An explosion point for the set of endpoints of the Julia set of exp.z/.
Ergodic Theory Dynam. Systems, 10(1): 177–183, 1990.
[22] Curtis T. McMullen, The Mandelbrot set is universal. In The Mandelbrot set, theme
and variations, volume 274 of London Math. Soc. Lecture Note Ser., pp. 1–17. Cam-
bridge Univ. Press, Cambridge, 2000.
[23] Rich Stankewitz, Wendy Conatser, Trey Butz, Ben Dean, Yun Li, and Kristopher
Hart, JULIA 2.0 Fractal Drawing Program.
http://rstankewitz.iweb.bsu.edu/JuliaHelp2.0/Julia.html.
[24] Norbert Steinmetz, Rational Iteration, volume 16 of de Gruyter Studies in Mathe-
matics. Walter de Gruyter & Co., Berlin, 1993.
[25] Hiroki Sumi, Random complex dynamics and semigroups of holomorphic maps.
Proc. London. Math. Soc. 102(1): 50–112, 2011.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 78 — #96

i i
1.A Appendix: Definitions and Properties of the Julia and

Fatou Sets
Here we present the formal definitions of the Fatou set and Julia set of a rational or en-
tire map. We also provide the statement of Montel’s theorem, a major tool in complex
dynamics, and a proposition stating some of the properties of the Fatou and Julia sets.
The spherical distance between the points z; w 2 C is .z; w/.
Definition 1.209 (Fatou Set). Let f be a rational or entire map. The Fatou set is the set
F .f / D fz 2 C W for every " > 0 there exists ı > 0 such that .z; w/ < ı implies
.f n .z/; f n .w// < " for all n 2 Ng:
Thus, if z 2 F .f / and " is small, then for w to have an orbit "-similar to the orbit of
z (by which we mean that corresponding orbit points are never more than " apart) we need
only to choose w close enough to z, i.e., within a distance ı. Another way to interpret this
definition is to say that F .f / is the set of points z such that for any " > 0 there exists
ı > 0 such that f n .4 .z; ı// 4 .f n .z/; "// for all n 2 N, i.e., a tiny neighborhood of
z (of size ı measured with the spherical metric) will have an orbit that stays tiny (of size no
greater than " measured with the spherical metric) along the entire orbit (see Figure 1.33).
This fails to happen with the map f .z/ D z 2 for any neighborhood of a point on the unit
circle C.0; 1/.
A major tool used in dynamics is Montel’s theorem, which we state without proof in
a setting most easily applied to the dynamics of interest to us here. (See [3, p. 9] to find it
stated in terms of normal families.)
f 5( ∆ σ (z 0 , δ))
f ( ∆ σ (z 0 , δ))
f 2 ( ∆ σ (z 0 , δ))
f f ...
∆ σ (z 5 ,ε)
∆σ ( z 1 , ε )
∆ σ (z 0 , δ ) ∆ σ (z 2 ,ε)
f
f f 3(∆ σ (z 0 , δ))
f 4 (∆ σ (z 0 , δ))
f
∆ σ (z 3 ,ε)
∆ σ (z 4 ,ε)
F IGURE 1.33. This picture illustrates how, for a point z0 2 F .f /, a tiny neighborhood .z0 ; ı/
has a forward image that stays tiny (within an " neighborhood of each orbit point zn ) for the entire
orbit. We illustrate this up to the fifth iterate, but the last arrow with the dots indicates that this
happens for all iterates f n .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 79 — #97

i i
1.A. Appendix: Definitions and Properties of the Julia and Fatou Sets 79
Theorem 1.210 (Montel’s Theorem). Let U C be an open set in the domain set of a
rational or entire map f . If the family of maps ff n W n 2 Ng omits three points z; w; v 2
C, i.e., f n .U / \ fz; w; vg D ; for all n 2 N, then U F .f /. In particular, if f .U / U
and C n U contains three or more points, then U F .f /.
Montel’s theorem can make it quick to show that f .z/ D z 2 has F .f / D CnC.0; 1/ D
.0; 1/ [ .C n .0; 1//. Since f ..0; 1// .0; 1/, the theorem gives .0; 1/ F .f /.
Similarly, we can show C n .0; 1/ F .f /. Also, from Section 1.3.2 or by considering
the definition of the Fatou set, we know that C.0; 1/ does not contain any points in the
Fatou set F .f /. Hence, we conclude F .f / D C n C.0; 1/.
Remark 1.211. Though its proof is too advanced for this text, it is true that if f is a rational
or entire map and z0 2 F .f /, then there exists r > 0 such that .z0 ; r / F .f /, i.e.,
F .f / is an open set. However, if f is not rational or entire, then F .f / need not be open.
To see this, consider the map f .z/ defined to be zero when z D x C iy with x; y 2 Q and
f .z/ D z otherwise. In this case f n D f for all n 2 N and F .f / D f0g. See Additional
Exercise 1.206 for such an example where f is continuous but not analytic.
Definition 1.212 (Julia Set). The Julia set of a rational or entire map f is defined to be
J.f / D C n F .f /:
For a point z to be in J.f / there must be points w that are arbitrarily close to z, but
which fail to have orbits similar to the orbit of z. More precisely, for a point z to be in
J.f / there must be " > 0 such that for every ı > 0, there exists w within a distance ı of z
such that the orbit of w is not "-similar to the orbit of z.
In Additional Exercise 1.207 you are asked to show the following properties (a)-(d).
Properties (e) and (f), however, require some tools from complex analysis which are a bit
beyond the level of this text (see [1, p. 148] and [24, p. 38] for details).
Proposition 1.213 (Properties of the Julia and Fatou Sets). Let f be a rational or entire
map. Then:
(a) F .f / is an open set in C and thus J.f / is a closed set in C.
1
(b) F .f / is completely invariant, i.e., f .F .f // F .f / and f .F .f // F .f /.
1
(c) J.f / is completely invariant, i.e., f .J.f // J.f / and f .J.f // J.f /.
(d) J.f / contains an open set if and only if J.f / D C.
(e) The set of repelling periodic cycles of f is dense in J.f /. That is, each repelling
periodic cycle of f is in J.f / and for every open set U that intersects J.f /, there is
a repelling periodic point z that lies in U .
(f) Let A be the set of z in J.f / such that the orbit of z is dense in J.f / (i.e., for every
open set U that intersects J.f /, there is an orbit point zn that lies in U ). Then A is
dense in J.f / (i.e., for every open set V that intersects J.f /, there is a point z 2 A
that lies in V ).
Remark 1.214. Additional Exercises 1.208 and 1.173 illustrate properties (e) and (f) for
f .z/ D z 2 . We see that, both in this example and in general, the Julia set contains a dense
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 80 — #98

i i
set of points that in some sense are the ultimate in predictable behavior. These are the
periodic points. Nothing could be more predictable than to have the orbit follow a finite
set of points over and over again since any orbit point can be only one of a few values.
However, the Julia set also contains points with an orbit that is dense in the Julia set. Such
orbits are unpredictable in the sense that an orbit point could be close to any of the infinitely
many points in the Julia set. Thus we see that inside the sensitive dependence that defines
chaos (and thus defines the Julia set J.f /) there lies an intriguing mix of predictability and
unpredictability.
1.B Appendix: Global Conjugation

and Möbius Map Dynamics
In this appendix we show that the global conjugation defined in Section 1.2.6 can be used
to quickly classify and understand the dynamics of a Möbius map, i.e., a map of the form
azCb
f .z/ D czCd where ad bc ¤ 0. The classification depends on the number and type of
fixed points of f . We begin with the following proposition.
Proposition 1.215. Non-identity Möbius maps can have only exactly one or exactly two
fixed points in C.
azCb
Proof. Let f .z/ D czCd be a non-identity Möbius map. Suppose that f .1/ ¤ 1, i.e,
c ¤ 0. Solving for fixed points of f , i.e., solving the equation f .z/ D z, yields the
equation cz 2 C .d a/z b D 0, which has two distinct roots or one double root in C. If
c D 0, then f is a linear map and so has one fixed point at 1 and possibly a second fixed
point in C.
We now describe the dynamics of a non-identity Möbius map f based on how many
fixed points it has.
Case 1: Suppose f .z/ fixes only 1. We can quickly show that f has the form f .z/ D
z C ˇ for some ˇ 2 C n f0g. Thus f n .z/ D z C nˇ, and hence f n .z/ ! 1 as n ! 1
for all z 2 C as in Example 1.49. We also see that J.f / D f1g.
Case 2: Suppose f .z/ fixes only w 2 C. Let .z/ D z 1w and define g.z/ D f 1
.z/,
noting that g is also Möbius. By Exercise 1.29 the map g.z/ has only one fixed point at
1. From Case 1 we see that g.z/ D z C ˛ for some ˛ 2 C n f0g and gn .z/ ! 1 for all
z 2 C. Hence f n .z/ D 1 n
g .z/ ! 1
.1/ D w for all z 2 C. So, if f is Möbius
n
with unique fixed point w, then f .z/ ! w for all z 2 C and J.f / D fwg.
Case 3: Suppose that f fixes 0 and 1. Then f .z/ D kz for some k 2 C n f1g (verify)
and thus f n .z/ D k n z. We have three categories based on jkj:
(a) If jkj < 1, then f n .z/ ! 0 for all z 2 C n f1g and J.f / D f1g.
(b) If jkj > 1, then f n .z/ ! 1 for all z 2 C n f0g and J.f / D f0g.
(c) If jkj D 1, then f is a rotation z 7! e i z for some 2 .0; 2 / whose dynamics are
easy to understand, and F .f / D C.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 81 — #99

i i
1.C. Appendix: Code for Drawing Random Dynamics Pictures 81
Case 4: Suppose that f fixes w1 and w2 where w1 ¤ w2 . Defining .z/ D zz w w2

1
and
1
setting g.z/ D f .z/, we see that g falls into Case 3. As in Case 2, we can now
understand the dynamics of f as a change of coordinates of the dynamics of g by noting
that f n .z/ D 1 n
g .z/. We have one of:
(a) f n .z/ ! w1 for all z 2 C n fw2 g and J.f / D fw2 g.
(b) f n .z/ ! w2 for all z 2 C n fw1 g and J.f / D fw1 g.
(c) f is conjugate to a rotation z 7! e i z for some 2 .0; 2 /, and thus F .f / D C.
Remark 1.216. In Cases 2 and 4, the Möbius map .z/ was used to move the fixed points
of f to more convenient locations so that the simple dynamics of Cases 1 and 3 could be
related to the dynamics of f . This is an advantage to using global conjugation: it allows
us to reposition special points in more convenient places before we do our analysis. The
technique can be used to simplify the analysis of Möbius map dynamics and we can use it
with higher degree maps as well. See Section 1.2.7 where a conjugation of the Newton map
of a quadratic function simplifies the analysis. Additional Exercises 1.167–1.170 explore
global conjugation further.
1.C Appendix: Code for Drawing

Random Dynamics Pictures
Because of its complexity, the following algorithm is not well suited for an applet. How-
ever, we provide Matlab code that can be used to generate pictures like the Devil’s Colos-
seum in Figure 1.31. The four files main.m, f.m, g.m, and proced2.m need to be created
separately, but stored in the same folder. When main.m is compiled using Matlab, a picture
will be generated.
% ********* This begins the file main.m *********************

% This file requires the files (functions) g.m, f.m, and
% proced2.m to be in the same folder when this file is
% compiled. Upon compilation this file will generate a graph
% of the probability that a random orbit generated by the
% maps f and g will escape to infinity.
h = 0.1; % Determines step size in mesh

% of points to be plotted
Maxd = 10; % Determines how many random steps can be taken
cntx = 0; % Counter for x coordinate

clear x1;
for x = -5:h:5
cntx = cntx + 1;
cnty = 0; % Counter for y coordinate
clear y1;
for y = -5:h:5
cnty = cnty + 1;
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 82 — #100

i i
n = 0;
x1(cntx) = x;
y1(cnty) = y;
[z(cntx,cnty), n] = proced2(x, y, 1, n, Maxd);
end
end
% % Uncomment the following lines to export/save data sets

% % containing the x(i) coordinates, y(j) coordinates, and
% % z(i,j) values which when plotted make up the graph.
% % These data sets can then be imported into another
% % application (such as Maple or Mathematica) and plotted
% % and otherwise manipulated there.
%
% save DevilDataX3.dat x1 -ASCII;
% save DevilDataY3.dat y1 -ASCII;
% save DevilDataZ3.dat z -ASCII;
figure
surfc(x1, y1, z)
lighting phong
shading interp %flat %interp
clear;
% ********* This ends the file main.m *********************
% ********* This begins the file f.m *********************

function z = f(x,y)
z1 = [x*x-y*y-1,2*x*y];
z = [z1(1)*z1(1)-z1(2)*z1(2)-1,2*z1(1)*z1(2)];
% ********* This ends the file f.m *********************
% ********* This begins the file g.m *********************

function z = g(x,y)
z1 = [(x*x-y*y)/4,x*y/2];
z = [(z1(1)*z1(1)-z1(2)*z1(2))/4,z1(1)*z1(2)/2];
% ********* This ends the file g.m *********************
% ********* This begins the file proced2.m *****************

% This recursively defined procedure/function will compute
% the value s which for proced2(x, y, 1, n, Maxd) (here I
% used q=1) represents the probability that the point (x,y)
% will have a random orbit escape (have modulus > K) in the
% first M steps of the orbit.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 83 — #101

i i
1.C. Appendix: Code for Drawing Random Dynamics Pictures 83
function [s, n] = proced2(x, y, q, n0, M)

n = n0 + 1;
p = 0.5; % Probability that map f is chosen at each step
K = 10; % Escape radius for random orbit
s = 0.0;
z=f(x,y); % file f.m must be in folder next to this file
w=g(x,y); % file g.m must be in folder next to this file
if n < M % M is the max number of recursion steps allowed

if z(1)*z(1)+z(2)*z(2) > K*K
s = s + p*q;
else
[d, n] = proced2(z(1), z(2), p*q, n, M);
s = s + d;
end
if w(1)*w(1)+w(2)*w(2) > K*K

s = s + (1-p)*q;
else
[d, n] = proced2(w(1), w(2), (1-p)*q, n, M);
s = s + d;
end
end
% ********* This ends the file proced2.m *******************
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 84 — #102

i i
i i
i i
i i
“ECA˙Book” — 2012/9/5 — 14:07 — page 85 — #103

i i
Soap Films, Differential Geometry,

2
and Minimal Surfaces
Michael J. Dorff (text),
2.1 Introduction
Minimal surfaces are beautiful geometric objects with interesting properties that can be
studied with the help of computers. Some standard examples of minimal surfaces in R3
are the plane, Enneper’s surface, the catenoid, the helicoid, and Scherk’s doubly periodic
surface (see Figure 2.1; the images shown are part of the surfaces, which continue on).
Enneper’s surface catenoid
helicoid Scherk’s doubly periodic surface

F IGURE 2.1. Examples of some minimal surfaces.
85
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 86 — #104

i i
86 Chapter 2. Soap Films, Differential Geometry, and Minimal Surfaces
Minimal surfaces are related to soap films that result when a wire frame is dipped in soap
solution. To get a sense of this connection, consider the following problem.
Steiner Problem: Four houses are located so that they form the vertices of a square
that has sides of length one mile. Neighbors want to connect their houses with a road of
least length. What should the shape of the road be?
F IGURE 2.2. What is the shortest path connecting the 4 vertices?
Possible solutions include:
Length = 3 Length » 3.41 Length » 2.83

F IGURE 2.3.
p
However, none is the solution. The correct solution in Figure 2.4 has length of 1 C 3
2:7 miles. For more information about Steiner problems see [5] or [13].
F IGURE 2.4. The shortest path connecting these 4 vertices.
How can we generalize this problem? One way is to have n vertices. So the problem
becomes, given n cities, find a connected system of straight line segments of shortest length
such that any pair of cities is connected by a path of line segments.
C C
B B
D D
A E A E
An n vertices Steiner problem. The solution.
F IGURE 2.5.
Another way to generalize is to move up a dimension. What is the analogue of the Steiner
problem in one dimension higher? The Steiner problem minimizes distance (a 1-dimensional
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 87 — #105

i i
2.1. Introduction 87
object) in a plane (a 2-dimensional object). Soap film minimizes area (a 2-dimensional ob-
ject) in space (a 3-dimensional object).
Is there a connection between the two- and three-dimensional optimization problems?
Consider the soap film created by dipping a cube frame into soap solution shown in Figure
2.6. The soap film creates the minimum surface area for a surface with a cube as its bound-
ary. If we projected it onto the plane, the resulting shape would look like the solution to the
Steiner problem with four vertices shown in Figure 2.4.
F IGURE 2.6. Soap film formed by a cube.
Water molecules exert a force on each other. Near the surface of the water there is a greater
force pulling the molecules toward the center of the water, creating surface tension that
tends to minimize the surface area of the shape. Soap solution has a lower surface tension
than water and this permits the formation of soap films that also tend to minimize geometric
properties such as length and area. For more information see [22].
Minimal surfaces can be created by dipping wire frames into soap solution.
Example 2.1. By dipping into soap solution a wire frame of a slinky (or helix) with a straw
connecting the ends of the slinky, we can create part of the minimal surface known as the
helicoid.
Example 2.2. By dipping a 3-dimensional version of the wire frame in Figure 2.8 (a box
frame missing two parallel edges on the top and two parallel edges on the bottom) into
F IGURE 2.7. The wire frame of a slinky can be used to create part of the helicoid.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 88 — #106

i i
F IGURE 2.8. The wire frame of a box missing four edges can be used to create part of Scherk’s
doubly periodic surface.
soap solution we can create part of the minimal surface known as Scherk’s doubly periodic
surface.
Exploration 2.3. The minimal surfaces shown in Figure 2.1 can be formed by dipping a
wire frame in soap solution. Determine the shape of the wire frame that creates (a) En-
neper’s surface and (b) the catenoid. Try it out!
Remark 2.4. To get a soap film of the part of Enneper’s surface shown in Figure 2.1, we
can dip a wire frame that matches the seams along a baseball. Dipping it in soap solution
produces two minimal surfaces. The first is half of the sphere of a baseball and the other is
the complementary half. If you start with one, you can deform it into the other by carefully
blowing air into the soap film. There is a third, mysterious and unseen minimal surface
one passes through while doing this, one that is unstable. It cannot remain in existence–
disturbances cause it to deform into another surface.
One active area of minimal surface theory is the study of complete embedded mini-
mal surfaces. These are minimal surfaces that are boundaryless (complete) and have no
self-intersections (embedded). The plane, the catenoid, the helicoid, and Scherk’s doubly
periodic surface are examples. Enneper’s surface is not embedded, because it has self-
intersections as its domain increases (see Exploration 2.11).
To begin to understand minimal surfaces, we need some tools from differential ge-
ometry discussed in Section 2. Section 3 defines a minimal surface and discusses some
examples and properties. Section 4 uses complex analysis to study minimal surfaces and
introduces the Weierstrass representation formula to describe and study their properties
efficiently. These sections are fundamental and should be read first. In Sections 5–6, we
explore ideas that lead to beginning research problems. Sections 5 and 6 are indepen-
dent. In Section 5 we present the Weierstrass representation in the form of the Gauss
map and height differential, which is the basis for much current research about minimal
surfaces in R3 . Section 6 connects ideas about minimal surfaces with planar harmonic
mappings in geometric function theory (i.e., the study of complex analysis from a geo-
metric viewpoint). In this chapter, there are three applets that can be accessed online at
www.maa.org/ebooks/EXCA/applets.html
1. DiffGeomTool is used to visualize and explore basic differential geometry concepts
in R3 such as the graph of a parametrization of a surface, curves on a surface, tangent
planes on a surface, and unit normals on a surface.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 89 — #107

i i
2.2. Differential geometry 89
2. MinSurfTool is used to visualize and explore minimal surfaces in R3 by using various

forms of the Weierstrass representation.
3. ComplexTool is used to plot the image of domains in C under complex-valued func-
tions.
Each section contains examples, exercises, and explorations that involve the applets. You
should do all the exercises and explorations, many of which present surfaces and concepts
that will be used later in the chapter (there are additional exercises at its end). There are
also short and long projects that are suitable as research problems. The goal of the chapter
is not to give a comprehensive treatment of the topic, but to engage the reader with the
general notions, questions, and techniques of the area, and to encourage the reader to pose
and pursue the reader’s own questions. To understand the nature and purpose of this text,
the reader should read the Introduction.
2.2 Differential geometry

Our goal is to develop the mathematics necessary to investigate minimal surfaces in R3 .
They minimize area locally and can be thought of as saddle surfaces. At each point, the
bending upward in one direction is matched with the bending downward in the orthogonal
direction. The bending is known mathematically as curvature. So, to understand and inves-
tigate minimal surfaces, we should first understand curvature, a quantity that is measured
using the tools of differential geometry. Differential geometry is a field of mathematics in
which calculus is applied to geometric shapes.
We begin our discussion of differential geometry by looking at a surface in R3. Every
point on a surface M R3 can be designated by a point, .x; y; z/ 2 R3 . It can also be
represented by two parameters. Let D be an open set in R2 . Then M can be represented by
a function x W D ! R3 , where x.u; v/ D .x1 .u; v/; x2.u; v/; x3 .u; v//, so M is the image
of x.D/. We will require that x be differentiable, so each coordinate function xk .u; v/ has
continuous partial derivatives of all orders in D. Such a function or mapping is called a
parametrization.
v z
u y
D Ì R2
x M Ì R3
F IGURE 2.9. The parameterization of a surface.
Let’s consider two examples.

Example 2.5. Enneper’s surface is a minimal surface formed by bending a disk into a
saddle surface. It can be parametrized by

1 3 1 3
x.u; v/ D u u C uv 2 ; v v C u2 v; u2 v 2 ;
3 3
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 90 — #108

i i
F IGURE 2.10. Enneper’s surface.
where u; v are in a disk of radius r . We can use the applet, DiffGeomTool, to graph the
parametrization of Enneper’s surface (Figure 2.10). Open DiffGeomTool and enter the co-
ordinate functions of the parametrization as
X.u; v/ D u 1=3 u ^ 3 C u v ^ 2
Y .u; v/ D v 1=3 v ^ 3 C u ^ 2 v
Z.u; v/ D u ^ 2 v^2
into the appropriate boxes. In the gray part on the right side of the applet, click on Circu-
lar grid with radius min: 0.0, radius max: 1.0, theta min: 0.0, and theta max: 2 (in
the applet enter pi for ). This is because we want our u; v values to be the unit disk. Then
click the Graph button. To rotate the graph, place the cursor on the image of the surface,
and then click on and hold the left button on the mouse as you move the cursor. To increase
the size of the image of the surface click on the left button on the mouse; to decrease it,
click on the right mouse button.
Example 2.6. If a heavy flexible cable is suspended between two points at the same height,
then it takes the shape of a curve whose equation is ˛.t/ D a cosh.t=a/ and is called a
catenary from the Latin word that means “chain”. In calculus, we discuss rotating a curve
about a line to get a surface of revolution. We can apply this idea to the catenary to get a
–y
F IGURE 2.11. Creating a catenoid by rotating a catenary.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 91 — #109

i i
surface known as the catenoid. A catenoid is a surface that can be generated by rotating a
catenary on its side about the z-axis (see Figure 2.11). A catenoid is also a minimal surface
as we will verify in Section 2.3. How do we parametrize this catenoid? If we let x1 D
a cosh v . 1 < v < 1/ and x3 D av, then r .v/ D .a cosh v; av/ is a parametrization
of the catenary curve on its side in the xz-plane. Rotating a line about an axis is a circular
motion, and a circle can be parametrized by .cos u; sin u/. So, we can parametrize the
rotation of the catenary curve about the z-axis by multiplying a cosh v by cos u for the
x1 -coordinate function, and multiplying a cosh v by sin u for the x2 -coordinate function.
Hence, we get a parametrization for the catenoid

x.u; v/ D a cosh v cos u; a cosh v sin u; av :
Using DiffGeomTool, we can graph it with a D 1 by clicking on Rectangular grid,

and setting the boxes to 0 u 2 and 2 3
v 2 3
(see Figure 2.12). Enter
F IGURE 2.12. The catenoid.
cosh v; cos u, and sin u as cosh.v/; cos.u/, and sin.u/. Check what happens if you change
the u; v values. For example, try
2 2
(a) u 2 , 3
v 3
2
(b) 0 u 2 , 0v 3

(c) 0 u 2 , 4
v 4
(d) 0 u 2 , v .
Exercise 2.7. A torus is a surface (but not a minimal surface) formed by rotating a circle
in the xz-plane about the z-axis. If the circle has radius b and is centered at a distance of a
from the origin, then its parametrization is

x.u; v/ D .a C b cos v/ cos u; .a C b cos v/ sin u; b sin v ;
where a; b are fixed, 0 < u < 2 , and 0 < v < 2 .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 92 — #110

i i
b x
a
y
F IGURE 2.13. Creating a torus by rotating a circle.
(a) Show how to derive the parametrization.

(b) Use DiffGeomTool to sketch the graph of a torus with a D 3 and b D 2. Use
Rectangular grid with 0 < u < 2 , and 0 < v < 2 .
Note: It will be helpful to keep a well organized notebook with answers to the exercises,
because the same calculations will be used several times. Try it out!
In pre-calculus, we say that a function of one variable y D F .x/ satisfies the vertical
line test. The graph of F .x/ is a 1-dimensional object in R2 , created by plotting points
.u; F .u// in the plane. Thus, it is parametrized by the map u ! .u; F .u// from R to R2 .
Analogously, we speak of a function of two variables z D f .x; y/, where .x; y/ lies in a
two-dimensional domain and f satisfies the vertical line test (here a line is vertical when
it is parallel to the z-axis). The graph of f .x; y/ is a two-dimensional surface in R3 with
a height of z D f .x; y/ at .x; y/. An example is the minimal surface known as Scherk’s
doubly periodic surface which can be parametrized by
cos u
x.u; v/ D u; v; ln :
cos v
Exercise 2.8.
(a) In the parametrization of Scherk’s doubly periodic surface, what are the restrictions on
the u and v values in the domain?
(b) Use DiffGeomTool and your answer from part (a) to sketch a graph of Scherk’s doubly
periodic surface with 0:48 u; v 0:48 .
(c) Scherk’s doubly periodic surface is an example of the graph of a function. For any
function f .x; y/, find a parametrization of its graph.
Try it out!
Exercise 2.9. Let r be a differentiable curve whose derivative does not vanish (i.e., r 0 .v/ ¤
0 for all v in the domain) and let r lie in a plane in R3. A surface of revolution is a surface
that is formed by rotating r about an axis in the plane such that the curve does not intersect
the axis. The catenoid and torus are examples. Let r .v/ D .f .v/; 0; g.v// be such a curve
in the xz-plane.
(a) Find a parametrization for the surface of revolution generated by rotating the curve
about the z-axis.
(b) Check that your answer in (a) matches the parametrizations of the catenoid and the
torus.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 93 — #111

i i
Try it out!
Exploration 2.10. For the torus Ta;b whose parametrization is given in Exercise 2.7, use
DiffGeomTool to plot T3;2 . Describe what happens to the shape of Ta;b as a gets smaller
and b gets larger. Explain this in terms of how we derived the parametrization of the torus.
Hint: In DiffGeomTool, plot the tori
T2:7;2 T2:4;2 T2;2 T3;2:4 T3;2:7 T3;3 :
What happens when a < b? Try it out!

Exploration 2.11. As mentioned earlier, Enneper’s surface is not embedded, because it
has self-intersections. Use DiffGeomTool and the parametrization in Example 2.5 to graph
Enneper’s surface with the domain being a disk of radius 1.
(a) What happens to Enneper’s surface as the radius r of the disk increases?
(b) Estimate the largest value of r for which Enneper’s surface has no self-intersections.
(c) Assuming that the intersection occurs on the z-axis, prove your result from part (b).
Try it out!
So far we have discussed how a function (i.e., a parametrization) models a surface. Our
goal is to determine the bending or curvature of curves on a surface. To do this, we will use
the parametrization of a surface to discuss the tangent plane and normal vector at a point on
the surface. Suppose x.u; v/ is a parametrization of a surface M R3. If we fix v D v0
and let u vary, then x.u; v0 / depends on one parameter and is known as a u-parameter
curve. Likewise, we can fix u D u0 and let v vary to get a v-parameter curve x.u0 ; v/ (see
Figure 2.14). Tangent vectors for the u-parameter and v-parameter curves are computed
by differentiating the component functions of x with respect to u and v, respectively. That
is, xu and xv are the tangent vectors defined by

@x1 @x2 @x3 @x1 @x2 @x3
xu D ; ; ; xv D ; ; :
@u @u @u @v @v @v
For each point p D x.u0 ; v0/ on the surface, we get two vectors by substituting u0 and
v0 in xu and xv . When we have a parametrization of a surface, we will require that xu and
xv be linearly independent (i.e., not be constant multiples of each other). The span of xu
and xv (i.e., the set of all vectors that can be written as a linear combination of xu , xv ) then
forms a plane called the tangent plane.
v xv
(u, v0) x (u, v) xu
u M Ì R3
v-param.
(u0, v) curve u-param.
D Ì R2 curve
F IGURE 2.14. The u-parameter and v-parameter curves.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 94 — #112

i i
Definition 2.12. The tangent plane of a surface M at a point p is

ˇ
Tp M D fvˇv is tangent to M at pg:
Definition 2.13. The unit normal to a surface M at a point p D x.a; b/ is

ˇ
xu xv ˇˇ
n.a; b/ D ˇ ˇ :
ˇxu xv ˇ ˇˇ
.a;b/
n
xv
Tp M
p
M Ì R3
xu
F IGURE 2.15. A tangent plane, Tp M , and unit normal vector, n.
Not every surface has a well-defined choice of a unit normal n. Those that do not are
called non-orientable. An example is the Möbius strip. An orientable strip can be modeled
by taking a strip of paper and taping the ends together. In this model, the strip has an
outside and an inside–you cannot get from one side to the other without crossing an edge.
A Möbius strip is modeled by taking the same strip of paper and twisting one end before
taping. This forms a strip in which you can get from one side to the other without crossing
over an edge. The relationship between a Möbius strip and its unit normal n is further
explored in Exercise 2.137 in the Additional Exercises at the end of this chapter. The unit
normal, n, is orthogonal to the tangent plane at p (see Figure 2.15). If the surface M is
oriented, then there are two unit normals at p 2 M , one pointing outward and one inward.
The definition of n chooses one of them.
Example 2.14. A torus is parametrized by

x.u; v/ D .3 C 2 cos v/ cos u; .3 C 2 cos v/ sin u; 2 sin v
where 0 < u; v < 2 . For v0 D 3 , the u-parameter curve is

p
x u; D .4 cos u; 4 sin u; 3/:
3

For u0 D 2
, the v-parameter curve is

x ; v D .0; 3 C 2 cos v; 2 sin v/:
2
We have
xu .u; v/ D . .3 C 2 cos v/ sin u; .3 C 2 cos v/ cos u; 0/
xv .u; v/ D . 2 sin v cos u; 2 sin v sin u; 2 cos v/:
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 95 — #113

i i

The u-parameter curve, x u; 3 and the v-parameter curve, x 2 ; v intersect on the torus

at p D x 2 ; 3 . The tangent vectors to the u- and v-parameter curves at p are

xu ; D . 4; 0; 0/
2 3

p
xv ; D .0; 3; 1/:
2 3
They span the tangent plane, Tp M . We compute that

p p
xu ; xv ; D . 4; 0; 0/ .0; 3; 1/ D .0; 4; 4 3/:
2 3 2 3
Hence, p
1 3
n ; D 0; ; :
2 3 2 2
We can use DiffGeomTool to display the u-parameter curve, the v-parameter curve, xu ,
xv , and n. Enter the parametrization in this example for the torus. Then click Curves.
A Point location box, a fixed u box, and a fixed v box will appear. In the
Point location box, enter 2 into the first box (i.e., the fixed u value) and 3 into the
second box (i.e., the fixed v value). Make sure that the fixed u box is clicked but not
the fixed v box. Then click the Graph button. The v-parameter curve will appear. If
you now click the track fixed u curve box, a slider will appear. Moving it with
the cursor will move the point along the v-parameter curve on the torus. Now, click on the
fixed u box and then click the Graph button again. The u-parameter curve will appear.
By clicking on the track fixed v curve box and moving the slider, the point along
the v-parameter curve will move. Next, click on the following boxes separately followed
by the Graph button: Tangent vectors box, Tangent plane box, and Normal
vector box. This will cause these geometric objects to appear. You should convince
F IGURE 2.16. The torus with u; v-parameter curves, the tangent vectors, the tangent plane, and the
normal vector.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 96 — #114

i i
yourself that the images of the vectors at .u; v/ D . 2 ; 3 / match the computed values in
Example 2.14.
Exercise 2.15. For a surface of revolution (see Exercise 2.9) parametrized by

x.u; v/ D f .v/ cos u; f .v/ sin u; g.v/
the u-parameter curves are called parallels and are the curves formed by horizontal slices,
while the v-parameter curves are called meridians and are the curves formed by vertical
slices. Describe the parallels and meridians for the catenoid in Example 2.6 and the torus
in Exercise 2.7. Try it out!
Exercise 2.16. A catenoid has parametrization

x.u; v/ D cosh v cos u; cosh v sin u; v
2 2
with 0 < u < 2 and 3
<v < 3
.
(a) Use DiffGeomTool to sketch the u-parameter curve, x.u; 0/, and the v-parameter curve,
x.0; v/, on the catenoid. Also, sketch the vectors xu .0; 0/, xv .0; 0/, and n.0; 0/.
(b) Compute the vectors xu .0; 0/, xv .0; 0/, and n.0; 0/.
Try it out!
Exercise 2.17. Scherk’s doubly periodic surface has parametrization
cos u
x.u; v/ D u; v; ln
cos v
with 0:48 < u; v < 0:48 .

(a) Use DiffGeomTool to sketch the u-parameter curve, x u; 4 and the v-parameter curve,

x 4 ; v on Scherk’s doubly periodic surface (make sure you use 0:48 < u; v <

0:48 ). Sketch the vectors xu 4 ; 4 , xv 4 ; 4 , and n 4 ; 4 . You can slide them by
clicking the track fixed u curve box. The collection of vectors, .xu , xv , n/, is
known as a moving frame or Frenet frame of the curve. The way they vary in R3 as the
frame moves along the curve describes how the curve twists and turns in R3. For more
details, see [18] or [21].

(b) Compute the vectors xu 4 ; 4 , xv 4 ; 4 , and n 4 ; 4 .
Try it out!
We will use the normal vector n to define the curvature of a curve on a surface. As we
mentioned in the introduction, at each point on a minimal surface, the bending upward in
one direction is matched with the bending downward in the orthogonal direction. We will
use this definition of curvature in our definition of a minimal surface. A plane containing
the normal n will intersect the surface M in a curve, ˛. For each ˛ we can compute its
curvature, which measures how fast the curve pulls away from its tangent line at p. A curve
in R3 can be parametrized by a function of one variable, say ˛.t/, where ˛ W Œa; b ! R3 .
The parametrization is not unique.
Exercise 2.18. Find two parametrizations of the unit circle in the xy-plane. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 97 — #115

i i
The lack of uniqueness can cause difficulties. To eliminate them we will require our pa-
rametrization to be a unit speed curve.
ˇ ˇ
Definition 2.19. A curve ˛ is a unit speed curve if ˇ˛ 0 .t/ˇ D 1.
If a parametrization of a regular curve ˛.t/ is not of unit speed, we can reparametrize it by
arclength to get a unit speed curve ˛.s/ (you probably saw this in third-semester calculus,
and you may want to review how to reparametrize a curve by arclength). We will assume
that curves are unit speed curves ˛.s/.
Given a curve ˛, we want to discuss its curvature (or bending). We quantify the amount
that the curve bends at each point p by measuring how rapidly it pulls away from its tangent
line at p. The amount of bending is the rate of change of the angle that tangents make
with the curve at p which is the second derivative.
ˇ ˇ
Definition 2.20. The curvature of a unit speed curve ˛ at s is ˇ˛ 00 .s/ˇ.
Example 2.21. A torus is parametrized by

x.u; v/ D .3 C 2 cos v/ cos u; .3 C 2 cos v/ sin u; 2 sin v ;
where 0 < u; v < 2 . Let’s compute the curvature for the u-parameter curves and the v-
parameter curves. The v-parameter curves (or meridians) are the curves formed by vertical
slices of the torus, and hence are circles of radius b D 2. To compute their curvature, we
take their parametrization

x.v/ D x.u0 ; v/ D .3 C 2 cos v/ cos u0 ; .3 C 2 cos v/ sin u0 ; 2 sin v ;
where u0 is a fixed value. Since jx0 .v/j D 2, we reparametrize x.v/ as a unit speed curve
by replacing v with 2s to get
s s s
x.s/ D 3 C 2 cos cos u0 ; 3 C 2 cos sin u0 ; 2 sin :
2 2 2
We then compute the curvature of the v-parameter curves to be
ˇ 00 ˇ
ˇx .s/ˇ D 1 :
2
The u-parameter curves (or parallels) are the curves formed by horizontal slices of the
torus, and so are circles of radius 3 C 2 cos v0 , where v0 2 .0; 2 / is fixed. The radii vary
between 1 and 5. The curves are parametrized by

x.u/ D x.u; v0 / D .3 C 2 cos v0 / cos u; .3 C 2 cos v0 / sin u; 2 sin v0 ;
which can be reparametrized as the unit speed curve

s s
x.s/ D .3 C 2 cos v0 / cos ; .3 C 2 cos v0 / sin ; 2 sin v0 :
3 C 2 cos v0 3 C 2 cos v0
Computing the curvature yields
ˇ 00 ˇ 1
ˇx .s/ˇ D ;
3 C 2 cos v0
1
which varies between 5
and 1.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 98 — #116

i i
curvature = 1
1
curvature = 1
5 curvature =
2
F IGURE 2.17. The curvature of the meridians and parallels on a torus.
s
Exercise 2.22. Verify that replacing v with 2 in x.u; v/ in the previous example gives a
unit speed curve. Try it out!
Exercise 2.23. Compute the curvatures of the meridians and parallels of the catenoid pa-
rametrized by
Try it out!
Exercise 2.24. The curve parametrized by ˛.t/ D .a cos t; a sin t; bt/ is a helix, which is
a spiral that rises with a pitch of 2 b on the cylinder x 2 C y 2 D a2 .
F IGURE 2.18. A helix in a cylinder.
We can create a surface by connecting a line from the axis .0; 0; bt/ through the helix
.a cos t; a sin t; bt/ at height bt. This ruled surface is a minimal surface known as a he-
licoid. All minimal surfaces, including the helicoid, can be parametrized in several ways.
We will use the parametrization of the helicoid

x.u; v/ D a sinh v cos u; a sinh v sin u; au :
(a) Compute the curvatures of the u-parameter curves and v-parameter curves of the
helicoid (making the v-parameter curve into a unit speed curve is not easy).
(b) Use DiffGeomTool to graph the helicoid with a D 1 (see Figure 2.19).
Try it out!
Exercise 2.25. From Example 2.21 you may have conjectured that the curvature of a circle
1
of radius r is . This is correct. Prove it. Try it out!
r
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 99 — #117

i i
F IGURE 2.19. Helicoid.
Let’s return to surfaces. Suppose we have a curve .s/ on a surface M . We can deter-
mine the unit tangent vector, w, of at p 2 M and the unit normal, n, of M at p 2 M .
Note that w and n span a plane P that intersects M creating a curve ˛.s/.
n plane P
s w
M Ì R3
a(s)
F IGURE 2.20. The normal curvature.
Definition 2.26. The normal curvature in the w direction is

k.w/ D ˛ 00 n:
ˇ ˇˇ ˇ
We know ˛ 00 n D ˇ˛ 00 ˇˇnˇ cos , where is the angle between n and ˛ 00 . Hence ˛ 00 n is the
projection of ˛ 00 onto the unit normal (whence, the name normal curvature). The normal
curvature measures how much the surface bends towards n as you travel in the direction of
the tangent vector w starting p. As we rotate the plane about n, we get a set of curves on
the surface each of which has a value for its curvature. Let k1 and k2 be the maximum and
minimum curvature values at p. The directions in which the normal curvature attains its
absolute maximum and absolute minimum values are known as the principal directions.
Definition 2.27. The mean curvature (i.e., average curvature) of a surface M at p is
k1 C k2
H D :
2
It turns out that k1 and k2 come from orthogonal tangent vectors. The mean curvature
depends on p. It can be shown that H does not change if we choose any two orthogonal
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 100 — #118

i i
vectors and use their curvature values to compute H at p. Note that from the definition we
see that if the principal curve with normal curvature kj is bending toward the unit normal
n, then kj > 0 and if it is bending away from n, then kj < 0.
Example 2.28. At a point on a sphere of radius a, all the curves ˛ are circles of radius
a and hence have the same curvature, 1=a. Since the curves are bending away from n,
k1 D 1=a D k2 , and the mean curvature is 1=a.
Exercise 2.29. Determine the mean curvature at points on the cylinder parametrized by
x.u; v/ D .a cos u; a sin u; bv/. Try it out!
Exercise 2.30. Determine the points on the torus

x.u; v/ D .a C b cos v/ cos u; .a C b cos v/ sin u; b sin v
where H > 0, H D 0, or H < 0. Try it out!

In the next section we will define a minimal surface in terms of mean curvature. Using
our current definition it is tedious to compute the mean curvature of a surface, because we
have to find the principle curvatures for every point on the surface. We would like to have
an explicit expression to compute mean curvature at a point p on the surface simply by
substituting the coordinates of p into the formula. There is a formula for mean curvature
using the coefficients of the first and second fundamental forms for a surface. Since ˛.s/ D
.x.u.s/; v.s/// the chain rule gives us that ˛ 0 .s/ D xu du C xv dv . From here on we will
ds ds
denote du by du and dv by dv. Because ˛ is a unit speed curve,
ds ds
1 Dj˛ 0 j2 D ˛ 0 ˛ 0
D.xu du C xv dv/ .xu du C xv dv/
Dxu xu du2 C 2xu xv dudv C xv xv dv 2
DE du2 C 2F dudv C G dv 2 : (2.1)
The terms E D xu xu, F D xu xv , and G D xv xv are known as the coefficients of the

first fundamental form. They describe how lengths on a surface are distorted as compared
to their measurements in R3.
Next, recall k.w/ D ˛ 00 n. Note that ˛ 0 n D 0, and so .˛ 0 n/0 D 0, which implies
˛ n C ˛ 0 n0 D 0, and thus ˛ 00 n D ˛ 0 n0 . Similarly, xu nu D xuu n. So
00
k.w/ D ˛ 0 n0
D .xu du C xv dv/ .nu du C nv dv/
D xu nu du2 .xu nv C xv nu / dudv xv nv dv 2
D xuu n du2 C 2xuv n dudv C xvv n dv 2
D e du2 C 2f dudv C g dv 2 :
The terms e D xuu n, f D xuv n, and g D xvv n are called the coefficients of
the second fundamental form. They describe how much the surface bends away from the
tangent plane.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 101 — #119

i i
Example 2.31. A catenoid can be parametrized by

Using this parametrization, we compute that
xu D . a cosh v sin u; a cosh v cos u; 0/

xv D .a sinh v cos u; a sinh v sin u; a/:
The coefficients of the first fundamental form are
E Dxu xu D a2 cosh2 vI
F Dxu xv D 0I
G Dxv xv D a2 cosh2 v:
What do E, F , and G tell us? Let .u0 ; v0 / 2 D be a point in the domain and let’s take a
small square with a vertex there. Because xu xv D F D 0, we know that the orthogonal
lines from the u-parameter curve and the v-parameter curve remain orthogonal on the
catenoid. That is, small squares will be mapped to small rectangles (technically, they are
not rectangles but the term serves our purpose). Because E D G we know that jxuj D jxv j
and adjacent sides of the image rectangle will have the same length. So small squares in
the domain D will be mapped to small squares on the catenoid. For simplicity take a D 1.
Then E D G D cosh2 v. When v D 0, E D G D 1 and as v gets farther away from
0, E and G get larger. This means that a small square containing the u-parameter curve
v D 0 will be mapped to a small square of the same size on the catenoid. As v gets farther
away from 0, the size of the side lengths of the image square will increase by a factor
of cosh2 v. This can be seen in Figure 2.21 where we have used the Transparency
of surface and the Transparency of the frame sliders to make the u- and
v-parameter curves more distinct (the u-parameter curve with v D 0 gets mapped to a
F IGURE 2.21. The catenoid.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 102 — #120

i i
parallel on the neck of the catenoid and the u-parameter curve with v D 2
3
gets mapped
to the edge of the catenoid, as displayed in the figure).
To compute the coefficients of the second fundamental form, we need to compute n D
xu xv
. Because
jxu xv j
xu xv D .a2 cosh v cos u; a2 cosh v sin u; a2 cosh v sinh v/;
we have
jxu xv j D a2 cosh2 v:
Hence
cos u sin u sinh v
nD ; ; :
cosh v cosh v cosh v
We can compute that
xuu D. a cosh v cos u; a cosh v sin u; 0/
xuv D. a sinh v sin u; a sinh v cos u; 0/
xvv D.a cosh v cos u; a cosh v sin u; 0/:
Therefore, the coefficients of the second fundamental form are:
e Dn xuu D a
f Dn xuv D 0
g Dn xvv D a:
What do e, f , and g tell us? Let .u0 ; v0 / 2 D be a point in the domain, and let p 2 M
be the image of .u0 ; v0 / on the surface. Then at p the vectors xu and xv create the tangent
plane Tp M and the unit normal n. For the catenoid, the u-parameter curve is bending away
from n while the v-parameter curve is bending toward n. Both curves are bending the same
amount away from the tangent plane, because e D g.
Exercise 2.32. A torus has the parametrization

x.u; v/ D .a C b cos v/ cos u; .a C b cos v/ sin u; b sin v :
(a) Compute the coefficients of the first and the second fundamental forms.
(b) Open DiffGeomTool and enter the parametrization with a D 2 and b D 1. Use the
Rectangular grid with 0 u 2 and 0 v 2 . And set the number of
u points to 20 and the number of v points to 20. Describe how the results from part
(a) match with the image of the torus in DiffGeomTool. Use the Transparency of
surface and the Transparency of the frame sliders and click off and on
the Wireframe button to make the wireframe clearer.
Try it out!
Exercise 2.33. Compute the coefficients of the first and second fundamental forms for
Scherk’s doubly periodic surface parametrized by

cos u
x.u; v/ D u; v; ln :
cos v
Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 103 — #121

i i
We want to express the mean curvature H in terms of the coefficients of the first and
second fundamental forms. We will show that
Eg C Ge 2Ff
H D :
2.EG F 2 /
We will establish the formula by a straightforward calculation from Oprea [22, pp. 40–
42]. This approach does not give much insight into the formula. There is a more elegant
way to derive this formula, but it requires some concepts that are beyond the scope of what
we will need (for a discussion involving this approach, see [3] or [18]).
Let w1 ; w2 be two perpendicular unit vectors in Tp M and k1 ; k2 be their normal
curvatures using the curves ˛1.s/ D .u1 .s/; v1 .s// and ˛2.s/ D .u2 .s/; v2 .s//. Let
p1 D du1 C idu2 and p2 D dv1 C idv2 . Then, using the second fundamental form
to compute k1 and k2 , we have
2H D k1 C k2 D e.du21 C du22 / C 2f .du1 dv1 C du2 dv2 / C g.dv12 C dv22/
D e.p1 p1 / C f .p1 p2 C p1 p2 / C g.p2 p2 /:
We want to further simplify this so that p1 and p2 do not appear in the expression. From
(2.1)
1 D E du2 C 2F dudv C G dv 2 :
Also Edu1du2 C F .du1 dv2 C du2 dv1 / C Gdv1dv2 D 0, because w1 ; w2 are perpen-
dicular and so w1 D w2 ˛10 .s/ ˛20 .s/ D 0. Hence,

Ep12 C 2Fp1 p2 C Gp22 D E du21 du22 C i 2du1du2

C 2F du1 dv1 du2 dv2 C i.du1 dv2 C du2 dv1 /

C G dv12 dv22 C i 2dv1dv2

D 2i Edu1 du2 C F .du1 dv2 C du2 dv1 / C Gdv1dv2

C Edu21 C 2F du1dv1 C Gdv12

Edu22 C 2F du2dv2 C Gdv22
D 0C1 1
D 0:
Thus,
q
p
2Fp2 ˙ 4F 2 p22 4EGP22 F EG F 2
p1 D D ˙i p2
2E E E
p
F EG F 2
p1 D i p2 ;
E E
so

F2 EG F 2 G
p1 p1 D 2
C 2
p2 p2 D p2 p2 ; (2.2)
E E E
and
2F
p1 p2 C p1 p2 D p2 p2 : (2.3)
E
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 104 — #122

i i
We have
2H D k1 C k2 D e.p1 p1 / C f .p1 p2 C p1 p2 / C g.p2 p2 /

G 2F
D e Cf C g p2 p2 :
E E
To get rid of p2 p2 , use (2.1) to derive
Ep1 p1 CF .p1 p2 C p1 p2 / C Gp2 p2
D E.du21 C du22 / C 2F .du1 dv1 C du2 dv2 / C G.dv12 C dv22/
D 1 C 1 D 2:
Using (2.2) and (2.3), we have

G 2F
2DE p2 p2 C F p2 p2 C Gp2 p2
E E

2F 2
) 2 D 2G p2 p2
E
E
) p2 p2 D :
EG F 2
Therefore,

1 G 2F Eg C eG 2Ff
H D e Cf C g p2 p2 D :
2 E E 2.EG F 2 /
2.3 Minimal surfaces

Now that we have some essential ideas from differential geometry, we can begin to ex-
plore minimal surfaces. We mentioned that minimal surfaces can be thought of as saddle
surfaces. That is, at each point the bending upward in one direction is matched with the
bending downward in the orthogonal direction. This can be described mathematically with
the definition:
Definition 2.34. A minimal surface is a surface M with mean curvature H D 0 at all
points p 2 M .
We can use the formula
Eg C Ge 2Ff
H D : (2.4)
2.EG F 2 /
to show that a surface with a specific parametrization is minimal.
Example 2.35. We will use (2.4) to show that the catenoid is a minimal surface. A catenoid
can be parametrized by

From Example 2.31
E Dxu xu D a2 cosh2 v;
F Dxu xv D 0;
G Dxv xv D a2 cosh2 v;
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 105 — #123

i i
2.3. Minimal surfaces 105
and
e Dn xuu D a;
f Dn xuv D 0;
g Dn xvv D a:
Hence
eG 2f F C Eg
H D D 0:
2.EG F 2 /
So the catenoid is a minimal surface.
Exercise 2.36. Using the parametrization for the helicoid
x.u; v/ D .a sinh v cos u; a sinh v sin u; au/;
prove that it is a minimal surface. Try it out!

Exercise 2.37. Using the parametrization for the torus

x.u; v/ D .a C b cos v/ cos u; .a C b cos v/ sin u; b sin v ;
prove that it is not a minimal surface. Try it out!

Exercise 2.38. Suppose a surface M is the graph of a function f .x; y/ of two variables
(see the paragraph before Exercise 2.8). Then M can be parametrized by

x.x; y/ D x; y; f .x; y/ ;
where its domain is the projection of M onto the xy-plane.

(a) Compute the coefficients of the first and second fundamental forms for M .
(b) A minimal graph is a minimal surface that is a graph of a function. Prove M is a
minimal graph if and only if

fxx 1 C fy2 2fx fy fxy C fyy 1 C fx2 D 0: (2.5)
Try it out!
Before Exercise 2.8, we stated that Scherk’s doubly periodic surface is a minimal
graph. We will now use (2.5) to prove that. Applying (2.5) is usually not easy, because
solving explicitly for f can be complicated. However, we can do this is when f can
be separated into two functions, each of which depends on only one variable. Suppose
f .x; y/ D g.x/ C h.y/. Then the minimal surface equation becomes
g00 .x/Œ1 C .h0 .y//2 C h00 .y/Œ1 C .g0 .x//2 D 0;
a separable differential equation that can be solved by separating the terms with x from
those with y by putting them on opposite sides, giving
1 C .g0 .x//2 1 C .h0 .y//2

00
D : (2.6)
g .x/ h00 .y/
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 106 — #124

i i
If we fix y, the right side remains constant even if we change x in the left side. The same
is true if we fix x and vary y. The only way this can occur is if both sides are constant. So
we have:
1 C .g0 .x//2
Dk H) 1 C .g0 .x//2 D kg00 .x/:
g00 .x/
d
To solve, let .x/ D g0 .x/. Then dx D g00 .x/ and so
Z Z
d
dx D k ;
1 C 2
xD k arctan C C;

xCC
D tan :
k
For convenience, let C D 0 and k D 1. Since D g0 , we can integrate again to get
g.x/ D lnŒcos x:
Completing the same calculations for the y-side of (2.6) yields
h.y/ D lnŒcos y:
Hence

cos x
f .x; y/ D g.x/ C h.y/ D ln ;
cos y
an equation for Scherk’s doubly periodic surface. Using DiffGeomTool we display the
graph of Scherk’s doubly periodic surface (see Figure 2.22). Because 2 < x; y < 2 , the
surface is defined over a square with side lengths , centered at the origin. By the Schwarz
Reflection Principle, we can fit pieces of Scherk’s doubly periodic surface together hori-
zontally and vertically to get a checkerboard domain (see Figure 2.23). Because one piece
F IGURE 2.22. Scherk’s doubly periodic surface.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 107 — #125

i i
F IGURE 2.23. A tiling of Scherk’s doubly periodic surface.
of the surface can be repeated or tiled in two directions, we get a doubly periodic surface.
Periodic minimal surfaces is an active area of research.
Let’s summarize the minimal surfaces we have encountered and some other well-
known examples.
1. The plane:
x.u; v/ D .u; v; 0/.
2. Enneper’s surface:
1 3 1 3
x.u; v/ D u 3u C uv 2 ; v 3v C u2 v; u2 v2 .
3. The catenoid:
x.u; v/ D .a cosh v cos u; a cosh v sin u; av/.
4. The helicoid:
x.u; v/ D .a sinh v cos u; a sinh v sin u; au/.
5. Scherk’s doubly periodic surface:
cos u
x.u; v/ D u; v; ln .
cos v
6. Scherk’s singly periodic surface:
x.u; v/ D .arcsinh.u/; arcsinh.v/; arcsin.uv//.
7. Henneberg surface:
1
x.u; v/ D . 1 C cosh.2u/ cos.2v/; sinh.u/ sin.v/ 3
sinh.3u/ sin.3v/;
sinh.u/ cos.v/ C 13 sinh.3u/ cos.3v//.
8. Catalan surface:
x.u; v/ D .1 cos.u/ cosh.v/; 4 sin. u2 / sinh. v2 /; u sin.u/ cosh.v//.
In addition to Enneper’s surface, the Henneberg surface and the Catalan surface are not
embedded.
There is an extensive list of minimal surfaces, but we have no way of listing all of
them. So, we often focus on trying to classify minimal surfaces. This means, we try to
find results that include all possibilities for minimal surfaces with specific properties. The
simplest example of this is Theorem 2.39.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 108 — #126

i i
D M Ì R3
F IGURE 2.24. An isothermal parametrization maps small squares to small squares.
Theorem 2.39. A nonplanar minimal surface in R3 that is also a surface of revolution is

contained in a catenoid.
As we have seen, using (2.5) to construct a minimal graph involves solving second
order differential equations. We can simplify matters if we use isothermal parametrizations.
Definition 2.40. A parametrization x is isothermal if E D xu xu D xv xv D G and
F D xu xv D 0.
Because E, F , and G describe how lengths on a surface are distorted as compared to their
usual measurements in R3 , if F D xu xv D 0 then xu and xv are orthogonal and if
E D G then the amount of distortion is the same in the orthogonal directions. Thus, we
can think of an isothermal parametrization as mapping a small square in the domain to a
small square on the surface. Sometimes an isothermal parametrization is called a conformal
parametrization, because the angle between a pair of curves in the domain is equal to the
angle between the corresponding pair of curves on the surface.
Example 2.41. The parametrization
x.u; v/ D .a cosh v cos u; a cosh v sin u; av/
for the catenoid is isothermal, because in Example 2.35 we derived that E D a2 cosh2 v D
G and F D 0. We can get a geometric sense that it is isothermal by using DiffGeomTool
to graph it. Open DiffGeomTool and enter the parametrization with a D 1 and click the
Graph button. Use the Transparency of surface and the Transparency of
the frame sliders and click off and on the Wireframe button to make the curves more
distinct. The grid of squares in the domain are mapped to a grid of squares as predicted
(see Figure 2.25).
Example 2.42. The parametrization

x.u; v/ D .a C b cos v/ cos u; .a C b cos v/ sin u; b sin v
for the torus is not isothermal (Note: because of Exploration 2.10, we assume that a > b).
This is because in Exercise 2.32, you derived that
E D.a C b cos v/2
F D0
G Db 2 :
Because F D 0, xu and xv are orthogonal on the torus. But E ¤ G except if v D
C 2 k; .k 2 Z/ and a D 2b. Thus the image of squares in the domain will be nonsquare
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 109 — #127

i i
F IGURE 2.25. This parametrization of the catenoid is isothermal.
rectangles whenever v ¤ C 2 k. Open DiffGeomTool and enter the parametrization for

the torus with a D 2 and b D 1. Set the Rectangular grid values to 0 u 2 and
0 v 2 . Again, use the Transparency of surface and the Transparency
of the frame sliders and click off and on the Wireframe button to make the curves
more distinct. The grid of squares in the domain are mapped to a grid of mostly nonsquare
le ngt h
rectangles. The ratio, height , of the sides of the rectangles is largest for the part of the torus
farthest away from the origin. This occurs when v D 0 (or v D 2 ), resulting in E D 4
while G D 1. The rectangles are squares for the part of the torus closest to the origin.
This occurs when v D , resulting in E D 1 while G D 1. This helps us see why the
parametrization is not isothermal (see Figure 2.26).
F IGURE 2.26. This parametrization of the torus is not isothermal.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 110 — #128

i i
Exercise 2.43. Using Definition 2.40 determine which of the parametrizations of minimal
surfaces is isothermal:
(a) Enneper’s surface parametrized by

x.u; v/ D u 31 u3 C uv 2 ; v 31 v 3 C u2 v; u2 v 2
(b) Scherk’s doubly periodic surface parametrized by
cos u
x.u; v/ D u; v; ln
cos v
(c) The helicoid parametized by
x.u; v/ D .a sinh v cos u; a sinh v sin u; au/.
Try it out!
Exploration 2.44. Use DiffGeomTool to check the reasonableness of your answers in Ex-
ercise 2.43 by graphing each parametrization as was done in Examples 2.41 and 2.42. Set
the number of u points to 20 and the number of v points to 20, and use the following values
for the U/V domain boxes:
(a) Enneper’s surface

3
u; v 3
(b) Scherk’s doubly periodic surface

2
C 0:1 u; v 2 0:1
(c) Helicoid
u; v .
From Exercise 2.43 and Exploration 2.44 you have seen there are parametrizations of
minimal surfaces that are not isothermal. However, requiring minimal surfaces to have an
isothermal parametrization is not a restriction because of the following theorem.
Theorem 2.45. Every minimal surface in R3 has an isothermal parametrization.
Remark 2.46. See [23] for a proof. In fact, every differentiable surface has an isothermal
parametrization, an interesting result whose proof is beyond the scope of this text. A proof
is given in [2, pp. 15-35].
In Example 2.35, we derived that the isothermal parametrization for the catenoid

x.u; v/ D a cosh v cos u; a cosh v sin u; av
has
eD g:
In general, we have
Theorem 2.47. Let M be a surface with isothermal parametrization. Then M is minimal
if and only if e D g.
Exercise 2.48. Prove Theorem 2.47. Try it out!
Exploration 2.49. If e D g for the coefficients of the second fundamental form, the
u-parameter curve and the v-parameter curve are bending the same amount away from the
normal n but in different directions. Use DiffGeomTool with Theorem 2.47 to geometrically
verify which of the following surfaces are minimal:
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 111 — #129

i i
(a) Enneper’s surface

1 3 1 3
x.u; v/ D u 3
u C uv 2 ; v 3
v C u2 v; u2 v2
(b) Cylinder
x.u; v/ D cos u; sin u; v
(c) Helicoid
x.u; v/ D .a sinh v cos u; a sinh v sin u; av/.
Try it out!
Here is an interesting and important result that uses complex analysis. We know that
if f .z/ D x.u; v/ C iy.u; v/ is an analytic function, then the Cauchy-Riemann equations
hold for f . That is,
xu D yv ; xv D yu ;
and y is called the harmonic conjugate of x. If f is analytic, then
f 0 .z/ D xu C iyu : (2.7)
This allows us to relate a minimal surface to another minimal surface, its conjugate minimal
surface.
Definition 2.50. Let x and y be isothermal parametrizations of minimal surfaces such that
their component functions are pairwise harmonic conjugates. That is,
xu D yv and xv D yu : (2.8)
Then x and y are called conjugate minimal surfaces.

Example 2.51. Let’s find the conjugate surface of the catenoid parametrized by
x.u; v/ D .a cosh v cos u; a cosh v sin u; av/:
Let y.u; v/ be its parametrization. By the first part of (2.8), we know
yv D xu D . a cosh v sin u; a cosh v cos u; 0/:
Integrating this with respect to v yields
y D . a sinh v sin u C F1 .u/; a sinh v cos u C F2 .u/; F3 .u//;
where Fk .u/ is a function independent of v. Similarly, by the second part of (2.8), we

derive
y D . a sinh v sin u C G1 .v/; a sinh v cos u C G2 .v/; au C G3.v//:
Equating, we get
y D . a sinh v sin u C K1 ; a sinh v cos u C K2 ; au C K3 /:
u 2 , v D e
Using the substitution u D e v , and letting K1 D 0, K2 D 0, and K3 D a 2 ,
does not affect the geometry of the minimal surface, and yields the parametrization of a
helicoid
y.e v/ D .a sinhe
u;e v coseu; a sinhev sine
u; ae
u/
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 112 — #130

i i
given in Exercise 2.24 (The negative sign in the third component function has the effect of
reflecting the surface through the xy-plane). Hence, the conjugate surface of the catenoid
is the helicoid.
The idea of conjugate minimal surfaces is really interesting. Two conjugate minimal
surfaces can be joined through a one-parameter family of minimal surfaces by
z D .cos t/x C .sin t/y;
where t 2 R. When t D 0 we have the minimal surface parametrized by x, and when

t D 2 we have the minimal surface parametrized by y. So for 0 t 2 , we have a con-
tinuous parameter of minimal surfaces known as associated surfaces. We can continuously
transform one minimal surface into another minimal surface so that all the in-between sur-
faces are also minimal. In Example 2.51, we saw that the helicoid and the catenoid are
conjugate surfaces. Images of them and associated surfaces are shown in Figure 2.27.
F IGURE 2.27. The helicoid, some associated surfaces, and the catenoid.
This is neat, and it is just the beginning. The rest of this section will explore properties
of conjugate surfaces.
Exercise 2.52. Find the conjugate minimal surface for Enneper’s surface

1 3 1 3
x.u; v/ D u u C uv 2 ; v v C u2 v; u2 v 2 :
3 3
If we try to determine the conjugate minimal surface for Scherk’s doubly periodic
surface with the parametrization
cos u
x.u; v/ D u; v; ln ;
cos v
the method will not work because the parametrization is not isothermal. Later we will see
that Scherk’s doubly periodic surface has a conjugate surface, Scherk’s singly periodic
surface (see Figure 2.28).
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 113 — #131

i i
F IGURE 2.28. Scherk’s singly periodic surface.
Exploration 2.53. You can see the associated surfaces that occur between Scherk’s dou-
bly periodic surface and Scherk’s singly periodic surface by using another applet, called
MinSurfTool that can be used to visualize and explore minimal surfaces in R3 . Open the
MinSurfTool applet. On the right-hand side near the top, there is a set of tabs for differ-
ent features. For this exploration, we use the W.E.(p,q) feature, so make sure that the
W.E.(p,q) tab is on top (if it is on top, it should be a different color than the other tabs).
In the Pre-set functions window, choose p.z/ D 1=.1 z 4 /, q.z/ D z. Click on
the Graph button, and one piece of Scherk’s singly periodic surface will appear. Then
move the slider arrow above the Graph button to the right to see how Scherk’s singly peri-
odic surface is transformed by way of the associated surfaces into Scherk’s doubly periodic
surface. Try it out!
Pieces of Scherk’s doubly periodic surface can be put together in the xy-plane in a
checkerboard fashion. They repeat (or are periodic) in two directions, x and y. Surfaces
that repeat in one direction are called singly periodic. Individual pieces of Scherk’s singly
periodic surface can fit together creating a tower in the z direction. You can visualize
adding two pieces together by taking one piece of Scherk’s singly periodic surface and
adding it to another piece that has been reflected across the xy-plane and shifted up in the
z direction. Continuing, you can create a tower of several pieces (see Figure 2.29). The
helicoid is a singly periodic surface too.
In Example 2.31 we saw that the coefficients of the first fundamental form for the pa-
rametrization of a catenoid are E D a2 cosh2 v, F D 0, and G D a2 cosh2 v. Exercise
2.140 from the Additional Exercises at the end of the chapter shows that for the param-
etrization of Enneper’s surface the coefficients are E D .1 C u2 C v 2 /2 , F D 0, and
G D .1 C u2 C v 2 /2 . The coefficients do not match up. However, for two conjugate mini-
mal surfaces and their associated minimal surfaces, the coefficients of the first fundamental
form are always the same. The following exercise will help you prove this surprising result.
Exercise 2.54.
(a) Prove that for two conjugate minimal surfaces, x and y, surfaces of the one-parameter
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 114 — #132

i i
F IGURE 2.29. Scherk’s singly periodic surface.
family
z D .cos t/x C .sin t/y
have the same fundamental form: E D xu xu D yu yu, F D 0, G D xv xv D yv yv .
(b) Prove that the surfaces z are minimal for all t 2 R.
Try it out!
The normal vector n at a point on a surface points orthogonally away from the surface.
Since minimal surfaces have different shapes, there is no reason to suppose that the normal
vectors on one surface will be related to the normal vectors on another surface. However,
for conjugate minimal surfaces and their associated minimal surfaces there is a strong
connection. For any point in the domain, the corresponding surface normal points in the
same direction on all these minimal surfaces. The next theorem establishes this.
Theorem 2.55. Let x; y W D ! R3 be isothermal parametrizations of conjugate minimal
surfaces. Then for .u0 ; v0 / 2 D, the surface unit normal is the same for the associated
surfaces.
Proof. Let .u0 ; v0 / 2 D. Let nx and ny represent the surface normal for x and for y,
respectively. Then by the definition of conjugate surfaces, x and y have the same unit
normal, because
xu xv y yu y yv
nx D ˇ ˇDˇ v ˇD ˇ u ˇ D ny :
ˇ xu xv ˇ ˇ yv yu ˇ ˇyu yv ˇ
To show that this is true for the associated surfaces, let z D .cos t/x C .sin t/y be the
parametrization of the associated surfaces. Then

zu zv D cos txu C sin tyu cos txv C sin tyv

D cos2 t xu xv C cos t sin t xu yv C cos t sin t yu xv C sin2 t yu yv

D cos2 t xu xv C cos t sin t xu xu C cos t sin t xv xv C sin2 t xv xu

D cos2 t xu xv C cos t sin t 0 C cos t sin t 0 C sin2 t xu xv
D xu xv :
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 115 — #133

i i
The following example and exploration helps us visualize this idea.

Example 2.56. Using DiffGeomTool we can graph the catenoid and its conjugate surface,
the helicoid, whose parametrizations are given in Example 2.51. If we plot the normal n at
. 3 ; 4 / on the conjugate surfaces, we see that both normals point in the same direction as
guaranteed by Theorem 2.55 (see Figure 2.30 and Figure 2.31).
F IGURE 2.30. The catenoid with n at . 3 ;

4 /.
F IGURE 2.31. The conjugate helcoid with n at . 3 ;

4 /.
Exploration 2.57. Open two windows of DiffGeomTool. In one plot the catenoid parame-
trized by
x.u; v/ D .cosh v cos u; cosh v sin u; v/;
2 2
where 0 u 2 and 3
v 3
. In the other plot its conjugate surface, the
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 116 — #134

i i
helicoid, with parametrization
y.u; v/ D . sinh v sin u; sinh v cos u; u/;
where 2 u 2 2 and 2 3
v 2 3
(The u values for the helicoid are different
than the values for the catenoid because we used the substitution u D e u 2 in Example
2.51). Make sure that the x; y; z-axes are positioned in the same directions. Then plot the
following unit normals, n, on each surface at the following points and observe that n points
in the same direction as prescribed by Theorem 2.55:
(a) at . 3 ; 0/ (b) at . 4 ;
2
/ (c) at . 6 ; 2 /.
Try it out!
2.4 Weierstrass representation

At the end of the last section, we saw that we could apply complex analysis to minimal
surface theory to define conjugate surfaces. In this section we will again use complex
analysis to learn more about minimal surfaces. First, we will use inherent properties of
an isothermal parametrization to give a necessary and sufficient condition for a surface to
be minimal. As we will see, the condition is useful for finding and classifying minimal
surfaces.
Theorem 2.58. If the parametrization x is isothermal, then
xuu C xvv D 2EH n;
where E is a coefficient of the first fundamental form and H is the mean curvature.
Exercise 2.59 (Proof of Theorem 2.58). The set fxu; xv ; ng forms a basis for R3 . Assume
F D 0. Then the vector xuu can be expressed in terms of the basis vectors. That is,
u v
xuu D uu xu C uu xv C en;
u v
where the coefficients, uu and uu , are known as Christoffel symbols and e comes from
the coefficient of the second fundamental form: e D n xuu .
(a) Show that uuu
D E v
2E and uu D
u Ev
2G by taking the inner product of xuu with the
basis vectors. In a similar manner, it can be shown that
Gu Gv
xvv D xu C xv C gn:
2E 2G
(b) Use the mean curvature equation (2.4) and the results from (a) to show that if the
parametrization x is isothermal, then
xuu C xvv D 2EH n:

Try it out!
Now, where do we go from here? For a minimal surface, H 0, so Theorem 2.58
tells us that xuu C xvv 0. What does that equation represent? It is Laplace’s equation
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 117 — #135

i i
2.4. Weierstrass representation 117
and relates to harmonic functions. Recall that '.u; v/ is a real-valued harmonic function if
'uu C'vv D 0 (for example, '.u; v/ D u2 v 2 is harmonic). This leads to the condition for
a surface to be minimal that we referred to in the beginning of this section. Corollary 2.60
will allow us to explicitly describe minimal surfaces using the Weierstrass representation,
which we will derive in this section.
Corollary 2.60. A surface M with an isothermal parametrization x.u; v/ D x1.u; v/;

x2 .u; v/; x3 .u; v/ is minimal if and only if x1 ; x2; and x3 are harmonic.
We need an isothermal parametrization for our surface, but this is not a difficulty because
of Theorem 2.45. Then this result tells us we will have a minimal surface if and only if
its coordinate functions are harmonic. This provides us another way to prove a surface is
minimal.
Proof. ()) If M is minimal, then H D 0 and so by Theorem 2.58 xuu C xvv D 0, and
hence the coordinate functions are harmonic. (() Suppose x1 ; x2 ; and x3 are harmonic.
Then xuu C xvv D 0. So by Theorem 2.58 we have that 2.xu xu /H n D 0. But n ¤ 0 and
E D xu xu ¤ 0. Hence H D 0 and M is minimal.
Exercise 2.61. Given the parametrization for Enneper’s surface

1 3 1 3
x.u; v/ D u u C uv 2 ; v v C u2 v; u2 v2 ;
3 3
use Corollary 2.60 to prove that it is a minimal surface. Try it out!

The importance of Corollary 2.60 is not in proving specific surfaces are minimal but lies
in establishing a formula that will guarantee that a surface created by it is minimal. The
formula is known as the Weierstrass representation for minimal surfaces. It provides us
with a simple way to construct many examples of minimal surfaces using functions from
complex analysis. After stating the Weierstrass representation in Theorem 2.66 we will
create more minimal surfaces. However, not all minimal surfaces are of equal interest. So
in section 2.5 we will investigate properties that make minimal surfaces interesting.
We now derive this important formula, the Weierstrass representation. Suppose M is a
minimal surface with an isothermal parametrization x.u; v/. Let z D u C iv be a point in
the complex plane, so z D u iv. Solving for u; v in terms of z; z we get
zCz z z
uD and vD :
2 2i
The parametrization of the minimal surface M can be written in terms of the complex
variables z and z as

x.z; z/ D x1 .z; z/; x2 .z; z/; x3 .z; z/ :
Exercise 2.62. Let f .u; v/ D x.u; v/ C iy.u; v/ be a complex function. Using u D zCz 2
and v D z2iz , we can express f in terms of z and z. In this exercise you will prove that
f is analytic if and only if f can be written in terms of z D u C iv alone, without using
z D u iv.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 118 — #136

i i
(a) Using the chain rule, derive

@f 1 @x @y i @y @x
D C C ;
@z 2 @u @v 2 @u @v

@f 1 @x @y i @y @x
D C C :
@z 2 @u @v 2 @u @v
@f
(b) Show that f is analytic ” @z D 0.
Try it out!
@f1 2
Example 2.63. The function f1 .z/ D z 2 is analytic because @z
.z / D 0. However,
f2 .z/ D jzj2 D zz is not analytic because @f
@z
1
D z ¤ 0.
Exercise 2.64. Prove that

@ @f
4 D fuu C fvv : (2.9)
@z @z
Try it out!
The next theorem uses Corollary 2.60 to establish the Weierstrass representation for
minimal surfaces.
Theorem 2.65. Let M be a surface with parametrization x D .x1 ; x2; x3 / and let D
.'1 ; '2 ; '3 /, where 'k D @x 2 2 2 2
@z . Let denote .'1 / C .'2 / C .'3 / . Then x is isothermal
k
2
” 0. If x is isothermal, then M is minimal ” each 'k is analytic.
Before we prove Theorem 2.65, let’s apply it to an example to help us understand what the
theorem says. Suppose we have the parametrization x D .x1 ; x2; x3 / D .z 13 z 3 ; i.z C
3
z /; z 2 /. Then '1 D @x
1 3
@z
1
D 1 z 2 , '2 D @x @z
2
D i.1 C z 2 /, and '3 D @x@z
3
D 2z.
2 2 2 2 2 2
Because D Œ1 z C Œ i.1 C z / C Œ2z D 0, the parametrization is isothermal
by the theorem. Each 'k is a polynomial and hence analytic. So x is a parametrization of
a minimal surface (it is Enneper’s surface). Make sure you understand how this example
relates to Theorem 2.65 before you read the proof of the theorem.
@f
Proof. Applying the complex differential operator @z
from Exercise 2.62 and squaring the
terms, we have
2 2 " 2 2 #
2 @xk 1 @xk @xk 1 @xk @xk @xk @xk
.'k / D D i D 2i :
@z 2 @u @v 4 @u @v @u @v
Also,
2 2 2 3
X
@x1 @x2 @x3 @xk 2
xu xu D C C D
@u @u @u @u
kD1
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 119 — #137

i i
P3 @xk 2
and similarly xv xv D kD1 . @v / . Hence,
2 D.'1 /2 C .'2 /2 C .'3 /2
3
1 X @xk 2 X @xk 2
3 3
X @xk @xk
D 2i
4 @u @v @u @v
kD1 kD1 kD1
1
D xu xu xv xv 2i.xu xv /
4
1
D .E G 2iF /:
4
Thus, x is isothermal ” E D G; and F D 0 ” 2 0.
Suppose that x is isothermal. By Corollary 2.60, it suffices to show that for each k, xk
is harmonic ” 'k is analytic. Using (2.9) and Exercise 2.62 this follows because

@2 xk @2 xk @ @xk @
C D4 D4 'k D 0:
@u@u @v@v @z @z @z
Let’s apply this theorem. Suppose we have analytic functions 'k and we want to find the
functions xk . If x is isothermal, then
ˇ ˇ ˇ ˇ ˇ ˇ 3
1 X @xk 2 X @xk 2
3
ˇ @x1 ˇ2 ˇ @x2 ˇ2 ˇ @x3 ˇ2
2
jj D ˇ ˇ ˇ Cˇ ˇ ˇ Cˇˇ ˇ D C
@z ˇ @z ˇ @z ˇ 4 @u @v
kD1 kD1
1 1 E
D xu xu C xv xv D .E C G/ D :
4 4 2
So if jj2 0, then the coefficients of the first fundamental form are zero and M is a
point.
We need to solve 'k D @x @z
k
for xk since the parametrization of the surface is given as
x D .x1 ; x2 ; x3/. The difficulty is that xk is a function of two variables, z and z, and we
want to have a representation where we have to integrate with respect to only one variable.
To overcome this, we will use some ideas about differentials (see [26] for an introduction
to differentials). Since xk is a function of the two variables u and v, we can write
@xk @xk
dxk D du C dv: (2.10)
@u @v
Also, dz D du C idv. Using Exercise 2.62 we have

@xk 1 @xk @xk
'k dz D dz D i .du C idv/
@z 2 @u @v
" #
1 @xk @xk @xk @xk
D du C dv C i dv du ;
2 @u @v @u @v

@xk 1 @xk @xk
'k dz D 'k dz D dz D Ci .du idv/
@z 2 @u @v
" #
1 @xk @xk @xk @xk
D du C dv i dv du :
2 @u @v @u @v
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 120 — #138

i i
Adding we get
@xk @xk
du C dv D 'k dz C 'k dz D 2 Ref'k dzg: (2.11)
@u @v
Combining (2.10) and (2.11), we have
dxk D 2 Ref'k dzg:

R
Therefore, xk D 2 Re 'k dz C ck . Since adding ck translates the image by a constant
amount and multiplying a coordinate function by 2 scales the the surface, the constants do
not affect the geometric shape of the surface. Hence, we do not need them and we will let
our coordinate function be Z
xk D Re 'k dz:
Summary: If we have analytic functions ’k (k D 1; 2; 3) such that 2 0 and jj2 6 0

and is finite, then the parametrization
Z Z Z
x D Re '1 .z/dz; Re '2 .z/dz; Re '3 .z/dz (2.12)
defines a minimal surface.
For example, consider the functions p.z/ and q.z/ such that
'1 Dp.1 C q 2 /
'2 D ip.1 q2/
'3 D 2ipq:
Then
2 DŒp.1 C q 2 /2 C Œ ip.1 q 2 /2 C Œ 2ipq2

DŒp 2 C 2p 2 q 2 C p 2 q 4 Œp 2 2p 2q 2 C p 2q 4 Œ4p 2q 2
D0;
and
jj2 Djp.1 C q 2 /j2 C j ip.1 q 2 /j2 C j 2ipqj2

Djpj2 Œ.1 C q 2 /.1 C q 2 / C .1 q 2 /.1 q 2 / C 4q q
Djpj2 Œ2.1 C 2q q C q 2 q 2 /
D4jpj2.1 C jqj2 /2 ¤ 0 (if p D 0, then 'k D 0 for all k):
For 'k to be analytic, p, pq 2 , and pq have to be analytic. If p is analytic with a zero of

order 2m at z0 , then q can have a pole of order no larger than m at z0 . A function that is
analytic in a domain D except possibly at poles is a meromorphic function in D. This leads
to the following result.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 121 — #139

i i
Theorem 2.66 (Weierstrass Representation (p; q)). Every regular minimal surface has a
local isothermal parametric representation of the form
x D .x1 .z/; x2 .z/; x3 .z//

Z z Z z Z z
D Re p.1 C q 2 /dz ; Re ip.1 q 2 /dz ; Re 2ipqdz ;
a a a
where p is an analytic function and q is a meromorphic function in a domain C,

having the property that where q has a pole of order m, p has a zero of order at least 2m,
and a 2 is a constant.
Example 2.67. For p.z/ D 1; q.z/ D iz, we get
nZ z o nZ z o nZ z o
2 2
x D Re .1 z / dz ; Re i.1 C z / dz ; Re 2z dz
0 0 0
n 1 3o n 1 o n o
D Re z z ; Re i z C z 3 ; Re z 2 :
3 3
Letting z D u C iv, this yields

1 3 1 3
x.u; v/ D u u C uv 2 ; v v C u2 v; u2 v2 ;
3 3
which gives Enneper’s surface.
You can use the applet MinSurfTool to graph an image of this surface using p and q.
After opening MinSurfTool, make sure that the W.E.(p,q) tab is on top. In the appropri-
ate boxes, put p.z/ D 1 and q.z/ D iz. Then click on the Graph button. You can increase
the size of the image of the surface by clicking on the left button on the mouse, and you can
decrease it by clicking on the right mouse button. You can rotate the surface by placing the
cursor arrow on its image and clicking, holding the left button on the mouse as you move
the cursor.
F IGURE 2.32. Enneper’s surface using p.z/ D 1 and q.z/ D iz.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 122 — #140

i i
Example 2.68. Let p.z/ D 1 and q.z/ D 1=z on the domain C f0g. We see that q is
meromorphic with a pole of order 1 at z0 D 0 and p does not have a zero of order 2 at z0 D
0. This does not violate the conditions of Theorem 2.66, because the domain is C f0g.
We will show that p and q generate a helicoid. Using the Weierstrass representation (p; q)
and letting z D u C iv, we get x.u; v/ D .x1 ; x2; x3 /, where
Z z
1 1 u
x1 D Re 1 C 2 dz D Re z Du
1 z z u C v2
2
Z z
1 1 v
x2 D Re i 1 2
dz D Im z C Dv
1 z z u C v2
2
Z z
1 v
x3 D Re 2i dz D 2 Im.log z/ D 2 arg z D 2 arctan :
1 z u
The parametrization is different than the parametrization we have been using for the heli-
coid,
e
x .e v / D .xe1 ; xe2 ; xe3/ D .a sinhe
u;e v cose
u; a sinhe
v sine
u; ae
u/:
To show that x also gives an image of the helicoid, we find a substitution that will change
x into e
x . We have
1
x12 C x22 D .u2 C v 2 / 2C
u2 C v 2
e v 2
2 2 2 2 2 ev e e
xe1 C xe2 D a sinh e
vDa :
2
Equating the right-hand sides and letting a D 2, we get that
u2 C v 2 D e 2e
v
:
With x3 D xe3 , we see that

v
D taneu:
u
Using the last two equations we can solve for u and v to get
u D ee
v
u
cose and v D ee
v
u:
sine
If we substitute u and v into x.u; v/ we get the parametrization e x .e

u;e
v / for the helicoid.
Using the W.E.(p,q) tab in MinSurfTool, we can get a graph of the helicoid by
choosing from the Pre-set functions the values p.z/ D 1 and q.z/ D 1=z. Since
the domain is C n f0g, the Disk domain: radius min: box is set to 0.2. Because of
the singularity at z D 0, there needs to be functions entered into the x, y and z boxes in the
Complex initial values for integration in radial direction sec-
tion. These functions come from explicitly solving the integrals for x1.z/, x2.z/, and x3 .z/
in the Weierstrass representation when p.z/ D 1 and q.z/ D 1=z and substituting in r e i
for z. If you use a choice from the Pre-set functions, the functions will be entered
automatically. However, if you enter your own p and q values, you will need to compute
the functions and enter them into these boxes to get the correct minimal surface image.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 123 — #141

i i
F IGURE 2.33. The helicoid using p.z/ D 1 and q.z/ D z1 .
Exercise 2.69. Show that the minimal surface generated by using p.z/ D 1 and q.z/ D 0
on the domain C in the Weierstrass representation is the plane. Try it out!
Exercise 2.70. Show that the minimal surface generated by using p.z/ D 1 and q.z/ D
i=z on the domain C f0g in the Weierstrass representation is the catenoid. Use the appro-
priate Pre-set function in the W.E.(p,q) tab of MinSurfTool to graph an image
of the surface. Try it out!
Exploration 2.71. Enneper’s surface can be constructed with p.z/ D 1 and q.z/ D z.
It has four leaves (two pointing up and two pointing down). The number of leaves can be
increased.
(a) Using p.z/ D 1 and q.z/ D z 2 on the domain C in the Weierstrass representation
gives Enneper’s surface with six leaves (see Figure 2.34). Compute its parametrization
x.u; v/.
(b) Conjecture the values of p and q for Enneper’s surface with n leaves.
(c) Use MinSurfTool with the W.E.(p,q) tab to check your conjecture.
Try it out!
Exploration 2.72. Use MinSurfTool with the W.E.(p,q) tab to graph an image of the
surface generated by the Pre-set functions p.z/ D 1 1z4 and q.z/ D z.
(a) What minimal surface is it?
(b) Click on the box “Multiply q.z/ by e^(i !)” and move the slider to generate a family
of minimal surfaces. They are associated surfaces (see the paragraph after Definition
2.50). When D 2 you get the conjugate surface. What is it?
(c) Experiment with MinSurfTool to view the associated family and find the conjugate
surface of the various minimal surfaces discussed above.
Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/29 — 12:57 — page 124 — #142

i i
F IGURE 2.34. Enneper’s surface with six leaves using p.z/ D 1 and q.z/ D z 2 .
Exploration 2.73. Scherk’s doubly periodic surface is generated with p.z/ D 1 1z4 and
q.z/ D iz. Use the appropriate Pre-set function in the W.E.(p,q) tab of Min-
SurfTool to graph an image of the surface generated by p.z/ D 1 1z2n and q.z/ D iz n 1
for various values of n D 2; 3; 4; : : :. The and C boxes above the Graph button can be
used to increase or decrease n.
(a) What happens to the surface as n increases?
(b) The surface has leaves that alternate between going up and going down. How is n
related to the number of leaves?
(c) What is the image of the projection of the surface onto the x1 x2 -plane for each n?
1
(d) Using the previous parts conjecture how many leaves the surface has if p.z/ D 1 z5
.
Why could such a surface not exist?
Try it out!
Exploration 2.74. Use the appropriate Pre-set function in the W.E.(p,q) tab of
MinSurfTool to graph an image of the surface generated by p.z/ D .1 1z4 /2 and q.z/ D
iz 3 . This surface is known as the 4-noid (see Figure 2.35).
(a) Try to create a 3-noid by changing the values of p and q and graphing the result in
MinSurfTool.
(b) Conjecture the values of p and q that will generate an n-noid.
Try it out!
While the Weierstrass representation will generate a minimal surface, there is no guar-
antee that it will be embedded. In the introduction of this chapter that we said that a sur-
face is embedded if it has no self-intersections. The plane, the catenoid, the helicoid, and
Scherk’s doubly periodic surface are embedded minimal surfaces, but Enneper’s surface is
not embedded. In Exploration 2.11 you saw that Enneper’s surface p intersects itself when
the domain contains a disk centered at the origin of radius R 3.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 125 — #143

i i
2.5. The Gauss map, G, and height differential, dh 125
F IGURE 2.35. Image of the 4-noid minimal surface.
Exploration 2.75. Using MinSurfTool with the W.E.(p,q) tab, get three sets of func-
tions p and q that create minimal surfaces that are not embedded. If the p and q functions
result in a singularity in the Weierstrass representation, then you will need to enter func-
tions into the x, y, and z boxes in the Complex initial values for integra-
tion in radial direction section. They come from explicitly solving the inte-
grals for x1 .z/, x2 .z/, and x3 .z/ in the Weierstrass representation for p.z/ and q.z/ and
then substituting in r e i for z. Try it out!
An important part of minimal surface theory is the study of complete (boundaryless)
embedded minimal surfaces. The following theorem tells us that a minimal surface without
boundary cannot be closed and bounded.
Theorem 2.76. If M is a complete minimal surface in R3 , then M is not compact.
Proof. By Theorem 2.45, we can assume that M has an isothermal parametrization. If

M were compact, then each coordinate function would attain a maximum. Since the real
part of an analytic function is harmonic, the coordinate functions in this parametrization
are harmonic by Theorem 2.65. Because harmonic functions attain their maximums on the
boundary of the set, M must have a boundary, which contradicts M being complete.
2.5 The Gauss map, G , and height differential, dh

We can use other representations for D .'1 ; '2 ; '3 / to form different Weierstrass rep-
resentations as long as 2 D 0 and jj2 ¤ 0 (see the Summary on page 120). An im-
portant representation employs the function known as the Gauss map, G, and the height
differential, dh. This representation is useful because G and dh describe the geometry of
the minimal surface. To develop the representation, we need some background about the
Gauss map. ˇ ˇ
The curvature of a unit speed curve, ˛, at a point s is ˇ˛ 00.s/ˇ. That is, the curvature
of a curve is described by the rate of change of its tangent vector. Similarly, the curvature
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 126 — #144

i i
of a surface is related to the rate of change of its tangent plane. Since a tangent plane is
determined by its unit normal vector, n, we can investigate the curvature of a surface by
studying the variation of n. This is the idea behind the Gauss map.
Definition 2.77. Let M W ! R3 be a surface with a chosen orientation (that is, a
differentiable field of unit normal vectors n). The Gauss map, np , translates the unit normal
on M at a point p to the unit vector at the origin pointing in the same direction as the unit
normal and thus corresponds to a point on the unit sphere S 2 .
np
p
Gauss
map
M Ì R3
S2
F IGURE 2.36. The Gauss map.
Example 2.78. Let’s determine the image of the Gauss map for the catenoid. A meridian
is a curve formed by a vertical slice on the surface. Take a meridian on the entire catenoid
(the image in Figure 2.37 is only part of a catenoid that extends on forever). Its Gauss
map, np , will be a meridian on S 2 from the north pole .0; 0; 1/ to the south pole .0; 0; 1/
that excludes the end points. They are excluded because no matter how far the catenoid
extends, the unit normal n never points exactly straight up or exactly straight down. Since
the catenoid is a surface of revolution, if we revolve the meridian, we get that the image of
the Gauss map for the catenoid is S 2 n f.0; 0; 1/; .0; 0; 1/g.
np
n1
n2 n p2
np
1
2
S
Mc
F IGURE 2.37. Image of a meridian on a catenoid under the Gauss map.
Exercise 2.79. Describe the image of the Gauss map for

(a) a right circular cylinder
(b) a torus
(c) Enneper’s surface defined only on D
(d) a helicoid
(e) Scherk’s doubly periodic surface.
Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 127 — #145

i i
np
A C′
B
B′
C A′
S2
Mc
F IGURE 2.38. np is orientation reversing.
Theorem 2.80 ([22]). Let M be a minimal surface with an isothermal parametrization.

Then the Gauss map of M preserves angles.
Though the Gauss map preserves angles, it reverses orientation. Such maps are called
anticonformal. To help visualize the reversal, take three points A, B, and C on a curved
path near the neck of the catenoid (see Figure 2.38). Since A is above the neck, the outward
pointing unit normal at A will be pointing downward and hence the Gauss map will put
it below the equator on S 2 at A0 . The point B is on the neck and so the outward pointing
unit normal at B will be horizontal. The Gauss map will put it on the equator of S 2 at B 0 .
Similarly, the normal at C will be mapped to C 0 . Thus, the curved path from A to B to C
in the positive direction on the catenoid gets sent by the Gauss map to a curve from A0 to
B 0 to C 0 in the negative direction on S 2 . That is, we have an orientation-reversing map.
Since the Gauss map associates a point on M with a point on S 2 , we can also associate
the original point with a point in the complex plane C by using stereographic projection.
Stereographic projection takes a point on S 2 to a point in the extended complex plane,
C [ 1. As in Figure 2.39, we place the complex plane through the equator of the sphere
and take a line connecting the north pole, .0; 0; 1/ 2 S 2 , with the point .x1 ; x2; x3 / 2 S 2 .
This line intersects the extended complex plane at a point, z D x C iy. In this setting the
unit sphere is known as the Riemann sphere.
Exercise 2.81. Describe the projections of the following sets on the Riemann sphere onto
the extended complex plane
(a) meridians
(b) parallels
(c) circles
(d) circles that contain (i.e., touch) the point .0; 0; 1/
(e) antipodal points (i.e., diametrically opposite points).
Try it out!
(0, 0, 1)
(x1, x2, x3)

z = x + iy
C È {¥ }
F IGURE 2.39. Stereographic projection.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 128 — #146

i i
Let be the projection of .x1 ; x2; x3 / 2 S 2 to the point x iy 2 C by stereographic

projection of .x1 ; x2; x3 / to z D x C iy followed by reflection across the real axis to
z D x iy. Note that is an anticonformal map.
Let G W D C ! C be the map defined by G D ı n ı x. It preserves angles because
, n, and x preserve angles and so the composition of these maps preserves angles. G is
orientation preserving because and n are orientation reversing and so their composition
is orientation preserving. Thus, G is a meromorphic function. It is also called the Gauss
map. While there are two different maps called the Gauss map, we will mean G when we
refer to “the Gauss map” in this chapter. See Figure 2.40
n
(x1 , x2, x3 )
np
p = x(u0 , v0)
M Ì R3
S2
x(u,v)
s
v
z = x – iy
(u0, v0 )
G
u
DÌC
F IGURE 2.40. The map G.
Example 2.82. Using the geometry of Enneper’s surface, ME , we can determine specific
values of a Gauss map, G, on ME even though we do not know what function G is. What is
F IGURE 2.41. Enneper’s surface.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 129 — #147

i i
V x3 s2
ME x2
x np s
x
0 u 0
x1
D D
n (0, 0, –1) Y
F IGURE 2.42. G.0/ D 0 for Enneper’s surface.
G.0/? Enneper’s surface is formed by bending a disk into a saddle surface. The point 0 2 C
should get mapped to the point in the center of Enneper’s surface. There are two normal
vectors on the surface. We will chose the downward pointing normal n. Hence, the unit
normal at the center of ME points straight down, and thus mapping it to S 2 under np gives
the vector pointing at .0; 0; 1/ (see Figure 2.42). Taking the stereographic projection, ,
results in the point z D 0 2 C and reflecting it across the real axis does not change 0, so
.0; 0; 1/ D 0. Hence, G.0/ D 0.
What is G.r / when r 2 Œ0; 1? The points r get mapped under x to a curve moving
upward along one of the upward pointing leaves of Enneper’s surface. The corresponding
downward pointing unit normal, nr , stays in the x1x3 -half plane (where x1 0) also
moving upward (i.e., x3 is increasing). As r approaches 1, nr approaches being parallel to
the x1 axis (see Figure 2.43). Thus, mapping the unit normals to S 2 , the curve fr 2 D W
0 r < 1g traces a meridian on S 2 from .0; 0; 1/ to .1; 0; 0/. Its stereographic projection
onto the complex plane gives fr 2 D W 0 r < 1g and reflecting this across the real axis
does not change the values. Hence, G.r / D r , where 0 r 1.
V x3 s2
ME x2
1 x n np s 1
u x
x1 (1, 0, 0)
D D
Y
F IGURE 2.43. G.1/ D 1 for Enneper’s surface.
What is G.e i / for 0 2 ? If we restrict the domain of Enneper’s surface to

D, these points get mapped to the edge of our image of Enneper’s surface in a positive
direction. At D 0, the unit normal is pointing outward (i.e., away from the opposite
leaf) and under the Gauss map, np , this corresponds to .1; 0; 0/ 2 S 2 . As moves from
0 to 2 , the unit normal moves from pointing outward to pointing inward (i.e., toward
the opposite leaf). At D 2 , the unit normal is mapped under np to .0; 1; 0/ 2 S 2
which projects under to i 2 C (see Figure 2.44). Reflecting across the real axis gives
i V x3 s2 i Y
ME x2
x np s
x
u
x1 (0, –1, 0)
D n D
F IGURE 2.44. G.i / D i for Enneper’s surface.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 130 — #148

i i
.0; 1; 0/ D i D i . A similar argument shows that the same thing happens for all
2 Œ0; 2 . That is, G.e i / D e i ; .0 2 /.
Example 2.83. Let’s determine some values for the Gauss map G for the singly periodic
Scherk surface with six leaves, MS . The domain used is D. The leaves are centered at rays
from the origin through each of the sixth roots of unity (i.e., e i k=3 ; .k D 0; :::; 5/) with the
leaf centered at the positive real axis pointing upward and the subsequent leaves alternating
between pointing downward and upward (see Figure 2.45).
F IGURE 2.45. The singly periodic Scherk surface with six leaves.
As in the previous example, we will use the downward pointing unit normal, and
so G.0/ D 0. Because the leaves are centered at the sixth roots of unity, let’s look at
G.e i k=3 /, where k D 0; :::; 5. The point 1 gets mapped under x to the edge of MS above
the positive real axis. By looking at the graph of MS , we see that this is in the middle of an
upward pointing leaf, and the corresponding unit normal n lies above the positive real axis
and points away from the origin in a plane parallel to the horizontal x1 x2 -plane. Mapping
this normal under np and then , results in the point 1. Hence, G.1/ D 1. Now, consider
G.e i =3/. The point e i =3 gets mapped to the edge of MS above the line r e i =3; r > 0.
Since the leaves alternate between pointing upward and downward, this is in the middle of
a downward pointing leaf and points toward the origin. Mapping it under np and then ,
results in the point e i 4=3 D e i 2=3. Hence, G.e i =3/ D e i 2=3.
In a similar way, we get
2
2 4
G .1/ D 1, G ei 3 D ei 3 , G ei 3 D ei 3 ,
4 2 8
5 4 10
G . 1/ D 1, G ei 3 D ei 3 D ei 3 , G ei 3 D ei 3 D ei 3 .
Exercise 2.84. A picture of the half catenoid on its side defined on D is shown in Figure
2.46 with the positive real axis on the right.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 131 — #149

i i
F IGURE 2.46. A view of the half catenoid on its side.
For the catenoid on its side determine

(a) G.0/ (b) G.1/ (c) G. 1/ (d) G.i / (e) G. i /.
Try it out!
Exercise 2.85. For the 4-noid (see Figure 2.35), determine:
(a) G.0/ (b) G.1/ (c) G. 1/ (d) G.i / (e) G. i /.
Try it out!
We will use the Gauss map G to form another Weierstrass representation of a param-
etrized minimal surface. In doing so, we will need the height differential, dh, which is so
called because it is locally (though not globally) the differential of the height coordinate.
We will not get into the definition of differential forms. They are in [26]. It is worth men-
tioning that at points where the Gauss map is vertical (i.e., G D 0 or G D 1), the height
function ought to have local minimums and maximums. Hence, dh ought to have a zero at
these points (for example, Figure 2.42).
Theorem 2.86 (Weierstrass Representation (G,dh)). Every regular minimal surface has a
local isothermal parametric representation of the form
Z z
1 1 i 1
x D Re G ; C G ; 1 dh; (2.13)
a 2 G 2 G
where G is the Gauss map, dh is the height differential, and a 2 is a constant.
Proof. From the Summary on page 120, we need
2 D 0 and jj2 ¤ 0 and be finite.
Comparing (2.13) and (2.12), we have that

1 1 i 1
'1 dz D G dh; '2 dz D C G dh; '3 dz D dh:
2 G 2 G
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 132 — #150

i i
In Exercise 2.87 you will show that
2 D 0 and jj2 ¤ 0:
Exercise 2.87. Prove that 2 D 0 and jŒj2 ¤ 0 in the proof of Theorem 2.86. Try it out!
We have
'1 C i'2
GD and dh D '3 dz:
'3
One advantage of using the Weierstrass representation with the Gauss map and height
differential is that the complex analytic properties of G and dh are related to the geometry
of a minimal surface. We will discuss this later, but first we will look at some examples.
Here is a list of the Weierstrass data for some common minimal surfaces.
(a) Enneper’s surface G.z/ D z dh D z dz on C.
1
(b) The catenoid G.z/ D z dh D z
dz on C n f0g.
i
(c) The helicoid G.z/ D z dh D z
dz on C n f0g.
z
(d) Scherk’s doubly periodic surface G.z/ D z dh D z4 1
dz on D.
iz
(e) Scherk’s singly periodic surface G.z/ D z dh D dz
z4 1
on D.
(f) Polynomial Enneper G.z/ D p.z/ dh D p.z/ dz on C.
(g) Wavy plane G.z/ D z dh D dz on C n f0g.
Example 2.88. For G.z/ D z k and dh D z k dz, where k D 1; 2; : : :, we get
Z z
1 1 k i 1
x D Re z ; C z ; 1 z k dh
k
0 2 zk 2 zk
1n 1 o 1n 1 o n z kC1 o
D Re z z 2kC1 ; Re i zC z 2kC1 ; Re :
2 2k C 1 2 2k C 1 kC1
This is Enneper’s surface with 2k C 2 leaves described in Exploration 2.71.
Exercise 2.89. The Weierstrass data for the catenoid and the helicoid, which are conjugate
surfaces (see Definition 2.50), have the same Gauss map, G, while the height diffferentials
dh differ by a multiple of i . Prove that this is true for any conjugate surfaces. Try it out!
Exercise 2.90. Let G.z/ D z 4 and dh D z 2 dz.
(a) Using (2.13), compute the parametrization.
(b) This minimal surface has a planar end (i.e., looks like a plane) and an Enneper end.
To graph the surface, use the entry under the Pre-set functions in the W.E.
(G,dh) tab of MinSurfTool.
Try it out!
The catenoid and the surface in Exercise 2.90 are examples of minimal surfaces with
ends. Loosely, an end of a minimal surface is a piece that goes on forever, or, more pre-
cisely, leaves all compact subsets of the minimal surface. From Theorem 2.76, all complete
minimal surfaces in R3 are not compact, and hence they must possess at least one end.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 133 — #151

i i
Exercise 2.91. Determine the number of ends of

(a) the catenoid (b) the plane (c) the helicoid (d) Enneper’s surface.
Try it out!
Ends occur in a deleted neighborhood (i.e., a disk with the center removed) centered at a
singularity. Three common types of ends for minimal surfaces are Enneper ends, catenoidal
ends, and flat or planar ends. In discussing ends, we represent ds, the metric (i.e., a way
to measure distance) on a minimal surface, in terms of G and dh. Using ds 2 D jj2 and
(2.13), we derive
1 1
ds D p jGj C jdhj: (2.14)
2 jGj
An Enneper end has ds jz k j jdzj, so the metric grows in the same manner as a poly-
nomial as we approach the end. A catenoidal end and a planar end have ds jdzj, so the
metric becomes Euclidean. A catenoidal end differs from a planar end because the residue
of dh is logarithmic.
Exercise 2.92. Prove (2.14). Try it out!
Example 2.93. We know that the catenoid has two catenoidal ends. Let’s show how we
could prove. Using the Weierstrass data, G.z/ D z and dh D 1z dz, for the catenoid, we
have the parametrization

1 1 i 1
x.z/ D Re z ; Re C z ; Re log z :
2 z 2 z
There is a singularity or pole of order 1 at 0 and a pole of order 1 at 1. To see that there
is a singularity at 1, we replace z with w1 and look at the limit as w goes to 0. Thus, the
catenoid will have two ends (one at 0 and one at 1). To determine their types, we look at
ds at these points. Because

1 1 1
ds D p jzj C jdzj
2 jzj jzj
as z ! 1, ds jdzj and because x3 is logarithmic, we have a catenoidal end.

At z D 0, substituting 0 does not work, so we let w D 1z and consider w ! 1. We
have 0
1 1 1 1
dh D dz D w Dw 2
dw D dw:
z w w w
Therefore
1 1 1
ds D p jwj C jdwj:
2 jwj jwj
As w ! 1, ds jdwj and because x3 is logarithmic, we have a catenoidal end.
Example 2.94. The surface in Exercise 2.90 has G.z/ D z 4 and dh D z 2 dz, and the
corresponding parametrization is

1 1 7 1 i 1 7 1 1 3
x.z/ D Re z ; Re z C ; Re z :
2 7 z 2 7 z 3
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 134 — #152

i i
There are singularities at 0 and at 1 and

1 1
4
ds D p jzj C 4 jzj2 jdzj:
2 jzj
As z ! 1, ds jzj6 jdzj and so we have an Enneper end.
At z D 0, we let w D 1z and consider w ! 1. Then G.w/ D w 4 and dh D w14 dw.
Hence,
1 1
ds D jwj4 C jdwj:
jwj4 jwj4
As w ! 1, ds jdwj, but because there is no logarithmic term, we have a planar end.
F IGURE 2.47. A minimal surface with a planar end and an Enneper end.
2 z 2 C3
Exercise 2.95. Let G.z/ D zz2C31 and dh D z2 1
dz. Show that the minimal surface has
one planar end and two catenoidal ends.
Try it out!
The Gauss map and height differential also tell us about two important types of curves
on a minimal surface, the asymptotic lines and the curvature lines. To understand what
these lines are, we will review normal curvature and principal directions discussed in Sec-
tion 2. Let p be a point on a curve on a minimal surface M . The tangent vector w and
normal vector n at p form a plane that intersects the surface in a curve, say ˛ (see Figure
2.20). The normal curvature in the direction w is ˛ 00 n and measures how much the surface
bends toward n as you move in the direction of w at point p. An asymptotic line is a curve
that is tangent to a direction in which the normal curvature is zero.
As we rotate the plane through the normal n, we get a set of curves on the surface each
of which has a value for its curvature. The directions in which the normal curvature attains
its absolute maximum and absolute minimum values are known as the principal directions.
Curvature lines are curves that are always tangent to a principal direction.
The relationship between these lines and the Weierstrass data is
dG
A curve z.t/ is an asymptotic line ” .z/ dh.z/ 2 i R
G
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 135 — #153

i i
F IGURE 2.48. The wavy plane.
dG
A curve z.t/ is a curvature line ” .z/ dh.z/ 2 R:
G
Often these equations are difficult to solve, but there are times when we can find the
asymptotic lines and curvature lines at specific points.
Example 2.96. Let G.z/ D z and dh.z/ D dz. This is a parametrization of the wavy
plane. Computing the Weierstrass representation, we get the parametrization

1 1 2 i i 2 ˚
.x1 .z/; x2 .z/; x3 .z// D Re log.z/ z ; Re log.z/ C z ; Re z :
2 4 2 4
We can plot an image of it using the corresponding function under Pre-set functions
in the W.E.(G,dh) tab in MinSurfTool.
For the wavy plane
dG dz
.z/ dh.z/ D dz:
G z
If we let z D e i (since we let the maximum radius be 1), then dz D i e i d and
ˇ
dG ˇ
.z/ dh.z/ˇˇ D e i .d/2 :
G zDe i
So, we get that for k 2 Z asymptotic lines occur at z D r e as r varies with D 2 C k

and curvature lines occur at z D r e as r varies with D k .
If we use MinSurfTool to plot the wavy plane with the minimum and maximum
values of 2 and 2 , we see that the asymptotic lines lie in the x1x2 -plane. If we plot the
minimum and maximum values of 0 and , we see that the curvature lines are reflection
lines through which the wavy plane can be reflected as if through a mirror to get a smooth
continuation of the minimal surface.
Exercise 2.97. Using G.z/ D z and dh D zdz for Enneper’s surface, show that asymp-
totic lines occur at z D r e as r varies with D 4 C k
2
, where k 2 Z, and curvature
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 136 — #154

i i
k
lines occur at z D r e as r varies with D 2
, where k 2 Z. Use MinSurfTool to plot the
lines on Enneper’s surface. Try it out!
Exercise 2.98. Prove that for conjugate surfaces, the asymptotic lines (curvature lines) of
one surface are the curvature lines (asymptotic lines) of the other surface. Try it out!
As mentioned, an advantage of using the Gauss map and height differential is that the
complex analytic properties of G and dh are related to the geometry of a minimal surface.
So, let’s look at the complex analytic properties of G and dh. From the Summary on page
120, we need jj2 to be finite and nonzero. Because of (2.14), we have the following
proposition.
Proposition 2.99. At a nonsingular point, G has a zero or pole of order n if and only if
dh has a zero of order n.
The integrals in the Weierstrass representation in (2.13) might depend upon the path of
integration if the domain of G and dh is not simply connected. If the representation does
not depend on the path of integration, then all closed paths in the domain satisfy the three
conditions
Z
1 1
Re G dh D 0
2 G
Z
i 1
Re C G dh D 0
2 G
Z
Re dh D 0:

If a curve has the property that one of these quantities is a nonzero constant, then the
surface will have periodic behavior. Example 2.101 illustrates this.
The three equations can be reduced to the two period conditions
Z Z
1
.i / G dh D dh (horizontal period condition);
G
Z (2.15)
.i i / Re dh D 0 (vertical period condition)

for closed paths in the domain.

Exercise 2.100. Show that the conditions
Z Z
1 1 i 1
Re G dh D 0 and Re C G dh D 0
2 G 2 G
are equivalent to
Z Z
1
G dh D dh:
G
Try it out!
The period conditions are useful in determining horizontal periods (e.g., Scherk’s dou-
bly periodic surface), vertical periods (e.g., Scherk’s singly periodic surface), and in deter-
mining constant values in G and dh.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 137 — #155

i i
Example 2.101. Scherk’s doubly periodic surface has the Weierstrass data G.z/ D z
R R
and dh.z/ D z4z 1 dz. The horizontal period condition is G dh D G1 dh for all
closed paths in the domain. Both integrands are meromorphic with poles of order 1 at
˙1; ˙i . So, the only paths that concern us are ones that enclose one, two, or three of these
poles. To calculate the integrals we use the residue theorem that states if is a simple
closed positively-oriented contour and f is analytic inside and on except at the points
z1 ; : : : ; zn inside , then
Z n
X
f .z/ dz D 2 i Res.f; zj /:
j D1
For poles of order 1,

Res.f; zj / D lim .z zj /f .z/:
z!zj
Z
Thus, for G dh, we have

z 3 zj z 2 3z 2 2zj z 3z 3 2zj z 2 zj3

Res.G dh; zj / D lim D lim D lim D :
z!zj z4 1 z!zj 4z 3 z!zj 4z 4 4
In particular,
1 i
Res.G dh; 1/ D Res.G dh; i / D
4 4
1 i
Res.G dh; 1/ D Res.G dh; i / D :
4 4
Similarly, we can compute that

1 1 1 i
Res dh; 1 D Res dh; i D
G 4 G 4

1 1 1 i
Res dh; 1 D Res dh; i D :
G 4 G 4
Now, if the path 1 contains only the pole at z1 D 1, then the horizontal period conditions
result in
Z Z
i 1 1 i
G dh D 2 i Res.G dh; 1/ D ; dh D 2 i Res dh; 1 D :
1 2 1 G G 2
The integrals should be equal, which occurs if the minimal surface is periodic in the imag-
inary direction with period . Likewise, if we take a path 2 that contains only the pole
z2 D i , then we get
Z Z
1
G dh D ; dh D ;
2 2 2 G 2
and the minimal surface is periodic in the real direction with period . All other paths
are covered by these two cases. If we look at the vertical period condition, we get that the
condition is automatically true for all paths and so the minimal surface is not periodic
in the vertical direction. This matches the image of Scherk’s doubly periodic surface from
MinSurfTool.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 138 — #156

i i
Exercise 2.102. Show that the period conditions given in (2.15) result in Scherk’s singly
periodic surface being periodic in the vertical direction. Try it out!
Let’s look at an example of how this can help us use the geometry of a minimal surface
to determine G and dh.
Example 2.103. From the Weierstrass data on page 132, we know that G.z/ D z and
dh D z dz for Enneper’s surface. We want to show how it can be determined by using the
geometric shape of the surface. Let’s determine a plausible candidate for G by making a
guess based on the value of G at a few points. From Example 2.82, we know that G.0/ D 0,
G.r / D r for 0 r 1, and G.e i / D e i for 0 2 . Therefore, it seems plausible
to let G.z/ D z. With this G, let’s determine dh. Because ds must be finite in (2.14)
and Enneper’s surface has no ends in C, dh cannot have poles in C. However, from the
sentence before Exercise 2.91, we know that Enneper’s surface must have at least one
end. It corresponds to the point at infinity, z D 1, and so dh has a pole at 1. Thus,
dh D z n dz for some n 2 N and 2 C. Since G.z/ D z has a zero of order 1
at 0, by Proposition 2.99 dh must have a zero of order 1 at 0 and no other zeros. Thus,
dh D z dz. For simplicity, we let D 1. The period conditions in (2.15) hold because
there are no poles in C, and so every integral along any closed path will equal 0. Hence,
the Weierstrass data
G.z/ D z; dh D z dz
generate a minimal surface.
Example 2.104. The singly periodic Scherk surface, MS , with six leaves that go off to
infinity (see Figure 2.45) will have six poles. Because of symmetry, we choose them to
be at the sixth roots of unity (i.e., e i k=3 ; .k D 0; :::; 5/). This means that dh will have
z 6 1 in its denominator. However, we will need to determine G first to know what should
be in the numerator of dh. From the results in Example 2.83, it seems reasonable that
G.z/ D z 2 . Since G.z/ D z 2 has a zero of order 2 at 0, by Proposition 2.99 dh must also
have a zero of order 2 at 0 and no other zeros. Thus, we so far have
z2
G.z/ D z 2 ; dh D dz;
z6 1
where 2 C. To determine values of , consider the period conditions in (2.15). There are
poles of order 1 at e i k=3 ; k D 0; : : : ; 5. We compute that

zj5 1 zj
Res.G dh; zj / D ; Res dh; zj D :
6 G 6
Hence, if contains the pole zj , then the horizontal period condition requires
Z
izj5
G dh D 2 i Res.G dh; zj / D
3
and
Z
1 1 i zj
dh D 2 i Res dh; zj D
G G 3
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 139 — #157

i i
2.6. Minimal surfaces and harmonic univalent mappings 139
to be equal (since there is no periodicity of MS in the horizontal direction).The integrals

will be equal for the poles zj5 D zj if D . That is, is purely imaginary. Without loss
of generality, we let D i and check that the vertical period condition holds. Hence, we
have that the Weierstrass data for singly periodic Scherk surface with six leaves is
iz 2
G.z/ D z 2 ; dh D dz:
z6 1
Exercise 2.105. Let M be Enneper’s surface with eight leaves. Using the approach of
Example 2.103 determine G and dh. Try it out!
Exercise 2.106. Let M be the 3-noid with ends symmetrically placed so that if the surface
is rotated by 2
3
it is unchanged. Determine G and dh. Try it out!
Small Project 2.107. Let M be a minimal surface that has six symmetrically-placed
catenoidal ends with four ends along the side (like a 4-noid), one end on the top, and
one end on the bottom. So, M will look the same if it is rotated horizontally by 2 or if it
is rotated vertically by 2 . Determine G and dh. Optional
Small Project 2.108. For Scherk’s singly periodic surface the four ends are symmetrically
placed so that if the surface is rotated by 2 you will get the same image. This is because the
denominator of dh is z 4 1, which has zeros equally spaced on the unit circle. It is possible
to create a variation of Scherk’s singly periodic surface that has four ends with rotational
symmetry of . That is, if the ends are labelled E1 ; : : : ; E4, then E2 will be closer to E1
than to E3 (and likewise, E4 will be closer to E3 than to E1 ) and if the surface is rotated
by you will get the same image. Determine G and dh. Optional
Large Project 2.109. Describe and classify the minimal surfaces where one of the coor-
dinates of the parametrization is fixed while the other two vary. For example, if x3 is fixed
to a specific function, what are the possible coordinate functions for x1 and x2 ? Generalize
this approach. Optional
Large Project 2.110. Describe and classify the minimal surfaces with G.z/ D z m and
dh D z n dz, for all n; m 2 N (see Example 2.88 and Example 2.94). There are several
cases to consider. Determine how to separate m; n into the cases remembering to discuss
types of surface, types of ends, lines of symmetry, etc. Optional
2.6 Minimal surfaces and harmonic univalent mappings

In the Summary on page 120, we learned that each coordinate
R function of the parametriz-
ation x of a minimal surface had the form xk D Re 'k dz with 'k analytic. Since the
real part of an analytic function is a harmonic function, we see that each xk is harmonic.
From the Summary, we have that .'1 /2 C .'2 /2 C .'3 /2 D 0. This means that if we know
'1 and '2 , we can determine '3 . So, a way to get a Weierstrass representation for minimal
surfaces is to use two harmonic functions x1 and x2 . Thus, we can investigate minimal
surfaces by studying harmonic mappings in the complex plane. They are known as planar
harmonic mappings and have been studied independently of minimal surfaces.
In this section we will develop another Weierstrass representation. We will use pla-
nar harmonic mappings instead of p and q as in Section 2.4 or G and dh as in Section
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 140 — #158

i i
2.5. What benefits do we obtain from this new approach? First, we can use results about
planar harmonic mappings to prove results about minimal surfaces. Since these two ar-
eas of mathematics developed independently, this approach can be fruitful. Second, in the
study of planar harmonic mappings, much of the work deals with one-to-one functions.
Geometrically, this means that the image f .G/ will not overlap or intersect itself. When
one-to-one functions are lifted from the complex plane into R3 , the resulting surfaces are
minimal graphs and hence have no self-intersections. This can be useful in establishing
the embeddedness of the minimal surfaces. For example, the one-to-one planar harmonic
function given by

i i Cz 1 1Cz
f .z/ D h.z/ C g.z/ D Re log C i Im log
2 i z 2 1 z
maps the unit disk onto a square that is the projection of Scherk’s doubly periodic surface
onto the plane. So, we can lift f from C into R3 to get Scherk’s doubly periodic sur-
face. Planar harmonic mappings that are one-to-one are also known as harmonic univalent
mappings, which can be studied on their own as is done in Chapter 4.
F IGURE 2.49. The image of f .D/ and Scherk’s doubly periodic surface.
Exercise 2.111.
(a) Although minimal graphs are embedded, the converse is not true. Give an example of
an embedded minimal surface that is not a minimal graph.
(b) Suppose you have a nonunivalent harmonic mapping. Why could it not be the projec-
tion of a minimal graph?
Try it out!
A planar harmonic mapping is a function f D u.x; y/ C iv.x; y/ where u and v
are real harmonic functions. Planar harmonic mappings are more general than analytic
functions, because we do not require u and v to be harmonic conjugates. The next theorem
allows us to relate planar harmonic mappings to analytic functions. We will assume that
the domain of f is the unit disk, D.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 141 — #159

i i
Theorem 2.112. Let a function f D u C iv, where u and v are real harmonic functions.
If D is a simply-connected domain and f W D ! C, then there exist analytic functions h
and g such that f D h C g.
Exercise 2.113.

(a) Show that f .x; y/ D u.x; y/ C iv.x; y/ D 13 x 3 xy 2 C x C i 13 y 3 x 2 y C y
is complex-valued harmonic by showing that u and v are real harmonic functions.

(b) Using x D 12 .z C z/ and y D 2i 1
.z z/, write f .x; y/ D 13 x 3 xy 2 C x C

i 13 y 3 x 2 y C y in terms of z and z.
(c) Determine the analytic functions h and g such that f D h C g.
Try it out!
Example 2.114. In the previous exercise, we saw that the planar harmonic map f W D !
C defined by

1 3 1 3
f .x; y/ D u.x; y/ C iv.x; y/ D x xy 2 C x C i y x 2y C y
3 3
can be written as
1
f .z/ D h.z/ C g.z/ D z C z 3 :
3
The image of D under f is a hypocycloid with four cusps. This can be seen by considering
f .e i / D u./ C iv./ and comparing the component functions, u./ and v./, to the
parametrized equation for a hypocycloid with four cusps. To help us visualize the image,
we can use the applet ComplexTool. To graph the image of D under the harmonic function
f .z/ D z C 13 z 3 , enter f in ComplexTool in the form z C 1=3 conj .z ^ 3/ (see Figure
2.50). We will later show that f is related to a minimal graph.
F IGURE 2.50. Image of D under the harmonic function f .z/ D z C 31 z 3 .
The harmonic function f .z/ D h.z/ C g.z/ can also be written as

˚ ˚
f .z/ D Re h.z/ C g.z/ C i Im h.z/ g.z/ (2.16)
because
1 1
Refh C gg D Œ.h C g/ C .h C g/ and Imfh gg D Œ.h g/ .h g/:
2 2i
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 142 — #160

i i

˚
Hence, in the previous example f .z/ D z C 31 z 3 can be written as f .z/ D Re z C 13 z 3 C
˚
i Im z 31 z 3 .
We are interested in harmonic functions that are one-to-one or univalent, because this
is a necessary condition to lift the harmonic mapping to a minimal graph. Theorem 2.119
establishes univalency, but it requires background material.
Definition 2.115. The dilatation of f D h C g is !.z/ D g0 .z/= h0 .z/.
Theorem 2.116 (Lewy [16]). f D h C g is locally univalent and orientation-preserving
if and only if jg0 .z/= h0 .z/j < 1, for all z 2 D.
See Chapter 4 for more background on the dilatation and on orientation-preserving
functions.
Exercise 2.117. Show that if z 2 D, then j!.z/j < 1 for
(a) !1 .z/ D e i z, where 2 R
(b) !2 .z/ D z n , where n D 1; 2; 3; : : :
zCa
(c) !3 .z/ D , where jaj < 1
1 C az
(d) !4 .z/, the composition of any of !1 , !2 , and !3 .
Try it out!
Creating nontrivial examples of harmonic univalent mappings that lift to minimal graphs
is not easy. One way to do this is to use the shearing technique of Clunie and Sheil-Small.
Before we proceed, we need to discuss a certain type of domain.
Definition 2.118. A domain is convex in the direction of the real axis (or convex in the
horizontal direction, CHD) if every line parallel to the real axis has a connected intersection
with .
CHD not CHD
Theorem 2.119 (Clunie and Sheil-Small). A harmonic function f D h C g locally uni-

valent in D is a univalent mapping of D onto a CHD domain ” h g is an analytic
univalent mapping of D onto a CHD domain.
Remark 2.120. This is known as the shear method, or shearing a function. In our situation,
suppose F D h g is an analytic univalent function convex in the real direction. Then the
corresponding harmonic shear is
f D hC g Dh gCgC g Dh g C 2 Refgg:
The harmonic shear differs from the analytic function by the addition of a real function.
Geometrically, you can think of this as taking F , the original analytic univalent function
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 143 — #161

i i
convex in the real direction, and cutting it into thin horizontal slices that are then translated
or scaled in a continuous way to form the corresponding harmonic function, f . This is why
the method is called shearing. Since F is univalent and convex in the real direction and we
are only adding a continuous real function to it, the univalency is preserved.
Example 2.121. Let
1 1Cz
h.z/ g.z/ D log ; (2.17)
2 1 z
which is an analytic function that maps D onto a horizontal strip convex in the direction of
the real axis (see Figure 2.51).

1 1Cz
F IGURE 2.51. Image of D under the analytic function log .
2 1 z
Let
!.z/ D g0 .z/= h0 .z/ D z 2 :
Applying the shearing method from Theorem 2.119 with the substitution g0 .z/ D z 2 h0 .z/,
we have
1 1
h0 .z/ g0 .z/ D ) h0 .z/ C z 2 h0 .z/ D
1 z2 1 z2
1 1 1 1 i i
) h0 .z/ D D C C C :
1 z4 4 1Cz 1 z i Cz i z
Integrating h0 .z/ and normalizing so that h.0/ D 0 yields

1 1Cz i i Cz
h.z/ D log C log : (2.18)
4 1 z 4 i z
We can use the same method to solve for normalized g.z/, where g.0/ D 0. We can also
find g.z/ by using (4.11) and (4.12). Either way, we get

1 1Cz i i Cz
g.z/ D log C log :
4 1 z 4 i z
So
i i Cz 1 1Cz
f .z/ D h.z/ C g.z/ D Re log C i Im log :
2 i z 2 1 z
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 144 — #162

i i
What is f .D/? We have

1 i Cz 1 1Cz
f .z/ D arg Ci arg D u C iv:
2 i z 2 1 z
Let z D e i 2 @D. Then
i Cz i C e i i e i
1 i.e i C e i
/ 1 cos
D D D i :
i z i e i i e i 1 C i.e i e i / C 1 1 sin
Thus,
ˇ (
1 i C z ˇˇ
4 if cos > 0;
uD arg ˇ D
2 i z ˇ
if cos < 0:
zDe i 4
Likewise, we can show that

(

4 if sin > 0;
vD
4
if sin < 0:
Therefore, we have that z D e i 2 @D is mapped to

8̂

ˆ z1 D p e i 4 D 4 C i 4 if 2 .0; 2 /
ˆ
ˆ 2 2
ˆ
<z2 D p 3
2 2
e i 4 D 4 C i 4 if 2 . 2 ; /
u C iv D 5
ˆz3 D p
ˆ

e i 4 D 4 i 4 if 2 .; 3 /
ˆ
ˆ 2 2 2
:̂z D p 7
ei 4 D i if 2 . 3 ; 2 /:
4 4 4 2
2 2
Thus, the harmonic function maps D onto the interior of the region bounded by a square
with vertices at z1 ; z2 ; z3, and z4 .

i i Cz 1 1Cz
F IGURE 2.52. Image of D under f .z/ D Re log C i Im log :
2 i z 2 1 z
1 3
Exercise 2.122. Verify that shearing h.z/ g.z/ D z 3
z with !.z/ D z 2 yields
f .z/ D z C 13 z 3 . Try it out!
To find the minimal graph that is associated with specific types of harmonic univalent
mappings we need to develop the appropriate Weierstrass representation as outlined in
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 145 — #163

i i
(2.12). It must satisfy 2 D 0 and jj2 ¤ 0, and we want it to use planar harmonic
mappings. A natural choice is to take
Z Z
x1 D Re.h C g/ D Re .h0 C g0 / dz D Re '1 dz
Z Z
0 0
x2 D Im.h g/ D Re i.h g / dz D Re '2 dz
Z
x3 D Re '3 dz
and then solve for '3 .

p p
Exercise 2.123. Derive '3 D 2ih0 g0 = h0 D 2i g0 h0 . Try it out!
We need '3 to be analytic and so we require the dilatation ! D g0 = h0 to be a perfect
square.
Theorem 2.124 (Weierstrass Representation (h; g)). If f D h C g is a sense-preserving
harmonic univalent mapping of D onto a domain C with dilatation ! D q 2 for a
function q analytic in D, then the isothermal parametrization

x.u; v/ D x1 ; x2; x3
nZ z p o
D Refh.z/ C g.z/g; Imfh.z/ g.z/g; 2 Im g0 ./h0 ./d
0
defines a minimal graph whose projection onto the complex plane is f . Conversely, if
˚
a minimal graph x.u; v/ D .u; v; F .u; v// W u C iv 2 is parametrized by sense-
preserving isothermal parameters z D x C iy 2 D, then the projection onto its base plane
defines a harmonic univalent mapping f .z/ D u C iv D Refh.z/ C g.z/g C i Imfh.z/
g.z/g of D onto whose dilatation is the square of an analytic function.
Summary: Let f D h C g defined on D be a harmonic univalent mapping such that
the dilatation ! D g0 = h0 is the square of an analytic function and j!.z/j < 1 for all z 2 D.
Then f lifts to a minimal graph using the Weierstrass formula given in Theorem 2.124.
Example 2.125. In Example 2.114 we had the harmonic univalent mapping

1 3 1 3 1 3
f .z/ D z C z D Re z C z C i Im z z ;
3 3 3
so h.z/ D z and g.z/ D 13 z 3 . Also, !.z/ D z 2 , the square of an analytic function. Hence
this harmonic mapping lifts to a minimal graph. We compute that
Z zp

x3 D 2 Im h0 ./g0 ./ d D Im z 2 :
0
This yields a parametrization of a surface that is the conjugate of Enneper’s surface given
in Example 2.67

1 1 3 ˚
x D Re z C z 3 ; Im z z ; Im z 2
3 3
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 146 — #164

i i
F IGURE 2.53. Side view of Enneper’s surface and the image of the unit disk under the harmonic
map.
and hence yields Enneper’s surface. The projection of Enneper’s surface onto the x1 x2 -
plane is the image of D under the harmonic mapping f . While Enneper’s surface is not
a graph over C, it is a graph over D as this result proves. You can see this by using Min-
SurfTool with the W.E. (h,g) tab. Enter the functions h.z/ D z, g.z/ D 31 z 3 , and
p
h0 g0 D z. Make sure the Surface and projection button is clicked allowing
you to see the minimal surface and its projection that is related to the harmonic mapping.
For the Disk domain use the unit disk (i.e., radius min = 0; radius max = 1; theta min
= 0; theta max = 2 ). The minimal surface is colored purple while f .D/ is colored green
(see Figure 2.53). As you move the image so that it is viewed from the top, the projection
of the minimal surface matches the image of f .D/ (see Figure 2.54).
F IGURE 2.54. The projection of Enneper’s surface is the image of the unit disk under the harmonic
map.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 147 — #165

i i
0 0
Exploration 2.126. In the Weierstrass representation (h; g), we require that
0
p ! D g =h
be the square of an analytic function. This is necessary because '3 D ih g0 = h0 , and if
g0 = h0 were not the square of an analytic function, then there would be two branches of
the square root. Geometrically, we can see that this is necessary. Use MinSurfTool with the
W.E. (h,g) tab to graph the following images and describe why the geometry of the
functions f D h C g in the left column lift to a minimal graph while those in the right
p
column do not. To enter z into the applet type sqrt(z).
(a) z C 13 z 3 p (b) z C 12 z 2 p p
(! D z 2 and h0 g0 D z) (! D z and h0 g0 D z)
(c) z 51 z 5 (d) z 14 z 4 . p
p p
(! D z 4 and h0 g0 D iz 2 ) (! D z 5 and h0 g0 D i z 5 )
Example 2.127. Consider the harmonic univalent mapping from Example 2.121 given by

i i Cz 1 1Cz
f .z/ D h.z/ C g.z/ D Re log C i Im log :
2 i z 2 1 z
Because !.z/ D z 2 is the square of an analytic function, we can lift it to a minimal

graph. We compute that
Z z
iz 1 1 C z2
x3 D 2 Im d D Im i log :
0 1 z4 2 1 z2
This yields a parametrization of Scherk’s doubly periodic minimal surface:

i i Cz 1 1Cz 1 1 C z2
x D Re log ; Im log ; Im i log :
2 i z 2 1 z 2 1 z2
We can use Pre-set functions from the W.E. (h,g) tab on MinSurfTool to plot
the minimal graph and the image of the unit disk under the planar harmonic mapping,
where

1 1Cz i i Cz
h.z/ D log C log ;
4 1 z 4 i z

1 1Cz i i Cz
g.z/ D log C log :
4 1 z 4 i z
Because of the singularities at ˙1; ˙i , the radius max is set to 0:999. The projection of
Scherk’s doubly periodic surface onto the x1 x2 -plane is a square that is the image of D
under the harmonic mapping f .
Exploration 2.128. Use the Pre-set functions from the W.E. (h,g) tab on
MinSurfTool to plot the minimal graphs associated with the functions h and g for the planar
harmonic mappings. Determine which minimal surfaces these are.
1 2nC1
(a) h.z/ D z, g.z/ D 2nC1 z (n D 1; 2; 3; : : :)
1
(b) h.z/ D z, g.z/ D z
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 148 — #166

i i
F IGURE 2.55. Side view of Scherk’s doubly periodic surface and the image of the unit disk under
the harmonic map.
1
(c) h.z/ D z, g.z/ D z
(d) h.z/ D 15 z 5 , g.z/ D 1

z
1 i Cz
i 1Cz
1 i Cz
i 1Cz

(e) h.z/ D 4
log i z 4
log 1 z
, g.z/ D 4
log i z
C 4
log 1 z
.
Try it out!
If we start with a harmonic univalent mapping that has a perfect square dilatation we
can use Theorem 2.124 to find the parametrization of the corresponding minimal graph. In
Example 2.125, we saw that the minimal graph corresponding to the harmonic univalent
map f .z/ D z C 13 z 3 is Enneper’s surface, and in Example 2.127 the minimal graph
corresponding to the harmonic univalent map f .z/ D ReŒ 2i log. iiCzz / C i ImŒ 12 log. 1Cz
1 z
/
is Scherk’s doubly periodic surface. There have been several research papers that have used
Theorem 2.124 to create minimal graphs from harmonic univalent mappings (e.g., [8], [9],
[10], [11], [17]). However, many of them have not identified the specific minimal graph
created, because it is often not as easy to as it was in Examples 2.125 and 2.127.
Question: Given a harmonic univalent mapping we can use Theorem 2.124 to find the
parametrization of a minimal graph. Can we determine which minimal graph this is?
The following examples show how we can determine the minimal graph when the
parametrization is not a standard parametrization for a known minimal surface.
Example 2.129. The analytic function
z
F .z/ D
1 z
is an important and interesting function in complex analysis. It maps D onto the right
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 149 — #167

i i

1 zC1 z z
F IGURE 2.56. Image of D under f D Re 4 log z 1 C 2.1 z/2
C Im 1 z .
ˇ
half-plane fz 2 C ˇ Refzg > 0g. If we shear
z
h.z/ g.z/ D with !.z/ D g0 .z/= h0 .z/ D z 2
1 z
then we get the harmonic univalent mapping f D h C g, where

1 zC1 3z 2z 2 1 zC1 z 2z 2
h D log C and g D log :
8 z 1 4.1 z/2 8 z 1 4.1 z/2
The image of D under f D h C g is shown in Figure 2.56.
Because !.z/ D z 2 , we can use Theorem 2.124 to find the parametrization of the
corresponding minimal graph:

1 zC1 z z 1 zC1 z
xD Re log C ; Im ; Im log :
4 z 1 2.1 z/2 1 z 4 z 1 2.1 z/2
What minimal graph is this? It is difficult to tell. The coordinate functions do not look like
a standard parametrization for a known minimal surface. Let us use a substitution to rewrite
the parametrization into a form that allows us to identify the minimal graph. If we use a
Möbius transformation for the substitution, it will not affect the geometry of the minimal
O
graph. Letting z 7! zC1
zO 1
, we get the parametrization

1 1 1 1 ˚ 1 1 2 1
xO D Re log .Oz / C zO 2 ; Im zO ; Im log .Oz / zO C :
4 2 2 2 4 2 2
This is useful, because it simplifies the log terms in x1 and x3. By switching the coordinate
functions and factoring out 21 we have

1 1 1 2 1 1 1 2 1
xD
e Im log .e
z/ e
z ; z/ C e
Re log .e z ; ŒIm fe
zg ;
2 2 2 2 2 2 2
(2.19)
and this looks close to the standard parametrization for the wavy plane. In Figure 2.57 we
have used the Complex P. Surf tab in MinSurfTool to graph the image of the surface
with this parametrization.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 150 — #168

i i
F IGURE 2.57. The conjugate of the wavy plane.
The coordinate functions correspond to the conjugate surface of the wavy plane scaled
by 21 . This is clear given the actual coordinates of the wavy plane

1 1 2 1 1
WD Re log .z/ z ; Im log .z/ C z 2 ; Re fzg :
2 2 2 2
Since the wavy plane is its own conjugate surface, this means that we can describe our
original minimal graph as the wavy plane.
How does this match with the image in Figure 2.56 of D under the original harmonic
univalent map? In deriving the parametrization of the wavy plane, we applied the trans-
O
formation z 7! zC1
zO 1
, switched the coordinate functions, and took the conjugate function.
These changes are equivalent to altering the original domain D, so we are looking at a
piece of the wavy plane coming from a different domain G. The projection of the wavy
plane coming from G is shown in Figure 2.58. There is similarity between the image in
Figure 2.56 and the image in Figure 2.58.
F IGURE 2.58. The wavy plane mapped from domain G.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 151 — #169

i i
z
Exercise 2.130. An important function in complex analysis is the Koebe function .1 z/2
.
By shearing
z
h.z/ g.z/ D with !.z/ D z 2 ;
.1 z/2
we derive the harmonic univalent mapping f D h C g, where
z z 2 C 31 z 3 1 3
z
hD 3
and g D 3 3 :
.1 z/ .1 z/
(a) Use Theorem 2.124 to find the parametrization of the minimal graph that f lifts to.
(b) Use the approach of Example 2.129 to show analytically that the minimal graph is
Enneper’s surface.
(c) Use ComplexTool to graph the image of D under f and MinSurfTool to sketch the
corresponding minimal graph.
Try it out!

Example 2.131. By shearing h.z/ g.z/ D 12 log 1Cz 1 z
with !.z/ D g0 .z/= h0 .z/ D
2 2
m z , where jmj 1, it was shown in [9] that the harmonic function f D h C g is
univalent, where

1 1Cz m 1 C mz
h.z/ D log C log
2.1 m2 / 1 z 2.m2 1/ 1 mz

m2 1Cz m 1 C mz
g.z/ D log C log :
2.1 m2 / 1 z 2.m2 1/ 1 mz

When m D e i 2 , the function f is the same as in Example 2.121 and the image of D under
f D h C g is a square. For every m with jmj D 1, the image of D under f D h C g is a
parallelogram.

F IGURE 2.59. Image of D under f D h C g when m D ei 2 .
Since !.z/ D g0 .z/= h0 .z/ D m2 z 2 , we can lift f to a minimal graph. We compute

that
m 1 m2 z 2
x3 D Im log :
1 m2 1 z2
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 152 — #170

i i

F IGURE 2.60. Image of D under f D h C g when m D ei 4 .
Hence, the corresponding parametrization of the minimal graph is

1 C m2 1Cz m 1 C mz
x D Re log C log ;
2.1 m2 / 1 z .m2 1/ 1 mz
!
1 1Cz m 1 m2 z 2
Im log ; Im log :
2 1 z 1 m2 1 z2

When m D e i 2 , the minimal graph is Scherk’s doubly periodic surface (see Figure 2.61).

F IGURE 2.61. Scherk’s doubly periodic square surface (m D ei 2 ).
For m D e i ; .0 < < 2 /, the minimal graphs are slanted Scherk’s surfaces (see
Figure 2.62).
What is the minimal graph when m D 1? In the limit (i.e., D 0) we have the equation

z i 1Cz iz 2
x D Re ; Re log ; Re :
1 z2 2 1 z 1 z2
This parametrization does not look like a standard parametrization of a known minimal
surface. We can use a substitution to rewrite it in a form that allows us to identify the
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 153 — #171

i i

F IGURE 2.62. A Scherk’s doubly periodic parallelogram surface (m D ei 4 ).

ez 1 iz 2
minimal graph. Using the substitution z 7 ! e z C1
and the fact that Re D
1 z2
1 1 C z2
Re , the equation is equivalent to
2i 1 z 2

1 1 1
XD sinh u cos v; v; sinh u sin v ;
2 2 2
which is an equation of a helicoid. Thus, we have a family of minimal surfaces that trans-
form Scherk’s doubly periodic surface continuously into the helicoid. This is neat!
F IGURE 2.63. Side view of the helicoid that is the limit function of the slanted Scherk’s surfaces.

iz 2 1 1 C z2
Exercise 2.132. Show that Re D Re . Try it out!
1 z2 2i 1 z 2
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 154 — #172

i i
Exercise 2.133. Consider the harmonic univalent map f .z/ D h.z/ C g.z/, where
1 1
1 1Cz 2
z 1 1Cz 2
z
4 1 z 1 z2 4 1 z 1 z2
(b) Use ComplexTool to graph the image of D under f and MinSurfTool with tab
W.E.(h,g) to sketch the corresponding minimal graph.
(c) Use the approach of Example 2.131 to show analytically that the minimal graph is the
catenoid.
Try it out!
Large Project 2.134. The analytic function F .z/ D z maps the unit disk D onto itself.
Shear h.z/ g.z/ D z with various perfect-square dilatations, !, that satisfy the
condition
2n i 2 z a 2
j!j < 1 for all z 2 D (e.g., ! D z .n 2 N/; ! D e z . 2 R/; ! D 1 az .jaj < 1/).
Determine the corresponding minimal graphs. Optional
Large Project 2.135. The analytic function F .z/ D 1 z z maps the unit disk D onto a right
half-plane and is an important function. Shear h.z/ C g.z/ D 1 z z with various perfect-
square dilatations, !, that satisfy the condition j!j < 1 for all z 2 D (e.g., ! D z 2n .n 2

a 2
N/; ! D e i z 2 . 2 R/; ! D 1z az .jaj < 1/). Determine the corresponding minimal
graphs. Optional
2.7 Conclusion
We have presented an introduction to minimal surfaces and described a few topics that
you can explore using the exercises, the exploratory problems, and the projects along with
the applets. For a deeper and thorough explanation of differential geometry consult [7],
[18], or [21] for beginners, and [3] for intermediates. Also, you should consider Spivak’s
five volume work [24]. Oprea’s book [22] is a nice source for an introduction to minimal
surfaces. For more background on minimal surfaces we recommend [25], [14], [15], [6],
[23], and [20].
2.8 Additional exercises

Differential geometry
Exploration 2.136. An oblique cylinder can be parametrized by
x.u; v/ D .cos u; sin u C v cos ; v sin /;

where 2 0; 2 is a fixed value. Use DiffGeomTool to explore what happens to it as
varies between 0 and 2 .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 155 — #173

i i
2.8. Additional exercises 155
Exercise 2.137. Use DiffGeomTool to graph the surface parametrized by

1 1 1
x.u; v/ D cos u 1 C v sin u ; sin u 1 C v sin u ; v cos u ;
2 2 2
where < u < , 12 < v < 12 . This surface is a Möbius strip and is nonorientable;
that is, the normal vector can change from pointing outward to pointing inward as it travels
along a closed path on the surface. You can see this in DiffGeomTool by clicking on the
Normal vector box and setting the Point locator: (u,v)= to . 0:1; 0/.
Next, change the u coordinate to u D 0:1 1, u D 0:1 2, u D 0:1 3,
u D 0:1 4, u D 0:1 5, and u D 0:1 6. Observe that n will make nearly
a complete path along a closed curve but it will change its direction.
Exercise 2.138. Describe the u-parameter and v-parameter curves on Enneper’s surface.
Exercise 2.139. In Exploration 2.11(c), you determined the largest value of r for which
Enneper’s surface has no self-intersections assuming that the intersection occurs on the
x3 -axis. In this exercise, prove the result without making that assumption.
Exercise 2.140. Compute the coefficients of the first and the second fundamental forms
for Enneper’s surface whose parametrization is

1 3 2 1 3 2 2 2
x.u; v/ D u u C uv ; v v C u v; u v :
3 3
Exercise 2.141. A CMC (Constant Mean Curvature) surface is a surface that has the
same mean curvature everywhere. Minimal surfaces are a subset of CMC surfaces. Using
DiffGeomTool sketch the following surfaces and determine which are CMC surfaces:
(a) x.u; v/ D .u v; u C v; 2.u2 C v 2 //, where 1 < u < 1, 1 < v < 1
(b) x.u; v/ D .cos u; sin u; v/, where < u < , 2 < v < 2

(c) x.u; v/ D .2 C cos v/ cos u; .2 C cos v/ sin u; sin v ; where 0 < u; v < 2
p p
(d) x.u; v/ D . 1 u2 cos v; 1 u2 sin v; u/, where 1 < u < 1, < v < .
Minimal surfaces
Exercise 2.142. Use (2.4) to show that Enneper’s surface parametrized by

1 3 1 3
x.u; v/ D u u C uv 2 ; v v C u2 v; u2 v 2
3 3
is a minimal surface.
Exercise 2.143. Prove Theorem 2.39 when the surface of revolution has the parametriz-
ation
x.u; v/ D f .v/ cos u; f .v/ sin u; v/:
Exercise 2.144. An oblique cylinder is a cylinder whose side forms an angle with the
x1 x2 -plane, where 0 < 2 . For a fixed it can be parametrized by
x.u; v/ D .cos u; sin u C v cos ; v sin /:
Determine the values of for which x is isothermal.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 156 — #174

i i
Exercise 2.145. Show that the parametrization

2u 2v
x.u; v/ D arctan ; arctan ;
1 .u2 C v 2 / 1 .u2 C v 2 /
2 !
1 .u v 2 C 1/2 C 4u2 v 2
ln
2 .u2 v 2 1/2 C 4u2 v 2
is an isothermal parametrization of Scherk’s doubly periodic surface (that is, show that it
is isothermal and that there is transformation that maps it to the parametrization given in
Exercise 2.43(b) for Scherk’s doubly periodic surface).
Exercise 2.146. Let X be a minimal surface that is not a plane and let Y be its conjugate
minimal surface. Is it possible that the plane is one of the associated surfaces for X and Y ?
If so, describe the geometry of X. If not, explain why.
Weierstrass representation
Exercise 2.147. In Example 2.67, show the details in going from Enneper’s surface pa-
rametrization
n 1 3o n 1 o n o
x D Re z z ; Re i z C z 3 ; Re z 2
3 3
to the parametrization

1 3 1 3
x.u; v/ D u u C uv 2 ; v v C vu2 ; u2 v2
3 3
that is also for Enneper’s surface.

Exercise 2.148. Compute the parametrization for the minimal surfaces generated by using
1
p.z/ D 2z and q.z/ D iz on the domain C f0g in the Weierstrass representation. Use
MinSurfTool with the W.E.(p,q) tab to graph an image of the surface, which is known
as the wavy plane. [Use radius min=0.001, radius max=1.3, min= , max= with
initial values x D Re.1=2 log.z/ 1=4z 2 /, y D Im.1=2 log.z/ C 1=4z 2 /, and z D Re.z/.]
Exercise 2.149. Compute the parametrization for the minimal surfaces generated by using
p.z/ D z 2 and q.z/ D zi2 on the domain C f0g in the Weierstrass representation. Use
MinSurfTool with the W.E.(p,q) tab to graph an image of this surface, which is known
as Richmond’s surface. [Use radius min=0.1, radius max=1, min= /24, max=2 + /24
with initial values x D Re.1=3z 3 C 1=z/, y D Im.1=3z 3 1=z/, and z D Re.2z/.]
The Gauss map, G , and height differential, dh

Exercise 2.150. Show that if z D x C iy is the projection of .x1 ; x2 ; x3/ on the Riemann
sphere onto to complex plane, then
x1 x2
xD ; yD :
1 x3 1 x3
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 157 — #175

i i
2.8. Additional exercises 157
Exercise 2.151. For Scherk’s doubly periodic surface find

(a) G.0/ (b) G.1/ (c) G. 1/ (d) G.i / (e) G. i /.
Exercise 2.152. The Weierstrass data for a 4-noid are
z3
G.z/ D z 3 and dh D dz:
.z 4 1/2
Show that the ends of the 4-noid are catenoidal.
Exercise 2.153. Determine asymptotic and curvature lines for Scherk’s doubly periodic
surface with G.z/ D z and dh.z/ D z4iz 1 dz.
Exercise 2.154. Determine the period conditions for the wavy plane with G.z/ D z and
dh.z/ D dz.
Exercise 2.155. Let M be the Scherk doubly periodic surface with six ends. Using the
approach of Example 2.104 determine G and dh.
Minimal surfaces and harmonic univalent mappings

Exercise 2.156. Prove that if f D u C iv is harmonic in a simply-connected domain G,
then f D h C g, where h and g are analytic.
˚
Exercise
˚ Prove that the representations f .z/ D h.z/C g.z/ and f .z/ D Re h.z/C
2.157.
g.z/ C i Im h.z/ g.z/ are equivalent.
Exercise 2.158. Shear h.z/ g.z/ D 1 z z with !.z/ D z 2 to get the harmonic univalent
function f D h C g given in Example 2.129, where

1 zC1 3z 2z 2 1 zC1 z 2z 2
8 z 1 4.1 z/2 8 z 1 4.1 z/2
Exercise 2.159. Shear h.z/ g.z/ D .1 zz/2 with !.z/ D z 2 to get the harmonic univalent
function f D h C g given in Exercise 2.130, where

1 zC1 3z 2z 2 1 zC1 z 2z 2
8 z 1 4.1 z/2 8 z 1 4.1 z/2
Exercise 2.160. Show that the parametrization

i i Cz 1 1Cz 1 1 C z2
x D Re log ; Im log ; Im i log
2 i z 2 1 z 2 1 z2
is equivalent to the parametrization in Exercise 2.145 that gives Scherk’s doubly periodic
minimal surface.
Exercise 2.161. For the harmonic univalent map f .z/ D h.z/ C g.z/, where

3 1Cz 3i 1 C iz 1 z
h D log log C
16 1 z 16 1 iz 4 1 z4

3 1Cz 3i 1 C iz 1 z3
gD log log C
16 1 z 16 1 iz 4 1 z4
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 158 — #176

i i
(b) Use ComplexTool to graph the image of D under f and MinSurfTool with tab
W.E.(h,g) to sketch the corresponding minimal graph.
(c) This minimal surface has four helicoidial ends and is not embedded (i.e., it does have
self-intersections). However, Theorem 2.124 states that it should be a minimal graph
and hence have no self-intersections. Explain why this is not a contradiction.
z
Large Project 2.162. The analytic function F .z/ D .1 z/2
maps the unit disk D onto
1
C n . 1; 4/ and is an important function. Shear h.z/ g.z/ D .1 zz/2 with various
perfect-square dilatations, !, that satisfy the condition
j!j < 1 for all z 2 D (e.g., ! D
z 2n .n D N/! D e ı z 2 ; . 2 R/; ! D 1z az a
.jaj < 1/). Determine the corresponding
minimal graphs.
2.9 Bibliography
[1] Fred J.Almgren and Jean Taylor, The geometry of soap films and soap bubbles, Sci.
Am. 235 (1976), 82–93.
[2] Lipman Bers, Riemann Surfaces, New York Univ., Institute of Mathematical Sci-
ences, New York, 1957–1958.
[3] William M. Boothby, An Introduction to Differentiable Manifolds and Riemannian
Geometry, Second edition. Pure and Applied Mathematics, vol 120. Academic Press,
Inc., Orlando, FL, 1986.
[4] James Clunie and Terry Sheil-Small, Harmonic univalent functions, Ann. Acad. Sci.
Fenn. Ser. A.I Math. 9 (1984), 3–25.
[5] Richard Courant and Herbert Robbins, What Is Mathematics?, Oxford University
Press, 1941.
[6] Ulrich Dierkes, Stefan Hildebrandt, Albrecht Kster, and Ortwin Wohlrab, Minimal
Surfaces I, Grundlehren der Mathematischen Wissenschaften, 295, Springer-Verlag,
Berlin, 1992.
[7] Manfredo do Carmo, Differential Geometry of Curves and Surfaces, translated from
the Portuguese, Prentice-Hall, Inc., Englewood Cliffs, NJ, 1976.
[8] Michael Dorff, Minimal graphs in R3 over convex domains, Proc. Amer. Math. Soc.
132 (2004), no. 2, 491–498.
[9] Michael Dorff and Jan Szynal, Harmonic shears of elliptic integrals, Rocky Mountain
J. Math. 35 (2005), no. 2, 485–499.
[10] Kathy Driver and Peter Duren, Harmonic shears of regular polygons by hypergeo-
metric functions, J. Math. Anal. App. 239 (1999), 72–84.
[11] Peter Duren and William Thygerson, Harmonic mappings related to Scherk’s saddle-
tower minimal surfaces, Rocky Mountain J. Math. 30 (2000), no. 2, 555–564.
[12] Walter Hengartner and Glenn Schober, On schlicht mappings to domains convex in
one direction, Comment. Math. Helv. 45 (1970), 303–314.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 159 — #177

i i
[13] Stefan Hildebrandt and Anthony Tromba, Mathematics and Optimal Form, Scientific
American Library, 1985.
[14] Karcher, Hermann, Construction of minimal surfaces, Survey in Geometry, Univ. of
Tokyo, 1989.
[15] Hermann Karcher, Introduction to the complex analysis of minimal surfaces, Lecture
notes given at the NCTS, Taiwan, 2003.
[16] Hans Lewy, On the non-vanishing of the Jacobian in certain one-to-one mappings,
Bull. Amer. Math. Soc. 42 (1936), no. 10, 689–692.
[17] Jane McDougall and Lisabeth Schaubroeck, Minimal surfaces over stars, J. Math.
Anal. Appl. 340 (2008), no. 1, 721–738.
[18] Richard Millman and George Parker, Elements of Differential Geometry, Prentice-
Hall Inc., Englewood Cliffs, NJ, 1977.
[19] Frank Morgan, Minimal surfaces, crystals, shortest networks, and undergraduate re-
search, Math. Intel., 14 (1992), 37–44.
[20] Johannes C. C. Nitsche, Lectures on Minimal Surfaces, vol. 1, Cambridge U. Press,
1989.
[21] John Oprea, Differential Geometry and Its Applications., 2nd ed., Classroom Re-
source Materials Series. Math. Assoc. of America, Washington, DC, 2007.
[22] ——, The Mathematics of Soap Films: Explorations with Maple, Student Mathemat-
ical Library, vol 10, Amer. Math. Soc., Providence, RI, 2000.
[23] Robert Osserman, A Survey of Minimal Surfaces, 2nd ed. Dover Publications, Inc.,
New York, 1986.
[24] Michael Spivak, A Comprehensive Introduction to Differential Geometry, vols 1–5,
3rd ed., Publish or Perish, Inc., Houston, TX, 2005.
[25] Matthias Weber, Classical minimal surfaces in Euclidean space by examples: geo-
metric and computational aspects of the Weierstrass representation in Global Theory
of Minimal Surfaces, 19–63, Clay Math. Proc., 2, Amer. Math. Soc., Providence, RI,
2005.
[26] Steven Weintraub, Differential Forms: A Complement to Vector Calculus, Academic
Press, Inc., San Diego, CA, 1997.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 160 — #178

i i
i i
i i
i i
“ECA˙Book” — 2012/9/5 — 14:07 — page 161 — #179

i i
Applications to Flow Problems

3
Michael A. Brilleslyper (text),
3.1 Introduction
This chapter developed from a series of lectures prepared for an undergraduate mathemat-
ical physics course. The lectures were designed to show students applications of complex
function theory that connected to familiar topics from calculus and physics. Two dimen-
sional flows of ideal fluids was a natural topic. Many ideas from vector calculus are used
and there are numerous applications of the methods that are developed. Modeling ideal
fluid flow is a standard application of conformal mappings and is readily found in most
undergraduate complex analysis texts (see [5] or [1]). However, in preparing the notes it
became apparent that there was a need for a unified treatment that included a variety of
applications and extensions such as including sources and sinks in the flow or accounting
for the role played by sources or sinks at infinity. The emphasis in many texts is on the
analytic aspects of the subject and not on the geometric or visual aspects of the flows.
This chapter combines both analytic and geometric aspects of flow problems. It is self-
contained and relies only on basic results from vector calculus and a standard first course
in complex variables. We created the easy-to-use applet FlowTool to encourage discovery
and experimentation.
FlowTool plots the streamlines for the flow of an ideal fluid. It permits the user to
select the number and location of sources and sinks on the boundary or in the interior of
the region and then allows the user to dynamically vary the strength of those sources and
sinks. The applet shows steady state fluid flow in four preset regions: the entire plane, the
half plane, the quadrant, and the strip.
The applet is limited to pre-set regions and particular types of sources. To deal with
the wide variety of problems that may be encountered in practice, we also use powerful
computer algebra systems such as Mathematica.
Vector fields arise naturally in many applications. They are used to model physical
phenomena such as the velocity field of a fluid flowing in a region or the electric force field
generated by a collection of charges. The geometry of the region, along with sources or
sinks in the field, determines the nature of the resulting vector field. Finding descriptions
of these fields is the focus of this chapter. It is also interesting to find the integral curves
of a vector field, which are curves that are everywhere tangent to the field. Integral curves
161
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 162 — #180

i i
162 Chapter 3. Applications to Flow Problems
have different names depending on the context including flow lines, streamlines, or lines of
force.
In this chapter we study physical situations in which complex function theory can be
used to help solve the problem of finding such vector fields and their integral curves. We
focus mainly on using techniques of complex analysis to solve flow problems. We start
with an example, omitting many of the details, but providing a glimpse at what lies ahead.
Example 3.1. Imagine an infinitely long, very shallow, channel in which fluid flows. Be-
cause the channel is shallow we assume the flow is a two dimensional. Real flows are three
dimensional, but we assume the flow is identical on all parallel planes. Orient the channel
on the complex plane so that one edge runs along the real axis and the other along the
horizontal line Im z D ; so the channel has width . Assume that fluid is being pumped
into the channel and drained from it at various points along the edges. For the time being
we assume that all pumps and drains operate at the same constant rate. Our goal is to de-
scribe the velocity of the fluid in the channel. We assume that the flow is in a steady state,
meaning that the velocity at a point of the domain does not change with time. Let the fluid
be pumped into the channel at equal rates at z D 0 and z D 2, and suppose fluid is drained
from the channel at the same rate at the point z D 2 C i . There is more fluid being
pumped into the channel than is being drained. See Figure 3.1. The curves shown are the
integral curves of the underlying vector field and they represent the path that a drop of dye
would follow if placed into the flow. A critical assumption in this model is that we are
dealing with an ideal fluid. This means that the fluid is incompressible, non-viscous, and
there is no loss of energy due to friction between the walls of the channel and the fluid.
Such fluids do not physically exist, but they frequently provide good approximations to
physical situations and they have mathematical properties that lend themselves to analysis
and modeling.
–4 –2 0 2 4 6
F IGURE 3.1. Flow lines for an ideal fluid in a channel with sources and sinks.
The plot in Figure 3.1 and most of the other plots in this chapter were generated us-
ing the computer algebra system Mathematica. Software packages such as Mathematica,
Maple, and Matlab are powerful and can be useful in generating complicated graphics.
However, they may not be easy to use or to adapt to a particular problem. To facilitate
understanding of this chapter we have provided the FlowTool applet. It allows you to ex-
periment with complicated flows for a fixed number of domains. We use it throughout the
chapter and in numerous exercises. We also provide the code in two examples for generat-
ing these types of plots in Mathematica.
Before moving on you should open the FlowTool applet and generate your own version
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 163 — #181

i i
3.1. Introduction 163
of Figure 3.1. Open the applet and select the strip domain from the drop down menu.
Uncheck the box marked equipotential lines since we wish to view only the flow
lines in this example. Use the mouse to place a source at z D 0, z D 2, and z D 2 C i
by clicking once near each of the three locations. A dialog box with a slider opens in the
right-hand panel each time a source is placed on the boundary of the channel. The sliders
are used to change the strength of a source. If the slider is moved to the left, the strength
will eventually become negative, signifying that the source has changed to a sink. Move
the slider to create a sink of strength 1 at z D 2 C i . The other two locations are
sources of strength 1. Figure 3.2 shows a graph similar to what you should see on the
applet. You can experiment further by adding or deleting sources and using the sliders to
vary the strengths. Observe how the flow lines run parallel to the edges of the channel,
indicating that the boundary of the region acts like a frictionless barrier to the flow.
F IGURE 3.2. Flow lines in a channel with sources and sinks generated with FlowTool.
There is interesting mathematics behind the plot of the flow lines in Figures 3.1 and 3.2.
This chapter develops the key ideas needed to solve flow problems in various regions with
combinations of sources and sinks along the boundary. In addition, later sections extend
the ideas and methods to more complicated settings.
Section 3.2 starts with a review of basic ideas from vector calculus including the im-
portant notions of the divergence and curl of a vector field and the definition of a harmonic
function. This material should be familiar to most students.
Section 3.3 provides a review of the main results needed from a standard course in
complex variables. It makes the critical connection between planar vector fields and com-
plex functions. We also introduce the Polya field and give a mathematical description of
sources and sinks.
Section 3.4 introduces the complex potential function that is the main tool for con-
structing ideal fluid flows. We make several connections between the complex potential
function and the underlying vector field. This leads directly to the construction of uniform
flows in various regions in Section 3.5. We continue the development in Section 3.6 by
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 164 — #182

i i
allowing sources and sinks along the boundary of the region. We then look at the uniform
flow in a channel in Section 3.7. This development sheds light on the important role played
by sources or sinks at infinity. Section 3.8 puts the previous material together by discussing
flows in any region with combinations of sources or sinks along the boundary.
The final five sections focus on applications or extensions of the material. Topics in-
clude flows inside disks, interval sources, steady state temperature problems, sources or
sinks in the interior of regions, and electric fields generated by dipoles and multipoles.
3.2 Background and Fundamental Results

We need several fundamental concepts from multivariable calculus. We briefly review the
main ideas. More background regarding vector fields and their associated operations can
be found in any standard calculus text such as [4].
We represent a 2-dimensional or planar vector field in Cartesian coordinates using two
real-valued functions of position: FE .x; y/ D < P .x; y/; Q.x; y/ >. Examples include
constant vector fields such as FE .x; y/ D < 3; 4 > and the field tangent to concentric
circles about the origin given by G.x; E y/ D < y; x >. Other examples include slope
fields for first-order differential equations or the magnetic field in a plane perpendicular
to a wire with a current flowing through it. Vector fields are represented graphically in
the plane by drawing arrows indicating the direction of the field at selected points. The
magnitude of the field at a point is
p
jFE j D P .x; y/2 C Q.x; y/2
and the direction satisfies

Q.x; y/
tan D
P .x; y/
for P .x; y/ ¤ 0. If Q.x; y/ ¤ 0 and P .x; y/ D 0, then the direction is ˙ 2 at .x; y/, the
choice being determined by the sign of Q.x; y/. The vector h0; 0i has no defined direction.
When graphing vector fields, it is difficult to draw vectors with their true magnitudes, so
3 3
2 2
1 1
–3 –2 –1 1 2 3 –3 –2 –1 1 2 3
–1 –1
–2 –2
–3 –3
F IGURE 3.3. The constant vector field given by F IGURE 3.4. Magnetic field generated by a cur-
FE .x; y/ D h4; 3i: rent in a wire. The circle represents an integral
curve of the field.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 165 — #183

i i
3.2. Background and Fundamental Results 165
graphical representations frequently scale their length down to a more manageable size.
Figure 3.3 shows a constant vector field and Figure 3.4 shows the magnetic field generated
by a current running perpendicular to the plane through the origin. We also show an integral
curve of the field.
The curl and divergence of FE are defined in terms of the vector differential operator
r D h@x ; @y i W
r FE D .Qx Py /kO (3.1)
r FE D Px C Qy (3.2)
The curl operator acts on 3-dimensional vector fields and yields another vector field
orthogonal to the original field. However, for 2-dimensional planar fields we assume that
O
the k-component of the vector field FE is zero, implying that the curl will always point in
O
the k-direction. Thus, it is sufficient to compute the scalar component of r FE ; namely
Qx Py :
A vector field is called irrotational at a point if the curl is zero and it is called incom-
pressible at a point if the divergence is zero. The standard physical description of a fluid
with zero curl is that an infinitesimally small paddle wheel placed horizontally into the
flow would not rotate. That is, the flow contains no vortices. The physical description of
zero divergence is that the fluid is incompressible. A consequence of this is that the amount
of fluid entering a region must be equal to the amount of fluid leaving it.
In this chapter we study vector fields that are both irrotational and incompressible
throughout a region. This may seem restrictive, but it includes many important examples in
addition to modeling ideal fluid flow. One example is the electric field generated in a plane
perpendicular to an infinitely long wire with a uniform current density. In this case, the
electric field is identical on all planes perpendicular to the wire. The electric field in this
scenario has magnitude inversely proportional to the distance from the wire. This is differ-
ent from the electric field generated by a point charge in space in which the magnitude of
the field is inversely proportional to the square of the distance from the wire, the familiar
inverse square law. A spherically symmetric field is not identical on parallel planes and
thus does not readily lend itself to analysis using complex variable methods. When we dis-
cuss electric fields in this chapter, we will always assume the field is generated by uniform
density current in a wire running perpendicular to the plane.
The requirement that both the curl and the divergence be zero imposes conditions on
the component functions of the vector field FE .x; y/ D < P .x; y/; Q.x; y/ >. From
(3.1) and (3.2) we obtain a pair of partial differential equations relating the components of
FE :
Px D Qy (3.3)
Py D Qx :
These equations are similar to the Cauchy-Riemann equations satisfied by analytic func-
tions. If f .z/ D u.x; y/ C iv.x; y/ is an analytic function, then the real and imaginary
parts satisfy the Cauchy-Riemann equations:
ux D vy (3.4)
uy D vx :
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 166 — #184

i i
Exercise 3.2. Determine if the following vector fields are incompressible, irrotational, or
both.
(a) FE .x; y/ D hax C by; cx C dyi, where a; b; c, and d are real
E
(b) G.x; y/ D he x cos y; e x sin yi
Try it out!
Given a vector field FE .x; y/ and a path C; the line integral of FE along C is denoted
R
by C FE d RE and measures the work done by the field on a particle traversing the path C:
The line integral is computed by letting r .t/ D .x.t/; y.t//; be a parameterization of C;
Rb
where a t b; and computing a FE .r .t// r 0 .t/ dt:
A vector field FE is said to be conservative in a simply connected domain D if the
R
line integral C FE d RE between any two fixed points in D is independent of the chosen
path C; provided the path lies entirely in D. There are numerous equivalent conditions that
guarantee that a field is conservative. See [4] for a complete discussion.
Theorem 3.3. Let FE be a vector field with component functions that are continuous and
have continuous first order partial derivatives throughout a simply connected region D.
The following are equivalent:
(a) FE is conservative.
(b) There exists a differentiable potential function such that r D FE ; where r D
hx ; y i:
R
(c) C FE d RE D 0 for every closed loop C in D:
(d) r FE D 0:
We make extensive use of the result that a curl-free vector field has a potential function.
The condition r D FE implies the system of partial differential equations
x D P .x; y/ (3.5)
y D Q.x; y/
is satisfied. The equations allow us to construct potential functions through partial inte-
gration. The function is unique up to an additive constant (why?) and is called the real
potential function of FE .
Another definition that plays an important role in our work is that of a harmonic func-
tion, a solution of Laplace’s equation. Harmonic functions play a critical role in much of
applied mathematics. They arise frequently as steady-state solutions to various physical
problems. Harmonic functions and their properties are closely tied to the theory of analytic
functions in complex analysis. In solving flow problems, harmonic functions play a central
role.
Definition 3.4. Let u W D R2 ! R have continuous second order partial derivatives.
If u satisfies Laplace’s equation u D r ru D uxx C uyy D 0; then u is a harmonic
function in D.
The following exercises explore vector fields, potential functions, and harmonic func-
tions. Exercise 3.6 shows that the potential function for an irrotational and incompressible
field is always harmonic.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 167 — #185

i i
3.3. Complex Functions and Vector Fields 167
Exercise 3.5. Let FE .x; y/ D hx 3 3xy 2 ; y 3 3x 2yi. Compute the curl and divergence
of FE . If FE is conservative, then find a potential function for it. If FE is incompressible, show
that the potential function is harmonic. (Optional) Use a program such as Mathematica or
Matlab to graph FE and several level curves of the potential function. What key geometrical
observation connects the direction of FE with the tangents to the level curves of the potential
function? Try it out!
Exercise 3.6. Suppose that FE is irrotational and incompressible. Let be a real potential
function of FE . Show that is a harmonic function. Try it out!
The result of Exercise 3.6 is particularly important. It is the irrotational feature of the
vector field that implies the existence of the potential function and the incompressibility
then implies that the potential function is harmonic.
Exercise 3.7. Consider an attracting force at the origin of the xy-plane whose magnitude
at a point is inversely proportional to the distance from the origin to the point. Determine a
formula for the vector field representing the force field and determine if the field is irrota-
tional and incompressible in a region not containing the origin. If possible, find a potential
function for the vector field. Change the field by assuming the force of attraction at a point
is inversely proportional to the square of the distance to the point. Compute the curl and
divergence of this field. How do your answers compare? Try it out!
3.3 Complex Functions and Vector Fields

In this section we make several connections between vector fields and complex functions.
A vector field that is irrotational and incompressible is closely related to the concept of an
analytic function.
A complex function of the complex variable z D x C i y can be expressed in terms of
its real and imaginary parts as f .z/ D u.x; y/ C i v.x; y/, where u and v are real-valued
functions of x and y. This is accomplished by setting z D x C i y and separating into
real and imaginary parts. The function f is said to be analytic at z D z0 if the complex
derivative f 0 .z/ exists for every z in a neighborhood of z0 . An immediate and far-reaching
consequence is that the functions u and v must satisfy the Cauchy-Riemann equations:
ux D vy (3.6)
uy D vx :
A basic result about analytic functions is

Theorem 3.8. Let f .z/ D u.x; y/ C iv.x; y/ be analytic in a domain D. Then the func-
tions u and v are harmonic in D.
This result follows directly from the Cauchy-Riemann equations and the reader is
strongly encouraged to prove it.
An equally important result is that a harmonic function on a simply connected domain
is the real part of an analytic function. For a complete discussion, see [1].
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 168 — #186

i i
Theorem 3.10. Let u.x; y/ be a harmonic function defined on a simply connected domain
D. Then, there exists a harmonic function v.x; y/ defined on D called the harmonic conju-
gate of u, such that f .z/ D u.x; y/ C iv.x; y/ is analytic on D: The harmonic conjugate
is uniquely determined up to an additive constant.
Exercise 3.11. Find a harmonic conjugate for u.x; y/ D 3x 2 2y 3y 2 : Try it out!
Exercise 3.12. Let u.x; y/ D 21 ln.x 2 C y 2 /. Show that u is harmonic on the punctured
plane. Next, omit an infinite ray emanating from the origin and construct a harmonic con-
jugate for u on the plane minus the ray. Try it out!
Next we establish a correspondence between the set of complex functions and the set
of planar vector fields. Since both complex functions and planar vector fields can be repre-
sented by a pair of real-valued functions, a complex function may be thought of as a vector
field and vice versa. We note the correspondence:
V .z/ D P .x; y/ C iQ.x; y/ ” FE .x; y/ D hP .x; y/; Q.x; y/i (3.7)
The correspondence may appear to be only a renaming of one object as another. How-
ever, looking at complex functions as vector fields sometimes yields insights that are not
otherwise apparent. When plotting the vector field of a complex function, the magnitude
of the vector field at z is jV .z/j and the direction is arg.V .z//. Figure 3.5 shows the vector
field corresponding to the complex function V .z/ D z 2
–3 –2 –1 1 2 3
–1
–2
–3
F IGURE 3.5. Vector field representation of V .z/ D z 2 :
For the remainder of this chapter we will use the notation V .z/ to refer to both complex
functions and their corresponding vector field representations.
The correspondence in (3.7), though simple, is not the most useful vector field repre-
sentation of a complex function. For reasons that will become clear, we instead associate a
complex function with its Polya vector field.
Definition 3.13. Let V .z/ D u.x; y/ C i v.x; y/ be a complex function. The vector field
given by the conjugate V .z/ D u.x; y/ i v.x; y/ is called the Polya vector field of V .z/.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 169 — #187

i i
N
Note that when plotting V .z/, we attach the arrow to the point z, not z.
The Polya vector field of V .z/ D z 2 is V .z/ D zN 2 and is shown in Figure 3.6. Con-
trast this field with the one shown in Figure 3.5. We will see the value of the Polya field
representation when we look at the vector field corresponding to an analytic function.
–3 –2 –1 1 2 3
–1
–2
–3
F IGURE 3.6. The irrotational, incompressible Polya field V .z/ D zN 2 :
The next theorem shows that if a complex function is analytic, then its Polya field is
irrotational and incompressible. The proof is a straightforward calculation involving the
Cauchy-Riemann equations and is left as an exercise.
Theorem 3.14. Let V .z/ be a complex function that is twice differentiable on a domain
D. V .z/ is analytic on D if and only if its Polya vector field V .z/ is irrotational and
incompressible on D.
Theorem 3.14 establishes a bijection between the set of irrotational, incompressible
planar vector fields and the set of analytic functions. It is this connection that allows us to
use the techniques of complex analysis to solve flow problems.
Exercise 3.16. Find the component functions on a suitably chosen domain of each irro-
tational and incompressible vector field corresponding to the following analytic functions.
Verify directly that the resulting vector fields have both curl and divergence equal to zero.
(a) V .z/ D z 2
(b) V .z/ D e z
1
(c) V .z/ D z
Try it out!
Exercise 3.17. Verify that the following vector field is irrotational and incompressible and
find the corresponding analytic function.
V .z/ D .x 2 y2 2x/ C i . 2xy C 2y/
Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 170 — #188

i i
The third example in Exercise 3.16 deserves some additional comments. The function
V .z/ D 1=z is not defined at z D 0; but is analytic at every point in any neighborhood
of z D 0: Such a point is called a singular point or a singularity of the function. The
corresponding Polya vector field V .z/ is also undefined at z D 0. Graphing the vector field
reveals that the field lines emanate radially from the origin; see Figure 3.7. We call such a
point a source. Conversely, if the field lines went towards the origin, we would refer to it
as a sink. This leads to the following definition.
–3 –2 –1 1 2 3
–1
–2
–3
F IGURE 3.7. Polya vector field of the analytic function f .z/ D z1 :
Definition 3.18. Let V .z/ be a function that is analytic except at z D 0. If V .z/ has a pole
of order 1 at z D 0; then zero is either a source or sink of the Polya field V .z/.
Sources and sinks have physical interpretations as a quantity being created or de-
stroyed. For flow problems a source is a location where fluid is pumped into the region,
whereas a sink is a drain, removing fluid from the region. For an electric field, sources and
sinks correspond to positive and negative charges respectively. Allowing vector fields to
have sources and sinks extends the examples to which we may apply our methods.
Exercise 3.19. Let V .z/ D 1=z: Show that the Polya field V .z/ emanates radially from the
origin and that the magnitude of the field at a point is inversely proportional to the distance
from the point to the origin. Try it out!
To have a vector field with a source at z D z1 rather than z D 0 we need only to trans-
late the analytic function to obtain V .z/ D 1=.z z1 /: A sink is obtained by changing the
sign of the function. The strength of a source or sink is changed by multiplying V .z/ by a
real scalar. To account for more than one source or sink we use the principle of superpo-
sition. This states that if several components act to generate a vector field, then the field is
the sum of the fields generated from each component separately. This is nothing more than
the linearity of vector addition, but its consequences are far-reaching: complicated systems
may be analyzed by studying the simpler components from which they are generated. For
example, Figure 3.8 shows the field generated by equal sources at z D 0 and z D 1 and
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 171 — #189

i i
sink of twice the strength located at z D 1 C 2i: The analytic function corresponding to
the field is
1 1 2
V .z/ D C : (3.8)
z z 1 z .1 C 2i /
–1
–2
–3
–3 –2 –1 0 1 2 3
F IGURE 3.8. Integral curves of a vector field with two point sources and one sink.
Qualitatively, only the relative strengths of sources or sinks matters in the analysis of
the situation. For completeness we compute the actual strength of a source or sink. We
define the strength of a source to be the flux of the vector field over any simple closed loop
C that encloses it, but no other singularities of the function.
Definition 3.20. Let FE .x; y/ D hP .x; y/; Q.x; y/i be an irrotational, incompressible
vector field with a source at z D z0 : The strength of the source is defined to be
Z
P .x; y/ dy Q.x; y/ dx; (3.9)
C
where C is a loop traversed in a counter clockwise direction containing z0 and no other

singularities of the field. If the strength is negative, then the singularity is a sink.
There is a relationship between the flux and circulation of a vector field and the complex
integral of the associated Polya field. The strength of the source can be determined directly
R
from the complex integral C V .z/ dz as shown in the following exercise.
Exercise 3.21. Let V .z/ D u.x; y/ C i v.x; y/ be a vector field. Let C be a simple loop
R
not passing through any singular points of the field. Expand the integral C V .z/ dz in real
and imaginary parts. (Hint: let dz D dx C i dy:) Show that
Z Z Z
V .z/ dz D u dx C v dy C i u dy v dx:
C C C
This shows that the flux of V .z/ outward across C is obtained from the imaginary part of
R
the integral C V .z/ dz. As a consequence, the circulation of V .z/ is given by the real part
of the integral. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 172 — #190

i i
This exercise gives a useful characterization and interpretation of complex integration

in terms of the familiar concepts of work and flux. An immediate consequence is that if
V .z/ is irrotational and incompressible inside and on C , then the work along C and the
flux across C must be zero. This implies that V .z/ is analytic, as was already established.
See Needham [2] for a thorough discussion of Polya vector fields and their relationship to
complex integration.
If V .z/ is analytic everywhere except for a finite number of simple poles located at
zi ; i D 1; 2; : : : ; n; then the strength of the source or sink of V .z/ at zk is given by the
R
imaginary part of C V .z/ dz; where C is a simple closed loop traversed in the counter-
clockwise direction such that zk is contained in the interior of C and none of the other zi
are inside C .
Let C be a simple closed curve containing zk but no other zi for i ¤ k. The function
can be represented as
g.z/
V .z/ D ;
z zk
where g.z/ is analytic inside and on C . By Cauchy’s integral formula,
Z
V .z/ dz D 2 ig.zk /:
C
Exercise 3.22. In constructing vector fields with sources or sinks, the most common case
is where V .z/ has the form
X n
aj
V .z/ D ;
z zj
j D1
where aj 2 R. Show that for the Polya field that the strength of the source or sink located
at z D zk is 2 ak . Try it out!
Exercise 3.23. Let a 2 R: Show that the vector field V .z/ D a=.Nz z0 / has a source of
R
strength 2 a at z D z0 by explicitly computing the imaginary part of C V .z/ dz; where
C is a circle centered at z0 . Do this by using a parameterization of C instead of Cauchy’s
integral formula. Try it out!
Exercise 3.24. The Polya field of the function
sin z
V .z/ D
z2
has a simple pole at the origin. What is the strength of the source? Try it out!
3.4 Complex Potential Functions

We have established a one-to-one correspondence using complex conjugation between ir-
rotational, incompressible vector fields in the plane with a finite collections of point sources
and sinks and complex functions that are analytic except at finitely many points where they
have poles of order one.
We now construct the complex potential function of the underlying vector field. Let
V .z/ be analytic on a simply connected domain so that V .z/ is irrotational and incom-
pressible, implying V .z/ has a potential function that is harmonic. By Theorem 3.10
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 173 — #191

i i
3.4. Complex Potential Functions 173
we know there exists a harmonic function such that the complex function .z/ D
.x; y/ C i .x; y/ is analytic. is determined only up to an additive constant. The
function is called a complex potential function of V .z/.
Definition 3.25. Given an irrotational, incompressible vector field V .z/ on a simply con-
nected domain, a complex potential function of V .z/ is the analytic function .z/ D
.x; y/ C i .x; y/; where is the real potential function of V .z/ and is a harmonic
conjugate of :
A globally defined complex potential function may not exist. For example, V .z/ D 1=zN
is irrotational and incompressible on the punctured plane, but a complex potential for this
field involves choosing a branch of the complex logarithm function that is not analytic on
any neighborhood containing the origin.
As we already know, the level curves of ; the real potential function, are orthogonal to
the direction of V .z/. Those curves form the equipotential lines of the field. For a velocity
field of an ideal fluid, they represent points where the velocity is constant. In the case of
an electric field, the level curves of represent curves of constant electrostatic potential.
What do the level curves of represent? The answer is found in the following standard
result from complex variables.
Theorem 3.26. Let .z/ D .x; y/Ci .x; y/ be analytic at z0 D x0 Ci y0 and suppose
that 0 .z0 / ¤ 0. Then the tangents to the level curves of and are orthogonal at the
point .x0 ; y0 /.
For a proof see Zill and Shanahan [5].
Theorem 3.26 implies that the level curves of are parallel to the underlying vector
field V .z/. Thus, they are the integral curves of the field and are exactly the curves that
were sketched by the FlowTool applet in the first example in this chapter. The function
is often referred to as the stream function for a flow problem. The level curves of are
called the streamlines.
The next exercise shows the vector field can be obtained from the complex potential
function.
Exercise 3.27. Let V .z/ be an irrotational and incompressible vector field with analytic
complex potential function .z/. Show that 0 .z/ D V .z/. (Hint: Write .z/ in terms of
its real and imaginary parts and differentiate using the Cauchy-Riemann equations.) Try it
out!
An equivalent statement to the last exercise is that
Z
.z/ D V .z/ dz: (3.10)
This formulation is useful in the next exercise.

Exercise 3.28. V .z/ D 3zN 2 4zN is an irrotational, incompressible vector field.
(a) Determine a complex potential function on C using (3.10).
(b) Find the real and imaginary parts of the complex potential function to easily find the
real potential function and the stream function.
Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 174 — #192

i i
The last exercise of this section illustrates a mapping property of the complex potential
function.
Exercise 3.29. Let be the complex potential for an irrotational, incompressible vector
field. Show that when viewed as a mapping of the z-plane to the w-plane, maps equipo-
tential curves to vertical lines and flow lines to horizontal lines. Try it out!
3.5 Uniform Flows in the Plane and other Regions

We now give examples that allow us to solve a wide variety of problems in different regions
of the plane. We start by considering simple flows in the entire plane. Consider the function
.z/ D z: Thinking of as a complex potential function, a natural problem is to determine
the underlying vector field. The following exercise asks you to do this.
Exercise 3.30. Show that .z/ D z is the complex potential function for a uniform flow
to the right. Explicitly compute the vector field V .z/. See Exercise 3.27. Show directly that
the level curves of the imaginary part of are stream lines for the flow. Try it out!
Exercise 3.31. How does the answer to Exercise 3.30 change if the complex potential
function is .z/ D .2 C 3i /z? Show that the stream lines are given by the family of linear
equations 3x C 2y D c. See Figure 3.9. Try it out!
–3 –2 –1 1 2 3
–1
–2
–3
F IGURE 3.9. Uniform flow with complex potential .z/ D .2 C 3i /z:
Our goal is to solve flow problems in regions such as sectors, strips, and disks. How-
ever, so far we have studied flows only in the entire complex plane. The extension to other
regions lies in the theory of conformal mappings. A conformal mapping is a one-to-one
complex function mapping a region in the z-plane to a region in the w-plane so that angles
and orientation are preserved. A key result is that an analytic function is conformal at all
points where its derivative is not zero. The critical property of conformal mappings for our
development is the theorem:
Theorem 3.32. Let .w/ be analytic on a domain D 0 2 C, and let f W D ! D 0 be
e
conformal for some domain D 2 C. Then the composition .z/ D . ı f /.z/ is analytic
on D.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 175 — #193

i i
3.5. Uniform Flows in the Plane and other Regions 175
The proof is found in many texts, such as [5].

The significance of Theorem 3.32 is that it allows a problem to be translated to a simpler
domain, solved, and then translated back to the original domain. Unfortunately, it is usually
impossible to map a given region conformally to the entire complex plane. Instead, we
make use of a deep and powerful theorem from complex analysis.
Theorem 3.33 (Riemann Mapping Theorem). Every simply connected domain in the com-
plex plane, with the exception of the entire plane, can be conformally mapped to the upper
half plane, H D fz j Im z > 0g:
The Riemann mapping theorem gives no indication of how to find the conformal map-
ping. There are tables of conformal mappings that give specific instances of the mappings
for a large collection of regions, as in [5]. Despite being an existence result, Theorem 3.33
indicates that it is valuable to know how to solve problems on the upper half plane. We will
translate different problems to equivalent problems in H in order to solve them.
Suppose an ideal fluid is flowing from left to right in H; so the real axis acts as a
boundary for the flow. That is, the real axis acts as a streamline for the flow. It is clear
that a constant vector field such as V .z/ D 1 is such a flow. It is easy to see the complex
potential of this field is .z/ D z. Thus, the identity function may be regarded as the
complex potential for a uniform flow to the right in H. Our goal is to use this and the
theory of conformal mappings to find the complex potential for flows in other regions and
in regions that have sources and sinks.
Our first example deals with a uniform flow in a quadrant, often called flow around a
corner.
Example 3.34. Suppose a vertical barrier is inserted along the imaginary axis in a uni-
form flow in H. The anticipated flow is shown in Figure 3.10. We want to find its complex
potential function for the underlying vector field. The key is Theorem 3.32. We seek a
conformal mapping that maps the region in Figure 3.10 to H with the property that the
boundaries of the region are mapped to the real axis. This is important because the bound-
aries of the region are always streamlines for the flow (that is the flow is parallel to the
–4 –3 –2 –1 0
F IGURE 3.10. Uniform Flow around a Corner.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 176 — #194

i i
boundaries and there is no friction). If we let h.z/ D z 2 then the function will map the
second quadrant to H. The negative real axis is mapped to the negative real axis and the
imaginary axis is mapped to the positive real axis. We know that the complex potential
for the uniform flow in H is given by .z/ D z, so we set .z/e D .h.z// D h.z/ D
2 2 2
z D x C y 2xyi; and that the underlying vector field is e0 .z/ D 2zN . Plotting the
streamlines gives Figure 3.10. The streamlines are the level curves of the imaginary part of
e namely 2xy D c, where c is a constant.
,
Exercise 3.35. Let R be the region in polar coordinates

R D f.r; / j r 0 and 0 g:
4
Find the complex potential function for a uniform, irrotational, incompressible flow with
no source or sinks in R. Try it out!
The discussion above along with Exercise 3.35 demonstrates the technique of using
conformal mappings to solve uniform flow problems. Any conformal mapping from a re-
gion D onto H is the complex potential for the velocity field of an ideal fluid flowing in
D.
Another example demonstrating these techniques is flow around a cylinder.
Example 3.36. The function h.z/ D z C 1=z is a conformal mapping from H minus the
upper half of the unit disk onto H. That is, h maps fz j jzj 1 and Im z > 0g onto H. The
streamlines shown in Figure 3.11 are the family of curves
y
y D c:
x2 C y2
–3 –2 –1 0 1 2 3
F IGURE 3.11. Streamlines around a Cylinder.
3.6 Sources and Sinks

Having seen how to solve for uniform flows in various regions, we now look at regions that
have sources or sinks on the boundary of the region, as the example of the channel in the
introduction. First, we will solve the problem in H:
Let a 2 R. We have already seen that V .z/ D 1=.Nz a/ is an irrotational, incompress-
ible vector field with a source at z D a of strength 2 . We are interested only in the part
of V .z/ that lies in H. Since this is half of the field, we define the effective strength of the
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 177 — #195

i i
3.6. Sources and Sinks 177
source on the boundary to be half of the strength, so the effective strength of the source
is . Only the relative strengths of the sources or sinks matter for the qualitative behavior
of the flow. However, if we wish to balance a source on the boundary with a sink in the
interior of the region, the notion of effective strength comes into play.
Before proceeding, we revisit the idea of a source in a vector field. The following
exploration shows how physical principles lead to our complex function representation of
a source or sink.
Exploration 3.37. Suppose that we have an irrotational and incompressible vector field
defined on the punctured plane with a source of strength S at z D 0. Assume that that the
flow lines are directed radially away from the origin. We wish to determine a formula for
the field. Let FE .x; y/ denote the vector field at points other than the origin. We assumed the
direction of FE is directly away from the origin, so we need only determine the magnitude
of FE : Since FE is incompressible, the amount of fluid crossing into a closed region must
equal the amount leaving it. We apply this principle to circles centered at the origin. The
source having strength S means that there are S units of fluid entering the region per unit
time, so S units of fluid cross the circle x 2 C y 2 D R2 per unit time. But the amount
crossing the circle is equal to the magnitude of the vector field multiplied by the length of
the circle. Setting these two quantities equal gives
S D jFE .x; y/j 2R: (3.11)

Solve this equation for the magnitude of the field and combine the result with the fact
that the direction is radial to find an explicit formula for the field. Show that its complex
representation is V .z/ D S=2 zN . Hint: 1=zN D z=jzj2 :
Immediate consequences of Exploration (3.37) are the results that we have already estab-
lished:
(a) If the source is at z D a instead of z D 0, then the complex representation of the
vector field is V .z/ D 1=.Nz a/.
N
(b) If there is a sink at z D a, then the function is V .z/ D 1=.Nz N
a/
(c) The strength of a source or sink is changed by multiplying V .z/ by a real scalar.
Exercise 3.38. Find the complex representation of the irrotational, incompressible vector
field in H if there is a source of effective strength at z D 3 and a sink of effective strength
2 at z D 2. Hint: The vector field at any point is the sum of the vector fields determined
from each source or sink. Use FlowTool to view the solution. Try it out!
If V .z/ D 1=.Nz a/;
N then the complex potential function on H can be obtained from
V .z/ by using (3.10): Z
.z/ D V .z/ dz D Log.z a/: (3.12)
Thus, we see that the complex logarithm is involved when there are sources and sinks along
the boundary. For the upper half plane, we use the principal value of the logairthm defined
by < Argz : The complex logarithm extends the real logarithm function, and .z/
is analytic for Im z > 0, but is not analytic in a domain containing z D a:
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 178 — #196

i i
Writing the logarithm in terms of real and imaginary parts

Log.z a/ D ln jz aj C i Arg.z a/;
the streamlines are given by the family of equations
y
arctan D c:
x a
This is easily manipulated into the family of rays emanating from z D a given by y D
.tan c/.x a/: The values c D 0 and c D correspond to the two streamlines that run
along the boundary of H in the positive and negative directions from the source at z D a.
This preceding development allows us to write the complex potential for an irrotational,
incompressible vector field in H with any combination of sources and sinks along the
boundary. The complete result is
Theorem 3.39. Let V .z/ be the irrotational, incompressible vector field in H generated by
a finite collection of simple sources or sinks. Assume there are sources located along the
boundary of H at z D ai , i D 1; : : : ; n, ai 2 R, with effective strengths, Si , i D 1; : : : ; n
(if Sj < 0, then there is a sink at aj ). Then the complex potential of V .z/ is given by
n
X
.z/ D Sj Log.z aj /:
j D1
The next exercise asks you to experiment with FlowTool to develop some intuition
regarding flows in H. You may go beyond the suggestions in the exercise and experiment
with other situations.
Exercise 3.40. Use FlowTool to view the flow in H for
(a) A source of strength 2 at z D 1 and a sink of strength 2 at z D 1:
(b) A source of strength 2 at z D 1 and a source of strength 2 at z D 1:
(c) A source of strength 4 at z D 1 and a sink of strength 2 at z D 1:
(d) Sources of strength 2 at z D 3 and z D 0 and a sink of strength 4 at z D 2:
Try it out!
In Exercise 3.40 sometimes all the fluid emanating from the sources is taken in by the
sinks, but at other times some of the fluid either escapes to infinity or seems to emanate
from infinity. If we think of infinity as a point on the boundary of H, then we are led to the
notion that there is a sink or source at infinity.
To examine the behavior at infinity, we revisit a flow in the entire plane with a finite
collection of sources or sinks. Consider the vector field
n
X Sj 1
V .z/ D :
2 zN zNj
j D1
Let C be a simple closed contour that encloses the singularities of V .z/ and consider
R
C V .z/ dz: By the residue theorem,
Z X n
Im V .z/ dz D Sj ;
C j D1
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 179 — #197

i i
3.7. Flow in a Channel 179

R
which gives the total flux across C . The integral C V .z/ dz can also be viewed as a line
integral around infinity traversed in the clockwise direction. When viewed in this way, its
value has the opposite sign of the total flux outward across C . For example, if V .z/ has a
source of strength 3 and a sink of strength 1 at points in the plane, then there must a sink of
strength 2 at infinity. If the net flux in the finite plane is zero, then the sum of the strengths
of the sinks and sources at infinity must be zero as well. That infinity can be both a source
and a sink will be explored further. Section 3.7 investigates flow in an infinite channel,
where infinity is a simultaneous source and sink.
3.7 Flow in a Channel

We return to a uniform flow in an infinitely long channel. Before dealing with any sources
or sinks along the boundary, we investigate the case of uniform flow to the right in the
channel. Assume the channel has one edge along the real axis and the other along the line
Im z D : Since the edges of the channel are horizontal, the flow is the restriction of the
uniform flow to the right in H restricted to the channel. Hence the complex potential is
.z/ D z:
On the other hand, the function f .z/ D e z is a conformal mapping of the strip onto
H such that boundary edges of the strip are mapped to the real axis. Using the example of
the quadrant as motivation and Theorem 3.32, we see that .z/ e D .f .z// D e z should
e
be the complex potential function for the uniform flow in the strip. However, Im .z/ D
e x sin y: Plotting the level curves gives the streamlines shown in Figure 3.12. They should
simply go from left to right parallel to the edges of the strip. What went wrong?
–3 –2 –1 0 1 2 3
F IGURE 3.12. Incorrect uniform flow in a channel.
The incorrect flow occurred from a failure to account for the behavior of infinity under
the conformal mapping. We may think of the flow in the channel as having a source at the
left end of the strip and a sink of equal strength at the right end. Points in the strip with large
negative real parts are mapped to points near zero. For all z D x C iy with 0 y ; we
have
lim e xCi y D 0:
x! 1
Thus, under the conformal mapping e z , we expect to see a source at z D 0 in H: On the

other hand, as x ! 1; the values of e z approach 1 and the sink is mapped to infinity in
H: This means that the complex potential in H has a source at 0. Its complex potential in
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 180 — #198

i i
H is .z/ D Log.z/. Thus, the complex potential for the uniform flow in the strip as given
e
by Theorem 3.32 is .z/ D .e z / D Log.e z / D z; as expected.
With an understanding of how sources and sinks at infinity must be accounted for, it
is easy to incorporate sources and sinks along the boundary. The following exercise asks
you to find the complex potential for various flows in a channel. In each case the sum of
all sources and sinks must be zero. Use FlowTool to see the streamlines for each flow and
to note the behavior at infinity.
Exercise 3.41. An infinitely long channel of width with its lower edge along the real
axis has a source of strength 6 at z D 0 and a sink of strength 6 at 3 C i: Find the
complex potential for the flow. Use FlowTool to visualize the streamlines of the flow. Does
any of the flow escape to infinity? Why or why not? Try it out!
Exercise 3.42. Use the same scenario as in Exercise 3.41, but suppose that the sink at
3 C i has strength 2: Describe the behavior at infinity. Find the complex potential and
use FlowTool to plot the streamlines. Does the result make physical sense? Try it out!
3.8 Flows in Other Regions

The methods used to solve flow problems in a strip apply to any region with sources or
sinks along the boundary. All that we need is a conformal mapping from the region to the
upper half-plane H such that the boundary of the region is mapped to the real axis. It is
then a matter of determining where the sources and sinks are mapped and accounting for
behavior at infinity. We can find the complex potential for the transformed problem in H
and compose it with the conformal mapping to obtain the solution.
Example 3.43. For the region R D H fz j jzj 1g suppose a flow in R is generated by
a source of strength 2 at z D 1 and a sink of strength 2 at z D 1: We find the complex
potential for this flow as follows:
(a) The mapping f .z/ D z C 1=z maps R conformally to H.
(b) f . 1/ D 2 and f .1/ D 2.
(c) The complex potential in H is .z/ D Log.z C 2/ Log.z 2/.
e
(d) Composing with f gives .z/ D Log.z C 1=z C 2/ Log.z C 1=z 2/.
e
(e) The flow lines are given by the level curves of Im .
The flow lines are shown in Figure 3.13.
–4 –2 0 2 4
F IGURE 3.13. Flow around cylinder with sources and sinks.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 181 — #199

i i
3.8. Flows in Other Regions 181
Apply the methods outlined in the example in the next exercise.

Exercise 3.44. Let R D fz j 0 arg.z/ =3 g. Assume there is a source of strength 4
at z D 0 and a sink of strength 2 at z D 2: Find the complex potential for the flow in R.
The streamlines are shown in Figure 3.14.
F IGURE 3.14. Streamlines for Exercise 3.44.
Even with a sophisticated graphing program it may not be easy to obtain a sketch of
the flow lines. The Mathematica code used to generate Figure 3.14 is
H[z_] = 4*Log[z] - 2*Log[z - 8];

F[z_] = H[zˆ3] // Simplify;
G[z_] = F’[z] // Simplify;
a = StreamPlot[{Re[G[x + I y]], -Im[G[x + I y]]},
{x, 0.01, 4}, {y, 0.01, 4},
AspectRatio -> Automatic, StreamPoints -> Fine,
FrameStyle -> Thick, Frame -> None,
RegionFunction -> Function[{x, y, z}, y < Sqrt[3]*x]];
b = Plot[Sqrt[3]*x, {x, 0, 4}, PlotStyle -> {Thick, Black}];
c = Graphics[{PointSize[.02], Point[{0, 0}]}];
d = Graphics[{PointSize[.02], Point[{2, 0}]}];
e = Plot[0, {x, 0, 4}, PlotStyle -> {Thick, Black}];
Show[a, b, c, d, e]
In the Mathematica code, the RegionFunction statement is used to plot streamlines

only in the region of interest. We have chosen to use the StreamPlot command rather
than the ContourPlot command. Both are useful, but StreamPlot uses arrows that
indicate the direction of the flow. Observe that StreamPlot requires a vector field as the
input which we find by differentiating the complex potential function and then taking the
conjugate, which gives us the Polya field. When the potential function contains the complex
logarithm function, the plot may have some breaks in the flow lines that do not look right.
Sometimes it is possible to fix this by combining the logarithm terms. For example, in the
code fragment, the function could have been defined as
H[z_]=Log[zˆ4/(z-8)ˆ2]
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 182 — #200

i i
The FlowTool applet provides the first quadrant as a domain available for study. In the
absence of any sources or sinks on the boundary, the uniform flow in this region is the flow
around a corner that we have already mentioned. By including sources or sinks some some
interesting flows can be seen.
Exploration 3.45. Develop a procedure for finding the complex potential for a flow in the
first quadrant with sources or sinks on the boundary. Use your method to find the complex
potentials for (a) - (c). Use the FlowTool applet to investigate the flow.
(a) Sources of equal strength at z D 1 and z D i:
(b) A source and a sink of equal strengths at z D 1 and z D i:
(c) Sources of equal strength at z D 1 and z D i , and a sink of double strength at the
origin.
3.9 Flows inside the Disk

The methods can be applied to fluid flow inside a disk. The next set of exercises explores
several flows inside a disk. Obtaining the graphical output for these flows can be challeng-
ing. We present the Mathematica code to generate one of the plots.
Exercise 3.46. Let D D fz 2 C j jzj < 1g. Show that
1 z
f .z/ D i
1Cz
is a conformal mapping from the unit disk D to the upper half plane H. Determine the
images of the points ˙1 and ˙i . Use the conformal mapping to find a uniform flow inside
D. The flow lines are shown in Figure 3.15. Explain why there appears to be both a source
and a sink at z D 1: Try it out!
F IGURE 3.15. Uniform flow in a disk. Observe the behavior at z D 1:
The Mathematica code for Figure 3.15 is
H[z_] = z;
F[z_] = H[I*(1 - z)/(1 + z)] // Simplify;
G[z_] = F’[z] // Simplify;
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 183 — #201

i i
3.10. Interval Sources and Sinks 183
a = StreamPlot[{Re[G[x + I y]], -Im[G[x + I y]]},

{x, -1.2, 1.2}, {y, -1.2, 1.2}, AspectRatio -> Automatic,
StreamPoints -> Automatic, Frame -> None, FrameStyle -> Thick,
RegionFunction -> Function[{x, y, z}, xˆ2 + yˆ2 < 1]];
b = ParametricPlot[{Cos[t], Sin[t]}, {t, 0, 2*Pi},
PlotStyle -> {Thick, Black}];
Show[a, b]
Exploration 3.47. This exploration considers flows inside a disk.

(a) Find the velocity field of the ideal fluid flow inside D if there is a source of strength 2
located at z D 1: The total strength of the sources in a problem must be balanced by a
sink or sinks of equal strength. Where is the sink in this problem? Can you control its
location?
(b) Find the complex potential for the flow in D if there is a sink of strength 2 at z D i:
(c) Suppose there are equal strength sources on the boundary of D located at z D 1 and
z D 1: Consider the location of the resulting sink. Why does this occur?
3.10 Interval Sources and Sinks

In this section we extend some of our methods. Instead of looking at point sources or sinks,
we consider interval sources or sinks. This will allow us to model phenomena such as flow
through a levee or the electric field generated by a line of charges. We present the material
as a series of exercises which lead to the key results.
We begin by examining the flow lines generated by a uniformly distributed source of
total strength 2 located along the interval a x b: We expect to see a vector field as
in Figure 3.16.
F IGURE 3.16. Integral curves of a vector field with a uniform interval source.
The next set of exercises shows that the complex potential for an interval source can
be derived as a limiting process of a collection of point sources as the number of points
approaches infinity.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 184 — #202

i i
Exercise 3.48. Let a D x0; x1 ; : : : ; xn D b be n C 1 equally spaced points in the inter-

val Œa; b. Assume there is a source of strength 2=.n C 1/ at each xj . Find the complex
potential .z/ for the flow in H. Try it out!
Now we let n ! 1 and recognize the limit as a definite integral.
Exercise 3.49. Let xj D xj C1 xj ; j D 0; 1; : : : ; n: Express the result of the last
exercise as a Riemann sum on Œa; b: Show that its limit is the definite integral
Z b
1
Log.z x/ dx: (3.13)
b a a
Try it out!
Exercise 3.50. Obtain the definition of the complex potential for a uniform interval source
by using integration by parts to show the that the integral (3.13) is
b z z a
ba .z/ D Log.z b/ C Log.z a/ 1: (3.14)
b a b a
Try it out!
Exercise 3.51. Use (3.14) to find the complex potential in H for a flow that has a uniform
source of strength 2 along the interval Œ0; 3 and a uniform sink of strength 2 along the
interval Œ 2; 1. The graph appears in Figure 3.17. Try it out!
–4 –2 0 2 4
F IGURE 3.17. Flow lines with interval source and sink.
Next we extend interval sources or sinks to the boundaries of other regions. We must
be sure the interval sources or sinks are still uniform, as we now show. Suppose in the first
quadrant R in C we have interval sources of equal strength located along the intervals Œ1; 3
and Œi; 3i : We proceed as in previous sections. The conformal mapping f .z/ D z 2 maps
R to H: We determine the behavior of the key intervals under the conformal mapping. The
interval Œ1; 3 is mapped to Œ1; 9 and Œi; 3i is mapped to Œ 9; 1: We know how to find
the complex potential in H from (3.14). We see that .z/ D 91 .z/ C 19 .z/: We then
find the complex potential in R by composition: .z/ e D .z 2 /: Unfortunately this result
is not quite correct because sources should be uniformly distributed across the intervals
in R: When we apply the conformal mapping the intervals in H are no longer uniformly
distributed.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 185 — #203

i i
3.10. Interval Sources and Sinks 185
To understand what is happening, focus on the interval Œ1; 3 under the mapping f .z/ D
z 2 : It is mapped to Œ1; 9: The first half, Œ1; 2; is mapped to Œ1; 4 and the second half, Œ2; 3;
is mapped to Œ4; 9: Hence the density of points is less in the second half of the interval.
Thus, when solving for the complex potential in H we cannot treat the intervals as though
the source is uniformly distributed. Recall that (3.14) was derived under the assumption
that the source was uniformly distributed along the interval. We must instead use (3.13)
and take into account the non-uniform distribution. Thus instead of computing the integral
R9 R3
1 Log.z x/ dx we compute 1 Log.z x 2 / dx:
Exercise 3.52. Find a complex potential for the flow in the first quadrant generated by
uniform interval sources along the intervals Œ1; 3 and Œi; 3i : Plot the streamlines.
Exercise 3.53. Find the complex potential for an ideal flow in the region

R D fr e i j r 0; 0 g
3
with a uniform source of strength 2 located along the interval Œ2; 4: Try it out!
Exercise 3.54. Find the complex potential for an ideal flow in the infinite channel
R D fz j 0 Im z 2g
with a uniform source of strength 2 located on the boundary along the interval Œ1C2i; 4C
2i : Try it out!
A natural extension of the previous material on interval sources is to consider intervals
with a non-uniform density. This small project was inspired by Potter [3].
Small Project 3.55. Suppose a function W Œa; b R ! R has the property that
Z b
.x/ dx D S;
a
where S is the total strength of the generalized source on Œa; b: We think of .x/ as giving
the source density at x 2 Œa; b: The goal is to find the complex potential for the interval
source with variable density. If .x/ D S=.b a/; for all x in Œa; b; then we obtain a
uniformly distributed source along the interval.
Rather than attempting to find the complex potential directly, it is better to first find the
underlying vector field V .z/: Subdivide the interval into n equal subintervals, where the
n th subinterval is Œxi ; xi C1: Now, consider the vector field Vi .z/ having a source at xi
with strength
b a
Si D .xi /:
n
Show by summing the individual vector fields and taking a limit as n ! 1 that the vector
field is given by
Z b
.x/
V .z/ D dx: (3.15)
a N
z x
From (3.15) it follows that the complex potential is
Z Z b !
.x/
.z/ D dx dz: (3.16)
a z x
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 186 — #204

i i
–1
–2
–3
–2 –1 0 1 2 3
F IGURE 3.18. Flow lines for variable density interval source.
For .x/ D x on the interval Œ 1; 2 the source density takes on both positive and
negative values, with the positive density increasing towards the right end of the interval.
The strength of the variable density source is
Z 2
3
x dx D :
1 2
Compute V .z/ by using the substitution w D z x in the integral. Use the result to find
the complex potential and plot the flow lines. The result is shown in Figure 3.18.
If is a polynomial, then a formula for V .z/ can be found explicitly using the same
substitution, w D z x: The reader is encouraged to find and plot the complex potential
for .x/ D x 2 on Œ0; 1:
3.11 Steady State Temperature Problems

A common problem in applied mathematics is to find the steady state temperature in a
region of the plane given the temperature distribution along the boundary of the region.
For the upper half plane H, the problem takes the form: Given a piecewise continuous,
bounded function f defined on the real axis (i.e., the temperature), find a function T in H
such that T D 0 and T agrees with f along the boundary. The solution involves finding
a harmonic function in H. We are seeking solutions that are physically meaningful. For
example, if the boundary values are identically zero, then T1 .x; y/ D 0 and T2 .x; y/ D
e x sin y are both solutions in H: However, T2 is unbounded, which does not correspond to
a real temperature distribution.
Every time we construct a complex potential function in H, both the real and imaginary
parts are harmonic. To see if our methods are useful, we discuss how to manipulate the
boundary values in an example.
We begin with the complex potential for an ideal flow in H with a single point source
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 187 — #205

i i
3.11. Steady State Temperature Problems 187
of strength 2 at z D 0: We know that the complex potential is .z/ D Log z; so
.z/ D ln jzj C i Arg.z/: (3.17)
Along the real axis, when x > 0; Arg.x/ D 0; and when x < 0; Arg.z/ D . Thus the
argument function is piecewise constant along the real axis.
Suppose the steady state temperature distribution along the boundary is
(
0 if x < 0
f .x/ D (3.18)
100 if x 0:
From (3.17), we see that multiplying Im .z/ by 100= gives a harmonic function in
H satisfying the boundary condition
(
100 100 if x < 0
Arg.x/ D (3.19)
0 if x 0:
This is almost correct. To get the boundary values switched to the correct intervals
we need some properties of the argument function. The next exercise leads to the correct
result.
Exercise 3.56. If Im z > 0; then show that reflecting Arg.z/ across the y-axis gives
Arg. zN /: Show that
Arg. zN / D Arg. z/:
Try it out!
From Exercise 3.56, we see that
100
g.x; y/ D Arg. z/

is harmonic and satisfies the correct boundary conditions. The graph of g.x; y/ is shown
in Figure 3.19. The solution g.x; y/ is the imaginary part of the complex potential
100
.z/ D Log. z/:

100
3
50
2
0
–2
y
1
0
x
2
0
F IGURE 3.19. Steady state temperature distribution with piecewise constant boundary values.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 188 — #206

i i
The streamlines correspond to isotherms—curves of constant temperature.

The next exercises explore slight variations on this problem. We then look at problems
where segments of the boundary are not insulated (i.e., there are segments that have non-
constant temperatures) and see how our methods can be applied.
Exercise 3.57. Solve the steady state temperature problem in H where the boundary tem-
perature is given below. Hint: This is similar to the above example: start with a slightly
different complex potential function.
(
100 if x < 2
f .x/ D (3.20)
0 if x 2:
Try it out!
perature is given below. Hint: Add a constant to the function, seeing the effect it has on the
other boundary value. (
100 if x < 2
f .x/ D (3.21)
50 if x 2:
Try it out!
The boundary conditions we have been considering can be extended to more segments.
The next exercise leads to the algorithm for solving the general problem.
perature is given by 8̂
<100 if x < 3
ˆ
f .x/ D 50 if 3 < x < 2 (3.22)
ˆ
:̂25 if x 2:
Make a contour plot of your solution and observe that the contours represent curves of
constant temperature. Try it out!
The general problem of finding the temperature distribution in the half-plane given
piecewise constant boundary conditions is a standard application in complex analysis texts,
as in [5]. The problem usually is presented as a boundary value problem for Laplace’s
equation.
Exploration 3.60. Solve:
Txx C Tyy D 0
subject to 8̂
ˆ
ˆ k0 if 1 < x < x1
ˆ
ˆ
<k1 if x1 < x < x2
T .x; 0/ D :
ˆ
ˆ ::
ˆ
ˆ
:̂k if xn < x < 1:
n
Using the techniques developed in the exercises, derive the general solution.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 189 — #207

i i
3.11. Steady State Temperature Problems 189
Next we consider a more general problem where segments of the boundary are not kept
at constant temperature. Suppose we have the boundary temperature distribution
(
100 if x < 0
f .x/ D (3.23)
0 if x > 1:
No temperature is specified on the interval .0; 1/: If it is not insulated then we expect
the temperature to change linearly from 100 degrees to 0 degrees (which can be seen by
solving the one-dimensional heat flow problem on the interval Œ0; 1 with the endpoints
held at 100 and 0 degrees).
We now investigate how our techniques can be applied to this problem. We find the
steady state temperature distribution in H for the boundary condition given in (3.23). The
complex potential for an ideal flow with an interval source uniformly distributed along the
interval Œ0; 1 is, from (3.14),
10 .z/ D .1 z/Log.z 1/ C zLog.z/ 1: (3.24)
We know that Im 10 .z/ must be constant along the intervals x < 0 and x > 1: We
determine the constants by choosing a test value from each interval:
Im 10 . 1/ D Im.2Log.2/ Log. 1/ 1/ D ;
and
Im 10 .2/ D Im. Log.1/ C 2Log.2/ 1/ D 0:
Hence multiplying our complex potential by 100= gives the correct temperatures on
the two intervals that are held constant. But what about the interval .0; 1/? In Figure 3.20,
we show the graph of Im 10 .x/ for 3 < x < 3 and observe the linear behavior on the
interval .0; 1/:
100
80
60
40
20
–3 –2 –1 1 2 3
F IGURE 3.20. Temperature along boundary.
The graph of the solution on H is shown in Figure 3.21. The surface representing the
temperature matches the boundary values.
The next small project asks you to generalize the example to a general problem of the
same type.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 190 — #208

i i
100
3
50
2
0
–2
y
1
0
x
2
0
F IGURE 3.21. Solution to heat equation.
Small Project 3.61. Find a general formula for the steady-state temperature distribution
T .x; y/ in the half-plane with the boundary data
8̂
ˆ
ˆ k1 if 1 < x < x1
ˆ
ˆ
<k2 if x2 < x < x3
T .x; 0/ D :
ˆ
ˆ ::
ˆ
ˆ
:̂k if x < x < 1
n n
We assume x1 < x2 < : : : < xn and that T .x; 0/ is linear on the intervals between the
xi locations.
3.12 Flows with Source and Sinks not on the Boundary

In this section we extend the ideas developed thus far to a wider array of applications.
Consider the upper half plane H, with a source located at z D i , not on the boundary. We
expect the flow to look like an ordinary source near z D i , but since it is constrained to
stay in H the real axis deflects it so that it runs parallel to the boundary. Figure 3.22 shows
the flow lines we expect.
–3 –2 –1 0 1 2 3
F IGURE 3.22. Flow in H with a single source at z D i .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 191 — #209

i i
3.12. Flows with Source and Sinks not on the Boundary 191
To obtain the complex potential that gives Figure 3.22, we balance the source at z D i
with a source of equal strength at z D i . By the principle of superposition, the ver-
tical components of the underlying vector fields will sum to zero along the real axis.
The resulting complex potential is .z/ D Log.z i / C Log.z C i / or, equivalently,
.z/ D Log.z 2 C 1/. It is not analytic in on any punctured neighborhood of i . Thus,
is not formally a complex potential in all of H: However, by choosing different branches
of the complex logarithm, it always possible to construct an analytic complex potential in
a simply connected domain not containing i .
Balancing sources or sinks across the boundaries, combined with our earlier work, can
be used to deal with sources or sinks in the interior of a region. Needham [2] refers to this
approach as the method of images. The next exercise finds the complex potential for a flow
in the first quadrant with a source located at the interior point z D 1 C 2i .
Exercise 3.62. Find the flow of an ideal fluid in the first quadrant with a single source
located at z D 1 C 2i . Hint: Use the conformal mapping f .z/ D z 2 to map the first
quadrant to H. Determine the location of source by computing f .1C2i /. Find the complex
potential in H by balancing the source with another source symmetrically located across
the real axis. Compose the result with f .z/ to obtain the complex potential of the flow in
the first quadrant. The result gives the flow shown in Figure 3.23. Try it out!
0
0 1 2 3 4
F IGURE 3.23. Flow in the first quadrant with a single source at z D 1 C 2i
When dealing with sources and sinks on the boundary and in the interior the notion of
effective strength comes into play. A source on the boundary must have twice the strength
as one in the interior to have the same effective strength.
Exercise 3.63. Find the complex potential for a uniform flow in H with a source at z D 0
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 192 — #210

i i
and a sink of equal magnitude at z D i: Answer: .z/ D Log.z 2 =.z 2 C1// The streamlines
are shown in Figure 3.24 Try it out!
–3 –2 –1 0 1 2 3
F IGURE 3.24. Balancing a source on the boundary with a sink in the interior.
The techniques developed thus far allow us to combine sources and sinks of various
relative strengths both on the boundary and in the interior of a region. The next exercise
shows that the complex potential for a complicated flow can be built up from simpler
pieces.
Exercise 3.64. Let R be the infinite strip 0 Im z . Suppose there is a source of
strength 2 at z D 1, a sink of strength 2 at z D 4; and a source of strength 2 in the
middle of the strip at z D i=2. Construct the complex potential of the velocity field. The
streamlines are shown in Figure 3.25. Try it out!
–2 0 2 4 6 8
F IGURE 3.25. Streamlines for the flow in Exercise 3.64.
3.13 Vector Fields with Other Types of Singularities

Another application arises when we interpret ideal flows as electric field lines generated
by uniform density current in a wire running perpendicular to the plane. We imagine the
plane being a copper plate with wires cutting through the plane at different locations. It is
convenient to think of these locations as positive or negative charges on the plane. A dipole
is obtained when a positive and negative charge of equal strength are separated by a small
distance. Suppose we have a source of strength 1 located at z D and a sink of strength
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 193 — #211

i i
3.13. Vector Fields with Other Types of Singularities 193
1 located at z D . The Polya vector field for the electric field is
1 1
f .z/ D :
zN zN C
As approaches zero, the field vanishes. To prevent the field from vanishing, we need to
increase the strengths of the source and sink inversely to the distance between them, so as
to keep the field strength constant while the distance between the charges approaches zero.
Exercise 3.65. For the Polya vector field

1 1 1
f .z/ D
2 zN zN C
compute the limiting field as goes to zero and the complex potential for the field. Hint:
Recall that 0 .z/ D f .z/. Try it out!
As shown in Exercise 3.65, the vector field generated by the dipole is f .z/ D 1=zN 2
and the complex potential is .z/ D 1=z: The electric field lines are the level curves of
Im .z/ and are shown in Figure 3.26.
Exercise 3.66. Show that the electric field lines for the dipole are circles with centers on
the imaginary axis. Try it out!
1.0
0.5
–1.0 –0.5 0.5 1.0
–0.5
–1.0
F IGURE 3.26. Electric field lines for a dipole.
Figure 3.26 shows one possible orientation of the dipole. In general, the orientation
of the dipole depends on the direction from which the source and the sink approach each
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 194 — #212

i i
other. In this case, the Polya field was 1=zN 2 and the dipole moment was 1. Other dipole
moments are obtained by considering the field d=zN 2 , where d is a complex number. For
example if d D 1 C i , then we obtain the dipole field shown in Figure 3.27.
1.0
0.5
–1.0 –0.5 0.5 1.0
–0.5
–1.0
F IGURE 3.27. Electric field lines for a dipole with dipole moment 1 C i .
Exploration 3.67. Investigate the behavior of the dipole whose field is given by f .z/ D
.a C bi /=zN 2 :
It is also possible to consider the problem of multiple charges approaching each other to
obtain multipoles. As an example, consider the complex potential function .z/ D 1=z 2 :
The electric field lines are shown in Figure 3.28.
Exercise 3.68. What charges are converging to give the electric field lines shown in Figure 3.28?
Try it out!
An interesting problem involves looking at sources or sinks and multipoles at the same
location. For example, consider the Polya field V .z/ D 1=zN 2 C 1=zN : The complex potential
function is .z/ D 1=z C Log z: As an exercise, plot the electric field lines near zero and
on a larger scale. The dipole dominates the behavior near zero and the source dominates
the behavior far from zero.
We close this chapter with a discussion of sources and sinks at infinity. Since we now
understand the idea of a multipole, we can classify the behavior of a flow at infinity.
Consider the example of uniform flow to the right in the entire plane C: Since fluid
appears from the left and disappears to the right it seems reasonable to say that there is
both a source and a sink at infinity. We now recognize this as a dipole at infinity.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 195 — #213

i i
1.0
0.5
–1.0 –0.5 0.5 1.0
–0.5
–1.0
F IGURE 3.28. Electric field lines for a multipole.
To study the behavior of a complex function at infinity, it is standard to replace z

with 1=z and study the behavior of the resulting function at z D 0. For the uniform flow,
the complex potential is .z/ D z: The behavior at infinity is determined by examining
.1=z/ D 1=z at z D 0: Thus we have a potential function of 1=z: Differentiating and
taking the conjugate to obtain the underlying vector field, we get 1=zN2 ; which we recognize
as a dipole.
More generally, for a vector field where the net sum of the sources and sinks is N; the
strength of the source or sink at infinity is N: If N D 0; then there could be a multipole
at infinity.
3.14 Bibliography
[1] Ruel V. Churchill and J. W. Brown, Complex Variables and Applications, 4th ed.,
McGraw-Hill Book Company, New York, 1984.
[2] Tristan Needham, Visual Complex Analysis, Oxford University Press, Oxford, UK,
1997.
[3] Harrison Potter, On Conformal Mappings and Vector Fields, Senior Thesis, Marietta
College, Marietta, OH, 2008.
[4] James Stewart, Multivariable Calculus, 6th ed., Thompson Brooks/Cole, 2008.
[5] Dennis G. Zill and P. D. Shanahan, A First Course in Complex Analysis with Applica-
tions, 2nd ed., Jones and Bartlett, Boston-Toronto-London-Singapore, 2009.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 196 — #214

i i
i i
i i
i i
“ECA˙Book” — 2012/9/5 — 14:07 — page 197 — #215

i i
Anamorphosis, Mapping Problems,

4
and Harmonic Univalent Functions
Michael J. Dorff (text),
4.1 Introduction
Complex-valued analytic functions have many nice properties that are not possessed by
real-valued functions. For example, we say a complex-valued function is analytic if we can
differentiate it one time. It is true that if a complex-valued function f is analytic, then we
can differentiate it infinitely many times. Complex-valued analytic functions can always be
represented as a Taylor series, and they are conformal (that is, they preserve angles when
f 0 ¤ 0), properties that are not true for real-valued functions that can be differentiated one
time. Why does an analytic function have these properties? If f D u C iv is an analytic
function, then its real part, u.x; y/, and its imaginary part, v.x; y/, satisfy Laplace’s equa-
tion and are both harmonic. Also, u and v satisfy the Cauchy-Riemann equations and are
harmonic conjugates of each other. In this chapter we discuss some ideas and problems re-
lated to a collection of univalent (i.e., one-to-one) complex-valued functions, f D u C iv,
where u and v satisfy Laplace’s equation but not necessarily the Cauchy-Riemann equa-
tions. These functions are known as harmonic univalent functions or mappings and contain
analytic univalent functions as a subset. Analytic univalent functions have been studied
since the early 1900s, and thousands of research papers have been written about them. The
study of harmonic univalent mappings is a fairly recent area of research. So, it is natural
to consider the properties of analytic univalent functions as a starting point for our study
of harmonic univalent mappings. A general question will be “What properties of analytic
univalent functions are true for the larger class of harmonic univalent functions?”
Section 4.2 discusses how to determine the image of domains in C under a collection of
complex-valued functions known as Möbius maps and introduces the applet ComplexTool
as an aid to visualize them. Section 4.3 presents some background about the family of uni-
valent analytic functions. Section 4.4 introduces the fundamentals of harmonic univalent
functions. The study of harmonic univalent functions from the perspective of univalent
complex-valued analytic functions is a new area of research. Finding examples of them
is not easy, but a useful method of doing so is discussed in Section 4.5. Sections 4.2–4.5
197
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 198 — #216

i i
198 Chapter 4. Anamorphosis, Mapping Problems, and Harmonic Univalent Functions
should be read first. After that, the remaining sections can be read in any order and are
independent of each other. Three applets are used in this chapter:
1. ComplexTool is used to plot the image of domains in C under complex-valued func-
tions.
2. ShearTool is used to plot the image of domains in C under a complex-valued har-
monic function that is formed by shearing an analytic function and a dilatation; the
user enters the analytic function and dilatation without having to solve explicitly for
the harmonic function.
3. LinComboTool is used to plot and explore the convex combination of complex-
valued harmonic polygonal maps.
They can be accessed online at www.maa.org/ebooks/EXCA/applets.html.
Each section contains examples, exercises, and explorations that involve using the applets.
You should do all of the exercises and explorations, many of which present functions and
concepts that will be used in the chapter. There are additional exercises at the end of the
chapter. In the study of harmonic univalent functions, there are many unsolved problems,
some of which are mentioned. There are also short and long projects that are suitable as
research problems for undergraduate students.
The goal of the chapter is not to give a comprehensive introduction to this topic, but
to engage readers with its general notions, questions, and techniques, and to encourage
readers to actively pose and answer their own questions. To understand the nature and
purpose of this text better, the reader should read the Introduction. The study of harmonic
univalent functions has many interesting problems that undergraduates can investigate with
computers and the applets. I hope that students will explore the ideas in this chapter and
will prove some new results in the field.
4.2 Anamorphosis and Möbius Maps

Numb3rs is a U.S. television show that ran from 2005 to 2010 that dealt with a group of
FBI special agents trying to solve crimes. One of the FBI agents is Don Eppes who has a
mathematical genius brother, Charlie, who helps the FBI solve their cases. In the episode
“Jack of All Trades” (season 5, episode 4), the FBI is trying to catch a thief who has
eluded them for two years. They do not even have an accurate image of his face. At one
point in this episode, the thief escapes by stealing a bus at night and driving away. Before
he escapes, one agent takes his picture with a flash camera. Unfortunately, the flash from
the camera reflects off the bus window obscuring the face of the thief. However, there is a
metal cylindrical thermos in the photo that displays a distorted image of the thief. Using
mathematics, Charlie recreates the image of the thief’s face, giving the FBI information
that leads to his capture.
The method of producing a distorted image that appears normal by viewing it through
a curved mirror or from an unusual angle is known as anamorphosis. Anamorphosis was
studied by painters in the 15–16th centuries as they were trying to understand perspec-
tive. Two famous paintings that display anamorphosis are Jan van Eyck’s “The Arnolfini
Marriage” (see Figure 4.1) and Hans Holbein’s “The Ambassadors” (see Figure 4.2). Van
Eyck’s painting is thought to depict the Italian merchant Giovanni Arnolfini and his wife.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 199 — #217

i i
4.2. Anamorphosis and Möbius Maps 199
F IGURE 4.1. Van Eyck’s “The Arnolfini Marriage.”
On the wall behind them is a curved mirror reflecting a distorted image of the scene. In
Holbein’s painting there is a distorted shape lying diagonally at the bottom. That it is a
skull can be seen more clearly when the viewer is looking at the painting from a certain
angle.
F IGURE 4.2. Holbein’s “The Ambassadors.”
In modern times anamorphosis has continued to be used. In the sketch “Mysterious

Island” by the Hungarian artist István Orosz (see Figure 4.3) the viewer sees the image of
a seashore with hills and the sun in the background, two men walking, and a ship being
tossed in the sea. If a cylindrical mirror is placed on top of the sun, the image of the author
Jules Verne appears on the cylinder.
In 2010 the group Preventable (also known as the Community against Preventable
Injuries) caused drivers to slow down while driving near a school in British Columbia,
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 200 — #218

i i
F IGURE 4.3. Orosz’s “Mysterious Island.” Images copyright István Orosz. Used with permission.
Canada by putting an anamorphosis image on the highway (see Figure 4.4). The image is
not clear until the driver reached a point on the highway when it appeared that a young girl
had darted into the road chasing after a pink ball. In the same way that the anamorphosis
image could appear to the driver, so could a child suddenly run in front of the driver.
F IGURE 4.4. Anamorphosis image of a child chasing a ball into the street. Used with permission.
This idea of how an image is distorted is used in complex analysis when trying to vi-
sualize how a domain is mapped by a complex-valued function. We begin by looking at
maps known as Möbius transformations. If we have a one-dimensional real-valued func-
tion, such as f .x/ D x 2 , the graph of the function tells us about some of the properties of f
(i.e., zeros, one-to-one, increasing, etc.). We want to do the same thing for complex-valued
functions, but we would need a 4-dimensional graph (two dimensions for the domain and
two for the range). We can represent some of the properties of a complex-valued function
by looking at specific sets in the domain and seeing where the complex-valued function
w D f .z/ takes them.
We can use the accompanying applet ComplexTool to graph the image of domains under
complex-valued functions or maps. Open ComplexTool (see Figure 4.5). Suppose that we
want to find the image of the unit disk under the map f .z/ D zC2 i . In the middle section
near the top there is a box that has f(z)= before it. In it, enter z C 2 i . Below it, there
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 201 — #219

i i
F IGURE 4.5. The applet ComplexTool.
is a window that states No grid. Click on the down arrow H and choose Circular
grid. An image of a circular grid will appear on the left, which we call the z-plane. Click
on the button Graph in the middle section below the function you entered. The image of
the circular grid will appear on the right, which we call the w-plane. To reduce the size of
the image, click on the down arrow H above the image and chose a different size, such as
Re: [-3,3] Im: [-3,3]. You can move the axes so that the image is centered by
positioning the cursor over the image, clicking on the left mouse button, and dragging the
image to the left (see Figure 4.6).
F IGURE 4.6. The image of the unit disk under the map f .z/ D z C 2 i.
ˇ
ˇ
Exploration 4.1. You can graph the halfdisk, S D fz ˇ jzj < 1; Im z > 0g, by using a
circular disk and restricting the the value to be between 0 and in the middle panel of
the applet. For the following functions determine the image of S . Make a conjecture about
how a domain is transformed under the maps f .z/ D zCA, f .z/ D Bz, and f .z/ D e i z,
where A D a1 C i a2 2 C, B > 0, and 2 R.
(a) f .z/ D z 1 (b) f .z/ D z C i (c) f .z/ D z 1Ci
(d) f .z/ D 0:5z (e) f .z/ D 2z (f) f .z/ D 2:5z
i =4 i =2 i =4
(g) f .z/ D e z (h) f .z/ D e z (i) f .z/ D e z.
Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 202 — #220

i i
The previous exercise helps us see characteristics of the maps:

(a) if A D a1 C i a2 2 C, then the map f .z/ D z C A moves the image domain a1 units
horizontally and a2 units vertically
(b) if B > 0, then the map f .z/ D Bz scales (i.e., expands or shrinks) the domain by a
factor of B
(c) if 2 R, then the map f .z/ D e i z rotates the domain about the origin in a counter-
clockwise direction by an angle of .
Exercise 4.2. Justify that f .z/ D z C A, f .z/ D Bz, and f .z/ D e i z map circles onto
circles and lines onto lines. Try it out!
If we think of a line as a circle with infinite radius, then we can say that the functions
preserve circles. That is, they map circles onto circles.
The inversion map is f .z/ D z1 and if we write z D r e i ,
1 1 1 i
f .z/ D D i D e :
z re r
1 1 i
Thus, the function f .z/ D z scales the domain by a factor of r and the factor e reflects
it across the real axis.
Exploration 4.3. Open ComplexTool. We want to explore the image of the unit circle
under f .z/ D 1z . To do this, choose the option Circular grid. In the middle panel
below this, select Interior circles: to be 1 and Rays: to be 0. This should give
the domain in the z-plane as the unit circle. Click on the down arrow H above the image
and chose Re: [-3,3] Im: [-3,3] for the z-plane. Do this also for the w-plane.
Left click on the Graph button to produce the image of the unit circle in the right box.
Left click on the domain circle and drag it around. See what happens to the image as you
do this.
(a) As you move the center of the domain circle away from the origin in the z-plane, what
happens to the size of the image circle in the w-plane?
(b) What shape does the image take when the domain circle intersects the origin?
F IGURE 4.7. The image of a circle not intersecting the origin under the map f .z/ D z1 .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 203 — #221

i i
F IGURE 4.8. The image of a circle intersecting the origin under the map f .z/ D z1 .
(c) What happens to the size of the image circle as the domain circle moves off the screen?
(d) Can you move the domain circle to a position so that the image is neither a circle nor
a line?
(e) Explain your observation in parts (a)–(d) in terms of the fact that the function f .z/ D
1
z
scales the domain by a factor of 1r and reflects it across the real axis
Try it out!
The inversion map f .z/ D z1 maps circles to either circles or lines. Let’s look at the
image of vertical lines under f .z/ D z1 . If the line L1 starts at the bottom of the imaginary
axis and travels upward, it can be described as z D 0 C iy, where y varies from 1 to 1.
The image of L1 is
1 1 1
D Di :
z iy y
Because the real part of the image is zero, L1 is mapped into the imaginary axis. As y starts
at 1 and increases, the image starts at 0 and moves upward. When y passes through 0,
the image will wrap around from the top of the imaginary axis and go to the bottom of
the imaginary axis. As y continues up along the positive imaginary axis, the image will
continue up from the bottom of the imaginary axis. You can use the applet to visualize this.
Open ComplexTool and enter the function f .z/ D 1=z. Chose the window size of Re:
[-3,3] Im: [-3,3] for the z-plane and the w-plane. Click on the box next to the
Sketch feature so that a checkmark appears. Then click on the Graph button. Place the
cursor arrow on the imaginary axis in the z-plane, hold down the left mouse button, and
drag the cursor upward. As you do so, the image in the w-plane should appear (see Figure
4.9).
Let’s see what the image of the vertical line with Re z D 12 is. In the z-plane, let
z D x C iy. Because
1 z 1 x y
D D 2z D 2 i
z zz jzj x C y2 x2 C y2
if we let w D u C iv in the image domain, then we have
x y
uD and vD :
x2 C y2 x2 C y2
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 204 — #222

i i
F IGURE 4.9. The image of a vertical line on the imaginary axis under the map f .z/ D z1 .
In the z-plane, the vertical line with Re z D 21 can be described by z D 1

2 C iy. So the
image of the vertical line under the inversion map is
1
1 2 y 2 4y
D u C iv D 1
Ci 1
D 2 Ci 2 :
z 4 C y2 4 C y2 4y C 1 4y C 1
What is the shape described by u and v? It is a circle centered at u D 1 of radius 1 (see

Figure 4.10). To see this, notice
2 2
2 2 2 4y
.u 1/ C v D 1 C
4y 2 C 1 4y 2 C 1
2 2 2
. 4y C 1/ C . 4y/
D
.4y 2 C 1/2
16y C 8y 2 C 1
4
D
.4y 2 C 1/2
D1
which is the equation of a circle centered at u D 1 of radius 1.
1 1
F IGURE 4.10. The image of a vertical line with Re z D 2 under the map f .z/ D z.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 205 — #223

i i
1
Exercise 4.4. Using this approach, show that f .z/ D z
maps the vertical line Re z D c

1 2

1 2
onto the circle u 2c C v 2 D 2c . Try it out!
1
The inversion map f .z/ D z
takes circles and lines onto circles and lines.
(a) A circle not intersecting the origin is mapped onto a circle not intersecting the origin.
(b) A line intersecting the origin is mapped onto a line intersecting the origin.
(c) A circle intersecting the origin is mapped onto a line not intersecting the origin, and
vice versa.
The maps we have discussed so far are special cases of Möbius transformations
Az C B
M.z/ D ;
Cz C D
where A; B; C; D 2 C and AD ¤ BC .
AzCB
Exercise 4.5. What happens to M.z/ D C zCD
if AD D BC ? Try it out!
If A D 1, C D 0, and D D 1, we have M.z/ D z C B, a translation map. If B D 0,
C D 0, and D D 1, we have M.z/ D Az, which is a scaling, a rotation, or both. If A D 0,
B D 1, C D 1, and D D 0, we have M.z/ D z1 , the inversion map. Because
A
Az C B C
.C z C D/ ADC
CB A B ADC
D D C
Cz C D Cz C D C Cz C D
if we let

1 AD A
f1 .z/ D C z C D; f2 .z/ D ; and f3 .z/ D B zC ;
z C C
then the Möbius transformation M.z/ D CAzCB zCD

can be expressed as .f3 ı f2 ı f1 /.z/.
We know how the maps f1 , f2 , and f3 affect a given domain, and since they map circles
(lines) to circles (lines), a Möbius transformation will also. By including the point at 1 in
our domain and range, the Möbius transformation is a one-to-one and onto function. Thus,
its inverse function exists and is
1 Dz B
M .z/ D ;
Cz C A
so the inverse of a Möbius transformation is a Möbius transformation.
1 Dz B
Exercise 4.6. Starting with a Möbius transformation M , compute that M is C zCA
.
Try it out!
We can determine the image of a circle under a Möbius transformation M since a circle
is determined by three distinct points. To do this, take three points on the circle, find their
images under M , and then determine the circle on which they lie.
Example 4.7. Let’s find the image of the unit circle fz j jzj D 1g under the map
z
M.z/ D :
1 z
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 206 — #224

i i
Choose three points on the unit circle. We will choose i , 1, and i . Then
1 1 1 1 1
M.i / D C i; M. 1/ D ; and M. i / D i;
2 2 2 2 2
and 12 C 12 i , 1
2
, and 1
2
1
2
i lie on the line, fz j Refzg D 1
2
g, which is the image of
the unit circle.
Question: What is the image of the unit disk fz j jzj < 1g under the map 1 z z ? Möbius
maps will send the interior of the domain into the interior of the image. Since M.0/ D 0,
we know that the unit disk is mapped onto the right half-plane fw j Refwg > 12 g.
z
F IGURE 4.11. The image of the unit disk under the Möbius map M.z/ D 1 z.
Exercise 4.8. Using the approach in Example 4.7 determine the image of the unit disk
fz j jzj < 1g under the transformation
z i
M.z/ D :
z
Use ComplexTool to check your answer. Try it out!
Suppose that we want to map a given circle or line C1 to another circle or line C2 . How
do we construct a function that will do this? Three points uniquely determine a circle and
Möbius transformations send circles (lines) to circles (lines). Choose three points z1 , z2 ,
and z3 on C1 and three points w1, w2 , and w3 on C2 . Then a Möbius transformation M.z/
with
M.z1 / D w1 ; M.z2 / D w2 ; and M.z3 / D w3
will map C1 onto C2 .
We can construct M.z/ by using the cross-ratio formula
.w w1 /.w2 w3 / .z z1 /.z2 z3 /
D : (4.1)
.w w3 /.w2 w1 / .z z3 /.z2 z1 /
If one of the terms zi or wi is 1 (i.e., one of the circles C1 or C2 is a line), then we omit
the expressions containing that term. For example, if w1 D 1, then (4.1) becomes
w2 w3 .z z1 /.z2 z3 /
D :
w w3 .z z3 /.z2 z1 /
Let’s use (4.1) to find a Möbius transformation that maps z1 to w1 , z2 to w2 , and z3 to w3 .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 207 — #225

i i
Example 4.9. Suppose that we want to determine a Möbius transformation that maps the
vertical line l D fz j Refzg D 12 g onto the unit circle, C . Pick three points on l . We
choose
1 1 1
z1 D ; z2 D C i ; and z3 D 1:
2 2 2
Next, we pick three points on C . Let’s choose
w1 D 1; w2 D i; and w3 D 1:
Using (4.1), we have

1
zC 2 .w C 1/.i 1/
1 1
D :
. 2
Ci 2
/ C 12 .w 1/.i C 1/
Solving for w yields
z
wD :
1Cz
z
Thus, the Möbius transformation M.z/ D 1Cz
maps l onto C . It is the inverse of the
transformation used in Example 4.7.
Exercise 4.10. In Example 4.9, work out the details to show that
1
zC 2 .w C 1/.i 1/ z
1 1 1
D simplifies to w D :
. 2 Ci 2/ C 2
.w 1/.i C 1/ 1Cz
Try it out!
In Example 4.9, we found a Möbius transformation M that maps the line l onto the
circle C . We see that M maps the right half-plane D D fz j Refzg > 12 g to the unit
disk because M.0/ D 0. It could have happened that we had constructed a Möbius trans-
formation that maps the right half-plane to the exterior part of the unit circle instead of the
unit disk. What can we do at the beginning to guarantee that the Möbius transformation
we construct will map the domain onto the region we want? We could have switched the
values of w2 and w3 so that w1 D 1, w2 D 1, and w3 D i , and constructed a different
Möbius transformation that maps l to C . Then, the right half-plane D would have been
mapped onto the exterior of the unit circle. To guarantee that the Möbius transformation
we construct will map our initial domain onto the region we want, we need to choose the
order of the points w1, w2, and w3 correctly. How do we do this? As we travel along l
from z1 D 12 to z2 D 12 C i 12 to z3 D 1, the right half-plane D is on the right side of
the line l . To guarantee that M maps D onto the unit disk, we need to choose the order of
the points w1 , w2 , and w3 so that the unit disk is also on the right side of the unit circle C .
This happens in Example 4.9, because we go from w1 D 1 to w2 D i to w3 D 1.
Exercise 4.11. Determine a Möbius transformation that maps the unit disk fz j jzj < 1g
onto the upper half-plane fz j Imfzg > 0g. Use ComplexTool to check your answer. Try it
out!
There is a nice short video called “Moebius Transformations Revealed” by Douglas
Arnold and Jonathan Rogness that illustrates some of the connections between Möbius
transformations and motions of the sphere. It can be found by searching the internet.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 208 — #226

i i
4.3 The Family S of Analytic,

Normalized, Univalent Functions
We will be discussing mapping problems in complex analysis. These problems deal with
the properties of a collection of functions that map one domain onto certain image domains.
We need some background material. Let G C be a simply-connected domain and let
D D fz W jzj < 1g, the unit disk.
Definition 4.12. A function f is univalent in G if f is one-to-one in G.
Univalent analytic functions have inverse functions that are analytic.
Example 4.13. Suppose we want to prove that f .z/ D .1 C z/2 is univalent in D. A
standard argument for that is to let z1 ; z2 2 D and suppose f .z1 / D f .z2 /. Then
f .z1 / D f .z2 / ) .1 C z1 /2 D .1 C z2 /2
) 1 C 2z1 C z12 D 1 C 2z2 C z22
) z12 z22 C 2.z1 z2 / D 0
) .z1 z2 /.z1 C z2 C 2/ D 0:
Since jz1 j; jz2 j < 1, we know that z1 C z2 C 2 ¤ 0. Hence, we must have z1 z2 D 0.

So, z1 D z2 . Thus, f is one-to-one.
The image of D under the map f .z/ D .1 C z/2 is shown in Figure 4.12.
F IGURE 4.12. The image of the unit disk under the map f .z/ D .1 C z/2 .
Exploration 4.14. From Example 4.13, we know that f .z/ D .1 C z/2 is univalent on
D. It can be shown that f .z/ D .1 C z/4 is not univalent on D (see Exploration 4.15).
Use ComplexTool to graph the image of D under the analytic function f .z/ D .1 C z/2
and under the analytic function f .z/ D .1 C z/4 . What aspects of the images suggest
that a function is univalent or is not univalent? Explore by plotting the following further
functions in ComplexTool and conjecture which are univalent:
1 2
(a) g1 .z/ D z z2 (b) g2 .z/ D z 2
z
3 2
(c) g3 .z/ D 2z z2 (d) g4 .z/ D z C 4
z
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 209 — #227

i i
4.3. The Family S of Analytic, Normalized, Univalent Functions 209
z z2
(e) g5 .z/ D 1 z
(f) g6 .z/ D 1 z
z
(g) g7 .z/ D .1 z/2
.
Try it out!
Exploration 4.15. Prove that f .z/ D .1 C z/4 is not univalent in D.
One way to do this is to find two distinct points z1 ; z2 2 D such that f .z1 / D f .z2 /. You
can use ComplexTool to help you find z1 and z2 . Plot the image of D under f .z/ D .1Cz/4 .
You can increase the size of the image by clicking on the left button on the mouse and
decrease it by clicking on the right mouse button. Check the Sketch box in the top middle
section. The Sketch command allows you to draw a shape in the domain on the left and
see its image under the function on the right. For example, draw a line along the imaginary
axis from 0 to i and a line along the imaginary axis from 0 to i ; the image curves meet
at f .i / D f . i /. Compute f .i / and f . i / to prove that this is true. This does not prove
that f is not univalent in D since i; i … D. Use the Sketch feature of ComplexTool
to help you find two points z1 ; z2 2 D such that f .z1 / D f .z2 /. To delete the shapes
you have drawn, use the Clear all button and then regraph your image. Hint: Using
ComplexTool find two lines in the original domain that are mapped to the line on the real
axis from 4 to 0. Parametrize the lines so that for each t, the image of the parametrized
lines under f give the same image point. Try it out!
In Exploration 4.14, you may have noticed that the image of D under g2 .z/ D z 21 z 2
is similar to its image under g3 .z/ D 2z z 2 . This is because g3 D 2g2 . We want to
avoid such repetitions. To do so, we will normalize all functions in the family of analytic
univalent functions defined on D. To do this suppose f1 is univalent and analytic in the
simply connected domain G ¤ C. The Riemann mapping theorem can be stated in the
following form:
Theorem 4.16 (Riemann Mapping Theorem). Let a 2 G. Then there exists a unique
function f2 W G ! C such that
(a) f2 .a/ D 0 and f20 .a/ > 0
(b) f2 is univalent
(c) f2 .G/ D D.
Thus f3 D f1 ı f2 1 maps D to f1 .G/ with f3 being univalent and analytic. So when
studying mappings of simply-connected domains, we can just let D be our domain. Let
f3 W D ! C be univalent and analytic. Since f3 is analytic, it has a power series about the
origin,
f3 .z/ D ˛0 C ˛1 z C ˛2 z 2 C ˛3z 3 C ;
that converges in D. Adding a constant translates the image domain and does not affect the
univalency. Hence
f4 .z/ D f3 .z/ ˛0 D ˛1 z C ˛2 z 2 C ˛3 z 3 C
is also univalent and analytic in D. Next, ˛1 ¤ 0 because f4 is univalent, so f40 .z/ ¤ 0
(for all z 2 D); but f40 .0/ D ˛1. Let
1 ˛2 2 ˛3 3
f5 .z/ D f4 .z/ D z C z C z C :
˛1 ˛1 ˛1
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 210 — #228

i i
Multiplying f4 by ˛11 rotates or stretches (or shrinks) the image domain. Hence f5 is
still univalent and analytic in D. We have normalized our original function f3 so that
f 0 .0/ D 1, and f .0/ D 0.
Definition 4.17. The family of analytic, normalized, univalent functions is denoted by S
(from the German word “schlicht” that means “simple” or “plain”); that is,
S D ff W D ! C j f is analytic and univalent with f .0/ D 0; f 0 .0/ D 1g:
Thus f 2 S implies f .z/ D z C a2 z 2 C a3 z 3 C .

Exercise 4.18. Show that f .z/ D z C a2 z 2 is univalent in D ” ja2 j 12 . Try it out!
Example 4.19. In Exploration 4.14, you graphed the image of D under g2 .z/ D z
1 2
2
z . While computer images are helpful, they can be misleading and inaccurate. So it is
important for us to determine them analytically. How can we determine g2 .D/ analytically?
We find the image of the boundary of D:
1 2i
w D g2 .e i / D e i e
2
1
D .cos C i sin / .cos 2 C i sin 2/
2
1 1
D cos cos 2 C i sin sin 2
2 2
D u C iv:
Thus, g2 .@D/ is parametrized by

1
u./ D cos cos 2
2
1
v./ D sin sin 2:
2
This is a cardiod, which is also known as an epicycloid with one cusp (see Figure 4.13).
Definition 4.20. An epicycloid is the path traced out by a point on a circle of radius b
rolling on the outside of a circle of radius a:

a
x./ D.a C b/ cos b cos C1
b

a
y./ D.a C b/ sin b sin C1 :
b
Exploration 4.21. In Exercise 4.18, you showed that f .z/ D z C a2 z 2 is univalent in

D ” ja2 j 21 . We want to make a conjecture about the generalization of this result. Use
ComplexTool to graph f .z/ D z C a3 z 3 for various values of a3 . What do you conjecture
is the bound on a3 for which f is univalent on D? Do the same for f .z/ D z C a4 z 4 ,
f .z/ D z C a5 z 5 , etc. What do you conjecture is the bound on an for which f .z/ D
z C an z n is univalent on D? What do you conjecture f .D/ is when an D n1 ? Try it out!
Let’s determine f .D/ analytically for a few examples that were included in Exploration
4.14.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 211 — #229

i i
1 2
F IGURE 4.13. The image of the unit disk under the map z 2z .
Example 4.22. If
z
fr .z/ D 2 S;
1 z
1 1
P
because D z n , we can multiply by z to get
1 z nD0
1
X
z
f .z/ D D zn D z C z2 C z3 C :
1 z
nD1
This is the Möbius transformation that maps D onto the right half-plane whose boundary
is the line Refwg D 12 (see Figure 4.14).
F IGURE 4.14. The image of the unit disk under the analytic right half-plane map in S.
Example 4.23. The power series for

z
fk .z/ D 2S
.1 z/2
1
can be found by differentiating the series for 1 z
and then multiplying by z:
1
X
z
D nz n D z C 2z 2 C 3z 3 C :
.1 z/2
nD1
For this function an D n for all n. We now show that the image of D under fk is a slit
domain. That is, it is a domain consisting of the entire complex plane except that a ray (or
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 212 — #230

i i
a slit) is cut out of it. To determine fk .D/, let
1Cz 1
u1 .z/ D ; u2 .z/ D z 2 ; u3 .z/ D Œz 1:
1 z 4
Then, " #
2
1 1Cz z
u3 ı u2 ı u1 .z/ D 1 D :
4 1 z .1 z/2
The Möbius transformation u1 maps D onto the right half-plane whose boundary is the
imaginary axis, u2 is the squaring function, and u3 translates the image one space to the
left and multiplies it by a factor of 41 .
Thus the image D is the entire complex plane except for a slit along the negative real
axis from w D 1 to w D 41 (see Figure 4.15). The function fk .z/ D .1 zz/2 is known
as the Koebe function.
1
-
4
F IGURE 4.15. The image of the unit disk under the Koebe function.
Exploration 4.24. It is difficult to interpret the image of D under the Koebe function using
ComplexTool. One way is to use increasing values that approach 1 for the radius of circles
in D. We can do this by using the left box in the “ radius ” feature in the center
panel of the applet. Graph D under the map .1 zz/2 using the values of 0:8, 0:85, 0:9, 0:95,
and 0:999 in the left box in the “ radius ” feature. Try it out!
Suppose we have an analytic function f with a Taylor series representation f .z/ D
z C a2 z 2 C a3 z 3 C . One question to ask is for what values of an is f in the family of
schlicht functions? If all an D 0 except possibly a2 , then f .z/ D z C a2 z 2 . By Exercise
4.18, we know that ja2 j 12 if and only if f 2 S . The function f .z/ D z 12 z 2 from
Example 4.19 is an extremal function. An extremal function is on the boundary between
those that satisfyˇ a condition
ˇ and those that do not. Here, f .z/ is extremal, because if we
increase ja2 j D ˇ 21 ˇ even just a little bit, then f .z/ D a C a2 z 2 is no longer schlicht. In
general, how large can jan j be and f .z/ D z C a2 z 2 C a3 z 3 C still be schlicht? For
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 213 — #231

i i
the Koebe function,

1
X
z
D nz n ;
.1 z/2 nD1
and so we have an D n. This led Bieberbach in 1916 to make his famous conjecture.
Bieberbach Conjecture. For f 2 S , jan j n, for all n. In particular, ja2 j 2.
Because the image of D under the Koebe function covers all of C except a slit along
the real axis, it seems plausible that the the Bieberbach conjecture is true with the Koebe
function being extremal. The conjecture is true but was not proved until 1984 by de Branges
[7].
We say an inequality involving a function sharp if it is impossible to improve it (that
is, we cannot decrease an upper bound or increase a lower bound). We can show that
an inequality is sharp by finding a function for which the inequality becomes equality,
an extremal function. For the Bieberbach conjecture or de Branges’ theorem, the Koebe
function is extremal.
There is another case in which the Koebe function is extremal. We know that if f 2 S ,
then f .D/ is not the entire complex plane, so there is a point a 2 C such that a … f .D/.
This leads to the question of how small can jaj be. For example, if f .z/ D z, then jaj D 1;
if f .z/ D 1 z z , the right half-plane mapping in Example 4.22, then jaj D 12 . The answer is
that for all f 2 S , jaj 14 . This is known as the Koebe 14 -Theorem. The Koebe function is
extremal, because jaj D 14 for the Koebe function. For these reasons and others, the Koebe
function is important in the study of schlicht functions.
z tz 2
Exploration 4.25. If f .z/ D , where 0 t 1, what is f .D/ when t D 0? What
.1 z/2
is f .D/ when t D 1? Using your answers to these two questions and not ComplexTool,
make a conjecture of what f .D/ is, when 0 < t < 1. Now, use ComplexTool to modify
or strengthen your conjecture. Using ComplexTool, what happens to f .D/ for t > 1? For
t D i k; 0 k 1? Try it out!
The family S has been studied extensively. Here are a few facts about normalized,
analytic univalent functions that will be used later:
(a) (uniqueness in the Riemann mapping theorem) Let G ¤ C be a simply-connected
domain with a 2 G. Because of the Riemann mapping theorem, the map f 2 S that
maps D onto G with f .0/ D a and f 0 .0/ > 0 is unique.
(b) (de Branges’ Theorem) For f 2 S , jan j n, for all n.
(c) (Koebe 14 -Theorem) The range of every function in class S contains the disk G D
fw W jwj < 14 g. This consequence of the fact that ja2 j 2 was proved by Bieberbach
in 1916.
(d) Let f 2 S . Then f .D/ omits a value on each circle fw W jwj D Rg where R 1.
There is no function f 2 S for which f .D/ contains @D, the unit circle.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 214 — #232

i i
4.4 The Family SH of Normalized,

Harmonic, Univalent Functions
About the same time that de Branges proved the Bierbach conjecture, Clunie and Sheil-
Small studied a family, SH , of complex-valued harmonic functions that contained S as a
proper subset and considered some of the properties of SH that had been investigated in S .
A function .x; y/ is harmonic if and only if xx C yy D 0.

Definition 4.26. A continuous function f D u C iv defined in G is a complex-valued har-
monic function in G if u and v are real harmonic (but not necessarily harmonic conjugates)
in G.
Example 4.27. The function
f .x; y/ D u.x; y/ C iv.x; y/ D .x 2 y 2 / C i 2xy

is complex-valued harmonic because
uxx C uyy D 2 2D0

vxx C vyy D 0 C 0 D 0:
Exercise 4.28. Show that

1 1 2
f .x; y/ D u.x; y/ C iv.x; y/ D x C x 2 y C i.y xy/
2 2
is complex-valued harmonic. Try it out!
Although harmonic functions are more general than analytic functions, some theorems
about analytic functions have equivalent forms for harmonic functions. These include the
mean-value theorem, the maximum-modulus theorem, Liouville’s Theorem, and the Ar-
gument Principle. However, by considering all harmonic functions instead of the subclass
of analytic functions we can sometimes get more information. For example, we can use
harmonic functions to study of minimal surfaces (see Chapter 2 on Minimal Surfaces).
One way of thinking of a function f .x; y/ D u.x; y/ C iv.x; y/ as being analytic is
that f can be expressed in terms of z D x C iy only without using z D x iy. Hence, the
function f D z 2 is analytic while f D zz is not. To explore this idea, let D z D x C iy
and D z D x iy. Then, we can formally write x D 21 . C / and y D 2i 1
. /.
Applying the chain rule on f .x.; /; y.; // with D z and D z, we can show that

@f 1 @u @v i @v @u
D C C (4.2)
@z 2 @x @y 2 @x @y

@f 1 @u @v i @u @v
D C C : (4.3)
@z 2 @x @y 2 @y @x
Exercise 4.29.
(a) Derive (4.2) and (4.3).
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 215 — #233

i i
4.4. The Family SH of Normalized, Harmonic, Univalent Functions 215
(b) Use these equations and the Cauchy-Riemann equations to prove that f .x; y/ D
@f
u.x; y/ C iv.x; y/ is analytic ” D 0.
@z
Try it out!
Exercise 4.30.
(a) Using x D 12 .z C z/ and y D 1
2i
.z z/, rewrite
1 1 2
2 2
in terms of z and z.
(b) Use Exercise 4.29 to determine if f is analytic.
(c) Show that all analytic functions are complex-valued harmonic, but not all complex-
valued harmonic functions are analytic.
Try it out!
The next theorem tells us that a complex-valued harmonic function defined on D is
related to analytic functions, and in fact can be expressed in terms of a canonical decom-
position.
Theorem 4.31. If f D u C iv is harmonic in a simply-connected domain G, then f D
h C g, where h and g are analytic.
Proof. If u and v are real harmonic on a simply-connected domain, then there exist analytic
functions K and L such that u D Re K and v D Im L. Hence,
KC K L L KCL K L
f D u C iv D Re K C i Im L D Ci D C D h C g:
2 2i 2 2
Exercise 4.32. Let

1 1 2
2 2
on D. In Exercise 4.28 you showed that f is harmonic. Find analytic functions h and g
such that f D h C g. Try it out!
We can use the applet ComplexTool to graph complex-valued harmonic functions. For
example, to graph the image of D under the harmonic function f .z/ D z C 21 z 2 , enter the
function in ComplexTool in the form z C 1=2 conj .z ^ 2/ (see Figure 4.16).
The harmonic function f .z/ D h.z/ C g.z/ can be written as
˚ ˚
f .z/ D Re h.z/ C g.z/ C i Im h.z/ g.z/ : (4.4)
˚
Hence, in Example 4.32, f .z/ D z C 12 z 2 can be written as f .z/ D Re z C 12 z 2 C
˚
i Im z 21 z 2 . In ComplexTool you can also enter the harmonic function in this form by
typing re.z C 1=2z ^ 2/C i*im .z 1=2z ^ 2/.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 216 — #234

i i
F IGURE 4.16. Image of D under the harmonic function f .z/ D z C 21 z 2 .

˚
Exercise
representations f .z/ D h.z/C g.z/ and f .z/ D Re h.z/C
4.33.˚ Prove that the
g.z/ C i Im h.z/ g.z/ are equivalent. Try it out!
Exploration 4.34. Graph the image of D under the following harmonic maps. Describe
characteristics that appear to be different for harmonic mappings as compared to analytic
mappings.
(a) f1 .z/ D z C 13 z 3

(b) f2 .z/ D Re 1 z z C i Im .1 zz/2
z 1 zC1
(c) f3 .z/ D 1 z 2
ez 1
i

pi 1Ce 3z 1 1CzCz 2
(d) f4 .z/ D Re ln i
C i Im 3
ln 1 2zCz 2
3 1Ce 3 z
z 1
(e) f5 .z/ D z C 2 ln.z C 1/ C .z C 1/e zC1 .
Try it out!
Since f D h C g where h and g are analytic, f has the series representation
1
X 1
X
f .z/ D an z n C bn z n :
nD0 nD1
Hence, we may normalize harmonic univalent functions as we normalized analytic univa-

lent functions.
Definition 4.35. Let SH be the family of complex-valued harmonic, univalent mappings
that are normalized on the unit disk. That is,
SH D ff W D ! C j f is harmonic, univalent
with f .0/ D a0 D 0; fz .0/ D a1 D 1g:
If we restrict this family by requiring b1 D 0, we have the family

O
SH D ff 2 SH j fz .0/ D b1 D 0g:
Notice S SH
O
SH .
Let’s look at some examples.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 217 — #235

i i
Example 4.36. In the next section we will prove that
1
f .z/ D h.z/ C g.z/ D z C z 2
2
O
is univalent and hence in SH . For now we will assume it and look at the image of D under
f (see Figure 4.16). The function f maps D onto the interior of the region bounded by a
hypocycloid with three cusps.
Definition 4.37. A hypocycloid is the curve produced by a fixed point on a small circle of
radius b rolling the inside of a larger circle of radius a. It has the parametric equations

a
x./ D.a b/ cos C b cos 1
b

a
y./ D.a b/ sin C b sin 1 :
b
In Example 4.53, we will see that f is a shearing of F .z/ D z 1=2z 2 which maps D
to an epicycloid (see Example 4.19).
Exploration 4.38.
(a) Use ComplexTool to plot the image of D under the analytic polynomial map F .z/ D

z 12 e i t 6 z 2 for t D 0; 1; 2; 3; 4; 5; 6. Describe what happens to the image as t varies.
(b) Use ComplexTool to plot the image of D under the harmonic polynomial map f .z/ D

z C 12 e i t 6 z 2 for t D 0; 1; 2; 3; 4; 5; 6. Describe what happens to the image as t varies.
(c) What differences do you notice between the images in (a) and (b) as t increases?
Explain why it is reasonable for these differences to occur.
Try it out!
Small Project 4.39.
(a) Use ComplexTool to plot the images of the following polynomials
Harmonic Functions Analytic Functions
(i) f1 .z/ D z C 31 z 3 (ii) F1 .z/ D z C 13 z 3
(iii) f2 .z/ D z C 14 z 2 C 14 z 2 C 13 z 3 (iv) F2 .z/ D z C 21 z 2 C 13 z 3
(v) f3 .z/ D z C 61 z 2 C 16 z 4 (vi) F3 .z/ D z C 16 z 6
(vii) f4 .z/ D z C 16 z 2 C 16 z 4 .
(b) List the similarities and differences between the images of D under harmonic and the
analytic functions.
(c) Questions to consider are: If a polynomial has three or more terms then how large can
its coefficients be in modulus to guarantee univalency? If a polynomial has three terms,
what difference does it make if the last two terms are z 2 and z 3 instead of z 2 and z 3 ?
(d) Plot your own examples of harmonic and analytic polynomials and see if the properties
from (b) are still valid.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 218 — #236

i i
Optional
Example 4.40. Let
1 2 1 2
z 2z 2z
f .z/ D h.z/ C g.z/ D
.1 z/2 .1 z/2

z z
D Re.h.z/ g.z// C i Im.h.z/ g.z// D Re C i Im :
1 z .1 z/2
We will prove that f is univalent in the next section. The image of D under f using
ComplexTool is shown in Figure 4.17.
z
z

F IGURE 4.17. Image of D under f .z/ D Re 1 z C i Im .1 z/2
.
ˇ
The image of D under the harmonic map f is the right half-plane fw 2 C ˇ Refwg
1 z
2 g, the same region as the image of D under the analytic map 1 z although the boundary
behavior is different.
Exploration 4.41.
(a) Use ComplexTool to plot D under the analytic right half-plane map 1 z z . Use the
Sketch box to draw radial lines from the origin to the boundary of D in the domain.
What are the images of points on the unit circle under this analytic map?
1 2 1 2
z z z
(b) Use ComplexTool to plot D under the harmonic right half-plane map .1 2z/2 .12 z/2 .
Use the Sketch box to draw radial lines from the origin to the boundary of D in the
domain. What is the image of points on the unit circle under this analytic map?
(c) Using (a) and (b), describe how the boundary behavior is different between the analytic
right half-plane map and this harmonic right half-plane map.
Try it out!
We are interested in harmonic univalent functions. However, results about univalent
functions can be difficult to obtain. So it is sometimes useful to consider locally univalent
functions instead of globally univalent. Locally univalent functions are functions that are
locally one-to-one. This means that there is a small neighborhood around a point z0 2 D
such as a small disk in D centered at z0 , and the function is one-to-one for points z in that
neighborhood. A locally univalent function may not be univalent in D.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 219 — #237

i i
Exploration 4.42. In Exploration 4.15, you proved that f .z/ D .1 C z/4 is not univalent
in D. This can be shown by seeing that the points 21 C 12 i and 12 12 i in D both map
to 14 under f .z/ D .1 C z/4 . However, f .z/ D .1 C z/4 is locally univalent in D. We
won’t prove this. To get a sense of what local univalence
ˇ means, use ComplexTool to graph
f .z/ D .1Cz/4 on the sector fz D r e i 2 D ˇ 2 < < 3 2
g. You can do this by entering
the values for into the left and right boxes for the “ ” feature in the center panel.
The point z0 D 21 C 12 i is in the sector. Also, the image indicates that f .z/ D .1 C z/4
is not univalent in it. This does not mean that f is not locally univalent. Local univalence
means that at z0 there is some neighborhood in which the function is one-to-one for points
z in the neighborhood. Use ComplexTool to determine a neighborhood of the points z0 in
which f is one-to-one. You can increase the size of the image by clicking on the left button
on the mouse and decrease the size by clicking on the right mouse button.
1
(a) z0 D 2
C 12 i (b) z0 D 1
2
1
2
i
5
(c) z0 D 6
C 16 i (d) z0 D 0:9 C 0:1i
Try it out!
Let’s look at the idea of local univalence.
Definition 4.43. A function f D h C g is locally univalent on G if Jf ¤ 0 on G, where
Jf is the Jacobian of f D u C iv:
ˇ ˇ
ˇu ˇ
ˇ x uy ˇ
Jf D det ˇ ˇ
ˇ vx vy ˇ
ˇ ˇ
ˇ.Re h/x C .Re g/x .Re h/y C .Re g/y ˇ
ˇ ˇ
D det ˇ ˇ:
ˇ .Im h/x .Im g/x .Im h/y .Im g/y ˇ
For analytic functions F , the Cauchy-Riemann equations yield .Re F /y D .Im F /x and
.Im F /y D .Re F /x . Hence we have
ˇ ˇ
ˇ.Re h/x C .Re g/x .Im h/x .Im g/x ˇˇ
ˇ
Jf D det ˇ ˇ
ˇ .Im h/x .Im g/x .Re h/x .Re g/x ˇ
D .Re h/2x .Re g/2x C .Im h/2x .Im g/2x
D jh0 j2 jg0 j2 :
Thus, we want jh0 j2 jg0 j2 ¤ 0.

Besides local univalence, another important property of these functions is being sense-
preserving. A continuous function f is sense-preserving (or orientation-preserving) if it
preserves orientation. Let f1 , f2 be defined on the punctured disk D f0g by
1
f1 .z/ D and f2 .z/ D z:
z

Both map the unit circle, @D, onto itself, and both map the points A D 1, B D e i 4 , and

C D i to the points A0 D 1, B 0 D e i 4 , and C 0 D i , respectively (see Figure 4.18). In
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 220 — #238

i i
C 1 C
B f1 (z) = z B f2 (z) = z
LHS A A¢ LHS A A¢
RHS
ID B¢ ID B¢
C¢ LHS C¢
F IGURE 4.18. f1 is sense-preserving and f2 is sense-reversing.
the domain, as we travel along the unit circle in a counterclockwise direction (i.e., going
from A to B to C ), the region to the left of this path is D. We will call this region the
left-hand side domain (LHS) and the region to the right of the path we will call the right-
hand side domain (RHS). So, C D is RHS. Where does D, a LHS, get mapped by the
functions? As we travel along @D in a counterclockwise direction in the domain set, the
image curve under both functions will be @D traversed in a clockwise direction. So, in the
image, D is now RHS while C D is LHS. The function f1 maps the point 12 2 D to
2 2 C D and so f1 maps the LHS onto the LHS. Functions that map the LHS onto the
LHS are sense-preserving. On the other hand, f2 maps 12 2 D to 12 2 D and so f2 maps
the LHS onto the RHS. Functions that map the LHS onto the RHS are sense-reversing.
When we travel counterclockwise along a simple closed contour 2 G, there exists
a left-hand side domain (LHS) and a right-hand side domain (RHS). Let the image curve
be f . /. The function f is sense-preserving if the original LHS domain with regard to
is mapped to the LHS domain with regard to f . /. The function f is sense-reversing if
the LHS domain with regard to is mapped to the RHS domain with regard to f . /. All
analytic functions are sense-preserving, while some complex-valued harmonic functions
are sense-preserving and some are sense-reserving.
Exploration 4.44. For the following harmonic functions, use ComplexTool to conjecture
if the function is (a) locally univalent and (b) sense-preserving. Hint: for sense-preserving,
you can use the Sketch feature to draw a counterclockwise curve on the unit circle and see
its image under the function.
(a) z C 2z (b) z C 21 z
(c) z C 2z 2 (d) z C 21 z 2
(e) 2z 2 C z (f) 12 z 2 C z
Try it out!
Now we will need the following definition.
Definition 4.45. !.z/ D g0 .z/= h0 .z/ is the dilatation of f D h C g.
There is a connection between the dilatation of a harmonic function and its locally
univalent and sense-preserving nature.
Theorem 4.46 (Lewy). The function f D h C g is locally univalent and sense-preserving
” j!.z/j < 1 for all z 2 G.
Exercise 4.47. Compute !.z/ for the functions in Exploration 4.44. Use your results from
Exploration 4.44 to verify Lewy’s Theorem for the functions in that Exploration. Try it
out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 221 — #239

i i
4.5. The Shearing Technique 221
Exercise 4.48. Show that for z 2 D, j!k .z/j < 1 for

(a) !1 .z/ D e i z, where 2 R
(b) !2 .z/ D z n , where n D 1; 2; 3; : : :

zCa
(c) !3 .z/ D , where jaj < 1
1 C az
(d) !4 .z/, the composition of any of !1 , !2 , and !3 .
Try it out!
Remark 4.49. Let f D h C g be a sense-preserving harmonic map with dilatation ! D
g0 = h0 . If j!.z/j D 1 for all z in an arc of @D, then the image of under f is either
(a) a concave arc (i.e., there are two points in f .D/ such that the line connecting them
goes outside of f .D/) (see Example 4.36); or
(b) a stationary point (see Example 4.40).
We discuss the dilatation more in Section 4.6.
4.5 The Shearing Technique

Finding examples of univalent harmonic mappings that are not analytic is not easy. The
shearing technique by Clunie and Sheil-Small is a useful way to construct new examples of
univalent harmonic mappings. Shearing starts with an analytic function F and a dilatation
! that have certain properties. Then, we can find a univalent harmonic function f D hC g
by writing F as F D h g and ! as ! D g0 = h0 , and solving for h and g. Before we
proceed, we need to discuss a property of F .
Definition 4.50. A domain is convex in the direction e i' if for every a 2 C the set
\ fa C te i' W t 2 Rg is either connected or empty. A domain is convex in the horizontal
direction (CHD) if every line parallel to the real axis has a connected intersection with .
CHD not CHD
Exercise 4.51. For which values of n D 1; 2; 3; : : : do the following functions map D onto
a CHD domain?
(a) f .z/ D z n
1 n
(b) f .z/ D z n
z (see Example 4.19 and Definition 4.20)
z
(c) f .z/ D .1 z/n (see Examples 4.22 and 4.23 to get you started).
Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 222 — #240

i i
We have a theorem by Clunie and Sheil-Small that forms the basis of the shearing
technique.
Theorem 4.52. Let f D h C g be a harmonic function that is locally univalent in D (i.e.,
j!.z/j < 1 for all z 2 D). The function F D h g is an analytic univalent mapping of D
onto a CHD domain ” f D h C g is a univalent mapping of D onto a CHD domain.
Before we prove Theorem 4.52, let’s look at an example.
Example 4.53. The shearing technique starts with a univalent analytic function F that
maps D onto a CHD domain and a dilatation ! where j!.z/j < 1 for all z 2 D. Then, we
construct a univalent harmonic function f D h C g by writing F as F D h g and ! as
! D g0 = h0 , and solving for h and g.
In Example 4.36, we claimed that the harmonic polynomial f .z/ D z C 21 z 2 (see
Figure 4.19) is univalent and is related to the analytic function F .z/ D z 21 z 2 (see Figure
4.20). We can use Theorem 4.52 to show this.
Let’s start with the analytic univalent function
1 2
F .z/ D h.z/ g.z/ D z z
2
that maps @D to an epicycloid with one cusp (see Example 4.19) resulting in a CHD do-
F IGURE 4.19. Image of D under the harmonic map f .z/ D z C 21 z 2 .
1 2
F IGURE 4.20. Image of D under the analytic map F .z/ D z 2z .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 223 — #241

i i
main. Choose !.z/ D g0 .z/= h0 .z/ D z and apply the shearing technique:
h0 .z/ g0 .z/ D 1 z ) h0 .z/ zh0 .z/ D 1 z
0
) h .z/ D 1
) h.z/ D z:
Since g0 .z/ D zh0 .z/ D z, we have g.z/ D 12 z 2 . Both h and g are normalized; that is,
h.0/ D 0 and g.0/ D 0. So the corresponding harmonic univalent function is
1
f .z/ D h.z/ C g.z/ D z C z 2 2 SH O
;
2
which was derived from shearing the analytic univalent function
1 2
F .z/ D z z 2S
2
with the dilatation !.z/ D z.
Remark 4.54. This technique is known as shearing. The word shear means to cut (as in
shearing sheep). As in Theorem 4.52, suppose F D h g is an analytic univalent function
convex in the horizontal direction. The harmonic shear is
f D hC g D h gCgC g D h g C 2 Refgg;
which differs from the analytic function by the adding of a real function. Geometrically,
this takes the image F .D/, that is convex in the horizontal direction and is formed from
the analytic univalent function F , and cuts it into horizontal slices. These slices are then
translated or scaled continuously to form the image f .D/ that is also convex in the hori-
zontal direction and is formed from the harmonic univalent function f (see Figure 4.21).
This is why the method is called shearing.
cut into horizonal

put all the
slices
horizontal
slices
together
each slice gets

scaled and
translated
horizontally
F IGURE 4.21. Shearing an analytic function to preserve univalency.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 224 — #242

i i
Exercise 4.55. Let f D h C g with h.z/ g.z/ D z and !.z/ D z. Compute h and g so
that f 2 SH
O
and use ComplexTool to sketch f .D/.
Exercise 4.56. Let f D h C g with h.z/ g.z/ D z 31 z 3 and !.z/ D z 2 . Compute h
and g so that f 2 SH
O
and use ComplexTool to sketch f .D/.
For exploring shears of functions, the applet ShearTool (see Figure 4.22) is better than
ComplexTool, because it allows you to see the image of D under a shear of h.z/ g.z/
without having to compute the harmonic function f D h C g. When using ShearTool,
enter an analytic function that is convex in the horizontal direction in h-g= box in the
upper left section and enter the dilatation in the ! box below it. The click on the Graph
button.
F IGURE 4.22. The applet ShearTool.
In Example 4.53, we sheared h.z/ g.z/ D z 12 z 2 with !.z/ D z. Entering them

into ShearTool we get the image of a hypocycloid with three cusps (see Figure 4.23).
Exploration 4.57.
(a) Use ShearTool to graph the image of D under f D h C g where h.z/ g.z/ D z and
!.z/ D z. Note the difference between this and Exercise 4.55.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 225 — #243

i i
1 2
F IGURE 4.23. The image of D when shearing h.z/ g.z/ D z 2z with !.z/ D z.
1 3
(b) Use ShearTool to graph the image of D under f D hC g where h.z/ g.z/ D z 3
z
and !.z/ D z 2 . Note the difference between this and Exercise 4.56.
Try it out!
To prove Theorem 4.52, we will use the lemma:
Lemma 4.58. Let C be a CHD domain and let be a real-valued continuous function
in . Then the map ‰.w/ D w C .w/ is one-to-one in ” ‰ is locally one-to-one. If
‰ is one-to-one, then its range is a CHD domain.
Proof. ()) Trivial.

(() Suppose the map ‰.w/ D w C .w/ is not one-to-one. Then, there are distinct
points w1 D u1 C iv1 , w2 D u2 C iv2 2 such that ‰.w1/ D ‰.w2/. Since is
real valued, we have that Im.‰.w1 // D Im.w1 C .w1 // D Im.w1 / D v1 . Similarly,
Im.‰.w2 // D v2 . Since ‰.w1 / D ‰.w2 /, we have v1 D v2 . Let’s say k D v1 D v2 for
e R ! R by ˆ.u/ D u C .u C i k/. Because ‰ is not
a constant k 2 R. Define ˆ W
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 226 — #244

i i
one-to-one, we have
ˆ.u1 / D u1 C .u1 C k/ D Ref‰.w1/g D Ref‰.w2/g D u2 C .u2 C k/ D ˆ.u2 /:
Thus, ˆ is not a strictly monotonic function. Suppose u0 is the point at which the mono-
tonicity of ˆ changes. Because ˆ maps into R, it cannot be locally one-to-one in any
neighborhood of u0 . Hence, ˆ is not locally one-to-one, and ‰ is not locally one-to-one.
Geometrically, ‰ acts as a shear in the horizontal direction and hence its range is CHD.
Proof of Theorem 4.52. ()) Assume f D h C g is one-to-one and D f .D/ is CHD

(see Figure 4.24). Then f D h g C g C g D h g C 2 Refgg. So
1 1 1
.h g/ ı f .w/ D .f 2 Refgg/ ı f .w/ D w 2 Refg.f .w//g D w C p.w/
may be defined in , where p is real valued and continuous. Since f is locally 1 1,

jg0 j < jh0 j ” g0 .z/ ¤ h0 .z/, 8z 2 D. Hence h g is locally 1 1 in D, and thus
w ! w C p.w/ is also locally 1 1 on since it is the composition of locally 1 1
functions. By Lemma 4.58, w ! w C p.w/ is univalent and its range is CHD. Hence,
.h g/.z/ D Œw C p.w/ ı f .z/ is univalent, being the composition of univalent functions,
and its image is CHD.
f –1(w) = z
W
f
D f –1
h–g w + p(w)
(h – g)(z)
F IGURE 4.24. Proof in the forward direction of Theorem 4.52.
.(/ Assume that F D h g is univalent on D and that D F .D/ is CHD (see Figure
4.25). Then f D F C 2Refgg and
1 1
f .F .w// D w C 2Refg.F .w//g
D w C q.w/:
is locally one-to-one (being the composition of locally one-to-one functions) in . By

Lemma 4.58, f ı F 1 is univalent in and has a range that is CHD.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 227 — #245

i i
w
F=h–g
W
D F –1
w + q(w)
f
F IGURE 4.25. Proof in the reverse direction of Theorem 4.52.
Example 4.59. Let’s shear the Koebe function with a standard dilatation. Let
z
h.z/ g.z/ D (4.5)
.1 z/2
and
!.z/ D g0 .z/= h0 .z/ D z:
Applying the shearing technique, we have
1Cz 1Cz
h0 .z/ g0 .z/ D ) h0 .z/ zh0 .z/ D
.1 z/3 .1 z/3
1Cz
) h0 .z/ D :
.1 z/4

1 2
z 2z C 61 z 3
h.z/ D : (4.6)
.1 z/3
We can use this method to solve for normalized g.z/, where g.0/ D 0. Instead we will find
g.z/ by using (4.5) and (4.6):
1 2
z 2zC 16 z 3
g.z/ D h.z/ D :
.1 z/2 .1 z/3
So
z C 13 z 3 z O
f .z/ D h.z/ C g.z/ D Re C i Im 2 SH :
.1 z/3 .1 z/2
The image of D under f is similar to the image of D under the analytic Koebe function
(see Figure 4.15) with the slit on the negative real axis except that the tip of the slit is at
1
6
instead of 41 . In Example 4.23 we used the transformation 1Cz1 z
D w D u C iv. Let’s
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 228 — #246

i i
use this transformation again (it is often useful in ˇhorizontal slit domains). Since z 2 D,
w D 1Cz
1 z
is the right half-plane fw D u C iv 2 C ˇ Re w D u > 0; 1 < v < 1g. Then
w 1
z D wC1 . Substituting into h.z/ and g.z/ and simplifying, we get

w 1 1 2 3 2 5
h D w Cw
wC1 8 3 3

w 1 1 2 3 2 1
g D w w C :
wC1 8 3 3
Because f D Re.h C g/ C i Im.h g/,
n o 1 n o
w 1 1
f D Re w 3 1 C i Im w 2 1 :
wC1 6 4
Using w D u C iv and taking the real and imaginary parts gives

w 1 1 3 1
f D u 3uv 2 1 C i uv:
wC1 6 2
If we let uv D 0 (so v D 0), then the imaginary part vanishes and, because u > 0, the real
part varies from 61 to C1. Thus, for uv D 0, f .D/ contains the line segment on the real
axis from 16 to C1. If we let uv D c ¤ 0, then the imaginary part is constant and the
3 2
real part is u6 c
2u
, which varies between 1 and C1. Thus, for c ¤ 0, f .D/ contains
the entire line parallel to the real axis that goes through i c. Therefore, f .D/ is the entire
complex plane except the slit on the negative real axis from 61 to 1.
Exercise 4.60. Let f D h C g with h.z/ g.z/ D .1 zz/2 and !.z/ D z 2 . Compute h
O
and g so that f 2 SH and determine f .D/. Try it out!
Caz 2
Exploration 4.61. If we shear h.z/ g.z/ D .1 zz/2 with !.z/ D z1Caz , where 1 <
a 1, then the image of D under f D h C g is a slit domain (like the image of D under
the analytic and harmonic Koebe maps) with the tip of the slit varying as a varies. Use
ShearTool to graph the domains decreasing a from 1 to 1 (You might find it beneficial
to set the Outer radius value to less than 1.0. Try to place the cursor on the tip of the
slit in the f .z/ domain box and look at the coordinates to estimate the distance of the
tip from the origin. How close to the origin can you get the tip of the slit? How far from
the origin can you get the tip of the slit? Try finding other ! expressions that result in slit
domains when shearing h.z/ g.z/ D .1 zz/2 ). Try it out!
Because of the importance and extremal nature of the analytic Koebe function (see
Example 4.23), it is conjectured that the harmonic Koebe function in Example 4.59 is
extremal. The coefficients of the harmonic function
1
X 1
X
f .z/ D an z n C bn z n
nD0 nD1 (4.7)

z C 13 z 3 z
D Re C i Im
.1 z/3 .1 z/2
ˇ ˇ
satisfy jan j D 16 .n C 1/.2n C 1/, jbn j D 16 .n 1/.2n 1/, and ˇjan j jbn jˇ D n.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 229 — #247

i i
Conjecture 1 (Harmonic Bieberbach Conjecture). Let

1
X 1
X
f .z/ D an z n C bn z n 2 S H
O
:
nD0 nD1
Then
1
jan j .n C 1/.2n C 1/;
6
1 (4.8)
jbn j .n 1/.2n 1/;
ˇ 6 ˇ
ˇjan j jbn jˇ n:
In particular,
5
ja2 j (4.9)
2
Exercise 4.62. Verify that the harmonic Koebe function given in (4.7) satisfies equality in
(4.8) and (4.9). Try it out!
Currently, the best bound is that for all functions f 2 SH
O
, ja2 j < 49 (see [12]). There
is room for improvement.
Large Project 4.63. Read and understand the proof that ja2 j 2 for analytic functions
in S (for example, see [1, Section 5.1]). Read and understand two proofs that give bounds
on ja2 j for harmonic functions in SHO
(see [6, Theorem 4.1] and [12, p. 96]). Investigate
ways to modify them, or other proofs, to establish that for f 2 SH
O
, ja2 j K for some K,
5
where 2 K < 49. Optional
Open Problem 4.64. Prove the Conjecture 1.
1
Because the tip of the the harmonic Koebe function is at 6
, we have a conjecture that
is the analogue of the analytic Koebe 41 -Theorem:
Conjecture 2 (Harmonic Koebe 16 -Conjecture). The range of a function in class SH
O
con-
˚ 1

tains the disk G D w W jwj < 6 .
1
-
6
1
F IGURE 4.26. The disk of radius 6 is contained in the image of D under the harmonic Koebe
function.
Open Problem 4.65. Prove Conjecture 2. The best result so far is that the range of f 2 SHO
˚ 1

contains the disk w W jwj < 16 (see [6]), so it would be interesting to increase the radius
1
to some K with 16 < K 16 .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 230 — #248

i i
A domain convex in the general direction ' was defined in Definition 4.50. The shear-
ing theorem by Clunie and Sheil-Small can be generalized to apply to such domains.
Corollary 4.66. A harmonic function f D h C g locally univalent in D is a univalent
mapping of D onto a domain convex in the direction ' ” h e 2i' g is an analytic
univalent mapping of D onto a domain convex in the direction '.
Example 4.67. The analytic right half-plane function 1 z z maps D onto a convex domain,
so it is convex in all directions '. Hence, it is convex in the direction of the imaginary axis
(' D 2 ). Let’s apply Corollary 4.66 with ' D 2 to 1 z z and use a dilatation that simplifies
calculations. With ' D 2 , we set F .z/ D 1 z z equal to h.z/ e 2i' g.z/ D h.z/ C g.z/
instead of h.z/ g.z/. So we start with
z
h.z/ C g.z/ D
1 z
and choose
!.z/ D g0 .z/= h0 .z/ D z:
Computing h and g as in the previous two examples yields
1 2
z 2
z
h.z/ D
.1 z/2
1 2
(4.10)
2z
g.z/ D :
.1 z/2
Hence, the harmonic function is

z z O
f .z/ D h.z/ C g.z/ D Re C i Im 2 SH :
1 z .1 z/2
This function is a harmonic right half-plane map.
z
Exercise 4.68. Verify that shearing h.z/ C g.z/ D 1 z
with !.z/ D z, yields (4.10).
Try it out!
The harmonic right half-plane map in Example 4.67 is the same harmonic map we
discussed in Example 4.40.
Exploration 4.69. Shear h C g D 1 z z using ! D z n , where n D 1; 2; 3; : : : and sketch
f .D/ using ShearTool. Describe what happens to f .D/ as n varies. Pay attention to the
number of points the green lines go to as n increases. Try it out!
One significance of D being mapped to the same domains under these two functions is
that the uniqueness of the Riemann mapping theorem for analytic functions does not hold
for harmonic functions. This leaves the open question:
Open Problem 4.70. What is the analogue of the Riemann mapping theorem for harmonic
functions?
z z 2 C az
Small Project 4.71. Let f D h C g with h.z/ g.z/ D and !.z/ D .
.1 z/2 1 C az
(a) Show that for 1 a 1, j!.z/j < 1; 8z 2 D.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 231 — #249

i i
(b) Compute h and g so that f 2 SH

O
.
(c) Show that for a D 1, f .D/ is a right half-plane.
(d) Show that for 1 < a 1, f .D/ is a slit domain like the Koebe domain. For each a,
determine where the tip of the slit is located.
Optional
Example 4.72. Here is a shearing example that we will use later in our discussion of
minimal surfaces. Let
1 1Cz
h.z/ g.z/ D log ; (4.11)
2 1 z
which is an analytic function that maps D onto a horizontal strip convex in the direction of
the real axis. Let
!.z/ D g0 .z/= h0 .z/ D z 2 :
Using ShearTool we see that the shear of h g with z 2 results in a univalent harmonic
function that maps onto the interior of the region bounded by a square (see Figure 4.27).
1 1Cz

F IGURE 4.27. The image of D when shearing h.z/ g.z/ D 2 log 1 z with !.z/ D z 2 .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 232 — #250

i i
Let’s compute h.z/ and g.z/ and prove that the image is a square region. Applying the
shearing method, we have
1 1
h0 .z/ g0 .z/ D ) h0 .z/ C z 2 h0 .z/ D
1 z2 1 z2
1 1 1 1 i i
) h0 .z/ D D C C C :
1 z4 4 1Cz 1 z i Cz i z

1 1Cz i i Cz
h.z/ D log C log : (4.12)
4 1 z 4 i z
We can use this method to solve for normalized g.z/, where g.0/ D 0, or we can find g.z/
by using (4.11) and (4.12). Either way, we get

1 1Cz i i Cz
g.z/ D log C log :
4 1 z 4 i z
So

i i Cz 1 1Cz O
f .z/ D h.z/ C g.z/ D Re log C i Im log 2 SH :
2 i z 2 1 z
What is f .D/? We have

1 i Cz 1 1Cz
f .z/ D arg Ci arg
2 i z 2 1 z
Du C iv:
Let z D e i 2 @D. Then

i Cz i C e i i e i
1 i.e i C e i
/ 1 cos
D D D i :
i z i e i i e i 1 C i.e i e i / C 1 1 sin
Thus,
ˇ (
1 i C z ˇˇ
4
if cos > 0
uD arg ˇ D
2 i z ˇ
if cos < 0:
zDe i 4
Likewise, we can show that

(

4 if sin > 0
vD
4
if sin < 0:
So, we have that z D e i 2 @D is mapped to

8̂

ˆ z D p ei 4 D C i 4 if 2 .0; 2 /
ˆ 1
ˆ 2 2 4
ˆ
<z3 D p 3
ei 4 D 4
C i 4 if 2 . 2 ; /
2 2
u C iv D 5
ˆ
z5 D 2p ei 4 D
i 4 if 2 .; 3
ˆ
ˆ 4 2 /
ˆ 2
:̂z D p 7
ei 4 D
i 4 if 2 . 3 ; 2 /:
7 4 2
2 2
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 233 — #251

i i
4.6. Properties of the Dilatation 233
Thus, this harmonic function maps D onto the interior of the region bounded by a square
with vertices at z1 ; z3 ; z5; and z7 .
This function can be obtained using the Poisson integral formula. Using the Poisson
formula we can find a harmonic function that maps D onto a regular n-gon for any n 3.
Let Z 2
1 1 C ze i t
f .z/ D Re e i .t /dt;
2 0 1 ze i t
.2kC1/ 2.kC1/
where .t/ D n . 2k
n t < n ;k D 0; :::; n 1/. We can derive that
1
X 1
h.z/ D z nmC1
mD0
nm C 1
1
X 1
g.z/ D z nm 1
mD1
nm 1
(4.13)
1
h0 .z/ D
1 zn
zn 2
g0 .z/ D ; and
1 zn
!.z/ D zn 2
:
!
1 1 C z C z2
Exercise 4.73. Let f D hC g with h.z/ g.z/ D log and !.z/ D z.
3 1 2z C z 2
1 z
(a) Show that h0 .z/ D 1 z3
and g0 .z/ D 1 z3
.
(b) According to the previous paragraph, what should be the image of D under f D hC g?
(c) Use ShearTool to sketch the image of f .
(d) Compute h and g so f 2 SH
O
.
Try it out!

1 1Cz
Exercise 4.74. Let f D h C g with h.z/ C g.z/ D log and !.z/ D z 2 .
2 1 z
Compute h and g explicitly so that f 2 SH O
and use ComplexTool to graph f .D/. Notice
the differences in the dilatations and the images of f .D/ in this exercise and in Example
4.72. Try it out!
4.6 Properties of the Dilatation

Because of the importance of the dilatation, we will examine some of its properties. We
know that an analytic function will map infinitesimal circles to infinitesimal circles at a
point where its derivative is nonzero. For harmonic functions, this does not hold, as can be
seen in the following exploration.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 234 — #252

i i
Exploration 4.75. In ComplexTool, enter values of 0.1, 0.4, 0.7, and 0.999 one at a time
into the Outer radius box in the center panel to plot disks under the analytic right
half-plane map
z
F .z/ D
1 z
and the harmonic right half-plane map
1 2 1 2
z 2
z 2
z
f .z/ D :
.1 z/2 .1 z/2
Look at the images of small circles under F and under f . What appears to be the image of
circles under the harmonic function f ? Try it out!
Now, let’s explore the ideas of the geometric dilatation Df and the analytic dilatation
!.z/ that we discussed earlier.
Exercise 4.76.
(a) Prove that the formulas

@f 1 @f @f @f 1 @f @f
D i and D Ci :
@z 2 @x @y @z 2 @x @y
are equivalent to the formulas (4.2) and (4.3) given in Section 4.4.
(b) Prove that if f D h C g is harmonic, then
@f @f
D h0 .z/ and D g0 .z/:
@z @z
Try it out!
Think of the differential of f as df D fz dz C f z d z. Then we have
jdf j D jfz dz C f z d zj D jh0 .z/dz C g0 .z/d zj:
Bound the differential, and use the fact that if dz is small, then d z will be small and
approximately equal to dz. When we examine upper and lower bounds, we have
.jh0 .z/j jg0 .z/j/jdzj jdf j .jh0 .z/j C jg0 .z/j/jdzj:
For sense-preserving harmonic functions, the left-hand side is always positive. The ratio of
the upper and lower bounds gives the geometric dilatation Df defined by
jh0 j C jg0 j
Df D :
jh0 j jg0 j
Since this is a ratio of the maximum to minimum jdf j, if we evaluate Df at z0 , we will

find a number that represents a ratio between the most and the least that an infinitesimal
circle will be deformed by the function. Thus, if the function maps an infinitesimal circle
to an infinitesimal ellipse, then the geometric dilatation gives the ratio of the major axis to
the minor axis of the ellipse.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 235 — #253

i i
Exercise 4.77.
(a) Prove that for analytic functions, Df D 1.
(b) Prove that for sense-preserving harmonic functions, Df 1.
z3
(c) Find a formula for Df for z C 3
. What are the maximum and minimum of this
function over the unit disk D?
Try it out!
3
Exercise 4.78. Examine the geometric dilatation for zC z3 in greater detail. For the points
z D 0; 0:5; 0:9; 0:9ei =4; and 0:9i , find Df .z/. Using ComplexTool, examine the images
of circles of radius 0.05 centered at the points. Estimate the ratio of the major axis to the
minor axis of the image ellipse. Does it match with your computation for Df ? Try it out!
While the geometric dilatation provides useful information about the function, some
information is lost when we take the moduli jh0 .z/j and jg0 .z/j. It is often useful to examine
the analytic part versus the anti-analytic part of f . We define what is sometimes called the
second complex dilatation of f ,
f z .z/ g0 .z/
!f .z/ D D 0 ;
fz .z/ h .z/
where the representation in the last equality makes sense only for harmonic functions.
When the function f is clear, we omit the subscript and refer to the dilatation as !. Because
!.z/ is analytic if and only if f .z/ is harmonic, the second complex dilatation is also called
the analytic dilatation of f .
Exercise 4.79.
f z .z/
(a) Prove that !.z/ D fz .z/
is analytic if and only if f .z/ is harmonic.
(b) Prove that !.z/ is identically 0 if and only if f is analytic.
(c) Prove that for sense-preserving non-analytic harmonic functions f , 0 < j!.z/j < 1.
Try it out!
We can ask what the relationship is between the geometric dilatation and the analytic
dilatation.
K 1
Exercise 4.80. Prove that Df .z/ K if and only if j!f .z/j KC1 . Try it out!
Exploration 4.81. Reexamine the function z C 13 z 3 , evaluating !.z/ at z D 0, 0:5, 0:9,
0:9e i =4, and 0:9i . What do you observe about the relationship between ! and images of
a small circle centered at the points? Try it out!
In Section 4.4, we remarked that if f D h C g is a sense-preserving harmonic map
that has j!.z/j D 1 for all z 2 arc of @D, then the image of the arc is either a concave arc
or stationary. To further explore this, we will use ShearTool to graph the image of D under
f D h C g, when h g D z and varying !.
Exploration 4.82.
(a) Shear h.z/ g.z/ D z using !.z/ D e i n=6z, for n D 0; : : : ; 6 and sketch f .D/ using
ShearTool. Describe what happens to f .D/ as n varies.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 236 — #254

i i
(b) Shear h.z/ g.z/ D z using !.z/ D z n , for n D 1; 2; 3; 4 and sketch f .D/ using
ShearTool.
(i) What patterns do you notice relating f .D/ and n?

(ii) Make a sketch of f .D/ for n D 5. Then graph the shear using ShearTool.
(iii) Make a sketch of f .D/ for n D 6. Then graph the shear using ShearTool.
zCa
(c) Shear h.z/ g.z/ D z using !.z/ D 1Caz for various values of a 2 D and sketch
f .D/ using ShearTool. Describe what happens to f .D/ as a varies.
Try it out!
Small Project 4.83. Investigate the shearing of h.z/ g.z/ D z n12 z n .n D 2; 3; 4; : : : /
with ! for various values of ! (the image of D under the analytic function z n1 z n is
not CHD for n D 4; 5; 6; : : : ; however, it is if we use z n12 z n ). Use the approach of
Exploration 4.82 as a starting point and then explore new approaches. Optional
So far, we have used only dilatations that are finite Blaschke products. A finite Blaschke
product B.z/ can be expressed in the form
n
Y mj
i z aj
B.z/ D e ;
1 aj z
j D1
where 2 R, jaj j < 1, and mj is the multiplicity of the zero aj . The dilatations given
in Exploration 4.82 and finite products of them are examples of finite Blaschke products.
Harmonic univalent mappings whose dilatations are finite Blaschke products have been
studied (see [17]). However, little is known about mappings whose dilatations are not finite
Blaschke products. One important type of mappings that are not finite Blaschke products is
a singular inner function, and we will now investigate harmonic univalent mappings with
dilatations that are singular inner functions.
First, we need to know what singular inner functions are. Let f W D ! C be an analytic
function and denote its radial limit by
f .e i / D lim f .r e i /:
r !1; r <1
Definition 4.84. A bounded analytic function f is called an inner function if jf .e i /j D

1 almost everywhere with respect to Lebesgue measure on @D. If f has no zeros on D,
then f is called a singular inner function.
Every inner function can be written as
Z e i C z
i
d(ei )
f .z/ D e i ˛ B.z/e e z ;
where ˛; 2 R, is a positive measure on @D, and B.z/ is a Blaschke product. The

zC1
function f .z/ D e z 1 is an example of a singular inner function.
zC1
Exercise 4.85. Show that if !.z/ D e z 1 , then j!.z/j < 1, for z 2 D. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 237 — #255

i i
It has been difficult to construct examples of harmonic mappings whose dilatations are
singular inner functions. For a while there were no known examples [21] but Weitsman [30]
provided two examples. We present them (see Example 4.88 and Example 4.92) giving a
shorter proof of his second example. The proof provides a method to find more examples.
One way to find an example of a harmonic map with a singular inner function as its
dilatation is to use the shearing technique. However, it is not often possible to find a closed
zC1
form for f D h C g. For example, let h.z/ g.z/ D z and the dilatation be !.z/ D e z 1 .
Then by the shearing technique
Z
1
h.z/ D zC1
dz:
1 ez 1
The integral does not have a closed form and so we cannot find an explicit representation
for f D h C g.
1 n
Exercise 4.86. Using the shearing technique with h.z/ g.z/ D z n
z and !.z/ D
zC1
ez 1 , express h as an integral. It is not possible to integrate h to get a closed-form solution.
Try it out!
Next we give an example by Weitsman in which the shearing technique allows us to
solve the integrals for h and g. But first, we will need a result by Pommerenke [22].
Theorem 4.87. Let f be an analytic function in D with f .0/ D 0 and f 0 .0/ ¤ 0, and let
z
'.z/ D ;
.1 C ze i /.1 C ze i /
where 2 R. If
zf 0 .z/
Re > 0; for all z 2 D;
'.z/
then f is convex in the direction of the real axis.
Example 4.88. Consider shearing the analytic function
z 1 zC1
h.z/ g.z/ D C ez 1
1 z 2
with
zC1
!.z/ D e z 1 :
So h i
1 zC1
h0 .z/ g0 .z/ D 1 ez 1 :
.1 z/2
Using D in Theorem 4.87, we have
h zC1
i
Re 1 ez 1 >0
ˇ zC1 ˇ
because ˇe z 1 ˇ < 1. Hence h g is convex in the direction of the real axis.
zC1
Shearing h g with !.z/ D e z 1 and normalizing yields
Z
1 z
h.z/ D dz D ;
.1 z/2 1 z
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 238 — #256

i i
and solving for g we get

1 zC1
g.z/ D ez 1:
2
The image given by the map is similar to the image given by the right half-plane map
z
1 z
except that there are an infinite number of cusps (see Figure 4.28).
z 1 zC1
F IGURE 4.28. Image of D under f .z/ D ez 1.
1 z 2
zC1
Exercise 4.89. Let f D h C g with h.z/ g.z/ D 1 z z and !.z/ D e z 1 . Use the
shearing method to compute h and g explicitly so f 2 SH O
and use ComplexTool to sketch
f .D/. Hint: In finding the h, use a u-substitution to evaluate the integral. Try it out!
Another technique to find harmonic mappings whose dilatations are singular inner
functions involves using the following theorem by Clunie and Sheil-Small [6].
Theorem 4.90. Let f D h C g be locally univalent in D and suppose that h C g is
convex for some jj 1. Then f is univalent.
Theorem 4.90 gives us a way to show that a harmonic function is univalent. To develop
the technique, we let D 0 in the theorem. This means that if h is analytic convex and if !
is analytic with j!.z/j < 1, then f D h C g is a harmonic univalent mapping. To establish
that a function f is convex, we will use the following theorem (see [13]).
Theorem 4.91. Let f be analytic in D with f .0/ D 0 and f 0 .0/ D 1. Then f is univalent
and maps onto a convex domain if and only if

zf 00 .z/
Re 1 C 0 0; for all z 2 D:
f .z/
In the next example we will show how we can use these ideas to construct a harmonic
univalent function whose dilatation is a singular inner function.
Example 4.92. Let
1 2 zC1
h.z/ D z z with !.z/ D g0 .z/= h0 .z/ D e z 1 :
4
We will use Theorem 4.91 to show that h is convex. Let
1
zh00 .z/ 2z 1 z
T .z/ D 1 C D1C 1
D 1
h0 .z/ 1 2
z 1 2
z
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 239 — #257

i i
zC1
1 2 1
F IGURE 4.29. Image of D under f1 .z/ D z 4z 4 .z 1/2e z1 .
which is a Möbius transformation. By the mapping properties of Möbius transformations

we can show that T maps D onto jz 23 j D 23 , the circle centered at 23 with radius 23 .
Hence, RefT .z/g > 0 and h is convex.
We can solve for g:
Z Z
0 1 zC1 1 zC1
g.z/ D h .z/!.z/ dz D .1 z/e z 1 dz D .z 1/2 e z 1 :
2 4
Hence,
1 2 1 zC1
f .z/ D h.z/ C g.z/ D z
z .z 1/2 e z 1 :
4 4
1 2
By Theorem 4.90, f D h C g is univalent. The image of D under f1 .z/ D z 4
z
zC1
1
4
.z 1/2 e is similar to the map of a harmonic polynomial but with an infinite number
z 1
of cusps in the middle section on the right side (see Figure 4.29).
This approach has been extended to find more examples of harmonic univalent func-
tions with singular inner function dilatations [2].
z 1
1 3 1
Exercise 4.93. Let h.z/ D z C 11 z and g.z/ D 11 .z 3/.z C 1/2 e zC1 . Show that
f D h C g is univalent and use ComplexTool to graph f .D/. Try it out!
Exercise 4.94. Use this approach to show that f D h C g is harmonic univalent, where
z 1
h.z/ D z C 2 log.z C 1/ and !.z/ D e zC1 . The graph of D under f is in Figure 4.30. Try
it out!
F IGURE 4.30. Image of D under f .z/ D h C g in Exercise 4.94.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 240 — #258

i i
We mentioned that it is not often possible to find a closed form for f D h C g when
the dilatation is a singular inner function. However, ShearTool can explore images of f .D/
when the dilatation is a singular inner product.
Exploration 4.95.
zC1
(a) If we shear h.z/ g.z/ D .1 zz/2 with !.z/ D e z 1 , then f D h C g will be
univalent and convex in the direction of the real axis. Use ShearTool to sketch f .D/.
z 2 C1
1 2
(b) Use ShearTool to sketch f .D/, where h.z/ g.z/ D z with !.z/ D e z2 1 .
2z
zC1
(c) Use ShearTool to sketch f .D/, where h.z/ g.z/ D 12 log 1Cz
1 z
with !.z/ D e z 1 .
Try it out!
Open Problem 4.96. What are the properties of harmonic univalent mappings whose di-
latation is a singular inner product?
4.7 Harmonic Linear Combinations

A common way to construct new functions with a given property is to take a linear com-
bination of two functions with that property. This is done with derivatives and integrals in
beginning calculus. In Exploration 4.25, this is done with the analytic Koebe mapping, fk ,
and the right half-plane mapping, fr , where
z z
fk .z/ D and fr .z/ D ;
.1 z/2 1 z
to derive the univalent analytic map
z tz 2
f3 .z/ D tfk .z/ C .1 t/fr .z/ D ;
.1 z/2
where 0 t 1.
Is it true that a linear combination of two one-to-one functions is a one-to-one function?
Let’s look at real-valued functions. Suppose f1 W R ! R and f2 W R ! R are one-to-one
functions. Will f3 .x/ D tf1 .x/ C .1 t/f2 .x/ also be one-to-one when 0 t 1? Not
necessarily. Let f1 .x/ D x 3 , f2 .x/ D x 3 , and t D 12 . Because they satisfy the horizontal
line test, f1 and f2 are one-to-one. But f3 .x/ D tf1 .x/ C .1 t/f2 .x/ D 0, which is not
one-to-one.
y y y
f 1(x) = x 3
1–1 f 2(x) = – x 3 f 3(x) = 0
1–1 not 1–1
x x x
F IGURE 4.31. f1 .x/ D x 3 and f2 .x/ D x 3 are one-to-one, but f3 .x/ D tf1 .x/C.1 t /f2 .x/ D
0 is not one-to-one when t D 12 .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 241 — #259

i i
4.7. Harmonic Linear Combinations 241
The difficulty is that f1 is an increasing one-to-one function, f2 is a decreasing one-to-

one function, and when t D 21 the increase of f1 cancels out the decrease of f2 . We can get
around this difficulty by requiring that f1 , f2 are either both increasing or both decreasing.
This idea can be applied to complex-valued functions.
Condition A. Suppose f is complex-valued harmonic and non-constant in D. There exists
sequences fzn0 g, fzn00 g converging to z D 1, z D 1, respectively, such that
lim Reff .zn0 /g D sup Reff .z/g

n!1 jzj<1
(4.14)
lim Reff .zn00 /g D inf Reff .z/g:
n!1 jzj<1
The normalization in (4.14) can be thought of as if f .1/ and f . 1/ are the right and
left extremes in the image domain in the extended complex plane.
Example 4.97. We will show that condition A is satisfied by f .z/ D z C 31 z 3 (see Figure
4.32).
F IGURE 4.32. Image of D under f .z/ D z C 31 z 3 .
One can see that f satisfies condition A by using the Sketch option in ComplexTool
to draw several paths of sequences fzn0 g 2 D that approach 1 and see that the images of
the paths approach the right-side cusp of the image of D, a hypocycloid of four cusps
(see Figure 4.32). Likewise, paths fzn00 g 2 D that approach 1 result in image paths that
approach the left-side cusp of the hypocycloid.
To prove that f satisfies condition A, we have
1 1 1
f .e i / D e i C e 3i
D .cos C cos 3/ C i.sin sin 3/:
3 3 3
So, Reff .e i /g D cos C 13 cos 3. Hence 43 Reff .e i /g 4
3
which implies
supjzj<1 Reff .z/g D 34 : Letting fzn0 g D f1 n1 g ! 1, we have that
4
lim Reff .zn0 /g D D sup Reff .z/g:
n!1 3 jzj<1
4
Similarly, lim Reff .zn00 /g D D inf Reff .z/g.
n!1 3 jzj<1
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 242 — #260

i i
Exercise 4.98. Use the Sketch option in ComplexTool to determine which of the follow-
ing harmonic functions satisfy condition A.

1 1Cz i i Cz
(a) f .z/ D z C 12 z 2 (b) Re log C i Im log
2 1 z 2 i z

z z i i Cz 1 1Cz
(c) Re C i Im (d) Re log C i Im log
1 z .1 z/2 2 i z 2 1 z

z C 31 z 3 z
(e) Re C i Im
.1 z/3 .1 z/2
Try it out!
To prove a result about linear combinations of harmonic functions, we will need the
following theorem by Hengartner and Schober [19] that employs condition A. However,
we won’t use Theorem 4.99 afterwards.
Theorem 4.99 (Hengartner and Schober). Suppose f is analytic and non-constant in D.
Then
Ref.1 z 2 /f 0 .z/g 0; z 2 D
if and only if
1. f is univalent in D,
2. f is convex in the imaginary direction, and
3. condition A holds.
We now study conditions under which f3 is globally univalent.
Theorem 4.100. Let f1 D h1 C g1 , f2 D h2 C g2 be univalent harmonic mappings
convex in the imaginary direction and !1 D !2 . If f1 , f2 satisfy condition A, then f3 D
tf1 C .1 t/f2 is convex in the imaginary direction (and univalent) .0 t 1/.
Proof. To see that f3 is locally univalent, use g10 D !1 h01 and g20 D !2 h02 D !1 h02 . Then
tg10 C .1 t/g20 t!1 h01 C .1 t/!1 h02

!3 D 0 D D !1 :
th01 C .1 t/h2 th01 C .1 t/h02
By Clunie and Sheil-Small’s shearing theorem (see Theorem 4.52), we know that hj C gj
(j D 1; 2) is univalent and convex in the imaginary direction. Also, hj C gj satisfies
condition A since Reffj g D Refhj C gj g. Applying Theorem 4.99 we have
Ref.1 z 2 /.hj0 .z/ C gj0 .z//g 0; .j D 1; 2/:
So
Ref.1 z 2 /.h03 .z/ C g30 .z//g

D Ref.1 z 2 /Œt.h01 .z/ C g10 .z// C .1 t/.h02 .z/ C g20 .z//g
D t Ref.1 z 2 /.h01 .z/ C g10 .z//g C .1 t/ Ref.1 z 2 /.h02 .z/ C g20 .z//g 0:
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 243 — #261

i i
By applying Theorem 4.99 in the other direction, we have that h3 C g3 is convex in the
imaginary direction, and so by the shearing theorem, f3 is convex in the imaginary direc-
tion.
Example 4.101. Let

i 1Cz 1 i Cz
f1 .z/ D Re log C i Im log
2 1 z 2 i z

1 1Cz i i Cz
f2 .z/ D Re log C i Im log :
2 1 z 2 i z
Notice f1 maps D onto a square region (see Figure 4.33). The image is the same as
for the harmonic square map in Example 4.72, but the function is different because f1
has different arcs of the unit circle mapped to the vertices and the dilatation for f1 is

i 1Cz 1 i Cz
F IGURE 4.33. Image of D under f1 .z/ D Re log C i Im log :
2 1 z 2 i z
!.z/ D z 2 which is different than the dilatation for the harmonic square map in Example
4.72. Condition A is satisfied for f1 (see Example 4.104 for more details). f2 maps D onto
a region similar to a hypocycloid with four cusps except instead of cusps the domain has
ends that extend to infinity (see Figure 4.34). The dilatation of f2 is also !.z/ D z 2 and

1 1Cz i i Cz
F IGURE 4.34. Image of D under f2 .z/ D Re log C i Im log :
2 1 z 2 i z
condition A is satisfied. By Theorem 4.100, f3 D tf1 C .1 t/f2 is univalent. The image

of D when t D 12 is shown in Figure 4.35.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 244 — #262

i i
F IGURE 4.35. Image of D under f3.z/ D 21 f1 .z/ C 12 f2 .z/.
Exploration 4.102. Let

i 1Ce i3z 1 1 C z C z2
f1 .z/ D Re p log C i Im log ;
3 1 C ei 3 z 3 1 2z C z 2

z z
f2 .z/ D Re C i Im :
1 z .1 z/2
Show that f1 and f2 satisfy the conditions of Theorem 4.100 and then use ComplexTool
to plot images of f3 D tf1 C .1 t/f2 for various values of t. Try it out!
Large Project 4.103. Theorem 4.100 gives sufficient but not necessary conditions on f1
and f2 for the linear combination f3 D tf1 C .1 t/f2 to be univalent. That f3 can be
univalent when f1 does not satisfy condition A is demonstrated by

z C 1=3z 3 z
f1 .z/ D Re C i Im ;
.1 z/3 .1 z/2

z z
1 z .1 z/2
Construct f3 for various values of t and use ComplexTool to see the images of D under f3 .
The following functions suggest that several of the hypotheses of Theorem 4.100 can
fail and f3 can still be univalent:
1 m 1 n
f1 .z/ D z z and f2 .z/ D z z ;
m n
where m; n 2. For various values of m, n, and t construct f3 and use ComplexTool to
see the images of D under f3 .
Investigate the examples and construct other examples in which f3 is univalent but
the hypotheses of Theorem 4.100 do not hold. Use them to make a conjecture about the
hypotheses of a new theorem that guarantees f3 will be univalent. Prove the new theorem.
Optional
Let us look at an example that is surprising and is related to the nonconvex polygons
described by Duren, McDougall, and Schaubroeck [15].
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 245 — #263

i i
Example 4.104. Let f1 D h1 C g1 be the harmonic square map in Example 4.101, where
! !
i 1Cz 1 i Cz
h1 .z/ D log log
4 1 z 4 i z
! !
i 1Cz 1 i Cz
g1 .z/ D log C log :
4 1 z 4 i z
We can write this as
" # " #
i 1Cz 1 i Cz
2 1 z 2 i z
Using the same approach as in Example 4.72, we see that z D e i 2 @D is mapped to

8̂

ˆ
ˆ z1 D 2p 2
ei 4 if 2 . 2 ; 0/
ˆ
ˆ
<z3 D p 3
e i 4 if 2 .0; 2 /
u1 C iv1 D 2 2
5
ˆ
ˆ z5 D p
e i 4 if 2 . 2 ; /
ˆ
ˆ 2 2
:̂z D p 7
e i 4 if 2 .; 3 /:
7 2
2 2
So, f1 maps D onto a square region with vertices at z1 ; z3 ; z5; and z7 (see Figure 4.36).
The dilatation for f1 is ! D z 2 and condition A is satisfied. For example, for a sequence
of points, fzn0 g, in the fourth quadrant approaching 1,

lim Reff1 .zn0 /g D p D sup Reff1 .z/g
n!1 2 2 jzj<1
and for a sequence of points, fzn00 g, in the second quadrant approaching 1,

lim Reff1 .zn00 /g D p D inf Reff1 .z/g:
n!1 2 2 jzj<1
F IGURE 4.36. Image of D under f1 .
Let f2 D h2 C g2 , where

! 3
!
1 i 3 ei 4 C z 1 i ei 4 Cz
h2 .z/ D e 4 log e 4 log 3
4 ei 4 z 4 ei 4 z

! 3
!
1 3 ei C z 4 1 ei C z 4
g2 .z/ D e i 4 log i C e i 4 log 3
:
4 e 4 z 4 ei 4 z
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 246 — #264

i i
Then
( i i 3 )
i e 4 Cz e 4 Cz
f2 .z/ D Re p log i C log 3
2 2 e 4 z ei 4 z
( i i 3 )
1 e 4 Cz e 4 Cz
C i Im p log i log 3
;
2 2 e 4 z ei 4 z
and z D e i 2 @D is mapped to
8̂

ˆ z D p if 2 . ; 4/
ˆ 0
ˆ 2 2 4
ˆ
<z2 ip

D 2 2
if 2 . 4 ; 3
4 /
u2 C iv2 D
ˆ
ˆ z4 D
p if 2 3 5
.4 ; 4 /
ˆ
ˆ 2 2
:̂z D ip

if 2 . 5 ; 7 /:
6 4 4
2 2
That is, f2 maps D onto a rotated square region with vertices at z0 ; z2 ; z4 ; and z6 (see
Figure 4.37) with ! D z 2 , and it also satisfies condition A.
By Theorem 4.100, f3 D tf1 C .1 t/f2 is univalent. What is the image of D under

1
f3 ? Let’s look at t D . You might think that f3 .D/ would be an overlay of f1 .D/ and
2
f2 .D/ (see Figure 4.38(a)), but it is not. It is the nonconvex star in Figure 4.38(b).
Why is the correct image the nonconvex star in Figure 4.38(b)? Let’s look at where
arcs of the unit circle are mapped under f3 . Notice f1 .e i / and f2 .e i / depend on which
(a) (b)
F IGURE 4.38. Which is the image of f3 .D/?
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 247 — #265

i i
of eight arcs is in. If 2 . 4 ; 0/, then f1 .e i / D z1 and f2 .e i / D z0 , and so in this

interval f3 .e i / D z1 Cz2
0
(that is, it is the midpoint between z1 and z0 ). Iif 2 .0; 4 /,
then f1 .e i / D z3 and f2 .e i / D z0 , and f3 .e i / D z3 Cz 2
0
. So
8̂
ˆ w1 D z1 Cz 0
D p
cos 8 e i 8 if 2 . 4 ; 0/
ˆ
ˆ
2 2 2
ˆ
ˆ z3 Cz0 3 i 3
ˆw2 D 2 D 2p2 cos 8 e 8
ˆ if 2 .0; 4 /
ˆ
ˆ
ˆ
ˆw3 D z3 Cz 2
D 2p 5
cos 8 e i 8 if 2 . 4 ; 2 /
ˆ
ˆ 2 2
ˆ
ˆ 7
<w4 D z5 Cz2 D p
cos 3 ei 8 if 2 . 2 ; 3 /
2 8 4
f3 .e i / D z Cz
2 2
9
ˆ
ˆ

w5 D 5 2 4 D p cos 8 e 8 i 3
if 2 . 4 ; /
ˆ
ˆ 2 2
ˆ
ˆ z Cz 11
ˆ
ˆ w6 D 7 2 4 D 2p cos 3
8 e
i 8
if 2 .; 5 4 /
ˆ
ˆ 2
ˆ
ˆ z7 Cz6 i 8 13 5 3
ˆ
ˆ w D 2 D p cos 8 e if 2 . 4 ; 2 /
ˆ 7 2 2
:̂w D z1 Cz6 D p 3 i 15
cos 8 e 8 if 2 . 3 ; 7 /:
8 2 2 4
2 2

The vertices w1; w3 ; w5; and w7 lie equally spaced on a circle of radius rout er D p cos 8
2 2
1:026, and the vertices w2; w4; w6 ; and w8 lie equally spaced on a circle of radius ri nner D

p cos 3 8
0:425. We can visualize the boundary of f3 .D/ by plotting the vertices
2 2
z0 ; z1 ; : : : z7 and drawing the midpoints w1; : : : ; w8 (see Figure 4.39).
w2 w3 w3
z2 z2 z2
z1 z3 z1 z1
z3 z3 w2 w1
w1
z4 z0 w4
z4 z0 z4 z0 w8
z5 z7 z5 z7 z5 z7 w5 w6
z6 z6 z6
w7
F IGURE 4.39. Visualizing the image of the boundary of f3 .D/.
We can also explore the linear combination of these two functions by using the applet,
LinComboTool (see Figure 4.40). Open LinComboTool. Make sure that at the top of the
page, the Number of Polygonal Panels is 2. In Panel #1 enter the left end
points of the intervals for the arcs of the unit circle used in f1 (they need to be positive
numbers). Then enter the real and imaginary values of the image of this arc under f1 . For
example, if we take the interval .0; 2 / for Arc 1 (we start with this interval because we
need to use nonnegative values), then for Arc 1 enter 0, for the real value of its image enter

p cos 34
, and for the imaginary value of its image enter p
sin 3
4
. Continue this for
2 2 2 2
the other arcs noting that for Arc 4 we will enter 3
2 in the first box. If there are not enough
boxes for the arcs, click on the Add button to add an arc. Similarly, click on Remove
if there are too many boxes for the arcs. When you are done entering the points, click
on Graph to produce the image f1 .D/. Then go to Panel #2 and enter the points for
f2 and graph f2 .D/. After these are graphed, click on Create LinCombogon and the
image will appear in the lower left-hand box (see Figure 4.41).
Exercise 4.105. Using LinComboTool, start with the same arc values and point values as
in Example 4.104. You can change the value of t by sliding the red dot near the top of the
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 248 — #266

i i
F IGURE 4.40. The applet LinComboTool.
page. Describe what happens as t varies from 0 to 1. In the example we showed that when
t D 12 , rout er D 2p

2
cos 8 1:026 and ri nner D 2p

2
cos 3
8 0:425. Compute rout er
and ri nner for any t .0 t 1/. Try it out!
Remark 4.106. In Theorem 4.100, we do not need that !1 D !2 . Looking over its proof
we see that all we need is that f3 is locally univalent. This occurs if
ˇ 0 ˇ
ˇ tg C .1 t/g20 ˇˇ
j!3 j D ˇˇ 10 < 1: (4.15)
th1 C .1 t/h02 ˇ
Exercise 4.107. We can have one pair of functions f1 , f2 mapping onto image domains
G1 , G2 , and another pair of functions fQ1 , fQ2 that map onto the same image domains G1 ,
G2 , but the linear combinations f3 and fQ3 map onto different image domains.
Repeat the steps in Example 4.104 using the same function for f2 but replacing f1 with
the harmonic square map in Example 4.72, where

1 1Cz i i Cz
h1 .z/ D log C log
4 1 z 4 i z

1 1Cz i i Cz
4 1 z 4 i z
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 249 — #267

i i
We have
1 z2
h01 .z/ D 4
; g10 .z/ D
1 z 1 z4
1 z2
h02 .z/ D ; g20 .z/ D :
1 C z4 1 C z4
(a) In this case, !1 .z/ D z 2 while !2 .z/ D z 2 . Using (4.15) in Remark 4.106, show
that f3 is locally univalent.
(b) Use LinComboTool find the image of f3 .D/ using this f1 and f2 .
(c) Explain why this happens by using the approach in Example 4.104 to compute the new
values of w1 ; : : : ; w8 and then use the visualization technique in the example to plot
the vertices z0 ; : : : ; z7 and draw the midpoints w1 ; : : : ; w8.
Try it out!
Exercise 4.108. Repeat the steps in Exercise 4.107 using the same function for f1 but
replacing f2 with the harmonic hexagon map that can be derived from (4.13) for h0 and g0
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 250 — #268

i i
F IGURE 4.42. Image of D under f3 in Exercise 4.107.
at the end of Example 4.72 , where
1
h02 .z/ D )
1 z6
i i i2 i2
1 1Cz e 3 1Ce 3 z e 3 1Ce 3 z
h2 .z/ D log C log i
C log i2
6 1 z 6 1 e3z 6 1 e 3 z
z4
g20 .z/ D )
1 z6
i i i2 i2
1 1Cz e3 1Ce 3 z e 3 1Ce 3 z
g2 .z/ D log log i
log i2
:
6 1 z 6 1 e3z 6 1 e 3 z
(a) In this case, !1 .z/ D z 2 while !2 .z/ D z 4 . Using (4.15) in the remark, show that
f3 is locally univalent.
values of the vertices of f3 .D/.
Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 251 — #269

i i
We can generalize Theorem 4.100 to include the linear combination of n functions

f1 ; : : : ; fn .
Theorem 4.109. Let f1 D h1 C g1 , : : :, fn D hn C gn be n univalent harmonic mappings
convex in the imaginary direction and !1 D D !n . If f1 , : : :, fn satisfy condition A,
then F D t1 f1 C C tn fn is convex in the imaginary direction, where 0 tn 1 and
t1 C C tn D 1.
Exploration 4.111. Using Theorem 4.109, create maps in three different panels of Lin-
ComboTool, where each map takes four arcs on the unit circle to four vertices of a square.
Make sure that they satisfy the conditions of the theorem. Then click on the Create
LinCombogon button to see the image domain. Explore this by using different maps in
the panels. For an example, see Figure 4.43. Try it out!
Large Project 4.112. In Example 4.104 and Exercise 4.107 we took linear combinations
of two harmonic square mappings and got fundamentally different images. Explore this
with other n-gons. Use LinComboTool to determine what and how many fundamentally
different (i.e., not rotations or scalings) images can be constructed when taking the linear
F IGURE 4.43. Example of image of D under the linear combination of three squares.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 252 — #270

i i
combination with t D 21 of two harmonic 5-gon maps, 6-gon maps, and n-gon maps. Make
sure that condition A holds and that j!3 j < 1. Optional
Large Project 4.113. In Exercise 4.108 we took linear combinations of a harmonic 4-
gon mapping and a harmonic 6-gon mapping with dilatations z 2 and z 4 , respectively.
Use LinComboTool to determine what combinations are possible and what images can be
constructed when taking linear combination with t D 21 of a harmonic m-gon and n-gon,
where m < n. Make sure that condition A holds and that j!3 .z/j < 1. Optional
For more on this topic see [10].
4.8 Convolutions
Another way of combining two univalent functions is with the Hadamard product or con-
volution. For analytic functions
1
X 1
X
f .z/ D an z n and F .z/ D An z n ;
nD0 nD0
their convolution is defined as

1
X
f .z/ F .z/ D an An z n :
nD0
Example 4.114. The convolution of the right half-plane function (see Example 4.22)
1
X
z
f .z/ D D zn
1 z
nD1
and the Koebe function (see Example 4.23)
1
X
z
F .z/ D D nz n
.1 z/2 nD1
convoluted results 0
0
with in
F IGURE 4.44. Right half-plane function convoluted with the Koebe function yields the Koebe
function.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 253 — #271

i i
4.8. Convolutions 253
is
z z
f .z/ F .z/ D
1 z .1 z/2
X1 1
X
D zn nz n
nD1 nD1
D.z C z 2 C z 3 C z 4 C / .z C 2z 2 C 3z 3 C 4z 4 C /
D.z C 2z 2 C 3z 3 C 4z 4 C /
z
D :
.1 z/2
z
Example 4.115. Take the Koebe function, f .z/ D , and the horizontal strip
.1 z/2
1 1Cz
map, F .z/ D log . What is their Hadamard product, f .z/ F .z/? We need to
2 1 z
compute the Taylor series for F . To do so, we use
Z Z X
1 1
X
1 1 nC1
log.1 z/ D dz D z n dz D z
1 z nD0 nD0
n C 1
and
1
X 1
log.1 C z/ D . 1/nC1 z nC1 :
nC1
nD0
Hence,
X 1 1
X
1 1Cz 1 1 nC1
log D . 1/nC1 z nC1 z
2 1 z nD0
n C 1 nD0
nC1
1
X 1
D z 2nC1
2n C 1
nD0
1 1
Dz C z 3 C z 5 C :
3 5
0
convoluted results
0
with in
0
F IGURE 4.45. The Koebe function convoluted with a horizontal strip map yields a double-slit map.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 254 — #272

i i
Thus,

z 1 1Cz
f .z/ F .z/ D log
.1 z/2 2 1 z
X1 1
X 1
D nz n z 2nC1
2n C 1
nD1 nD0
1 1
D.z C 2z 2 C 3z 3 C 4z 4 C 5z 5 C / .z C z 3 C z 5 C /
3 5
Dz C z 3 C z 5 C :
1 1
Since D 1 C z C z 2 C z 3 C , we have that D 1 C z2 C z4 C z6 C
1 z 1 z2
z
and D z C z 3 C z 5 C . That is,
1 z2

z 1 1Cz z
f .z/ F .z/ D log D :
.1 z/2 2 1 z 1 z2
z
Exercise 4.116. Let f .z/ D log.1 z/ and F .z/ D . Determine f .z/ F .z/.
.1 z/2
Try it out!
Proposition 4.117.
z
(a) The right half-plane mapping, f .z/ D , acts as the convolution identity: if F is
1 z
z
an analytic function, then F .z/ D F .z/.
1 z
z
(b) The Koebe function, f .z/ D , acts as a differential operator: if F .z/ is an
.1 z/2
z
analytic function, then F .z/ D zF 0 .z/.
.1 z/2
(c) Convolution is commutative: if f1 and f2 are analytic functions, then f1 f2 D f2 f1 .
(d) If f1 and f2 are analytic functions, then .f1 .z/ f2 .z//0 D zf10 .z/ f2 .z/.
Exercise 4.118. Prove Proposition 4.117 (a)-(d). Try it out!
If f1 ; f2 2 S , then f1 f2 may not be in S . For example,
1
X 1
X
z z
D nz n nz n
.1 z/2 .1 z/2 nD1 nD1
X1
D n2 z n … S:
nD1
1
X
Why do we know that n2 z n … S ?
nD1
However, we have a result by Ruscheweyh and Sheil-Small [25]. If the analytic func-
tion, f 2 S , maps onto a domain that is convex, then we will denote that by writing
f 2 K.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 255 — #273

i i
Theorem 4.119 (Ruscheweyh and Sheil-Small). If f1 ; f2 2 K, then f1 f2 2 K.

Now let’s consider harmonic convolutions.
Definition 4.120. For harmonic univalent functions
1
X 1
X
f .z/ D h.z/ C g.z/ D z C an z n C bn zn
nD2 nD1
1
X X1
F .z/ D H.z/ C G.z/ D z C An z n C B n zn;
nD2 nD1
define the harmonic convolution by

1
X 1
X
f .z/ F .z/ D h.z/ H.z/ C g.z/ G.z/ D z C an An z n C bn B n z n :
nD2 nD1
If the harmonic function, f 2 SH , maps onto a domain that is convex, then we will
write f 2 KH . As we mentioned it is known that if f1 ; f2 2 K, then f1 f2 2 K. Is there
a similar result for harmonic univalent convex mappings? There are a few known results
about harmonic convolutions of functions on D.
Theorem 4.121 (Clunie and Sheil-Small). If f 2 KH and ' 2 S , then the functions
f .˛' C '/ 2 SH
map D onto a close-to-convex domain, where j˛j 1.

Clunie and Sheil-Small posed the open problem (see [6]).
Open Problem 4.122. For f 2 KH , what are the harmonic functions F such f F 2
KH ?
As partial answers, there are the following results.
Theorem 4.123 (Ruscheweyh and Salinas). Let g be analytic in D. Then
fe
g D Reff g g C Imff g g 2 KH
for all f 2 KH ” for each 2 R; g C i zg0 is convex in the imaginary direction.

Theorem 4.124 (Goodloe). Let fm ; fn 2 KH be the canonical harmonic functions that
map D onto the regular m-gon and n-gon, respectively. Then fm fn 2 KH and the image
of D is a p-gon, where p D lcm.m; n/.
Exercise 4.125. Compute fk D f4 f6 , where f4 D h4 C g4 is the canonical square
map (see Example 4.101) given by
! ! Z
1 1Cz i i Cz 1
h4 .z/ D log C log D dz
4 1 z 4 i z 1 z4
! ! Z
1 1Cz i i Cz z2
g4 .z/ D log C log D dz:
4 1 z 4 i z 1 z4
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 256 — #274

i i
and f6 D h6 C g6 is the canonical regular hexagon map (see Exercise 4.108) given by
i i i2 i2
1 1Cz e 3 1Ce 3 z e 3 1Ce 3 z
h6 .z/ D log C log i
C log i2
6 1 z 6 1 e3z 6 1 e 3 z
Z
1
D dz
1 z6
i i i2 i2
1 1Cz e 3 1Ce 3 z e 3 1Ce 3 z
g6 .z/ D log log i
log i2
6 1 z 6 1 e 3 z 6 1 e 3 z
Z 4
z
D dz:
1 z6
Sketch fk .D/ using ComplexTool. Try it out!
In considering Open Problem 4.122, let’s look at a simple problem: if f1 ; f2 2 KH ,
then when is f1 f2 2 SH ?
We need Lewy’s theorem that f D h C g with h0 .z/ ¤ 0 in D is locally univalent and
sense-preserving if and only if !.z/ D g0 .z/= h0 .z/j < 1, 8z 2 D.
Theorem 4.126. Let f1 D h1 C g1 ; f2 D h2 C g2 2 KH with hk .z/ C gk .z/ D 1 z z for
k D 1; 2. If f1 f2 is locally univalent and sense-preserving, then f1 f2 2 SH and is
convex in the direction of the real axis.
z z
Proof. Since h.z/ C g.z/ D and F .z/ D F .z/ for any analytic function
1 z 1 z
F , we have that
h2 g2 D .h1 C g1 / .h2 g2 /
D h1 h2 h1 g2 C h2 g1 g1 g2
h1 g1 D .h1 g1 / .h2 C g2 /
D h1 h2 C h1 g2 h2 g1 g1 g2 :
Thus,
h1 h2 g1 g2 D 12 Œ.h1 g1 / C .h2 g2 /: (4.16)
We will now show that .h1 g1 / C .h2 g2 / is convex in the direction of the real axis.
We have

h0 .z/ g0 .z/
h0 .z/ g0 .z/ D h0 .z/ C g0 .z/
h0 .z/ C g0 .z/

0 0
1 !.z/ p.z/
D h .z/ C g .z/ D ;
1 C !.z/ .1 z/2
where Refp.z/g > 0; 8z 2 D. So, letting '.z/ D z=.1 z/2 , we have

( z )
zŒ.h01 .z/ g10 .z// C .h02 .z/ g20 .z// .1 z/2
Œp1 .z/ C p2 .z/
Re D Re z
'.z/ .1 z/2
n o
D Re p1 .z/ C p2 .z/ > 0:
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 257 — #275

i i
Therefore, by Theorem 4.87 in Section 4.6 and (4.16), h1 h2 g1 g2 is convex in the

direction of the real axis.
Since we assumed that f1 f2 is locally univalent, we apply Clunie and Sheil-Small’s
shearing theorem (see Theorem 4.52) to get that f1 f2 D h1 h2 g1 g2 is convex
in the direction of the real axis.
It is known (see [9]), that for any right half-plane mapping f D h C g 2 KH ,

z
h.z/ C g.z/ D :
1 z
Hence, Theorem 4.126 applies to harmonic right half-plane mappings.
Example 4.127. Let f0 D h0 C g 0 be the canonical right half-plane mapping given in
Example 4.40 with h0 .z/ C g0 .z/ D 1 z z with !.z/ D z. Then
1 2
z 2
z
h0 .z/ D
.1 z/2
1 2
2z
g0 .z/ D :
.1 z/2
z
Let f1 D h1 C g 1 , where h1 .z/ C g1 .z/ D 1 z
with !.z/ D z. Then

1 1Cz 1 z
h1 .z/ D log C
4 1 z 21 z

1 1Cz 1 z
g1 .z/ D log C :
4 1 z 21 z
Note that f1 is a right half-strip mapping (see Figure 4.46).
Let F1 D f0 f1 D H1 C G1 . We have
3
1 0
1 1Cz 4
z 14 z 3
H1 .z/ D h0 .z/ h1 .z/ D h1 .z/ C zh1 .z/ D log C
2 8 1 z .1 z/2 .1 C z/
1
1 1 1Cz z 21 z 2 41 z 3
G1.z/ D g0 .z/ g1 .z/ D g1 .z/ zg10 .z/ D log C 4 ;
2 8 1 z .1 z/2 .1 C z/
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 258 — #276

i i
with
2z 2 C z C 1
! .z/ D z
e :
z2 C z C 2
The image of D under F1 D f0 f1 is shown in Figure 4.47.
F IGURE 4.47. Image of D under F1 D f0 f1 .
Exercise 4.128. Compute F D H C G, where F D f0 f0 . Sketch F .D/ using

ComplexTool. Try it out!
In the rest of this section we will consider for which ! D g0 = h0 is f D h C g locally
univalent. Let
z 21 z 2 1 2
2
z
f0 .z/ D h0 .z/ C g0 .z/ D 2
.1 z/ .1 z/2
be the canonical right half-plane mapping used in Example 4.127.
Also, as mentioned after the proof of Theorem 4.126, the collection of functions f D
h C g 2 SH O
that map D onto the right half-plane, R D fw W Re.w/ > 1=2g, have the
form
z
h.z/ C g.z/ D :
1 z
We will use the following method to prove that local univalency holds:
Method 1. (Cohn’s Rule, see [23, p 375]) Given a polynomial
f .z/ D a0 C a1 z C C an z n
of degree n, let
f .z/ D z n f .1= z/ D an C an 1z C C a0 z n :
Denote by p and s the number of zeros of f inside the unit circle and on it, respectively.
If ja0 j < jan j, then
an f .z/ a0 f .z/
f1 .z/ D
z
is of degree n 1 with p1 D p 1 and s1 D s the number of zeros of f1 inside the unit
circle and on it, respectively.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 259 — #277

i i
Theorem 4.129. Let f D h C g 2 KH O

with h.z/ C g.z/ D 1 z z and !.z/ D e i z n
.n 2 N and 2 R/. If n D 1; 2, then f0 f 2 SHO
is convex in the direction of the real
axis.
! D .g0 g/0 =.h0 h/0 . By Theorem 4.126

Proof. Let the dilatation of f0 f be given by e
and by Lewy’s theorem, we need only show that je ! .z/j < 1; 8z 2 D.
If F is analytic in D and F .0/ D 0, then
1
h0 .z/ F .z/ D F .z/ C zF 0 .z/
2
1
g0 .z/ F .z/ D F .z/ zF 0 .z/ :
2
Since g0 .z/ D !.z/h0 .z/, we know g00 .z/ D !.z/h00 .z/ C ! 0 .z/h0 .z/.
Hence,
zg00 .z/ z! 0 .z/h0 .z/ z!.z/h00 .z/
! .z/ D
e D :
2h0 .z/ C zh00 .z/ 2h0 .z/ C zh00 .z/
z
Using h.z/ C g.z/ D 1 z
and g0 .z/ D !.z/h0 .z/, we can solve for h0 .z/ and h00 .z/ in
terms of z and !.z/:
1
h0 .z/ D
.1 C !.z//.1 z/2
2.1 C !.z// ! 0 .z/.1 z/

h00 .z/ D :
.1 C !.z//2 .1 z/3
Substituting h0 and h00 into the equation for e

! , we get
z! 0 .z/h0 .z/ z!.z/h00 .z/
! .z/ D
e
2h0 .z/ C zh00 .z/
(4.17)
! .z/ C Œ!.z/ 21 ! 0 .z/z C 12 ! 0 .z/
2
D z :
1 C Œ!.z/ 12 ! 0 .z/z C 12 ! 0 .z/z 2
If !.z/ D e i z, then (4.17) yields

2i
z 2 C 12 e i
z C 21 e i
p.z/
! .z/ D
e ze 1 i 1 i 2
D ze 2i :
1C 2
e z C 2
e z q.z/
1
Note that q.z/ D z 2 p.1=z/. Hence, if z0 is a zero of p, then z0
is a zero of q. So,
.z C A/.z C B/
! .z/ D
e ze 2i :
.1 C Az/.1 C Bz/
Using Method 1, we have

a2 p.z/ a0 p .z/ 3 1 i 1
p1 .z/ D D zC e :
z 4 2 4
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 260 — #278

i i
So, p1 has one zero at z0 D 13 32 e i 2 D. By Cohn’s Rule, p has two zeros, namely A
and B, with jAj; jBj < 1.
Next, consider the case in which !.z/ D e i z 2 . Then
ˇ ˇ
ˇ ˇˇ z 3 C e i ˇ
! .z/j D ˇz 2 ˇˇˇ
je ˇ D jzj2 < 1:
1 C e i z 3 ˇ
Example 4.130. Let f2 D h2 C g2 be the harmonic mapping in D such that h1 .z/ C

g2 .z/ D 1 z z and !2 .z/ D z 2 . Then we can compute

1 1Cz 1 z 1 z
h2 .z/ D ln C C ;
8 1 z 21 z 4 .1 z/2

1 1Cz 1 z 1 z
g2 .z/ D ln C ;
8 1 z 2 1 z 4 .1 z/2
ˇ
and the image of D under f2 is the right half-plane, R D fw 2 C ˇ Refwg 12 g. Note
that f2 .e i t / D 12 C i 16

, if 0 < t < , and f2 .e i t / D 12 i 16

, if < t < 2 (see Figure
4.48).
Let
F2 D h0 h2 C g0 g2 D H2 C G2:
Then we can compute that

1 1Cz 1 z 1 z 1 z
H2 .z/ D ln C C C
16 1 z 41 z 8 .1 z/2 2 .1 z/3 .1 C z/
(4.18)
1 1Cz 1 z 1 z 1 z3
G2.z/ D ln C C :
16 1 z 4 1 z 8 .1 z/2 2 .1 z/3 .1 C z/
It can be shown analytically that F2 .D/ is the entire complex plane except for two half-

lines x ˙ 16 i , x 14 . This is not clear if we use ComplexTool with the standard settings to
view this image (see Figure 4.49). However, using both this image and the image of the unit
circle (see Figure 4.50), it seems reasonable. To graph the image of @D in ComplexTool,
change the settings in the middle box of ComplexTool to Interior circles: 0 and
Rays: 0.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 261 — #279

i i
F IGURE 4.49. Image of D under F2 D f0 f2 .
F IGURE 4.50. Image of @D under F2 D f0 f2 .
Remark 4.131. If we assume the hypotheses of Theorem 4.129 with the exception of
making n 3, then for each n we can find !.z/ D e i z n such that f0 f … SH
O
. For
n
example, if n is odd, let !.z/ D z and then (4.17) yields

n
z nC1 C n2 1 z n2
! .z/ D z
e :
1 C n2 1 z n n2 z nC1
n
It suffices to show that for some point z0 2 D, je
! .z0 /j > 1. Let z0 D nC1 2 D. Then
n
n nC1 n
n
n
n nC1 2
1 nC1 2
! .z0 / D
e
nC1 1 n n n n n nC1
2 1 nC1 2 nC1
(4.19)
h nC1 i h i
nC1 n n n
n nC1
C 1 nC1
D1C n
n
n
n :
2
1 C 2 nC1 C nC1n
n
n nC1
n n
Note that nC1 n nC1
C 1 nC1 n
> 0. Also, n2 1 C n2 nC1 C nC1 n
>0
n
since n2 1 C n2 nC1 n
> n 32 > e and nC1 n converges to e as n ! 1. Thus, if n 5
ˇ 4 1 3 3 4ˇ
3 ˇ 3 34 4 ˇ
is odd, je
! .z0 /j > 1. If n D 3, it is easy to compute that je! .z0 /j D 43 ˇ 4 21 3 32 4 ˇ >
4 2 3 4 2 3
n
2. If n is even, let !.z/ D z n and z0 D This simplifies to the same e
. ! .z0 / given
ˇ ˇ
nC1
4 4 ˇ 45 454 255 ˇ
(4.19) and the argument also holds for n 6. If n D 4, je
! .z0 /j D 5 ˇ 55 44 5 245 ˇ >
15.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 262 — #280

i i
Exploration 4.132. Using ComplexTool, graph e ! .D/ given in (4.17) for !.z/ D z n ,
where n D 1; 2; 3; 4. Explain how the images support Theorem 4.129 and Remark 4.131.
Try it out!
Theorem 4.133. Let f D h C g 2 KH O
with h.z/ C g.z/ D 1 z z and !.z/ D zCa
1Caz
with
a 2 . 1; 1/. Then f0 f 2 SH
O
is convex in the direction of the real axis.
zCa
Proof. Using !.z/ D 1Caz
, where 1 < a < 1, we have
1C3a

z2 C 2
z C 1Ca
2
! .z/ D
e z 1C3a 1Ca 2

1C 2
z C 2
z
f .z/
D z
f .z/
.z C A/.z C B/
D z
.1 C Az/.1 C Bz/
p.z/
D z :
q.z/
Using Method 1,
a2 p.z/ a0 p .z/ .a C 3/.1 a/ .1 C 3a/.1 a/
p1 .z/ D D zC :
z 4 4
1C3a
So p1 has one zero at z0 D aC3
that is in the unit circle since 1 < a < 1. Thus, jAj,
jBj < 1.
Large Project 4.134. In Theorem 4.126, we require that the convolution function satisfy
the dilatation condition ˇ 0 ˇ
ˇ g .z/ ˇ
j!.z/j D ˇˇ 0 ˇˇ < 1; 8z 2 D:
h .z/
Determine ! functions for which the dilatation condition holds and ones for which it does
not hold. See Theorem 4.129, Theorem 4.133, and Remark 4.131 for examples. Optional
Large Project 4.135. Similar to the right half-plane
map, f D h C g is an asymmetric
1 1Cze i˛
vertical strip map if h.z/ C g.z/ D 2i sin ˛ log 1Cze i˛ , where 0 < ˛ < . Theorem
4.126 can be stated in terms of asymmetric vertical strip mappings instead of right half-
plane mappings.
O 0 0
Theorem. Let f D h C g 2 KH with ! D g = h be such that h C g D
1 1Cze i˛
2i sin ˛ log 1Cze i˛ , where 0 < ˛ < . Then f0 f 2 SH and is convex in
O
the direction of the real axis.

Determine ! functions for which the dilatation condition holds for this theorem and
ones for which it does not hold. Optional
4.9 Conclusion
We have presented an introduction to harmonic univalent mappings and described a few
topics to entice a beginner. Our emphasis has been on the geometric aspects of harmonic
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 263 — #281

i i
univalent mappings that students can explore using the exercises, the exploratory prob-
lems, and the projects along with the applets. There are more interesting and deeper topics
in harmonic univalent mappings. Here is a short list along with some resources: coeffi-
cient estimates and conjectures ([6], [12], [27]); a generalized Riemann mapping theorem
([12], [18]); properties of special classes of functions such as convex, close-to-convex, star-
like, and typically real ([6], [12], [27]); harmonic polynomials ([5], [28], [31]); extremal
problems ([12]); harmonic meromorphic functions ([20], [29]); inner mapping radius ([3],
[11]); and multiply connected domains ([12], [14]). Another topic is the connection be-
tween harmonic mappings and minimal surfaces. It is discussed in Chapter 2. In addition,
there are several nice general resources that can be used to learn more about harmonic uni-
valent functions. These include Peter Duren’s book [12], Clunie and Sheil-Small’s original
article [6], Bshouty and Hengartner’s article [4], and Schober’s article [26]. Bshouty and
Hengartner complied a list of open problems and conjectures [3].

Anamorphosis and Möbius Maps
Exercise 4.136. Determine the explicit image of horizontal lines Im z D k under the
inversion map f .z/ D 1z and analytically prove your result.
Exercise 4.137. Analytically determine the image of the left half-plane fz j Refzg < 0g
under the map
z i
M.z/ D :
z
Use ComplexTool to check your answer.
Exercise 4.138. The strip D D fz j0 < Imfzg < 1g can be thought of as the intersection
of the domains D1 D fz j 0 < Imfzgg and D2 D fz j Imfzg < 1g. Analytically determine
z i
the image of the strip D under the Möbius transformation M.z/ D zCi .
Exercise 4.139. Determine a Möbius transformation that maps the left half-plane fz j Refzg <
0g onto the disk fz j jz 1j < 1g. Use ComplexTool to check your answer.
Exercise 4.140. Prove that if z1 , z2 , z3 are distinct points and w1, w2 , w3 are distinct
points, then the Möbius transformation T satisfying T .z1 / D w1 , T .z2 / D w2, T .z3 / D
w3 is unique.
azCb
Exercise 4.141. Let M.z/ D czCd with ad ¤ bc be a Möbius transformation. Prove that
M is a rotation if and only if M.0/ D 0 and M preserves distances. Preserving distances
means jz1 z2 j D jM.z1 / M.z2 /j.
The Family S of Analytic, Normalized, Univalent Functions

Exploration 4.142.
(a) Using ComplexTool guess the smallest k > 0 such that .z C k/2 is univalent in D.
(b) Prove your guess.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 264 — #282

i i
(c) Using ComplexTool guess the smallest k > 0 such that .z C k/3 is univalent in D.
(d) Prove your guess.
Exercise 4.143. Show that f .z/ D z C a3 z 3 is univalent in D ” ja3 j 13 . Determine
f .D/ analytically when a3 D 13 .
P
Exercise 4.144. Work out the details to show that .1 zz/2 D 1 n
nD1 nz D z C 2z C
2
3z 3 C :
z cz 2
Exercise 4.145. Determine f .D/ analytically f .z/ D , where 0 < c < 1.
.1 z/2

1 1Cz
Exercise 4.146. Prove that f .z/ D log is univalent. Determine f .D/ analyt-
2 1 z
ically.
Exploration 4.147. Let " #
c
1 1Cz
fc .z/ D 1 :
2c 1 z
(a) Show that if c D 2, then fc .z/ is the Koebe function.
(b) Show that if c D 1, then fc .z/ is the right half-plane mapping.
(c) Use ComplexTool to view the image of D under fc for various values c, 0 < c < 2.
For what values of c does fc appear to be univalent?
Exercise 4.148. Find the image of D analytically under the univalent function f .z/ D
z
.
1 z2
The Family SH of Normalized, Harmonic, Univalent Functions

Exercise 4.149. Determine if f .x; y/ D u.x; y/ C iv.x; y/ D .x 3 C xy 2 / C i.x 2 y C y 3 /
is complex-valued harmonic.
@2 f
Exercise 4.150. Prove that f .x; y/ D u.x; y/ C iv.x; y/ is harmonic ” D 0.
@z @z
1 2
Exercise 4.151. Rewrite f .x; y/ D u.x; y/ C iv.x; y/ D .x 2
x C 12 y 2 / C i.y xy/
in terms of z and z and determine if f is analytic.
1
Exercise 4.152. Prove that for all functions f 2 SH
O
, the sharp inequality jb2 j 2 holds.
Exercise 4.153. Verify that the image of D under the harmonic function f .z/ D z C 12 z 2
is a hypocycloid with three cusps.
Exercise 4.154. If a domain is convex in the direction e i' for every value of ' 2 Œ0; /,
then it is called a convex domain. For example, a disk is a convex domain. For which values
of n D 1; 2; 3; : : : do the following functions map D onto a convex domain?
(a) f .z/ D z n ,
1 n
(b) f .z/ D z n
z (see Example 4.19 and Definition 4.20),
z
(c) f .z/ D .1 z/n
(see Examples 4.22 and 4.23 to get you started).
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 265 — #283

i i
The Shearing Technique

Exploration 4.155. Let f D h C g with h.z/ g.z/ D z n1 z n and !.z/ D z n 1 . Use
ShearTool to sketch the graph of f .D/ for different values of n and then compute h and g
explicitly so that f 2 SH
O
.
z
Exercise 4.156. Let f D h C g with h.z/ g.z/ D .1 z/2
and !.z/ D z. Compute h
and g explicitly so that f 2 SH
O
and determine f .D/.
z z C 12 z
Exercise 4.157. Let f D h C g with h.z/ g.z/ D and !.z/ D z .
.1 z/2 1 C 21 z
(a) Show that j!.z/j < 1; 8z 2 D.
(b) Compute h and g explicitly so that f 2 SH
O
.
(c) Show that f .D/ is a slit domain like the Koebe domain. Determine where the tip of
the slit is located.
(d) What is the significance of this example in relationship to the Riemann mapping theo-
rem?
z
Exercise 4.158. Let f D h C g with h.z/ C g.z/ D and !.z/ D e i z, where
1 z
2 Œ0; 2 /. Use ShearTool to sketch the graph of f .D/ for different values of n and
compute h and g explicitly so that f 2 SH O
.
Exploration 4.159. We can find harmonic functions fn D hn C gn that map onto regular
n-gons by generalizing the ideas from Example 4.72. Use ShearTool to explore the images
of D under f D h C g, where f comes from shearing
n
X1 2 cos 2k 2 k
n
hn .z/ gn .z/ D log 1 ze i n
n
kD0
with !.z/ D z n 2
.

1 1Cz
Exploration 4.160. Let f D hC g with h.z/ g.z/ D log and !.z/ D m2 z 2 ,
2 1 z
where m D e i .0 2 /. Use ShearTool to sketch the graph of f .D/ for different val-
ues of n and compute h and g explicitly so that f 2 SH O
Note: This Exploration fits
nicely with minimal surfaces, because when m D i , f lifts to a canonical minimal sur-
face, Scherk’s doubly-periodic, and when m D 1, f lifts to a different canonical minimal
surface, the helicoid.
Properties of the Dilatation

Exploration 4.161. Shear h.z/ g.z/ D 1 z z using !.z/ D az, where 1 a 1 and
sketch f .D/ using ShearTool. Describe what happens to f .D/ as a varies.
Exploration 4.162. Shear h.z/ g.z/ D 1 z z using !.z/ D z n , where n D 1; 2; 3; 4; 5
and sketch f .D/ using ShearTool. Describe what happens to f .D/ as a varies.
1 z

Exploration 4.163. Shear h.z/ g.z/ D log 1Cz using !.z/ D e i n=6z, where n D
0; : : : ; 6 and sketch f .D/ using ShearTool. Describe what happens to f .D/ as n varies.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 266 — #284

i i
zC1 zC1
Exploration 4.164. Shear h.z/ g.z/ D 1 z z C ae z 1 using !.z/ D e z 1 , where
0:5 a 0:5 and sketch f .D/ using ShearTool. Describe what happens to f .D/ as a
varies.
i˛

i zCe
Exercise 4.165. Let h˛ .z/ D 1Czez i˛ , where 0 < ˛ < , and !˛ .z/ D e 1Cze i˛ .
Compute f˛ D h˛ C g ˛ and show that f˛ 2 SH O
. Use ComplexTool to sketch f˛ .D/ for
various values of ˛. As ˛ approaches 0, you should get the image shown in Figure 4.28.

1 1Cze i
Exercise 4.166. Let h .z/ D 2i sin log 1Cze i
, where 2 < , and ! .z/ D
2 sin. /

e . / h.z/ 1 . Compute f D h C g and show that f 2 S O . Use ComplexTool to
H
sketch f .D/ for various values of . As approaches , you should get the image shown
in Figure 4.28, but each f .D/ is different than any f˛ .D/ in Exercise 4.165.
Harmonic Linear Combinations
Exercise 4.167. Let

˚ ˚
f1 .z/ D Re z2 log.1 z/ C i Im z ;

z C 1=3z 3 z
.1 z/3 .1 z/2
(a) Show that f1 can be derived by shearing h.z/ g.z/ D z with g0 .z/ D zh0 .z/.
(b) Use ComplexTool to plot the image of D under f1 . Recall that f2 is the harmonic
Koebe function. What is the image of D under f2 ?
(c) Use ComplexTool to see that f3 D tf1 C .1 t/f2 is not univalent for at least one
value of t; 0 t 1. Why does this not contradict Theorem 4.100?

z 1 1Cz
f1 .z/ D Re C i Im log ;
.1 z/2 2 1 z

i 1 iz 1 1Cz
2 1 C iz 2 1 z
Show that f1 and f2 satisfy the conditions of Theorem 4.100 and then use ComplexTool to
plot images of f3 D tf1 C .1 t/f2 for various values of t.

z 1 zC1 z 1 zC1
f1 .z/ D Re e z 1 C i Im C ez 1 ;
1 z 2 1 z 2

1 2 1 zC1 1 2 1 zC1
f2 .z/ D Re z z .z 1/2 e z 1 C i Im z z C .z 1/2 e z1 :
4 4 4 4
Show that f1 and f2 satisfy the conditions of Theorem 4.100 and then use ComplexTool to
plot images of f3 D tf1 C .1 t/f2 for various values of t.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 267 — #285

i i
Exercise 4.170. Repeat the steps in Example 4.104 using the same function for f1 but
replacing f2 with the harmonic square map in Example 4.72, where

1 1Cz i i Cz
h2 .z/ D log C log
4 1 z 4 i z

1 1Cz i i Cz
4 1 z 4 i z
(a) In this case, !1 .z/ D z 2 while !2 .z/ D z 2 . Using (4.15) in Remark 4.106, show
that f3 is locally univalent.
values of w1 ; : : : ; w8 and then use the visualization technique in the example to plot
the vertices z0 ; : : : ; z7 and draw the midpoints w1 ; : : : ; w8.
Exploration 4.171. Using LinComboTool, start with the same arc values and correspond-
ing point values as in Example 4.104. In Panel #1 increase the arc values by increments

of 16 while not changing the point values, and decrease the arc values in Panel #2 by
the same amount. You can do this either by changing the value in the Arc n box or by us-
ing the cursor to move the four blue dots the same amount in the same direction in Panel
#1 and the same amount in the opposite direction in Panel #2 on the unit circle of the
domain in each panel. Describe how the image domain changes as the arc values in Panel
#1 increase by a total of 4 and decrease in Panel #2 by the same amount.
Exploration 4.172. Using LinComboTool, create a map in Panel #1 that maps three
arcs on the unit circle to three vertices of an equilateral triangle. Then create a second
map in Panel #2 that maps three different arcs on the unit circle to three vertices of
a rotated equilateral triangle. Make sure that the maps satisfy the conditions of Theorem
4.100. Click on the Create LinCombogon button to see the image domain. Explore
this idea by using different maps in the panels to get at least three different image domains.
Exploration 4.173. Using LinComboTool, create a map in Panel #1 that maps six arcs
on the unit circle to six vertices of a regular hexagon. Then create a second map in Panel
#2 that maps six different arcs on the unit circle to six vertices of a rotated regular hexagon.
Make sure that the maps satisfy the conditions of Theorem 4.100. Click on the Create
LinCombogon button to see the image domain. Explore this idea by using different maps
in the panels to get at least three different image domains.
Convolutions
Exercise 4.174. Let
Z
1 1 1Cz
f .z/ D dz D log
1 z2 2 1 z
and
Z
1 1 i5 i 1 i5 i5 1
F .z/ D dz D e 3 log 1 C e 3 z C e 3 log 1 C e 3 z log 1 z :
1 z3 3 3 3
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 268 — #286

i i
Using (4.13) at the end of Example 4.72, determine f F and the image of D under this
convolution. In general, what is f F when f 0 .z/ D 1 1zm and F .z/ D 1 1zn ?
Exercise 4.175. In Theorem 4.121, let f be the canonical right half-plane mapping f0 2
KH and let '.z/ D .1 zz/2 2 S . Compute F D f0 . ' C '/ and use ComplexTool to
sketch F .D/.
Exercise 4.176. Derive the expressions for H2 and G2 given in (4.18).
Small Project 4.177. Compute fa D ha C g a , where ha .z/ C ga .z/ D 1 z z and !.z/ D
zCa
1Caz
. From Theorem 4.133, we know that Fa D f0 fa 2 SH for 1 < a < 1. Compute
Fa and use ComplexTool to sketch Fa .D/ for various values of a; 1 < a < 1. Describe
what happens as a varies between 1 and 1.
4.11 Bibliography
[1] Lars Ahlfors, Conformal Invariants: Topics in Geometric Function Theory, McGraw-
Hill, Inc., New York, 1973.
[2] Zach Boyd, Michael Dorff, Rachel Messick, Matthew Romney, and Ryan Viertel,
Harmonic univalent mappings with singular inner function dilatation, preprint.
[3] Daoud Bshouty and Walter Hengartner (editors), Problems and conjectures for har-
monic mappings, from a workshop held at the Technion, Haifa, 1995.
[4] ——, Univalent harmonic mappings in the plane, Handbook of Complex Analysis:
Geometric Function Theory, Vol. 2, 479–506, Elsevier, Amsterdam, 2005.
[5] Daoud Bshouty and Abdallah Lyzzaik, On Crofoot-Sarason’s conjecture for har-
monic polynomials, Comput. Methods Funct. Theory 4 (2004), no. 1, 35–41.
[6] James Clunie and Terry Sheil-Small, Harmonic univalent functions, Ann. Acad. Sci.
Fenn. Ser. A.I Math. 9 (1984), 3–25.
[7] Louis de Branges, A proof of the Bieberbach conjecture, Acta Math. 154 (1985), no.
1–2, 137-152.
[8] Michael Dorff, Convolutions of planar harmonic convex mappings, Complex Vari-
ables Theory Appl., 45 (2001), no. 3, 263–271.
[9] ——, Harmonic mappings onto asymmetric vertical strips, in Computational Meth-
ods and Function Theory 1997, (N. Papamichael, St. Ruscheweyh, and E. B. Saff,
eds.), 171–175, World Sci. Publishing, River Edge, NJ, 1999.
[10] Michael Dorff, Missy Lucas, Rachel Messick, and Chad Witbeck, Convex combina-
tions of harmonic mappings and minimal graphs, preprint.
[11] Michael Dorff, and Ted Suffridge, The inner mapping radius of harmonic mappings
of the unit disk, Complex Variables Theory Appl. 33 (1997), no. 1–4, 97–103.
[12] Peter Duren, Harmonic Mappings in the Plane, Cambridge Tracts in Mathematics,
156, Cambridge University Press, Cambridge, 2004.
[13] ——, Univalent Functions, Springer-Verlag, New York, 1983.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 269 — #287

i i
[14] Peter Duren, and Walter Hengartner, Harmonic mappings of multiply connected do-
mains, Pacific J. Math. 180 (1997), no. 2, 201–220.
[15] Peter Duren, Jane McDougall, and Lisabeth Schaubroeck, Harmonic mappings onto
stars, J. Math. Anal. Appl. 307 (2005), no. 1, 312–320.
[16] Mary Goodloe, Hadamard products of convex harmonic mappings, Complex Var. The-
ory Appl. 47 (2002), no. 2, 81–92.
[17] Paul Greiner, Geometric properties of harmonic shears, Comput. Methods Funct. The-
ory, 4 (2004), no. 1, 77–96.
[18] Walter Hengartner and Glenn Schober, Harmonic mappings with given dilatation, J.
London Math. Soc. (2) 33 (1986), no. 3, 473–483.
[19] ——, On schlicht mappings to domains convex in one direction, Comment. Math.
Helv. 45 (1970), 303–314.
[20] ——, Univalent harmonic functions, Trans. Amer. Math. Soc. 299 (1987), no. 1, 1–31.
[21] Richard Laugesen, Planar harmonic maps with inner and Blaschke dilatations, J. Lon-
don Math. Soc. (2) 56 (1997), 37–48.
[22] Christian Pommerenke, On starlike and close-to-convex functions, Proc. London
Math. Soc. (3) 13 (1963), 290–304.
[23] Qazi Ibadur Rahman and Gerhard Schmeisser, Analytic Theory of Polynomials, Lon-
don Mathematical Society Monographs New Series, 26, Oxford University Press,
Oxford, 2002.
[24] Stephan Ruscheweyh and Luis Salinas, On the preservation of direction-convexity
and the Goodman-Saff conjecture, Ann. Acad. Sci. Fenn., Ser. A. I. Math. 14 (1989),
63–73.
[25] Stephan Ruscheweyh and Terry Sheil-Small, Hadamard products of schlicht func-
tions and the Polya-Schoenberg conjecture, Comment. Math. Helv., 48 (1973), 119–
135.
[26] Glenn Schober, Planar harmonic mappings, Computational Methods and Function
Theory (Valparaso, 1989), 171–176, Lecture Notes in Math., 1435, Springer, Berlin,
1990.
[27] Terry Sheil-Small, Constants for planar harmonic mappings, J. London Math. Soc.
(2) 42 (1990), no. 2, 237–248.
[28] Ted Suffridge, Harmonic univalent polynomials, Complex Var. Theory Appl., 35
(1998), no. 2, 93–107.
[29] John Thompson, A family of meromorphic harmonic mappings, Complex Var. Theory
Appl. 48 (2003), no. 8, 627–648.
[30] Allen Weitsman, Harmonic mappings whose dilatations are singular inner functions,
unpublished manuscript.
[31] Alan Wilmshurst, The valence of harmonic polynomials, Proc. Amer. Math. Soc. 126
(1998), no. 7, 2077–2081.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 270 — #288

i i
i i
i i
i i
“ECA˙Book” — 2012/9/5 — 14:07 — page 271 — #289

i i
Mappings to Polygonal Domains

5
Jane M. McDougall and Lisbeth E. Schaubroeck (text),
5.1 Introduction
A rich source of problems in analysis is determining when, and how, we can create a
one-to-one function of a particular type from one region onto another. In this chapter, we
consider the problem of mapping the unit disk D onto a polygonal domain by two different
classes of functions. For analytic functions we give an overview and examples of the well
known Schwarz-Christoffel transformation. We then diverge from analytic function theory
and consider the Poisson integral formula to find harmonic functions that will serve as
mapping functions onto polygonal domains. Proving that harmonic functions are univalent
requires us to explore some less widely known theory of harmonic functions and to use
relatively new techniques.
Because of the Riemann mapping theorem, we can simplify our mapping problem for
either class of function to asking when we can map the unit disk D D fz W jzj < 1g
univalently onto a target region. This is because if we want to map one domain (other than
the entire set of complex numbers) onto another, we can first map it to D by an analytic
function, and then apply an analytic or harmonic mapping from D to the other domain
(recall that the composition of a harmonic function with an analytic function is harmonic).
We begin in Section 5.2 by using the Schwarz-Christoffel formula to find univalent
analytic maps onto polygonal domains, and so set the stage for the corresponding prob-
lem for harmonic functions with the Poisson integral formula in Section 5.3. Perhaps be-
cause of their importance in applications, many first books on complex analysis introduce
Schwarz-Christoffel mappings through examples, without emphasis on subtleties of the
deeper theory. Our approach will be the same, and the examples we include are chosen
to bring together ideas found elsewhere in this book, such as the shearing technique from
Chapter 4 and the construction of minimal surfaces from Chapter 2.
We also include an example of a Schwarz-Christoffel map onto a regular star, a poly-
gonal domain that is highly symmetric but not convex. The problem of using the Poisson
integral formula to construct a univalent harmonic function onto a non-convex domain is
not at all well understood. In Section 5.3, after developing the theory for convex domains,
we carefully construct univalent harmonic maps onto regular star domains, and lead the
student to further investigation.
271
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 272 — #290

i i
272 Chapter 5. Mappings to Polygonal Domains
We use the term univalent for one-to-one, and take domain to mean “open connected
set in the complex plane.” The applets used in this chapter are
1. ComplexTool, to plot the image of domains in C under complex-valued functions.
2. PolyTool, to visualize the harmonic function that is the extension of a particular
kind of boundary correspondence. The user can dynamically change the boundary
correspondence and watch the harmonic function change.
3. StarTool, to examine the functions that map the unit disk D onto an n-pointed star.
The user can modify the shape of the star (by changing n and r ) and the boundary
correspondence (by changing p).
5.2 Schwarz-Christoffel Maps

In this section we consider conformal maps from the unit disk and the upper half-plane onto
various simply connected polygonal domains. By the Riemann mapping theorem, we can
map the unit disk conformally onto any simply connected domain that is a proper subset
of the complex numbers with a mapping function that is essentially unique.
While the Riemann mapping theorem tells us that there exists a univalent analytic func-
tion to map D onto our domain, finding an actual mapping function is no easy task. Even
for a simple domain such as a square, the mapping function from the disk cannot be ex-
pressed in terms of elementary functions. One situation in which this problem is relatively
simple is in mapping a region bounded by a circle or line in the complex plane to another
such region using fractional linear transformations. For this problem, we need only pick
three points on the bounding line or circle in the domain, and map them (in order) to three
arbitrarily chosen points on the boundary of the target region (see for example [10], [14],
[18], or [19]). The selection of three pairs of points determines the fractional linear trans-
formation completely, and works, for example, when finding a conformal map from D onto
any planar region bounded by a line or circle.
Exercise 5.1. Show that the linear fractional transformation (or Möbius transformation)
z 0 D .z/ D i 1Cz
1 z
maps the unit disk to the upper half-plane by finding the images of
0
three boundary points. Then show that its inverse function 1 .z 0 / D z D zz0 Cii maps the
upper half-plane to the unit disk. Try it out!
How can a mapping function be found when the target region is more complicated?
This question is relevant to solving the heat equation or the study of fluid flows, as dis-
cussed in Chapter 3.
The Schwarz-Christoffel transformation frequently enables us to find a function map-
ping onto a polygonal domain. In most texts the formula is presented as a mapping from
the upper half-plane H onto the target polygon. We now develop this formula—for a more
thorough treatment see [12], [3], and [15]. Suppose our target polygon has interior angles
˛k and exterior angles ˇk , where ˛k Cˇk D 1 and ˛k > 0. The exterior angle measures
the angle through which a bug, traversing the polygon in the counterclockwise direction,
would turn at each vertex. The exterior angle is signed, following the usual convention that
a counterclockwise rotation is positive while a clockwise rotation is negative. We obtain
the exterior angle by extending the side on which the bug approaches the vertex, and then
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 273 — #291

i i
5.2. Schwarz-Christoffel Maps 273
seeing through what angle the bug must rotate to get to the next side of the polygon. For ex-
ample, in Figure 5.1 the angle marked by ˇ2 is 2 , so ˇ2 D 1=2, which coincides with
the description of ˇ2 as a clockwise turn on the boundary. We can see from ˛2 C ˇ2 D 1
that ˛2 D 3=2. For a simple closed polygon (that is, one with no self-intersections), it is
always possible to describe the interior and exterior angles using coefficients ˛k and ˇk
as described. We refer to a vertex such as the one with ˇ1 > 0 in Figure 5.1 as a convex
corner and a vertex such as the one with ˇ2 < 0 as a non-convex corner.
b2 p = – 2p b1 p
3p
a1 p
a2 p = 2
F IGURE 5.1. A sample polygon with both a convex and a non-convex corner.
The Schwarz-Christoffel formula for mapping the upper half-plane H to the polygon
with exterior angles described by coefficients ˇk as above is
Z z
1
f .z/ D A1 dw C A2 ; z 2 H. (5.1)
0 .w x1 / .w x2 /ˇ2 .w xn /ˇn
ˇ1
The real values xi are preimages of the n vertices of the polygon, which we will refer to
from now on as prevertices. Different choices of the constants A1 and A2 rotate, scale, or
translate the target n-gon.
In (5.1), we use w as the variable of integration, and the limits of integration are chosen
to make the definite integral into a function of z. The (arbitrary) choice of 0 as a fixed point
might have to be altered if it corresponds to a point of discontinuity of the integrand.
Exercise 5.2. You may be familiar with the sine and arcsine functions on the complex
plane. Verify that the Schwarz-Christoffel mapping of H onto the infinite half strip de-
scribed by j Re.z/j < 2 and Im.z/ > 0 is given by the arcsine function. Use the prevertices
x1 D 1, x2 D 1 in formula 5.1. Try it out!
We can observe that the angles at the vertices are represented in the formula, but
nowhere do we see an accomodation for the side lengths of the target polygon. They are
influenced by the choice of prevertices xi , but in a nonlinear and non-obvious way. Many
texts include a map onto a triangle as a first example, but from elementary geometry we
know that the ratios of the side lengths of a triangle are determined by its interior angles.
To illustrate how side length depends on the choice of prevertices, we first consider the
Schwarz-Christoffel map of the upper half-plane onto a rectangle. We will make a some-
what arbitrary choice of prevertices, and then evaluate the resulting integral to determine
the target rectangle. We will find that just as the prevertices are arbitrarily chosen, so are
the side lengths of our target rectangle. In computing the integral, we will encounter a first
example of a special function known as an elliptic integral.
Example 5.3. We map the upper half-plane H onto a rectangle. We choose prevertices
x1 D 3; x2 D 1; x3 D 1; and x4 D 3. Since our target image is a rectangle, all exterior
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 274 — #292

i i
angles are =2, and ˇi D 1=2 for each i . Using (5.1), we obtain the mapping function
Z z
1
f .z/ D A1 dw C A2 ; z 2 H:
0 .w 1/ .w 3/ .w C 1/1=2 .w C 3/1=2
1=2 1=2
The constant A1 allows us to scale and rotate the image of H and A2 allows for a
translation. By choosing A1 D 1 and A2 D 0 we simplify to
Z z
1
f .z/ D p dw: (5.2)
0 .w 2 1/ .w 2 9/
The choice of constants does not affect the aspect ratio (ratio of adjacent sides) of the
rectangle. However the integral cannot be evaluated using techniques in standard calculus
texts. Instead it is a special function known as an elliptic integral (of the first kind, with
parameter k D 31 ).
Definition 5.4. An elliptic integral of the first kind is an integral of the form
Z sin
1
F .; k/ D p dw:
0 .1 w 2/.1 k 2 w 2 /
R 1
An alternate form is F .; k/ D 0
p d:
1 k 2 sin2
p
The change of variables w D sin ; dw D cos d D 1 w 2d shows that the
integrals in Definition 5.4 are identical. (The computer algebra system Mathematica uses
an alternate form, representing the integral by EllipticF[; m], where m D k 2 .)
p
Exercise 5.5. Carry out the change of variables w D sin ; dw D cos d D 1 w 2d
to show that
Z sin Z
1 1
F .; k/ D p dw D p d:
0 2
.1 w /.1 k w / 2 2 0 1 k 2 sin2
Try it out!
Returning to our integral from (5.2), we manipulate the integrand to recognize it as an
elliptic integral of the first kind.
Z z
1
f .z/ D p dw
0 .w 2 1/ .w 2 9/
Z z
1 1
D q q dw
0 1 2 1 2
1=9 .w 1/ 9
w 1
Z
1 z 1
D q dw
3 0 .1 w 2 / 1 19 w 2
1 1
D F .arcsin z; /:
3 3
As we will see, our choice for the prevertices ( 3; 1; 1, and 3) directly affects the
aspect ratio of the rectangle.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 275 — #293

i i
Exercise 5.6. Follow these steps to determine the aspect ratio of the rectangle that is the
image of H under the function
1 1
f .z/ D F .arcsin z; /:
3 3
(a) Explain why the integral
Z 1
1
K1 D q dw
0 1 2
.1 w2/ 1 9w
is a real number. Hint: use the geometry of the integrand. Conclude that

1 1 1 1
f .1/ D F arcsin 1; D F =2; D K1 =3:
3 3 3 3
By symmetry, show that f . 1/ D K1 =3. Thus the length of one side of the rectangle
is 2jK1 j=3:
(b) Determine that since f .3/ is the next vertex of the target rectangle (moving counter-
clockwise), then f .3/ D f .1/ C iK2 where K2 is a real constant. Combine this with

1 1
f .3/ D F arcsin 3;
3 3

to show that iK2 D 13 F .arcsin 3; 13 / F .=2; 13 / : p
The choice of sign for K2 could
be positive or negative, depending on our choice of 1. For consistency with our
choice of angles, K2 should be positive (Mathematica uses the “wrong” branch of the
square root for the function on the real axis).
(c) Combine the findings above to determine that the aspect ratio of the rectangle is

2F =2; 13
1:279:::
F arcsin 3; 13 F =2; 13
2.5
5
2.0
4
3 1.5
2 1.0
1
0.5
0
–4 –2 0 2 4 0.0
–1.5 –1.0 –0.5 0.0 0.5 1.0 1.5
F IGURE 5.2. A portion of upper half-plane (left) and a portion of the target rectangle (right) to
which it maps under the function of Exercise 5.3.
The mapping for Example 5.3 is illustrated in Figure 5.2. Only a portion of the upper
half-plane and its image are shown. The figure caption explains why the target rectangle is
incompletely filled in the upper central area. However, we can see that the aspect ratio is
approximately what we calculated.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 276 — #294

i i
Small Project 5.7. Rework Example 5.3 for a more general situation. Use the prevertices
x1 D ; x2 D 1; x3 D 1; and x4 D ; where > 1: You can find an equation
involving as a variable that, chosen correctly, would force the target rectangle to be a
square. The equation cannot be explicitly solved, but its solution can be approximated
numerically. This is a standard problem in some introductory complex analysis books (see
for instance example 22 of section 14 in [18]).
One observation we can make based on this example is that while it is straightforward
to write down a mapping function that has the correct angles, there is no simple way to pre-
scribe the side lengths. Also, there are only a few cases when our integral can be expressed
in terms of elementary functions, and in general the integral is not easy to evaluate. To find
and evaluate specific Schwarz-Christoffel mappings, it is usually helpful to use symmetry
of the target polygon (and of the prevertices) to simplify the computations.
The theory of Schwarz-Christoffel maps is intricate, and there is no guarantee that the
Schwarz-Christoffel formula will result in a univalent function (see [9]). All we can say
for sure is that a map from the upper half plane to a simply connected polygonal domain
that is conformal must take the form of (5.1) for some choice of constants and prevertices
(for details, see [3]).
A simplification to the Schwarz-Christoffel formula that is frequently employed is to
set one of the prevertices to be 1, which removes one factor from the denominator in the
integrand. Due to the symmetry of our target region, a rectangle, we have not used this
simplification in our examples here. Rather, we made use of symmetry in Exercise 5.6
by choosing the prevertices to be ˙1 and ˙3. However we were unable to find explicit
values of the prevertices to map onto a square. In addition to the problem of prescribing
the lengths of the sides of the target polygon, a further problem arises for target polygons
more complicated than a rectangle. Typically we will produce an integral that cannot be
evaluated, even if we utilize special functions.
In mapping onto a regular polygon, these two problems can be resolved if we obtain a
Schwarz-Christoffel formula that maps from the unit disk instead of the upper half plane,
using symmetrically placed points i on the unit circle in place of the xi in our existing
formula. This modified formula can be found by precomposing our mapping function (5.1)
with the Möbius transformation from the unit disk to the upper half-plane discussed in
Exercise 5.1. The prevertices can be chosen to be the symmetrically placed nth roots of
unity. If appropriate symmetry is exhibited by the angles of the target polygon, then the
side lengths of the target polygon will be equal. The following exercise asks you to show
that the transformed formula from the upper half plane turns out to have the same form as
our original Schwarz-Christoffel formula.
Exercise 5.8. Carry out the change of variables w D .z/ D i 1Cz 1 z
; which maps the
disk in the z-plane to the upper half w-plane (see Exercise 5.1). Show that the Schwarz-
Christoffel formula retains the same form as (5.1). Try it out!
The Schwarz-Christoffel map that we will use on the unit disk is then
Z z
1
f .z/ D C1 dw C C2 ; z 2 D, (5.3)
0 .w 1 / 1 .w 2 /ˇ2 .w n /ˇn
ˇ
where ˇi is the exterior angle of the i th vertex of the target polygon, and the pre-images
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 277 — #295

i i
i of the vertices are on the unit circle. We use i instead of xi to emphasize that the
prevertices are not on the real axis. As with (5.1), the complex constants C1 and C2 with
C1 ¤ 0 rotate, resize, and translate the polygon.
Example 5.9. We obtain the Schwarz-Christoffel map onto a regular n-gon. The exterior
angles of a regular n-gon are 2=n; so ˇi D 2=n; and the mapping function is
Z z
1
f .z/ D dw:
0 Œ.w 1 /.w 2 / .w n /2=n
Suppose that the i are the nth roots of unity. Because

n
Y
.w i / D w n 1
i D1
Rz
the integral becomes 0 n 1 2=n dw: By factoring out . 1/2=n we can adjust the multi-
.w 1/
plicative constant and choose our mapping function to be
Z z Z z
2=n 1 1
f .z/ D . 1/ 2=n
dw D dw: (5.4)
0 .w n 1/ 0 .1 w n /2=n
Here f has been defined from the Schwarz-Christoffel formula with choices of constant
C1 D . 1/ 2=n (which rotates the figure by 4=n radians) and C2 D 0: The last formula
cannot be evaluated using the usual methods from calculus, but can once again be evaluated
using special functions known as hypergeometric functions.
5.2.1 Basic Facts about Hypergeometric Functions

The integral in the last example cannot be expressed in terms of elementary functions, but
can be evaluated and plotted using a computer algebra system by using some facts about
power series known as hypergeometric functions. Hypergeometric functions have many
applications, and can be used to evaluate the integrals obtained above. We introduce the
most widely utilized hypergeometric functions—the so-called “two F ones,” where each
coefficient of the power series is a rational function with numerator and denominator of
the second order in n.
Definition 5.10. The hypergeometric function 2 F1 .a; bI cI z/ is the power series
X1
.a/n .b/n z n
2 F1 .a; bI cI z/ D ;
nD0
.c/n nŠ
where a, b, and c 2 C and
.x/n D x.x C 1/ .x C n 1/
is the shifted factorial, or Pochhammer symbol.

Exercise 5.11. Use algebra to check that .x/nC1 = .x/n D x C n. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 278 — #296

i i
P
If we compute the ratio of successive terms in the geometric series 1 n n
nD0 r z we
obtain the ratio r z. In the next exercise we carry out the computation for a hypergeometric
series.
Exercise 5.12. Show that the ratio of successive terms in the series 2 F1 .a; bI cI z/ is
.a C n/.b C n/z
:
.c C n/.n C 1/
Try it out!
The formula obtained motivates the use of the term “hypergeometric.” In a geometric
series the ratio of successive terms is a constant times z, whereas in a hypergeometric
function the ratio is a rational function of n, multiplied by z.
Exercise 5.13. Apply the ratio test to show the hypergeometric function 2 F1 .a; bI cI z/
converges on compact subsets of the unit disk. Try it out!
Functions which are useful or widely applicable typically earn the status of “special
function.” A number of well known special functions can be written as hypergeometric
series. For example,
1Cz
log D 2z 2 F1 1=2; 1I 3=2I z 2
1 z
.1 z/ a D 2 F1 .a; bI bI z/

arcsin z D z 2 F1 1=2; 1=2I 1I z 2 :
The functions sin.z/ and cos.z/ themselves can be obtained as limiting cases of 2 F1 series.
Now we consider an example that involves a 2 F1 hypergeometric series, z 2 F1 . 12 ; 14 I 45 I z 4 /.
We will see that it is a Schwarz-Christoffel transformation that maps the unit disk onto a
square.
Exercise 5.14. Use Definition 5.10 to find the first few terms in the series of
z 2 F1 . 21 ; 14 I 54 I z 4 /. The following table gives the first values of the Pochhammer symbols
that are needed. If you graph your result using ComplexTool, you should get a picture
similar to Figure 5.3.
n .1=2/n .1=4/n .5=4/n Coefficient of z 4nC1

0 1 1 1 1
1 1/2 1/4 5/4 1/10
2 3/4 5/16 45/16 1/24
3 15/8 45/64 585/64 5/208
4 105/16 585/256 9945/256 35/2176
5 945/32 9945/1024 208845/1024 3/256
Try it out!
The rate of convergence in this example is fast—even with just a few non-zero terms
of the series, we obtain a map whose image is approximately a square. We now see how to
derive the formula for a Schwarz-Christoffel map onto the square.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 279 — #297

i i
F IGURE 5.3. ComplexTool image of an approximation of the conformal map (using the first five
terms).
Definition 5.15 (Euler representation). The hypergeometric function 2 F1 .a; bI cI z/ can

be written in integral form as
Z 1 b 1
.c/ t .1 t/c b 1
2 F1 .a; bI cI z/ D dt;
.b/.c b/ 0 .1 tz/a
which is known as the Euler representation of the 2 F1 function.
The symbol stands for the gamma function, defined for z in the right half plane by
Z 1
.z/ D t z 1 e t dt:
0
It can be extended analytically to the whole plane except for the negative integers 1; 2;
3; : : : :
Exercise 5.16. For integer values of n, the gamma function is related to the factorial by
.n/ D .n 1/Š. Prove this by directly evaluating .1/ D 1 and then showing that
.z C 1/ D z.z/: Hint: use integration by parts. Try it out!
Exercise 5.17. In this exercise, you will show that Definitions 5.10 and 5.15 are equivalent.
a
(a) Expand .1 tz/ using the binomial theorem to obtain a power series
X1
a .a/n n n
.1 tz/ D t z :
nD0
nŠ
(b) Use the formula (see [13], Theorem 7, p. 19 for a proof)

Z 1
.p/ .q/ D .p C q/ tp 1
.1 t/q 1
dt
0
(where p and q have positive real parts) to show
Z 1
.b C n/ .c b/
t nCb 1 .1 t/c b 1 dt D :
0 .c C n/
The integral on the right is also an important special function known as the beta func-
tion in the variables p and q.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 280 — #298

i i
(c) Show
.b/n .c/ .b C n/ .c b/
D :
.c/n .b/ .c b/ .c C n/
(d) Put the previous facts together to obtain the required formula. Substitute the power
series for the denominator, and interchange summation and integral to show that
Z 1
tb 1
.1 t/c b 1
.b/ .c b/
dt D 2 F1 .a; bI cI z/ :
0 .1 tz/a .c/
Try it out!
It takes a little work to establish the relationship of the beta function with the gamma
function used in part (b). For a detailed exposition and an introduction to special functions
in general, see [13].
Example 5.18. For a square (a regular 4-gon), the Schwarz-Christoffel map from the unit
disk is given by z 2 F1 . 12 ; 14 I 54 I z 4 /: The numbers may seem arbitrary, but when we apply
the Euler integral representation we will see that they are the numbers we need (after a
transformation) to evaluate the integral. With n D 4 in the integral representation (5.4) we
obtain Z z
1
p dw:
0 1 w4
Let a D 1=4; b D 1=2; and c D 5=4: We use the Euler integral representation for
2 F1 .a; bI cI z/ to get
Z 1 3=4
4
.5=4/ t .1 t/0
2 F1 1=2; 1=4I 5=4I z D dt
.1=4/.1/ 0 .1 tz 4 /1=2
Z 1
1 1
D 1=4 3=4
p dt:
0 t 1 tz 4
Change variables by letting w 4 D tz 4 (so t D .w 4 =z 4 /) and 4w 3dw D z 4 dt: Then
Z z 3
4
z 1 4w 3dw
2 F1 1=2; 1=4I 5=4I z D 1=4 3
p
0 w 1 w 4 z4
Z z
1 1
D p dw:
z 0 1 w4
Thus Z z
1
f .z/ D p dw D z 2 F1 1=2; 1=4I 5=4I z 4 :
0 1 w 4
Contrast the map in Figure 5.4, where the domain is the (bounded) unit disk, with
the map onto a rectangle in Figure 5.2. An advantage of the disk map is that we see the
entire mapping domain and that the entire target square is filled. Rotational and reflectional
symmetry is obtained by using the symmetrically placed nth roots of unity on the unit circle
as prevertices.
Exercise 5.19. Show that the conformal map from the disk onto the regular n-gon is, up to
rotations, translations and scalings, given by z 2 F1 .2=n; 1=nI .n C 1/ =nI z n /. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 281 — #299

i i
1.0
0.5
0.5
0.0 0.0
–0.5
–0.5
–1.0
–1.0 –0.5 0.0 0.5 1.0 –0.5 0.0 0.5
F IGURE 5.4. The unit disk (left) and the target square (right) to which it maps under the function
of Exercise 5.18.
For readers who have worked through Chapters 2 or 4, we describe another situation
where the technique of integration illustrated in Example 5.18 can be useful. In Chapter 4,
Section 4.5 discusses the shear construction, and in Chapter 2, Section 2.6 discusses the
shear construction and its relationship with minimal surfaces.
We now examine the conformal map onto a non-convex polygon.
Small Project 5.20. Construct a non-convex hexagon P with ˇ1 D ˇ4 D 1=3 and ˇ2 D
ˇ3 D ˇ5 D ˇ6 D 2=3. (Draw it!) Find the representation for the Schwarz-Christoffel
transformation 5.3 that maps the unit disk D onto such a hexagon P , with the prevertices
i being the sixth roots of unity. Specifically, let 1 D 1 and 4 D 1, with the other sixth
roots of unity going in order counterclockwise around the circle. Verify that the analytic
function f .z/ W D ! P is given by

2 1 7 6 z3 2 1 3 6
f .z/ D z 2 F1 ; I Iz 2 F1 ; I Iz :
3 6 6 3 3 2 2
In the language of Chapter 4, let h.z/ g.z/ be f .z/ and let the dilatation be !.z/ D
z 2 : Find the harmonic function h.z/ C g.z/. Verify that by using

2 1 7 6
h.z/ D z 2 F1 ; I Iz
3 6 6
and

z3 2 1 3 6
g.z/ D 2 F1 ; I Iz ;
3 3 2 2
h0
we obtain !.z/ D g0 D z 2 : Use a computer algebra system to create a picture of the image
of D under h.z/ C g.z/.
If you have studied Chapter 2, you can find the minimal surface that lifts from this
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 282 — #300

i i
harmonic function. You should find that it is defined by

2 1 7 6 z3 2 1 3 6
x1 D Re z 2 F1 ; I Iz C F
2 1 ; I I z
3 6 6 3 3 2 2
3
2 1 7 6 z 2 1 3 6
x2 D Im z 2 F1 ; I Iz 2 F1 ; I Iz
3 6 6 3 3 2 2

2 1 3 6
x3 D Im z 2 2 F1 ; I Iz :
3 3 2
We now examine the conformal map onto a symmetric non-convex polygon in the
shape of a star. In the next section we intend to find harmonic maps onto the same figure,
and will find that while harmonic mappings onto polygons with convex corners are rela-
tively easy to construct, there is little or no theory when non-convex corners are involved.
Example 5.21. Suppose we want to map onto a (non-convex) m-pointed star, so there are
2m vertices. The interior angles alternate between ˛1 and ˛2 where ˛1 < 1 < ˛2
(see Figure 5.5). Assuming we have a non-convex star, exterior angles alternate between a
positive value ˇ1 and a negative value ˇ2 :
b1 p
b2 p
a1 p
a2 p
F IGURE 5.5. Interior and exterior angles of a symmetric star.
Since the exterior angles of a simple closed polygon must add to 2 , ˇ1 C ˇ2 must
satisfy m .ˇ1 C ˇ2 / D 2 so ˇ1 C ˇ2 D 2=m: For odd indices i we have ˇi D ˇ1 > 0
and for even i , ˇi D ˇ2 < 0. Using the Schwarz-Christoffel formula for the unit disk we
obtain the mapping function
Y
ˇ2
Z .w i /
z
i even
f .z/ D Y dw:
0 .w i /ˇ1
i odd
Letting i be nth roots of unity,

n
Y n
Y
.z i / D z m 1 and .z i / D z m C 1;
i even i odd
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 283 — #301

i i
so we can rewrite the mapping function as

Z z
.w m 1/ ˇ2
f .z/ D m ˇ
dw:
0 .w C 1/ 1
Disregarding constants chosen to expand or rotate the figure as necessary, we can set
Z z
.1 w m/ ˇ2
f .z/ D m ˇ
dw;
0 .1 C w / 1
a conformal mapping from the disk onto the star shape in Figure 5.5.
Example 5.22. This appears as an exercise in [12], Chapter V. Prove that the integral that
maps the unit disk onto a 5-pointed star with interior angles alternating between =5 and
7=5 is given by
Z z
.1 w 5 /2=5
f .z/ D 5 4=5
dw:
0 .1 C w /
The corresponding exterior angles are 4=5 and 2=5; so ˇ1 D 4=5, ˇ2 D 2=5, and
m D 5.
To compute this integral we use the Appell F1 function of two variables defined below.
Definition 5.23. The Appell F1 function is defined by
1 X
X 1
.a/nCm .b1 /m .b2 /n m n
F1 .aI b1; b2 I cI x; y/ D x y ;
mŠnŠ .c/nCm
nD0 mD0
In Mathematica we can evaluate the Appell F1 function using the command

AppelF1(a,b1,b2 ,c,x,y). Furthermore, there is an integral representation formula,
analogous formula to that of the hypergeometric function, given by
F1 .aI b1 ; b2I cI x; y/
Z 1
.c/
D ua 1
.1 u/c a 1
.1 ux/ b1
.1 uy/ b2
du:
.a/ .c a/ 0
We omit the proof, but a derivation of the integral representation can be found in [2, Chapter
9]. Substituting we find that

F1 1=5I 4=5; 2=5I 6=5I z 5; z 5
Z 1
.6=5/ 4=5 2=5
D u 4=5 .1 u/0 1 uz 5 1 C uz 5 du
.1=5/.1/ 0
Z 1 2=5
1 uz 5
D 1=5 4=5
du:
0 u4=5 .1 C uz 5 /
To evaluate the integral in the Schwarz-Christoffel formula we change variables, letting

w 5 D uz 5 (so 5w 4dw D z 5 du). Then
Z z 4 2=5
5 5
z 1 w5 5w 4dw
F1 1=5I 4=5; 2=5I 6=5I z ; z D 1=5 4
0 w .1 C w 5 /
4=5 z5
Z 2=5
1 z 1 w5
D dw:
z 0 .1 C w 5/4=5
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 284 — #302

i i
Thus
Z 2=5
z 1 w5
f .z/ D dw D z F1 1=5I 4=5; 2=5I 6=5I z 5; z 5 I
0 .1 C w 5/4=5
the mapping is illustrated in Figure 5.6.
–1
–2
–1 0 1 2
F IGURE 5.6. Image of the conformal map of the unit disk onto the 5-pointed star.
Exercise 5.24. Show that the conformal map from the disk onto the m-pointed star with
exterior angle ˇ1 > 0 and ˇ2 D 2=m ˇ1 (up to rotations, translations and scalings) is
given by z F1 .1=nI ˇ1 ; ˇ2; .n C 1/ =nI z n ; z n / where F1 is the Appell F1 function. Try
it out!
5.3 The Poisson Integral Formula

While the Schwarz-Christoffel formula gives analytic and thus angle-preserving (confor-
mal) functions from D to a polygon, it often starts with an integral that requires advanced
mathematics to evaluate. If our goal is not an analytic function, we could work with the
Poisson integral formula, which gives a harmonic function from the unit disk to the target
domain. The definition of a harmonic function is
Definition 5.25. A real-valued function u.x; y/ is harmonic provided that
uxx C uyy D 0:
A complex-valued function f .z/ D f .x C iy/ D u.x; y/ C iv.x; y/ is harmonic if u and

v are harmonic.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 285 — #303

i i
5.3. The Poisson Integral Formula 285
The definition of a complex-valued harmonic function does not require that u and v
be harmonic conjugates, so while all analytic functions are harmonic, a complex-valued
harmonic function is not necessarily analytic. The functions we work with for the rest of
the chapter will not be analytic, and thus not conformal.
You may be familiar with the Poisson integral formula as a way of constructing a real-
valued harmonic function that satisfies certain boundary conditions. For example, if the
boundary conditions give the temperature of the boundary of a perfectly insulated plate,
then the harmonic function gives the steady-state temperature of its interior. Another ap-
plication is finding electrostatic potential given boundary conditions. We give a brief sum-
mary. For more detailed discussion, see [14] or [19].
Theorem 5.26 (Poisson Integral Formula). Let the complex-valued function fO.e i / be
piecewise continuous and bounded for in Œ0; 2 : Then the function f .z/ defined by
Z 2
1 1 jzj2 O i t
f .z/ D f .e /dt (5.5)
2 0 je i t zj2
is the unique harmonic function in the unit disk that satisfies the boundary condition
lim f .r e i / D fO.e i /
r !1
for all where fO is continuous.

We present the proof when the boundary function fO.e i / is the real part of a function
that is analytic on a disk with radius larger than 1. This proof can be found in any standard
complex analysis textbook, for example, [10] or [14]. The full result appears in [1, Chapter
6] and [10, Chapter 8].
Proof. Cauchy’s integral formula tells us that if we have a function f .z/ that is analytic
inside and on the circle jzj D R, then, for jzj < R,
Z
1 f ./
f .z/ D d : (5.6)
2 i jjDR z
We use the Greek letter zeta () as the variable of integration in the integral. In the
discussion that follows, we will be thinking about evaluating the function f .z/ at some
fixed value of z, so the variable under consideration is now : We observe, for reasons that
will soon be obvious, that for fixed z with jzj < 1, the function
f ./ z
1 z
is analytic in the variable on and inside jj < 1, since the denominator is non-zero.
(Exercise: Why is the denominator non-zero?) Thus, by the Cauchy integral theorem, we
know that
Z
1 f ./ z
D 0: (5.7)
2 i jjD1 1 z
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 286 — #304

i i
Combining (5.6) and (5.7), we see that

Z
1 f ./
f .z/ D d C 0
2 i jjD1 z
Z
1 f ./ f ./ z
D C d
2 i jjD1 z 1 z
Z
1 1 z C z. z/
D f ./d
2 i jjD1 . z/.1 z/
Z
1 1 jzj2
D f ./d :
2 i jjD1 . z/.1 z/
Parameterize the circle jj D 1 by .t/ D e i t , giving d D i e i t dt so
Z 2
1 1 jzj2
f .z/ D i t
f .e i t /i e i t dt
2 i 0 .e z/.1 e i t z/
Z
1 jzj2 2 1
D it i t .e i t
f .e i t /e i t dt
2 0 .e z/e z/
Z
1 jzj2 2 f .e i t /
D dt:
2 0 .e i t z/.e i t z/
Taking the real part of both sides gives us a harmonic function since the real part of an
analytic function is harmonic, and also yields (5.5).
1 jzj2
Exercise 5.27. Verify that the Poisson kernel, , can be written as
je i t zj2
it
e Cz 1 C ze i t
Re D Re :
ei t z 1 ze i t
Try it out!
The integral in (5.5) is, in general, difficult to integrate. However, if there is an arc on
which fO.e i t / is constant, the integration is easy.
Exercise 5.28. Verify that
Z b it !
1 e Cz b a K 1 ze i b
K Re dt D K C arg
2 a ei t z 2 1 ze i a
" ! #
K ei b z b a
D arg i a : (5.8)
e z 2
Hint: Interchange integration and taking the real part of a function. The identity 1Cw
1 w D
1 C 12ww will be helpful. Try it out!
The last formulation of (5.8) can be visualized geometrically. Consider a situation with
e i a and e i b on the unit circle and z inside the unit disk. Then the vector from z to e i a
is e i a z and ib
! the vector from z to e is e
ib
z, so that the angle between them is
ib
e z
arg i a , as is shown if Figure 5.7.
e z
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 287 — #305

i i
e ib
q
e ia
z
!
ei b z
F IGURE 5.7. Geometric interpretation of D arg i a .
e z
Example 5.29. Assume that the unit disk is a thin insulated plate with a temperature along
the boundary of 50 degrees for the top semicircle and 20 degrees along the bottom semi-
circle. From physics, we know that the function that describes the temperature within the
unit disk is harmonic. Find it.
Solution: Apply (5.8) for a1 D 0; b1 D ; K1 D 50 and then add it to the result where
a2 D ; b2 D 2; K2 D 20; to get

1 1Cz
f .z/ D 70 C 60 arg :
2 1 z
(70 is 50 C 20 and 60 D 2.50 20/.) As z ranges across the unit disk, 1Cz 1 z
covers the
right half-plane (you can check this by graphing 1Cz
1 z
using ComplexTool), so its argument
is between =2 and =2. This gives values for f .z/ between 20 and 50, which makes
sense. The solution gives the average temperature plus or minus half the difference between
the maximum and minimum temperatures.
Exercise 5.30. Referring to

1 1Cz
f .z/ D 70 C 60 arg
2 1 z
in Example 5.29, find f .0/, f .i=2/, and f . i=2/. Do your answers make sense?
Using the result of Exercise 5.28, we can see that computing the Poisson integral for-
mula for a piecewise constant boundary is simple. Many of its applications come from
having the boundary correspondence remain constant on arcs of the unit circle.
Most introductory analysis books give the Poisson integral formula for real-valued
fO.e i /. It can also be applied to create a harmonic function for complex-valued fO.e i /,
but the univalence of the harmonic function is not at all apparent. Let’s see what happens
if we use the Poisson integral formula with complex boundary values.
Example 5.31. The simplest example is obtained by letting the first third of the unit circle
(that is, the arc from 0 to e i 2=3) map to 1, the next third to e i 2=3, and the last third to
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 288 — #306

i i
e i 4=3. Let’s work through the details of this integration, working from (5.8). We compute
!
1 2 1 ze i 2=3
f .z/ D . 0/ C 2 arg
2 3 1 ze 0
!
i 2=3 4 2 i 2=3 1 ze i 4=3
Ce . / C 2e arg
3 3 1 ze i 2=3

4 1 ze 2 i
Ce i 4=3.2 / C 2e i 4=3 arg
3 1 ze i 4=3
2
D 1 C e i 2=3 C e i 4=3
3.2 /
! !
1 1 ze i 2=3 i 2=3 1 ze i 4=3
C arg Ce arg
1 ze 0 1 ze i 2=3

1 ze 2 i
Ce i 4=3 arg
1 ze i 4=3
! !
1 1 ze i 2=3 1 ze i 4=3
D 0C arg C e i 2=3 arg
1 z 1 ze i 2=3

1 z
Ce i 4=3 arg :
1 ze i 4=3
Figure 5.8 shows the image of the unit disk as graphed in ComplexTool. It appears to be
one-to-one on the interior of the unit disk. It is not one-to-one on the boundary! (Entering
this formula into ComplexTool is unwieldy, so the function is one of the Pre-defined
functions, the one called Harmonic Triangle. We will soon use the PolyTool ap-
plet to graph similar functions.)
F IGURE 5.8. ComplexTool image of the harmonic function mapping to the triangle.
Exercise 5.32. Find a formula that maps the unit disk harmonically to the interior of a
convex regular n-gon. Try it out!
Small Project 5.33. Refer to Chapter 4 and its discussion of the shear construction. Find
the pre-shears of the polygonal mappings in Exercise 5.32. What analytic function do you
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 289 — #307

i i
shear to get the polygonal function? A first step is to determine the dilatation of the func-
tion. See [4] for more details.
Exercise 5.34. For a non-convex example, show that the function that maps quarters of the
unit circle to the four vertices f1; i; 1; 2i g is

3i 1 1 C iz 1Cz 1 iz i 1 z
C arg C i arg arg C arg :
8 1 z 1 C iz 1Cz 2 1 iz
When we graph this in ComplexTool, it appears not to be one-to-one. Furthermore, f .0/ D
3i
8
, so that the image of D is not the interior of the polygon. This is another of the Pre-
defined functions in ComplexTool. Try it out!
F IGURE 5.9. The PolyTool Applet.
At this point, you should start using the PolyTool applet. In it, you can specify which
arcs of the unit circle will map to which points in the range, and the applet will compute
and graph the harmonic function defined by extending the boundary correspondence to a
function on D. When you open this applet, you see a circle on the left and a blank screen
on the right. You can create a harmonic function that maps portions of the boundary of the
circle to vertices of a polygon in one of two ways. First, you can click on the unit circle in
the left panel to denote an arc endpoint, continue choosing arc endpoints there, and then
choose the target vertices by clicking in the right graph. (As you click, text boxes between
the panels fill with information about where you clicked.) Once you have the boundary
correspondence you want, click the Graph button. Second, you can click the Add button
to get text boxes for input. For example, to create the function in Exercise 5.34, click Add,
then fill in the first row of boxes for Arc 1: with 0 maps to 1+0i. When you want
another set of arcs, click Add again. The Arc boxes denote the starting point of the arc
(i.e., for the arc from 0 to =3, use 0 in the Arc box). Continue filling, and when you
are ready to compute the Poisson integral to get the harmonic function, click the Graph
button. Once you have a function graphed, you can drag around either the arc endpoints
(in the domain on the left) or the target points (in the range on the right) and watch the
function change.
Exploration 5.35. Are there ways of rearranging the boundary conditions to make the
function in Exercise 5.34 univalent? What if the bottom half of the unit circle gets mapped
to i=2 and the top half of the unit circle is divided into thirds for the other three vertices?
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 290 — #308

i i
This new function isn’t univalent, but is in some sense is closer to univalent than the map-
ping in Exercise 5.34. Can a modification be made so that it is univalent? Try it out!
Exercise 5.36. This is an extension of Exploration 5.35. Sheil-Small [16] proved (by tech-
niques other than those discussed so far) that the harmonic extension of the boundary
correspondence below maps the unit disk univalently onto the desired shape:
arc from to maps to
i i i
i 3=5 C 4i=5 1
3=5 C 4i=5 3=5 4i=5 i=2
3=5 4i=5 i 1
For this function, first convince yourself that it appears to be univalent, and then find
f .z/. Try it out!
We will be working with harmonic functions that are extensions of a piecewise con-
stant boundary correspondence as in Example 5.36. For future discussions, we make the
definition:
Definition 5.37. Let fe i tk g be a partition of @D, where t0 < t1 < : : : < tn D t0 C 2 . Let
fO.e i t / D vk for tk 1 < t < tk . We call the harmonic extension of this step function (as
defined by the Poisson integral formula) f .z/.
Example 5.38. To understand the definition better, we demonstrate how the notation in
Definition 5.37 is used for the function in Exercise 5.36. Since the arc from i to i can be
thought of as the arc along the unit circle from e i =2 to e i =2, we say that t0 D =2 and
t1 D =2. These points map to the vertex at i , so v1 D i . Finding the next arc set is more
difficult, because we need to find the angle t2 that goes with the point z D 53 C 54 i D e i t2 .

We can get only a numerical estimate, found by Carctan 4=5 3=5
D Carctan . 4=3/
2:2143: (We add because values of the arctangent function are in the first or fourth
quadrant, while the angle is in the second quadrant.) Thus we have t2 2:2143 and
v2 D 1: Continuing, we have t3 4:0689 and v3 D i=2. We finish with t4 D 3=2 and
v4 D 1: As in the definition, t4 D t0 C 2: Then from Definition 5.37 the function f .z/ is
the harmonic function that appears to be univalent when graphed in PolyTool.
Exercise 5.39. Combine the result of Exercise 5.28 (on p. 286) with Definition 5.37 to
show that the function f .z/ in Definition 5.37 can be written

t1 t0 v1 1 ze i t1
f .z/ D v1 C arg
2 1 ze i t0

t2 t1 v1 1 ze i t2
Cv2 C arg
2 1 ze i t1

tn tn 1 vn 1 ze i tn
C C vn C arg :
2 1 ze i tn 1
Try it out!
Since the Poisson integral formula gives rise to a harmonic function, we must learn
some basics of the theory of harmonic functions before proceeding.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 291 — #309

i i
5.4. Harmonic Function Theory 291
5.4 Harmonic Function Theory

Chapter 4 gives a detailed explanation of harmonic functions, as does [5]. Much of that
material will be helpful for our investigations, so we repeat it here.
5.4.1 The Basics

A harmonic function f can be written as f D h C g, where h and g are analytic functions.
0 .z/
The analytic dilatation !.z/ D gh0 .z/ is a measure of how much the harmonic function does
not preserve angles. A dilatation of !.z/ 0 means that the function is analytic, and so
must be conformal. A dilatation with modulus near 1 indicates that the function distorts
angles greatly. (For more intuition about the dilatation, see Section 4.6 of Chapter 4.) A
result of Lewy states that a harmonic function has nonzero Jacobian (denoted Jf .z/ D
jh0 j2 jg0 j2 ) if it is locally univalent. This result is in line with our understanding of the
relationship between local univalence and a nonvanishing derivative for analytic functions.
Theorem 5.40 (Lewy’s Theorem). For a harmonic function f defined on a domain , if
f locally univalent in , then Jf .z/ ¤ 0 for all z 2 .
This is equivalent to Lewy’s theorem in Chapter 4.
A consequence of Lewy’s theorem is that if a function is locally univalent in , then
its analytic dilatation satisfies either j!.z/j < 1 for all z 2 , or j!.z/j > 1 for all z 2 .
In our work, we will study only functions that are locally univalent in a domain and
satisfy j!.z/j < 1 for all z 2 . They are called sense-preserving because they preserve
the orientation of curves in .
Exercise 5.41. Verify that the condition that Jf .z/ ¤ 0 is equivalent to j!.z/j ¤ 1 as long
as h0 .z/ ¤ 0. Conclude that a function that is locally univalent and sense-preserving has
Jf .z/ > 0 and j!.z/j < 1. Try it out!
Of particular interest is determining how to split the argument function, which is har-
monic and sense-preserving, into h and g.
Exercise 5.42. Show that f .z/ D K arg.z/ has canonical decomposition
1 1
h.z/ D K log.z/ and g.z/ D K log.z/.
2i 2i
Try it out!
Another consequence of the canonical decomposition of a harmonic function is that
we can write the analytic functions h and g defined in a domain in terms of their power
series expansions, centered at z0 2 , as
1
X 1
X
f .z/ D a0 C ak .z z0 /k C b0 C bk .z z0 /k : (5.9)
kDn kDm
If f is sense-preserving, then we have that either m > n or m D n with jbn j < jan j. In
either case, when f is represented by (5.9), we say that f has a zero of order n at z0 .
Exercise 5.43. In this exercise, we prove that the zeros of a sense-preserving harmonic
function are isolated.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 292 — #310

i i
(a) Assume that f .z/ is a sense-preserving locally univalent function with series expan-
sion as given in (5.9). Show that if f .z0 / D 0, there exists a positive ı and a function
such that for 0 < jz z0 j < ı we can write
f .z/ D h.z/ C g.z/ D an .z z0 /n .1 C .z// (5.10)
where
anC1 anC2 1
.z/ D .z z0 / C .z z0 /2 C C bm .z z0 /m C ;
an an an .z z0 /n
where the ellipses denote a continuation in m.
(b) Show that (a) implies that j .z/j < 1 for z sufficiently close to z0 , since m n and
jbn =an j < 1 if m D n:
(c) Show that (b) implies that the zeros of a sense-preserving harmonic function are iso-
lated, since f .z/ ¤ 0 near z0 (except, of course, at z0 ).
Try it out!
5.4.2 The Argument Principle

The argument principle for analytic functions gives a nice way to describe the number
of zeros and poles inside a contour. It is in many introductory complex analysis courses.
However, we explore this topic to emphasize the geometric nature of the result before
studying the related result for harmonic functions. We first provide a formal definition of
the winding number of the image of a contour about the origin.
Definition 5.44. The winding number of the image under f .z/ of a simple closed contour
about the origin is the net change in argument of f .z/ as z traverses in the positive
1
(counterclockwise) direction, divided by 2: It can be denoted by 2 arg f .z/:
To explore the relationship between the winding number of the image of a contour
about the origin and the number of zeros and poles contained within the contour, do the
following exploration.
Exploration 5.45. Open ComplexTool. Change the Interior circles to 1 and Rays
to 0. This will graph only the boundary of the circle. As you graph the following functions,
examine how the image of the circle winds around, or encloses, the origin. Count how
many times the image of the circle winds around the origin, counting the counterclock-
wise direction as positive and the clockwise direction as negative. If the image of the circle
winds around the origin once, you know that there must be a zero of f .z/ inside that circle.
(Think about this last sentence and make sure you understand it.)
(a) Graph f .z/ D z 2 , using a circle of radius 1. Use the Sketch button and trace around
the circle in the domain to get a feeling for how many times its image winds around
the origin. You should already know the answer. (Any other radius will work too. Why
is that?)
(b) Graph f .z/ D z.z 0:3/. Use circles of radius 0:2; 0:3, and 0:5. (You may have to
zoom in on the image of the circle of radius 0.2 to understand what it is doing.) You
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 293 — #311

i i
may check the Vary radius checkbox and use the slider to change the radius of the
circle.
(c) Graph f .z/ D z 4 6z C 3. Use circles of radius 0:9; 1:5; 1:7, and 2.
z 4 6zC3
(d) Graph f .z/ D z 1
, using circles of radius 0:9 and 1:5.
z4 6zC3
(e) Graph f .z/ D .z 1/2
, using circles of radius 0:9; 1:5; and 2.
(f) Go back through the previous three items, now changing the function while keeping
the radius fixed.
(g) For the previous functions, find the locations of the zeros and poles, paying attention
to how far they are from the origin.
(h) Make up your own function and do some more experiments.
Based on the exploration, what is your connection between the winding number of the
image of a circle about the origin and the number of zeros and poles inside the circle?
Try it out!
Theorem 5.46 (Argument Principle for Analytic Functions). Let C be a simple closed
contour lying entirely within a domain D. Suppose f is analytic in D except at a finite
number of poles inside C and that f .z/ ¤ 0 on C . Then
I
1 f 0 .z/
dz D N0 Np ;
2 i C f .z/
where N0 is the number of zeros of f inside C and Np is the number of poles of f inside
C , counted according to their order or multiplicity.
z2
z1 0
F IGURE 5.10. A path around a branch cut.
Before proving Theorem

I 5.46, we explore the connection betweenZ the winding number
z2
1 f 0 .z/ f 0 .z/
and the integral dz. We start with a related integral, dz, where
2 i C f .z/ z1 f .z/
z1 and z2 are points close to each other, but lying on opposite sides of a branch cut of
log f .z/, and we take a counterclockwise path along C from z1 to z2 . (See Figure 5.10.)
We have:
Z z2 0
f .z/
dz D log f .z/jzz21
z1 f .z/
D ln jf .z2 /j ln jf .z1 /j C i.arg.f .z2 // arg.f .z1 ///:
When we take the limit as z1 ! z2 , we get that ln jf .z2 /j ln jf .z1 /j ! 0 and
i.arg.f .z2 // arg.f .z1 /// ! 2 i .the winding number/. Think about this last statement
and make sure you understand it.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 294 — #312

i i
Proof. The proof of the argument principle relies on the Cauchy integral formula and de-
formation of contours. We begin by deforming the contour C to a series of smaller contours
0 .z/
around the isolated zeros and poles of f . If there are no zeros or poles, then ff .z/ is ana-
lytic, so the integral is zero, as desired. We analyze the zeros and poles individually, and
add the results to get the conclusion. More formally, when f has zeros or poles inside C
they must be isolated, and because f is analytic on C , there are only a finite number of
distinct zeros or poles inside C . Denote the zeros and poles by zj , for j D 1; 2; : : : ; n.
Let j be a circle of radius ı > 0 centered at zj , where ı is chosen small enough that the
circles j lie in D and do not meet each other. Join each circle j to C by a Jordan arc
j in D. Consider the closed path formed by moving around C in the positive direction
while making a detour along each j to j , running once around this circle in the negative
direction, then returning along j to C . See Figure 5.11.
zj
gj
lj C
F IGURE 5.11. The contour .
This curve contains no zeros or poles of f , so

I
1 f 0 .z/
arg f .z/ D dz D 0
2 i f .z/
by the argument above. When considering the total change in argument of f .z/ along ,
the contributions of the arcs j along cancel out, so that
n
X
C arg f .z/ D j arg f .z/;
j D1
where the circles j are now traversed in the positive direction. Thus now we may consider
each j and sum the results.
Suppose that f has a zero of order n at z D zj . Then f .z/ D .z zj /n fn .z/, where
fn .z/ is an analytic function satisfying fn .zj / ¤ 0, as shown in any introductory complex
analysis book. Then
f 0 .z/ D n.z zj /n 1
fn .z/ C .z zj /n fn0 .z/
and
f 0 .z/ n.z zj /n 1 fn .z/ C .z zj /n fn0 .z/
D
f .z/ .z zj /n fn .z/
n f 0 .z/
D C n :
z zj fn .z/
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 295 — #313

i i
f 0 .z/
When we integrate this along j , we get n.2 i / C 0, because fnn.z/ is analytic inside the
contour.
Suppose that f has a pole of order m at z D zk . This means that f can be written as
f .z/ D .z zk / m fm .z/, where fm is analytic an nonzero at z D zk . Then, as previously,
we have
f 0 .z/ m.z m 1
fm .z/ C .z zk /
zk / m
fm0 .z/
D
f .z/ .z zk / m fm .z/
0
m f .z/
D C m :
z zk fm .z/
When we integrate this along k , we get m.2 i / C 0.

Summing over j D 1; 2; : : : n gives us the integral over and the result.
There are many versions of the argument principle for harmonic functions. We need
only the simple proof presented in this section, developed by Duren, Hengartner, and
Laugesen [6].
Theorem 5.47 (Argument Principle for Harmonic Functions). Let D be a Jordan domain
with boundary C . Suppose f is sense-preserving harmonic function on D, continuous in
D and f .z/ ¤ 0 on C . Then C arg f .z/ D 2N , where N is the number of zeros of
f .z/ in D, counted according to multiplicity.
Proof. Suppose that f has no zeros in D. This means that N D 0 and the origin is not
an element of f .D [ C /. A fact from topology says that in this case, C arg f .z/ D 0,
and the theorem is proved. We prove this fact. Let be a homeomorphism of the closed
unit square S onto D [ C with the restriction of to the boundary, O W @S ! C , also a
homeomorphism. See Figure 5.12.
S
f DÈC
f
F
F IGURE 5.12. The composition of f and .
The composition F D f ı is a continuous mapping of S onto the plane with no

zeros, and we want to prove that @S arg F .z/ D 0. Subdivide S into finitely many small
squares Sj so that on each Sj , the argument of F varies by at most =2. Then, since
F .Sj / cannot enclose the origin, @Sj arg F .z/ D 0. When we consider @S arg F .z/, it
P
is the sum j @Sj arg F .z/ because the contributions to the sum from the boundaries of
each Sj cancel out, except where the boundary of Sj agrees with the boundary of S . Thus
@S arg F .z/ D 0, as desired.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 296 — #314

i i
Now suppose f has zeros in D. Because they are isolated (as proven in Exercise 5.43),
and because f is not zero on C , there is only a finite number of distinct zeros in D. We
proceed as in the proof of the analytic argument principle, and denote the zeros by zj for
j D 1; 2; : : : ; n. Let j be a circle of radius ı > 0 centered at zj , where ı is chosen so
small that the circles j lie in D and do not meet each other. Join each circle j to C by a
Jordan arc j in D. Consider the closed path formed by moving around C in the positive
direction while making a detour along each j to j , running once around this circle in the
negative direction, then returning along j to C . (See Figure 5.11.) This curve contains
no zeros of f , so arg F .z/ D 0 by the first case in this proof. When considering the
total change in argument along of f .z/, the contributions of the arcs j along cancel
out, so that
Xn
C arg f .z/ D j arg f .z/;
j D1
where each of the circles j is now traversed in the positive direction. Thus now we may
consider each j and sum the results.
Suppose that f has a zero of order n at a point z0 . Then, as in Exercise 5.43 on page
291, f can be locally written as
f .z/ D an .z z0 /n .1 C .z//
where an ¤ 0 and j .z/j < 1 on a sufficiently small circle defined by jz z0 j D ı. This

shows that
arg f .z/ D n arg.z z0 / C arg.1 C .z// D 2 n: (5.11)
Therefore, if f has zeros of order nj at the points zj , the conclusion is that

n
X n
X
C arg f .z/ D j arg f .z/ D 2 nj D 2N;
j D1 j D1
and the theorem is proved.
Exercise 5.48. Justify why arg.1 C .z// D 0 in (5.11). Try it out!

In the next section, we look at the polygonal maps defined in Section 5.3. We will use
the argument principle for harmonic functions in the proof of the Rado-Kneser-Choquet
theorem.
5.5 Rado-Kneser-Choquet Theorem

As you examine the image of the unit disk using the examples in Section 5.3, you may
notice that some of the functions seem to be one-to-one on the interior of the domain, while
others do not seem to be univalent. Look again at the examples, and compare functions that
map to convex domains with functions that map to non-convex domains.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 297 — #315

i i
5.5. Rado-Kneser-Choquet Theorem 297
Exploration 5.49. Make a conjecture about when functions are one-to-one, using the ex-
ercises from Section 5.3. Do this before reading the Rado-Kneser-Choquet theorem! Try it
out!
In general, we completely understand the behavior of harmonic extensions (as defined
in Definition 5.37) that map to convex regions:
Theorem 5.50 (Rado-Kneser-Choquet Theorem). Let be a subset of C that is a bounded
convex domain whose boundary is a Jordan curve : Let fO map @D continuously onto
and suppose that fO.e i t / runs once around monotonically as e i t runs around @D. Then
the harmonic extension given in the Poisson integral formula is univalent in D and defines
a harmonic mapping of D onto :
For the proof of this important theorem, we use the following lemma.
Lemma 5.51. Let be a real-valued function harmonic in D and continuous in D. Sup-
pose has the property that, after a rotation of coordinates, .e i t / .e i t / 0 on
it it
the interval Œ0; , with strict inequality .e / .e / > 0 on a subinterval Œa; b with
0 a < b . Then has no critical points in D.
The condition on may seem mysterious at first, so we will discuss it. One kind of
function for which the property holds is that is at most bivalent on @D. What does “at
most bivalent” mean? Univalent means that a function is one-to-one. Bivalent means that
a function is two-to-one, or that there may be z1 ¤ z2 such that f .z1 / D f .z2 /, but
that if f .z1 / D f .z2 / D f .z3 /, then at least two of z1 ; z2 ; z3 are equal. Another kind of
function described in Lemma 5.51 is one that is continuous on @D where .e i t / rises
from a minimum at e i ˛ to a maximum at e i ˛ , then decreases to its minimum at e i ˛ as
e i t runs around the unit circle, without having any other local extrema but allowing arcs of
constancy. See Figure 5.13.
y(eit)
t
–p –a a p
F IGURE 5.13. Boundary condition of satisfying the hypotheses of Lemma 5.51.
@
Proof of Lemma 5.51. To show that has no critical points in D, we show that @z
¤ 0 in
D. This is equivalent to saying that

1 @ @
i ¤ 0:
2 @x @y
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 298 — #316

i i
We will simplify the proof by proving that z .0/ ¤ 0, and claim that will be sufficient. If
z0 is a point in D, consider the function '.z/ D 1z0 z0zz that is a conformal self-map of D
with '.0/ D z0 , and let F ./ D .'.//. Observe that F is harmonic in D, continuous
in D, and satisfies the same condition about F .e i t / F .e i t / as does. Applying the
chain rule to F ./ gives F ./ D z .'.//' 0 ./; since ' is analytic and thus has ' D 0.
(In general, the chain rule is more complicated for harmonic functions. Here, since ' is
analytic, it has takes its familiar form.) Substituting 0 for gives F .0/ D z .z0 /' 0 .0/;
implying that if F .0/ D 0 then .z0 / D 0. Thus when we have proven that z .0/ ¤ 0,
we will be able to generalize to z .z0 / ¤ 0 for all z0 in D.
We use the Poisson integral formula to prove that z .0/ ¤ 0. Substituting in (or
O .e i t / D limr !1 .r e i t / on @D) gives
Z 2 Z 2
1 1 jzj2 1 1 zz
.z/ D O .e i t /dt D O .e i t /dt:
2 0 je i t zj2 2 0 .e i t z/.e i t z/
From
O .e i t / @ 1 z z
@ 1 zz
O .e i t / D
@z .e i t z/.e i t z/ e it z @z e i t z
!
O .e i t / e i t .e i t z/
D
e it z .e i t z/2

ei t
D O .e i t /
.e i t z/2
we get
Z 2
1 O .e i t /e it
z .0/ D dt:
2 0
From the hypotheses of the lemma, we know that there is a t 2 .0; / such that .e i t /
.e i t / > 0: Thus
Z 2
1 O .e i t /e i t dt
Im z .0/ D Im
2 0
Z 2
1 O .e i t / sin.t/dt:
D
2 0
Since O is periodic,
Z Z 0
1 O .e i t / sin.t/dt C O .e i t / sin.t/dt
Im z .0/ D
2 0
Z Z
1 O .e i t / sin.t/dt O .e i t / sin.t/dt
D
2 0 0
Z
1
D . O .e i t / O .e i t // sin.t/dt < 0:
2 0
The last inequality relies on the fact that sin.t/ is non-negative on Œ0; . We have shown
that Im z .0/ ¤ 0, proving the lemma.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 299 — #317

i i
5.5. Rado-Kneser-Choquet Theorem 299
Proof of Theorem 5.50. Without loss of generality, assume that fO.e i t / runs around coun-
terclockwise as t increases. (Otherwise, take conjugates.) We will show that if f is not
locally univalent in D, then Lemma 5.51 gives a contradiction.
Suppose that f D u C ivis not locally
univalent, so that the Jacobian of f vanishes
ux vx
at z0 in D. This means that has a determinant of 0 at z0 , so the system of
uy vy
equations
aux C bvx D 0
auy C bvy D 0
has a nonzero solution .a; b/. Thus the real-valued harmonic function D au C bv has
a critical point at z0 (since .a; b/ ¤ .0; 0/). However, the hypothesis of Theorem 5.50
implies that satisfies the hypothesis of Lemma 5.51. Thus we have a contradiction, so f
must be locally univalent.
Now that we see that f is locally univalent, we apply the argument principle to show
that f is univalent in D. Since f is sense-preserving on @D and locally univalent, f is
sense-preserving throughout D. If f is not univalent, there are points z1 and z2 in D such
that f .z1 / D f .z2 /. That would imply that f .z/ f .z1 / has two zeros in D, so that the
winding number of f .z/ f .z1 / about the origin is 2, which contradicts the hypotheses
about the boundary correspondence. This completes the proof.
Exercise 5.52. Give a detailed proof that the hypothesis of Theorem 5.50 implies that
satisfies the hypothesis of Lemma 5.51. Try it out!
The description of fO in Theorem 5.50 does not require that it be one-to-one on @D, but
permits arcs of constancy. Furthermore, the Rado-Kneser-Choquet theorem is true when fO
has jump discontinuities as long as the image of @D is not contained in a straight line. This
requires some additional justification, so we state it separately as a corollary.
Corollary 5.53. Let f .z/ be defined as in Definition 5.37 on p. 290. Suppose the vertices
v1 ; v2 ; : : : ; vn , when traversed in order, define a convex polygon whose interior is denoted
by . Then f .z/ is univalent in D and defines a harmonic mapping from D onto .
Here is some motivation for the proof of Corollary 5.53. Take a sequence of func-
tions fOm .e i t / that are continuous and converge to the boundary correspondence fO.e i t / of
Definition 5.37. (One can be described by having fOm .e i t / D vk for t-values in the interval
t tk
.tk 1 C tk 2m tk 1
; tk tk 2m
tk 1
/ while for t-values in the interval .tk tk 2m
tk 1
; tk C kC1
2m
/,
fOm .e i t / maps the interval linearly to the segment between vk and vkC1 .) Each fOm .e i t / sat-
isfies the conditions of the Rado-Kneser-Choquet theorem, and so extends to a univalent
harmonic function, fm .z/, in the unit disk. But the sequence fm converges uniformly on
compact subsets of D, so the entire sequence converges uniformly to f in D. Therefore,
f .z/ inherits the univalence from the sequence. The fact that the limit function is univalent
is not immediately apparent. Full details may be found in [8].
The theorem does not guarantee anything about univalence if the domain is not
convex. In fact, we may expect that univalence will not be achieved. For example, look at
Exercise 5.34 on page 289.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 300 — #318

i i
Exploration 5.54. Extend Exploration 5.35 on page 289. Instead of modifying the bound-
ary correspondence, start with the correspondence in Exercise 5.36. Then move the vertex
9i
at i=2 closer to i . A nice picture comes from having the vertex set be f1; i; 1; 10 g. There
we see the lack of univalence clearly. Try it out!
5.5.1 Boundary behavior

In this section, we explore what seems to be true in some of the above examples, that there
appears to be interesting boundary behavior of our harmonic extensions of step functions.
Examine this behavior in the next exploration.
Exploration 5.55. Using ComplexTool or PolyTool, graph the function from Example 5.31
that maps D to a triangle. Investigate the behavior of the boundary using the sketching tool
of the applet. Approach the break point between arcs, such as z D 1, along different paths.
Approach radially, then approach along a line that is not a radius of the circle. Observe how
the paths that approach 1 cause the image of the path to approach different points along the
line segment that makes up a portion of the boundary of the range. (As you get close to an
arc endpoint, the image of the sketch may jump to a vertex: here, examine where the image
is immediately before that jump.) In PolyTool and ComplexTool, you can hit the Graph
button to clear previous sketching but keep the polygonal map. Repeat the exercise with
some other examples of polygonal functions. Try to answer:
(a) Given a point on the boundary of the polygon, is it possible to find a path approach-
ing @D such that .z/ approaches ‹
(b) As you approach an arc endpoint in @D radially, what point on the boundary of the
polygon do you approach?
Try it out!
As you performed the exploration, you probably discovered some of the known prop-
erties of the boundary behavior of harmonic extensions of step functions, originally proven
by Hengartner and Schober [8], who proved a more general form of the next theorem.
We state the theorem as it applies to the step functions of Definition 5.37. The cluster
set of f at a point e i tk is the set of limits of sequences fzn g, where zn is inside , and
limn!1 .zn / D e i tk .
Theorem 5.56. Let f be the harmonic extension of a step function fO.e i tk / in Definition
5.37. Denote by the polygon defined by the vertices vk . By definition, the radial limits
limr !1 f .r e i t / lie on for all t except those in ftk g. The unrestricted limit
fO.e i t / D lim f .z/
z!e it
exists at every point e i t 2 @Dnfe i t1 ; e i t2 ; :::; e i tn g and lies on : Furthermore,

(a) the one-sided limits as t ! tk are
lim fO.e i t / D vk and lim fO.e i t / D vkC1

t !tk t !tkC
(b) the cluster set of f at e i tk 2 fe i t1 ; e i t2 ; : : : ; e i tn g is the linear segment joining vk to

vkC1 .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 301 — #319

i i
5.6. Star Mappings 301
Proof. Part (a) of the theorem follows directly from the definition of f and the Poisson
integral formula. Since the boundary correspondence is defined in Definition 5.37, the
limits follow. Now we need to show part (b). Let us consider e i tk . If z approaches e i tk
along the circular arc

e i tk C z
arg D ; 1 < < 1; (5.12)
e i tk z 2
then f .z/ converges to

1 1
.1 /vk C .1 C /vkC1 :
2 2
Therefore the cluster set of f at tk is the line segment joining vk and vkC1 , and part (b) is
proven.
Exercise 5.57. Use analytic geometry to see that (5.12) is a circular arc. Hint: Consider
the cases where z is either on the circle or is the center of the circle for some intuition. Try
it out!
Exercise 5.58. Theorem 5.56 holds for non-univalent mappings. Go back to previous ex-
amples and identify the line segment that connnects the vertices. Regraph the example from
Exercise 5.34. Using the Sketch utility in either ComplexTool or PolyTool, check that the
limit as you approach one of the tk does appear to be that line segment. Try it out!
5.6 Star Mappings

From the Rado-Kneser-Choquet theorem, we see that harmonic functions mapping the unit
disk D to convex polygons are well understood. That is, if we define a harmonic function
mapping from the unit disk to a convex polygon as in Definition 5.37, the function is
univalent in D. Theorem 5.56 describes the boundary behavior fully, showing that the limit
of the function as we approach one of the break points between vertex pre-images, tk , gives
the line segment joining the vertices.
Non-convex polygons are not nearly as well understood. We first examine non-convex
polygons in their simplest mathematical form: the regular stars.
Definition 5.59. By an n-pointed star, or r -star, we mean an equilateral 2n-gon with the
vertex set n o
r ˛ 2k ; ˛ 2kC1 W k D 1; 2; : : : ; n and ˛ D e i =n ;
where r is some real constant.

When r D 1, the n-pointed star is a regular 2n-gon and when r < cos.=n/ or
r > sec.=n/ it is a strictly non-convex 2n-gon. Our preimages of the vertices of the
2n-gon will be arcs centered at the 2nth roots of unity (this is different from our previous
examples).
Example 5.60. We will find a harmonic mapping of the unit disk into the 0.3-star. That is,
we will find the harmonic extension of the following boundary correspondence:
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 302 — #320

i i
F IGURE 5.14. The 0:3-star for n D 3.
for t from to fO.e i t /

=6 =6 0:3
=6 =2 e i =3
=2 5=6 0:3e i 2=3
5=6 7=6 1
7=6 3=2 0:3e i 4=3
3=2 11=6 e i 5=3
After going through details as in previous examples, we discover that the harmonic
extension is
" !
1 1 ze i =6
f .z/ D 0:3 arg
1 ze i =6
!
i 5=6
1 C iz 1 ze
C e i =3 arg C 0:3e i 2=3 arg
1 ze i =6 1 C iz
!
1 ze i 7=6 i 4=3 1 iz
arg C 0:3e arg
1 ze i 5=6 1 ze 7i =6
!#
1 ze i11=6
Ce i 5=3 arg :
1 ze i 3=2
Graph this using ComplexTool (it is one of the Pre-defined functions). It appears
to be univalent, but we have not yet proved that.
Exercise 5.61. Prove that if f .z/ is the harmonic extension to the r -star as defined in
Definition 5.59, then f .0/ D 0. Interpret this result geometrically. Try it out!
Exercise 5.62. Modify the function in Exercise 5.60 to have r D 0:15 and see whether
it appears univalent. To graph it in ComplexTool, choose the previous star as one of the
Pre-defined functions and then modify the equation in the function box.
Try it out!
To work with these stars, we may sometimes want to vary the boundary correspon-
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 303 — #321

i i
5.6. Star Mappings 303
dence. That is, we may want not to split @D evenly among the 2n vertices. It will become
useful to us to have an unequal correspondence in the boundary arcs but maintain some
symmetry. To do this, we still consider arcs centered at the 2n-th roots of unity, but alter-
nate between larger and smaller arcs. If we examine the geometry, we realize that an even
split would make each arc have length 2 2n
D =n. Two consecutive arcs would together
have length 2=n. To maintain some symmetry but let the arcs alternate in size, we want
two consecutive arcs to add to 2=n, but not split evenly. We introduce a parameter p,
with 0 < p < 1, to split consecutive arcs into p2=n and .1 p/2=n, whose sum is still
2=n. We make the definition:
Definition 5.63. Let n 2 be an integer, r a positive real number, and ˛ D e i =n. Define
a boundary correspondence from @D the vertices of the r -star for all but a finite number of
points on @D by mapping arcs with endpoints f˛e ip=n; ˛e ip=n; 0 k n 1g as:

r ˛ 2k ; e i t 2 .˛ 2k e ip=n ; ˛ 2k e ip=n /
fO.e i t / D 2kC1 : (5.13)
˛ ; e 2 .˛ 2kC1 e i.1 p/=n ; ˛ 2kC1 e i.1 p/=n /
it
Let f be the Poisson extension of fO.
e ip/n
2(1 – p)p
n
2pp/n r
F IGURE 5.15. The first two arcs and their images from Definition 5.63. The dots on the left-hand
side indicate points of discontinuity of the boundary correspondence.
The arc .˛ 2k e ip=n ; ˛ 2k e ip=n / centered at ˛ 2k is mapped to the vertex r ˛ 2k and

the arc .˛ 2kC1 e i.1 p/=n ; ˛ 2kC1e i.1 p/=n / centered at ˛ 2kC1 is mapped to the vertex
˛ 2kC1 :
Exercise 5.64. Show that the interval in (5.13),
.˛ 2kC1 e i.1 p/=n

; ˛ 2kC1e i.1 p/=n
/;
can be written as .˛ 2k e ip=n ; ˛ 2kC2e ip=n

/. Try it out!
Now use the StarTool applet. The default for this applet is the 3-pointed star from Ex-
ample 5.60. The arcs and their target vertices are colored (with a light blue arc mapping to
a light blue vertex, for example). The default p-value is 0:5, which corresponds to evenly
spaced arcs. You can use the slider bars (the plus/minus buttons for n) or type in the text
boxes to change the values for n, p, and r: The maximum n-value allowed 18, which is
sufficient for the explorations below. As with ComplexTool, there is the option to Sketch
on the graph to see the mapping properties of the stars. There is also an option to Show
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 304 — #322

i i
F IGURE 5.16. A star with n D 5, r D 0:15, and p D 0:9.
roots of !.z/. The roots of !.z/ can be helpful but are not essential for the first ex-
plorations. In general, try to see what happens for small values of n, such as 4; 5; 6; or 7.
Exploration 5.65. Explore different values of n and r . See if you can find a pattern for the
univalence of the star function.
(a) What is the relationship between the r that you choose and the p-value necessary for
univalence? Is there a range of p that works?
(b) What happens as p goes to 0 or 1?
(c) For a given p-value, can you determine the minimal r ? That is, how small can you
make r and maintain univalence?
(d) Is there a minimum r -value, one so that no p-value that will give univalence?
(e) As you change n, what happens to the p that you need to achieve univalence?
(f) What is the relationship between r , n, and p? (It is unlikely that you will be able to
answer this question now, but make some conjectures.)
Try it out!
5.7 Dilatations of Polygonal Maps are Blaschke Products

We now use the tools of harmonic functions to study the polygonal maps. To understand
when the star maps are univalent, we first examine their dilatations. As we discover, the
dilatation of a polygonal map is a Blaschke product.
Exercise 5.66. For the function generated in Exercise 5.34 on page 289:
(a) Find the formulas for h.z/ and g.z/. (Use the result of Exercise 5.42.)
(b) Find the derivatives h0 .z/ and g0 .z/, verifying that they simplify to

1 3 3 2 5 5
h0 .z/ D . i /z 3iz C . C i /
.2 i /.1 z 4 / 2 2 2 2
and
0 1 5 5 2 3 3
g .z/ D 4
. i /z C 3iz C . C i / :
.2 i /.1 z / 2 2 2 2
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 305 — #323

i i
5.7. Dilatations of Polygonal Maps are Blaschke Products 305
(c) Show that the zeros of g0 .z/ are z1 0:9245 0:9245i and z2 0:3245 C 0:3245i
and that the zeros of h0 .z/ are 1= z1 and 1= z2 . Thus we are able to write the dilatation
as
z1 z z2 z
!.z/ D C ;
1 z1 z 1 z2 z
where C is a constant.
(d) What are jz1 j; jz2 j; and jC j? Remember these when you read Theorem 5.81 in Section
5.8.
Try it out!
Motivated by the results of Exercise 5.66, we examine functions of a particular form.
Exploration 5.67. We examine the properties of functions of the form
z0 z
Bz0 .z/ D .
1 z0 z
0:5 z
(a) Using ComplexTool, graph the image of the unit disk under B0:5.z/ D 1 0:5z
,
i=4
B0:5 exp.i =4/ .z/ D 1 0:5e z
0:5e i=4 z
0:5i z
, and B0:5i .z/ D 1C0:5iz . What is B.0/? What is
the image of the unit disk under B.z/‹ Does B.z/ appear to be univalent?
i=4
(b) Graph the image of the unit disk under B2 exp.i =4/ .z/ D 1 2e2e i=4z z . What is B.0/?
What is the image of the unit disk under B.z/‹ Does B.z/ appear to be univalent?
(c) What is the general formula for Bz0 .0/? What effect does arg.z0 / have on the location
of Bz0 .0/? What effect does jz0 j have on the image of the unit disk?

z 2
(d) Graph the image of the unit disk under f1 .z/ D .B0:5 .z//2 D 10:50:5z : What is
f1 .0/? What is the image of the unit disk under f1 .z/‹ Does f1 .z/ appear to be uni-
valent? How does this compare to f .z/ D z 2 ‹
(e) Graph the image of the unit disk under f2 .z/ D B0:5 .z/B0:5i .z/, f3 .z/ D
B0:5 .z/B0:2i .z/, and f4 .z/ D B0:5 exp.i =4/ .z/B0:2i .z/. What are f2 .0/, f3 .0/, and
f4 .0/? How do f2 .0/, f3 .0/, and f4 .0/ relate to the values of z0 in the functions
Bz0 .z/‹
Try it out!
Definition 5.68. A Blaschke factor is Bz0 .z/ D 1z0 z0zz , and a finite Blaschke product of
order n is a product of n Blaschke factors, possibly multiplied by a constant with jj D 1:
n
Y zk z
:
1 zk z
kD1
Geometrically, the multiplication by is a rotation.

Remark 5.69. This definition is non-standard. The standard definition of Blaschke prod-
uct, as in Chapters 4 and 6, assumes that jzk j < 1. We do not place that restriction on zk
to simplify our computations.
We use the result of Exercise 5.39 on page 290 to see that the dilatation of the harmonic
polygonal functions to an n-gon is a Blaschke product of order at most n 2. This was
proved by Sheil-Small in [16], and is discussed in [5].
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 306 — #324

i i
Let fk .z/ denote the contribution to f .z/ that arises from applying the Poisson integral
P
formula to the boundary correspondence for tk 1 < t < tk . Then f .z/ D nkD1 fk .z/.
On the interval tk 1 < t < tk ,

vk vk 1 ze i tk
fk .z/ D .tk tk 1 / C arg
2 1 ze i tk 1
by Definition 5.37. The canonical decomposition of fk .z/ is hk .z/ C gk .z/. By Exercise

5.42, we have

vk vk 1 ze i tk
hk .z/ D .tk tk 1 / C log
2 2 i 1 ze i tk 1
vk vk
D .tk tk 1 / C log.1 ze i tk / log.1 ze i tk 1 /
2 2 i
and

vk 1 ze i tk
gk .z/ D log
2 i 1 ze i tk 1
vk
D log.1 ze i tk / log.1 ze i tk 1 / :
2 i
The computations that follow give some rigor to our intuition: since h and g are sums of
logarithms, their derivatives are sums of terms that have 1 ze i tk in their denominators.
We will combine the factors and write the derivatives as Blaschke products. From h.z/ D
Pn
kD1 hk .z/ we have:
Xn
0 vk e i tk e i tk 1
h .z/ D
2 i 1 ze i tk 1 ze i tk 1
kD1
Xn
vk 1 1
D : (5.14)
2 i z e i tk z e i tk 1
kD1
The function g0 .z/ is identical except for having vk instead of vk :
Xn
0 vk e i tk e i tk 1
g .z/ D
2 i 1 ze i tk 1 ze i tk 1
kD1
Xn
vk 1 1
D : (5.15)
2 i z e i tk z e i tk 1
kD1
Combining like factors gives

n n
1 X vk vkC1 1 X vk v kC1
h0 .z/ D and g0 .z/ D : (5.16)
2 i z e i tk 2 i z e i tk
kD1 kD1
Pn
Because vnC1 D v1 , kD1 .vk vkC1 / D 0, a result we use later.
Pn
Exercise 5.70. Prove that kD1 .vk vkC1 / D 0, since vnC1 D v1 . Interpret the result
geometrically. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 307 — #325

i i
5.7. Dilatations of Polygonal Maps are Blaschke Products 307
We rely on the observations that
h0 .1= z/ D z 2 g0 .z/ or g0 .1= z/ D z 2 h0 .z/; (5.17)
which arise from:
n n
1 X vk v kC1 z2 X vk v kC1
h0 .1= z/ z 2 g0 .z/ D
2 i 1=z e i tk 2 i z e i tk
kD1 kD1
n n
z X e i tk . v k v kC1 / z2 X vk v kC1
D
2 i e i tk z 2 i z e i tk
kD1 kD1
Xn i tk 2 Xn
z e . vk v kC1 / z vk v kC1
D i t
2 i z e k 2 i z e i tk
kD1 kD1
Xn i tk 2
1 . vk vkC1 /.ze z /
D
2 i z e i tk
kD1
n
1 X . vk vkC1 /.z/.e i tk z/
D
2 i z e i tk
kD1
Xn
1
D . vk vkC1 /.z/
2 i
kD1
D 0:
Exercise 5.71. We proved the first half of (5.17). Using that result, prove the second part
of (5.17) with minimal computation. Try it out!
Exercise 5.72. Interpret the result geometrically. If z0 2 D is such that g0 .z0 / D 0, then
what do we know about the zeros of h0 ? How are the locations of the zeros of h0 related to
the locations of the zeros of g0 ? Completing this exercise will give intuition about a later
result. Try it out!
Exercise 5.73. Show that h0 .z/ and g0 .z/ of Exercise 5.66 satisfy (5.17). Do the conclu-
sions of the last two exercises hold true for this example? Try it out!
We can already tell that if we have a common denominator for h0 .z/ or g0 .z/ that
n
Y
it is .z e i tk /, and we would guess that the ratio g0 .z/= h0 .z/ would be a product
kD1
of rational functions. That is all we know now—it is not obvious that this product is a
Blaschke product, although we may expect that from Exercise 5.66. The remainder of this
section will be devoted to determining that it is, indeed, a Blaschke product, and we will
find the order of the product.
Exploration 5.74. Based on the results of Exercise 5.66 and other examples, make a con-
jecture about the number of Blaschke factors that are in the dilatation of a harmonic func-
tion from D to an n-gon. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 308 — #326

i i
n
Y
The common denominator of h0 and g0 is S.z/ D .z e i tk /, giving:
kD1
P .z/ Q.z/
h0 .z/ D and g0 .z/ D :
S.z/ S.z/
We want to find P and Q. Each term of the P .z/ has the form
n
Y
.vk vkC1 / .z e i tj /;
j D1Ij ¤k
a polynomial of degree at most n 1. The coefficient of the z n 1 term of P .z/ is vk vkC1

n
X
for each piece of the sum in (5.14), so this coefficient is .vk vkC1 /, which is 0. Thus
kD1
P .z/ has degree at most n 2. The same argument works for Q.z/, because it has the
same structure as P .z/.
n n
Y
1 n
Exercise 5.75. Show that S.1= z/ D . 1/ S.z/ e i tk : Try it out!
z
kD1
We could write this as

n n
Y
1
S.1= z/ D . 1/n S.z/ e i tk
: (5.18)
z
kD1
Exercise 5.76. Show that (5.18) holds for the denominator of the derivatives in Exercise
5.66, 2 i.1 z 4 /: Try it out!
We combine (5.17) and (5.18) to get a relationship between P .z/ and Q.z/. Substitut-
ing into h0 .1= z/ D z 2 g0 .z/, we see that
h0 .1= z/ D z 2 g0 .z/
P .1= z/ Q.z/
D z2
S.1= z/ S.z/
P .1= z/ Q.z/

1 n Q D z2 :
z
. 1/n S.z/ nkD1 e i tk S.z/
This leads to
n
Y
zn 2
P .1= z/ D . 1/n Q.z/ e i tk
: (5.19)
kD1
Exercise 5.77. Using (5.19), show that

n
Y
zn 2
Q.1= z/ D . 1/n P .z/ e i tk
: (5.20)
kD1
Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 309 — #327

i i
5.8. An Important Univalence Theorem 309
Since f is orientation-preserving, we know that h0 .z/ ¤ 0 in D. This implies that

P .z/ ¤ 0 in D. In particular, P .0/ ¤ 0. Substituting 0 into (5.20), we find that the left-
hand side is not zero, which forces the degree of Q to be at least n 2. We determined that
the degree of Q is at most n 2, so its degree is precisely n 2. Similarly, the degree of
P is n 2. Since the degree of Q is n 2, we write
nY
m 2
Q.z/ D z m .z zk /
kD1
to show that Q may have m zeros at the origin and n m 2 zeros elsewhere (the zk need
not be distinct). Using (5.20), we write
m n Y
m 2 n
Y
1 1
zn 2
zk D . 1/n P .z/ e i tk
:
z z
kD1 kD1
The left-hand side of the equation may be written as
nY
m 2
1
zn m 2
.1 z zk / ;
zn m 2
kD1
nY
m 2
so we see that since the zeros of Q are zk , the zeros of P are the zeros of .1 z zk /,
kD1
which are precisely 1= z k . Now we can see what the Blaschke product is.
Exercise 5.78. Find the relationship between the number of zeros of Q and the number of
zeros of P . If Q has degree n 2 with m zeros at the origin and n m 2 zeros away
from the origin, then how many of the zeros of P are at the origin? How many of the zeros
of P are away from the origin? Try it out!
We summarize the results of our work (as originally proved by Sheil-Small [16, Theo-
rem 1]; see also [5])
Theorem 5.79. Let f be the harmonic extension of the step function fO.e i t / in Definition
5.37. Then
Q.z/ P .z/
g0 .z/ D and h0 .z/ D ;
S.z/ S.z/
where Q.z/; P .z/, and S.z/ are defined as above, and P and Q are polynomials of degree
at most n 2. Their ratio !.z/ D Q=P satisfies j!.z/j D 1 when jzj D 1, so is a Blaschke
product of degree at most n 2.
5.8 An Important Univalence Theorem

We examine a theorem of Sheil-Small that tells when the harmonic function in Definition
5.37 is univalent. It is enough to know the location of the zeros of the analytic dilatation
0 .z/
!.z/ D gh0 .z/ .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 310 — #328

i i
Exploration 5.80. Open the StarTool applet. Check the box in front of Show roots
of !.z/. You will see extra dots appear in the right-hand pane (the range of the function),
as well as a unit circle for reference. The dots give the locations of the zeros of !.z/.
Experiment with the values of p and r to see if there is a relationship between the roots of
!.z/ and the univalence of the star. Do this for various values of n. Does your conjecture
agree with what you learned in Exercise 5.66? Try it out!
The theorem below was proved by Sheil-Small, [17, Theorem 11.6.6].
Theorem 5.81. Let f be a harmonic function of the form in Definition 5.37. It is the har-
monic extension of a piecewise constant boundary function with values on the m vertices of
a polygonal region , so that, by Theorem 5.79, the dilatation of f is a Blaschke product
with at most m 2 factors. Then f is univalent in D if and only if all zeros of ! lie in D.
In this case, f is a harmonic mapping of D onto :
Proof. Suppose that f is univalent in D. If a Blaschke factor is 'z0 .z/ D 1z0 z0zz with
the constant z0 not having modulus 1, then, since the dilatation ! is a product of a finite
number of Blaschke factors, !.z/ ¤ 0 on the unit circle. This is because the zero of the
Blaschke factor is z0 , and if jz0 j D 1; we get that 'z0 .z/ D z0 for all z. If ! has a zero
at z0 outside of D, then it has a pole at 1= z0 2 D. If it also has zeros in D, then there
are points in D where j!.z/j < 1 and other points where j!.z/j > 1. This implies that the
Jacobian of f changes sign in D, which would force the Jacobian to equal 0 at some point
in D, contradicting Lewy’s theorem, which says that the Jacobian is non-zero since f is
locally univalent. Thus there are two possibilities for a univalent f : Either all of the zeros
of !.z/ lie in D, or all lie outside D. If the zeros of ! lie outside of D, then j!.z/j > 1
in D and f has negative Jacobian, contradicting its construction as a sense-preserving
boundary function. Therefore, all the zeros of ! lie in D.
Conversely, assume all the zeros of ! lie within D. By the mapping properties of
Blaschke products, j!.z/j < 1 in D. We use the argument principle to show that f is
univalent in D and maps D onto . Choose w0 2 . Let C" be the path in D consisting
of arcs of the unit circle along with small circular arcs of radius " about the points bk (the
points bk are the arc endpoints in the domain disk), as shown in Figure 5.17.
If " is sufficiently small, the image of C" will not go through w0, and will have winding
number C1 around w0 . Since j!.z/j < 1 inside C" , it follows from the argument principle
for harmonic functions that f .z/ w0 has one simple zero inside C" (or, put another way,
bk
e
ce
F IGURE 5.17. Tiny circle around bk .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 311 — #329

i i
5.9. The Dilatation for Star Mappings 311
f .z/ D w0 has exactly one solution for z 2 D). Thus f .D/. A similar construction
with w0 … to shows that w0 … f .D/. Thus f maps D univalently onto .
To apply the theorem to the star mappings of Section 5.6, we study the dilatation of the
star mappings in detail.
5.9 The Dilatation for Star Mappings

We use Definition 5.63 of Section 5.6 as a starting point. We will build on the basic formula
for the functions h0 .z/ and g0 .z/, and then will simplify the dilatation !.z/ as a Blaschke
product. By doing so, we completely determine which star functions are univalent.
Using (5.14), (5.15), and Definition 5.63, we find that for the star functions:

0 r ˛0 1 1
h .z/ D
2 i z ˛ 0 e ip=n z ˛ 0 e ip=n

˛ 1 1
C
2 i z ˛ 2e ip=n z ˛ 0e ip=n

r ˛2 1 1
C C :::
2 i z ˛ 2e ip=n z ˛ 2e ip=n
n 1
r X 2k 1 1
D ˛ (5.21)
2 i z ˛ 2k e ip=n z ˛ 2k e ip=n
kD0
n 1
1 X 2kC1 1 1
C ˛
2 i z ˛ 2kC2 e ip=n z ˛ 2k e ip=n
kD0
and
n 1
r X 2k 1 1
g0 .z/ D ˛ (5.22)
2 i z ˛ 2k e ip=n z ˛ 2k e ip=n
kD0
n 1
1 X 2kC1 1 1
C ˛ :
2 i z ˛ 2kC2 e ip=n z ˛ 2k e ip=n
kD0
Our goal is to express !.z/ D g0 .z/= h0 .z/ as the ratio of Blaschke products guaranteed
by Sheil-Small’s work. We first establish some identities involving sums of the quantities
of the type found in (5.21) and (5.22).
If is a primitive mth root of unity, then
m
Y
.z k / D zm 1: (5.23)
kD1
Exercise 5.82. Prove (5.23). Interpret it geometrically. Try it out!

Exercise 5.83. Let be a primitive mth root of unity. Show that
m
Y1
.z k a/ D z m am : (5.24)
kD0
Hint: Replace z with z=a in (5.23). Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 312 — #330

i i
Now the hard question: How do we add the sums in (5.21) and (5.22), given that their
numerators are not 1? As a step toward an answer, we establish
Lemma 5.84. If is a primitive mth root of unity, then
m
X1 k mam 1
D : (5.25)
z k a z m am
kD0
Exercise 5.85. Prove the lemma, using partial fraction decomposition:

(a) See how (5.24) relates to (5.25).
(b) Since we have n distinct linear factors in the denominator, we can expect to find that
m
X1 ak
mam 1
D :
z m am z k a
kD0
(c) We will find ak0 . By setting z D k0 a, establish that
mam 1 m
ak0 D Q k k
DQ k0
:
k¤k0 . a
0 a/ k¤k0 . k /
Y
(d) Show that . k0 k / D m k0
: Remember that k0 is an mth root of unity, so
k¤k0
k0 m D 1.
(e) Conclude that ak0 D k0 :
(f) (5.25) follows.
Try it out!
From Exercise 5.64 on page 303 we know that
˛ 2kC2 e ip=n
D ˛ 2kC1 e i.1 p/=n
and ˛ 2k e ip=n D ˛ 2kC1 e i.1 p/=n
:
Thus !
n 1
1 X ˛ 2kC1 1 ne i.1 p/.n 1/=n
D
2 i z ˛ 2kC2 e ip=n 2 i z n C e i.1 p/
kD0
and !
n 1
1 X ˛ 2kC1 1 ne i.1 p/.n 1/=n
2k ip=n
D :
2 i z ˛ e 2 i z n C e i.1 p/
kD0
Exercise 5.86. Prove that

n 1
!
1 X ˛ 2kC1 1 ne i.1 p/.n 1/=n
D :
2 i z ˛ 2k e ip=n 2 i z n C e i.1 p/
kD0
Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 313 — #331

i i
5.9. The Dilatation for Star Mappings 313
Combining all these,

n 1
i. n n 1 /p
0 n r e i. n /p re
h .z/ D n ip
C n ip
2 i z e z e
n 1 n 1
!
e i. n /.1 p/
e i. n /.1 p/
C n C n : (5.26)
z C e i.1 p/ z Ce i.1 p/
0
We know from Theorem 5.79 that gh0 .z/ .z/
can be written as a Blaschke product. To do
this, we will find a common denominator and combine the four terms of h0 .z/ to see the
quotient. First, we find that we can write the common denominator more simply.
Exercise 5.87. Prove that .z n e ip /.z n e ip
/ D .z n C e i.1 p/
/.z n C e i.1 p/
/,
which we call Sn .z/. Try it out!
Exercise 5.88. Using algebra to show that

0 n n .n 1/p .n 1/.1 p/
h .z/ D z r sin sin (5.27)
Sn .z/ n n
p
.1 p/
Cr sin C sin :
n n
Hint: find a common denominator, simplify, and use properties of z C z. Try it out!
With similar methods, we can prove that
p
nz n 2 .n 1/.1 p/
g0 .z/ D z n r sin C sin (5.28)
Sn.z/ n n

.n 1/p .n 1/.1 p/
Cr sin sin :
n n
If we let
sin .n 1/.1 p/
n
r sin .n 1/p
n
cD (5.29)
r sin p
n
C sin .1 p/
n
we have n
nz n 2 p .1 p/ z c
g0 .z/ D r sin C sin : (5.30)
n n Sn .z/
We combine the result of Theorem 5.81 with the dilatation. When the zeros of the
dilatation are in the unit disk, then the harmonic function f D h C g that defines the star
is univalent. By a straightforward computation, we find that the dilatation of f is
zn 2
.z n c/
!.z/ D : (5.31)
1 zn c
Exercise 5.89. Verify that Theorem 5.81 holds for the star function. Try it out!
Exploration 5.90. Using the fact that f is univalent when jcj < 1, do the following:
(a) Use the StarTool applet to explore graphically what relationship there is between n, p,
and c.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 314 — #332

i i
(b) Given n, find the range of p-values that make jcj < 1.
(c) Given p, find the range of n-values that make jcj < 1.
Try it out!
Exercise 5.91. For a given n, consider the formula for c in (5.29) to be a function of p
alone. Prove that any star configuration is possible; that is, prove that for any r , p can be
found to make jcj < 1. What ranges of p makes this happen? Conversely, prove that if
r < cos.=n/ or r > sec.=n/, p can be chosen to make the function not univalent. Why
is this not true for cos.=n/ r sec.=n/? For more information, see [7, Theorem 4].
Try it out!
Small Project 5.92. Refer to Chapter 2. For what values of c is the dilatation a perfect
square? Find and describe the associated minimal surfaces. Can they be described as ex-
amples of other well-known surfaces? For more information, see [11].
5.10 Open Questions

Large Project 5.93. If you move one vertex of the star, do the same results hold for the
relationship between n and p? (For example, take the vertex at r , and move it to r C " or
r ". Is the star still univalent?)
Large Project 5.94. Can we map to any polygon univalently? The star takes full advantage
of the symmetry. Once you lose that advantage it is harder to discover whether the zeros of
the dilatation have modulus less than 1. This question is known as the mapping problem,
proposed by Sheil-Small in [16].
Small Project 5.95. Look at a function f that is not univalent. Now look at the set S D
of points on which the function f is univalent. How do you find S ? What is its shape? Is
it starlike? Is it convex? Is it connected? Is it simply connected? Can DnS be connected?
Small Project 5.96. We discussed one way of proving that a harmonic function is univalent
by looking at zeros of the analytic dilatation !.z/. In Chapter 4, there is another set of
criteria for univalence in Section 4.7. Connect the two. Does one imply the other? How
does the work with stars generalize to the approach in Chapter 4? Are there results in this
chapter that could not be found using the methods of Chapter 4?
5.11 Bibliography
[1] Lars Ahlfors, Complex Analysis, third edition, McGraw-Hill, New York, 1979.
[2] W. N. Bailey, Generalized Hypergeometric Series, Cambridge Tracts in Mathematics
and Mathematical Physics, No. 32, Stechert-Hafner, Inc., New York, 1964.
[3] Tobin A. Driscoll and Lloyd N. Trefethen, Schwarz-Christoffel Mapping. Cambridge
University Press, Cambridge, 2002.
[4] Kathy Driver Peter and Duren, Harmonic shears of regular polygons by hypergeo-
metric functions, J. Math. Anal. Appl., 239(1): 72–84, 1999.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 315 — #333

i i
[5] Peter Duren, Harmonic Mappings in the Plane, Cambridge Tracts in Mathematics,
v. 156, Cambridge University Press, Cambridge, 2004.
[6] Peter Duren, Walter Hengartner, and Richard S. Laugesen, The argument principle
for harmonic functions, Amer. Math. Monthly, 103(5): 411–415, 1996.
[7] Peter Duren, Jane McDougall, and Lisbeth Schaubroeck, Harmonic mappings onto
stars. J. Math. Anal. Appl., 307(1):312–320, 2005.
[8] Walter Hengartner and Glenn Schober, Harmonic mappings with given dilatation, J.
London Math. Soc. (2), 33(3): 473–483, 1986.
[9] Elgin Johnston, A “counterexample” for the Schwarz-Christoffel transform, Amer.
Math. Monthly, 90(10): 701–703, 1983.
[10] Serge Lang, Complex Analysis, fourth edition, Graduate Texts in Mathematics, v.
103, Springer-Verlag, New York, 1999.
[11] Jane McDougall and Lisbeth Schaubroeck, Minimal surfaces over stars, J. Math.
Anal. Appl., 340(1): 721–738, 2008.
[12] Zeev Nehari, Conformal Mapping, Dover Publications Inc., New York, 1975.
[13] Earl D. Rainville, Special Functions, The Macmillan Co., New York, 1960.
[14] E. B. Saff and A. D. Snider, Fundamentals of Complex Analysis for Mathematics,
Science, and Engineering, second edition, Prentice Hall, Upper Saddle River, NJ,
1993.
[15] Roland Schinzinger, and Patricio A. A. Laura, Conformal Mapping: Methods and
Applications, Dover Publications Inc., New York, 2003.
[16] T. Sheil-Small, On the Fourier series of a step function, Michigan Math. J., 36(3):
459–475, 1989.
[17] T. Sheil-Small Complex Polynomials, Cambridge Studies in Advanced Mathematics,
v. 75, Cambridge University Press, Cambridge, 2002.
[18] Richard A. Silverman, Complex Analysis with Applications, Dover Publications Inc.,
New York, 1984.
[19] Dennis G. Zill and Patrick D. Shanahan, A First Course in Complex Analysis with
Applications, Jones and Bartlett, Sudbury, MA, 2003.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 316 — #334

i i
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 317 — #335

i i
Circle Packing
6
Kenneth Stephenson (text and software)
Complex analysis is in many ways the ultimate in continuous mathematics. It presents

you with a smooth world: continuous variables, infinitely differentiable functions, smooth
surfaces, backed up by a complex arithmetic with its power series, line integrals, and all
sorts of handy formulas. In this chapter, however, we will develop a different view of the
topic. Here the geometry behind analytic functions moves strongly to the fore as we see
how one might discretize complex analysis.
An analogy is that of a mountain stream. We normally treat the stream as a continuous
medium and use continuous variables and functions to understand it — pressure, velocity,
vorticity — modeled, perhaps, by the Navier-Stokes and other partial differential equations.
Yet we know that the stream is in fact a collection of individual water molecules, a discrete
medium. The macro behavior, the waves, eddies, and currents, emerge from myriad local
interactions among the discrete water molecules. So one might ask: what purely local rules
of interaction among molecules could lead to the observed global behavior?
In this chapter, we address this issue for continuous analytic functions. Circles will be
our molecules, and “packing” conditions will provide the local rules for their interaction.
Circle packings, configurations of myriad individual circles, each interacting only with its
neighbors, will manifest macro behavior that we will recognize as a version of analyticity.
Later in the chapter we will also see that our discrete objects converge under appropriate
refinement to the familiar continuous counterparts. In effect, then, circle packing provides
a quantum complex analysis, one that is classical in the limit.
This is a topic ideal for visualization and experimentation. In fact, it is accessible only
through the computer, so you will find yourself tethered to the software more closely than
in the earlier chapters. Persevere and you may develop a new appreciation for complex
analysis.
I want you to jump right in with an experiment from the foundations of the topic.
The software and scripts are accessible through the book’s web site. Download the Java
application CirclePack as a jar file to your desktop and double click to start it. Us-
ing File ! Load script, choose the file Cookie.xmd and run through it using the
NEXT button. Try it out!
317
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 318 — #336

i i
318 Chapter 6. Circle Packing
6.1 First Impressions

Circle packings were introduced to analysis by William Thurston in a talk in 1985 ([9]).
The context was the conformal mapping of bounded, simply connected, plane regions
onto the unit disc D. The central result is the famous Riemann mapping theorem stated in
the Appendix, page 355. (Alternate statements: Theorem 3.33, p. 175, and Theorem 4.16,
p. 209). For us, the preferred statement is this: Given and designated points w0; w1
within , there exists a unique one-to-one analytic mapping F from D onto so that
F .0/ D w0 and so that w1 lies on the image of the positive real axis. Figure 6.1 therefore
illustrates the discrete Riemann mapping theorem.
F IGURE 6.1. A conformal map display, domain (left) and range (right).
Figure 6.1 shows the mapping window as displayed by our experimental testbench, the
software package CirclePack. In such side-by-side displays, the domain circle packing will
be on the left, the range packing on the right. In this instance, the domain is D and the range
is the given region . The function itself can be interpreted in various ways. Each circle on
the left corresponds to a circle on the right, so we can map circles to circles. Alternately,
as we will see later, we can map the triangles formed by triples of circles on the left to
those formed by the corresponding triples of circles on the right. The packings provide, in
essence, a map from D onto , a discrete conformal map. We will see the details shortly,
but assuming you have run Cookie and have this example before you on the computer,
try clicking the left mouse button on a circle in one packing: both it and the corresponding
circle of the other packing will be highlighted (likewise, for faces, using the middle mouse
button). Try it out!
We say a few words about CirclePack before we go on. Figure 6.2 shows a typical
screen. In addition to the mapping window with the two packings, it shows various frames
and panels that you will use to manage CirclePack. Unlike the software in earlier chap-
ters, CirclePack is a Java application rather than an applet: it accesses you computer’s file
system, it is larger, more comprehensive (and reconfigurable), and it calls shared C/C++ li-
braries for computationally intense experiments. It also has an important scripting feature:
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 319 — #337

i i
6.1. First Impressions 319
F IGURE 6.2. CirclePack in Mapping Window mode.
a script refers to a file with prepared sequences of commands. Scripts for this chapter are
provided on the book’s web site; they have file extension .xmd and are indicated in the text
by Name. To run an experiment, then, start CirclePack, from within it load Name.xmd,
then click through the prepared commands.
To recreate Figure 6.1, run the script Cookie. This script takes you through a very
structured experiment: simply press the NEXT button (or press enter when the mouse is in
the display canvas) to run through the demonstration steps. Try it out! (If things become
jumbled in CirclePack— and they sometimes will! — it may be that you have outrun the
computations: click the up icon to restart at the beginning of the script, or if things have
really gone sour, exit and restart CirclePack.)
Even in the passive experimental mode of Cookie you can interact with the images by
moving, focusing, clicking circles and faces, and so forth, as we describe later. However,
it is hoped that as your experience and curiosity grow, you move through the following
additional stages to gain more experimental independence:
(1) The Open Script button brings up the CirclePack Script Window. The com-
mands of the script are laid out linearly and are typically accompanied by text expla-
nations. If you didn’t understand what the experiment was purporting to show, read
along as you execute the commands. Figure 6.3 illustrates the Script Window, with
text, icons that encapsulate the commands, included files, etc.
(2) In the script you can open a command icon by clicking its small “+” box to see the
string of commands that it executes. For example.,
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 320 — #338

i i
F IGURE 6.3. A typical script for CirclePack.
infile path kurve.g;cookie;disp -w -c -cf b

tells CirclePack to read the path kurve.g from the script’s data section, cookie-cut
the portion of the resident packing inside that path, then clear the canvas and display
the circles of the resulting packing, with the boundary circles filled. Modify these
commands and you can change the action in the experiment. In many scripts, hints
are given about alterations you might try. In this first experiment, for example, after
overlaying with a hexagonal circle packing, optional commands ’i’ or ’d’ will
increase or decrease, respectively, the sizes of the circles in that overlay. After you
have run through the experiment once, re-run it and try these adjustments to see what
happens.
(3) If you want to dig deeper, click the Advanced button to bring up CirclePack’s main
control frame, which gives access to its full functionality. The program holds up to
three packings, has panels with messages, history, and rudimentary error feedback,
Commands: lines where you can enter strings of commands, and various droppable
icons: drag one onto a packing window and a prepared string of commands will be
applied to that packing.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 321 — #339

i i
6.1. First Impressions 321
(4) For the ambitious, the Java code for CirclePack is available under the GNU open
source license. It includes a PackExtender class that allows anyone (with enough
work) to create specialized data structures and commands having full access to the
core functionality of CirclePack. Several existing PackExtenders will be used in later
experiments.
Figure 6.4 provides an annotated screen shot of CirclePack with some of its screens
and auxiliary panels. (The screenshots are current as of publication, but CirclePack may
change over time.)
F IGURE 6.4. An annotated screenshot.
Bare-bones scripts are easy to write for yourself, but following, reading, and modi-
fying those of others can be a challenge. The prepared scripts are built for ease of use,
so the commands are often long and involved. For example, disp -w -c will clear a
screen and draw the circles. However, disp -w -ct3c20 i -cfc180 b will clear
the screen then draw the interior circles thicker and in blue and the boundary circles filled
with red. Don’t let the complications bother you — go ahead and start your own scripts
with simple commands and small steps. When things go wrong, adjust the commands. As
you learn CirclePack your scripts may become more involved and sophisticated, but also,
more effective.
We now give the definitions and notations we will use in our discrete analytic function
theory. Although static images are helpful, there is nothing like running experiments, even
in automatic mode, to bring the topic to life. The conformal mapping experiment we did
may help, and I hope it motivates the theory we will be investigating later. However, we
have next a sequence of experiments tailored to display and explain the basics of the theory.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 322 — #340

i i
6.2 Basics of Circle Packing

Definition 6.1. A circle packing is a configuration of circles with a prescribed pattern of
tangencies.
F IGURE 6.5. A small menagerie of circle packings.
Figure 6.5 illustrates a few circle packings from a tour you can take in Menagerie.
Try it out! Among the things you might observe:
(a) The circles form mutually tangent triples and hence have triangular interstices.
(b) The packings can live in the Euclidean plane, C, in the hyperbolic plane, D (the unit
disc in C), or on the Riemann sphere, C, represented as the unit sphere at the origin
in R3 . (See Appendix B.)
(c) The pattern of tangencies shows up in the carrier of the packing, the geometric
triangles formed by connecting the centers of tangent circles.
(d) This pattern can be simply or multiply connected.
(e) There are interior circles, those completely surrounded by their neighbors, and bound-
ary circles, those on the edge of the configuration.
(f) There are numerous manipulations, colors, and display options for displaying circle
packings and their carriers which are invaluable in highlighting the structures you
need to study.
We will have to make formal definitions shortly, but I suggest that you first play around
with the CirclePack Owl in OwlPlay. You might run sequentially through the commands
first, but by pressing the Script button you can open the Script Window and follow de-
scriptions associated with the commands. There are suggestions along the way for changes
that you can make to see how the packings react. Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 323 — #341

i i
6.2. Basics of Circle Packing 323
Mathematical formalities are essential both for understanding what is going on and for
establishing results that you and others can depend upon. Let me lay out the formal objects
involved in circle packing. Follow along in Objects, where these are illustrated explicitly
in turn, and read the descriptions in the script as you proceed. In other words, Try it out!
Complex, K: The “prescribed pattern” of tangencies for a circle packing is rep-

resented as a triangulation K of a topological surface (technically, a simplicial 2-
complex). This is a combinatorial object: K has vertices v, edges hv; wi, and faces
hv; u; wi formed by triples of vertices, but does have a geometric setting. It may be
finite or infinite, may or may not have boundary, and may be simply or multiply
connected. We always assume K is oriented, so in every face hv; u; wi, the vertices
are listed in positive (counterclockwise) order, and these orientations are globally
consistent, meaning that if faces f and g share a directed edge e D hv; wi, then e is
positively oriented in one face and negatively oriented in the other. The degree of a
vertex v is the number of edges from v (hence the number of neighboring vertices).
As K is a “triangulation”, every vertex will necessarily have finite degree.
Of course, I cannot display an abstract complex, but Objects begins with a
random geometric triangulation, one having no a priori connection with any circle
packing — it is a simple graph that you might have sketched by hand and entered
into a drawing program.
Circle Packing, P : A circle packing for K is a configuration P D fcv gv2K of
circles such that:
(1) There is a circle cv associated with every vertex v of K.
(2) If vertices v and u are neighbors (denoted v u and meaning that hv; ui is an
edge of K), then the circles cv and cu are (externally) tangent to one another.
(3) If hv; u; wi is an oriented face of K, then the circles cv ; cu; cw form a triple of
mutually tangent circles that is positively ordered.
You will observe in Objects that the circles are indexed sequentially starting at
1. There is an icon tool in the script that displays the flower of a selected vertex;
namely, the circle itself and the chain of neighboring petal circles. The script high-
lights other distinctions: Interior circles are those having closed flowers (their petals
wrap completely around them), while the boundary circles are those on the edge,
with flowers that do not close up. Likewise, boundary and interior edges and faces
are illustrated. A circle packing P is said to be univalent if its circles have mutually
disjoint interiors.
Carrier, carr (P ): The carrier of P is the union of triangles formed by connecting
the centers of the triples of circles in P . In the script, the carrier serves to display
the abstract complex. P is locally univalent if for every interior vertex v the faces in
carr.P / corresponding to the flower of v have mutually disjoint interiors.
Label, R: A label for K is a vector R D frv gv2K of putative radii, one for each
vertex of K. As you might imagine, circle radii are of central importance in circle
packing. They are maintained in CirclePack and can be found using the Info button.
(Radii and center data depend on which of the three geometries you are using.)
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 324 — #342

i i
If you try to create a circle packing by hand, say by manipulating circle objects
in some graphics program, you quickly learn how tightly choreographed their radii
must be. The computational heart of circle packing is computing compatible radii;
the neutral term “label” is used in place of “radii” because the algorithms typically
start with values that do not fit together and then apply iterative adjustments to ap-
proach a solution.
Angle Sums, R : Given a complex K and an associated label R, how might you
determine if circles with those radii would fit together in the pattern prescribed by K?
The key lies with angle sums. For an interior vertex v let fv1 ; v2 ; : : : ; vk g be a list,
in counterclockwise order, of its neighbors. Since the list is closed, it is convenient
to write vkC1 D v1 .
The faces containing v are hv; vj ; vj C1 i; j D 1; 2; : : : ; k. Suppose r , rj ,
rj C1 are the associated labels taken from R. For each face we could place circles
cv ; cj ; cj C1 in the appropriate geometry to form a triple. The triangle formed by
their centers would have edge lengths that are the sums of radii, and we could appeal
to the law of cosines to compute the angle at v in the triangle. In Euclidean geometry,
for instance, it would be
.r C rj /2 C .r C rj C1 /2 .rj C rj C1 /2
˛.r I rj ; rj C1 / D arccos :
2.r C rj /.r C rj C1 /
The sum of the angles at v in all the faces containing v is known as the angle sum:
X
R .v/ D ˛.r I rj ; rj C1z/:
hv;vj ;vj C1 i
The neighbors of v will wrap precisely around v if and only if R .v/ is an integral
multiple of 2 . This is illustrated in Objects, but note that we do not need to place
any circles to compute an angle sum, as computations are done entirely with labels.
Packing Label: A label R for K is known as a packing label if R .v/ is an integral

multiple of 2 for every interior vertex v. If P is a circle packing for K (in one of
our geometries), then the list R of its radii will necessarily be a packing label. The
converse is true for simply connected complexes K: if R is a packing label for K,
then one can lay out circles, using R for radii, to form a circle packing P for K.
CirclePack is a platform for creating, manipulating, analyzing, and displaying circle

packings. Its computational core lies in computing packing labels, and although the pack-
ing algorithms themselves are interesting and revealing mathematics, they are not our target
here. We will instead treat CirclePack as a black box computational engine so that we can
concentrate on the underlying geometry. (Recall that there are several levels of usage for
scripts; the more skill you develop, the more you will learn from experiments: (1) click
NEXT to simply watch the action; (2) press Open Script to follow commands and de-
scriptions; (3) modify the given script commands; (4) add new commands to a script; and
(5) create new scripts from scratch.)
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 325 — #343

i i
6.3. Circle Packing Manipulations 325
6.3 Circle Packing Manipulations

We base our introduction to circle packing on a sequence of scripts revealing the objects
of study and their principal manipulations available within CirclePack.
1. Patterns concentrates on combinatorics. Starting with an empty canvas, a seed
7 command creates the simplest type of packing, namely a single flower — an interior
circle surrounded by, in this case, seven petal circles. As you click through the script the
first time, disregard the occasional repackings that are needed. The emphasis is on building
and adjusting the combinatorics. You will see how to add generations of new neighbors,
how to add and delete boundary circles, how to cut open or zip up a string of boundary
circles, how to double a packing across a boundary segment, how to adjoin two packings,
how to cookie-cut one packing from a larger one, and how the process of hex (hexagonal)
refinement works. Try it out!
2. Geometries demonstrates that CirclePack operates in the three standard geome-
tries: Euclidean, represented by the familiar complex plane C; hyperbolic, represented
by the unit disc D C with the Poincaré metric (density jdsj D 2jdzj=.1 jzj2 /);
and spherical, represented by the Riemann sphere C, namely, the unit sphere f.x; y; z/ W
x 2 Cy 2 Cz 2 D 1g in R3 . (See Appendix B.) The geometries are said to have zero, negative,
or positive curvature, respectively.
The three geometries are nested, D C C, with this last inclusion via stereographic
projection. Moreover, circles in one geometry are circles in the others, though with differ-
ent centers and radii. The script illustrates changes in geometry, displays circles and faces
in each, and shows the effects of automorphisms (i.e., Möbius transformations). The plane
is the most comfortable setting for most of us. The sphere is also quite familiar — we live
on one — but in circle packing it is by far the most challenging. Hyperbolic geometry is
perhaps least familiar, but it is in many ways the richest. One detail to note in hyperbolic
geometry is that we can have circles of infinite radius. Namely, a horocycle in D is a circle
which is internally tangent to the unit circle. It can be treated in a natural way as a circle of
infinite hyperbolic radius whose center is at the point of tangency with the unit circle. Try
it out!
3. Layout may help you understand the repacking/layout process; though we treat
them as black box operations in general, you should see a little about them. In the script,
an embedding of a complex K is displayed without any reference to circles. A randomly
chosen label R is generated (R will be different each time you run the script) and circles are
drawn using the labels as radii. The repack command adjusts the labels iteratively based
on their values and the neighbor relationships encoded in K. (One option for repacking
lets you see the intermediate radii in stages.) Boundary radii (red circles) are fixed, interior
radii (blue circles) change to meet the packing condition, but in the display the centers
are temporarily fixed. Only when the radii are right (i.e., comprise a packing label), are
we ready for the layout command to move the circles to consistent locations. First, the
two circles for some edge are laid out to be tangent to one another; then the rest of the
circles can be placed unambiguously in succession. The original combinatorial pattern is
preserved, but now the locations for its vertices are determined by the circle centers. We
have a packing! Try it out!
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 326 — #344

i i
4. BdryControl illustrates manipulations critical to our later work: solutions of

boundary value problems for radii and boundary angle sums. The first section emphasizes
boundary angle sum manipulations. The script provides tools allowing you with the cursor
and key presses to prescribe boundary angle sums of , =2, or 3=2 at various boundary
vertices; other boundary vertices have unrestrained angle sums that adjust during repack-
ing. The key lesson here is that while you control some things, the packing pushes back on
others — there is a mixture of flexibility and rigidity. Here are some things to think about:
Can you prescribe boundary angle sums arbitrarily? Do you encounter problem situations?
What happens if you prescribe all but one of the boundary angle sums? In this case, can
you compute that remaining one? Can you prescribe angles that will force the region to
overlap itself? Some of these issues will come up in Exercise 6.5.
It is interesting to contemplate the boundary situation in the other geometries as well.
The same script tools work to set boundary angle sums in hyperbolic geometry but you
will see that the constraints of the geometry are different. Try building some quadrilaterals.
Try it out! On the sphere, there is little we or CirclePack can do. As there is no packing
algorithm known in spherical geometry, we can not generate or manipulate boundary val-
ues. (If you study spherical geometry, you can show that there are boundary constraints.
For example, were you to build a spherical quadrilateral you would find that its boundary
angle sums exceeded 2 .)
5. Type567 investigates connections between combinatorics and geometry using
constant-degree packings, those in which every circle has the same degree (the same num-
ber of neighbors). The takehome message involves these intuitive associations: degree six
circles, zero curvature (flat); degree five or less, positive curvature; degree seven or more,
negative curvature. Try it out!
6. Maximal packings are particularly important to our work and are investigated in
three scripts. A maximal packing for a complex K is a univalent circle packing whose
carrier fills the underlying geometric space.
The theorem at the very foundation of circle packing is the Koebe-Andreev-Thurston
(KAT) theorem. (We say K is a sphere when we mean that it is a triangulation of a topo-
logical sphere.)
Theorem 6.2 (KAT). Let K be a sphere. Then there is univalent circle packing PK for K
living on the Riemann sphere, C. Moreover, PK is unique up to Möbius transformations
and inversions of C.
We assume our complexes K and their associated packings P have the same orienta-
tion, so we will not allow inversions. Also, we say a circle packing for K is essentially
unique if it is unique up to conformal automorphisms of the underlying space. In our ter-
minology, then, the KAT Theorem states: Every complex K which triangulates a sphere
has a maximal packing PK of the Riemann sphere, and PK is essentially unique.
If K is simply connected but does not triangulate a sphere, then it necessarily triangu-
lates a topological disc, and the central case, both in theory and computations, is when K
triangulates a closed topological disc (we will say that K is a closed disc). Specifically,
the conditions are that K be finite, simply connected, and have one connected boundary
component (a single chain of boundary edges and vertices).
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 327 — #345

i i
6.4. Discrete Function Theory 327
Theorem 6.3. If K is a closed disc, then there is a univalent packing PK for K in the
hyperbolic plane whose boundary circles are horocycles. This packing is unique up to
Möbius transformations of D.
MaxPackHyp explores this theorem. As this is key to much of our later work, I recom-
mend that you go through this script first. In addition to seeing the practical computations
in action, you might also glean the intuition behind the formal proof.
MaxPackSph should be run next. It illustrates that the KAT Theorem and Theorem 6.3
are equivalent, each implies the other. Since packing computations cannot (yet) be carried
out directly in spherical geometry, the hyperbolic computations are a prerequisite.
MaxPackInf is the third script. If K is simply connected but neither a sphere nor
a closed disc, then it is necessarily infinite (possibly having boundary). This third script
investigates some infinite maximal packings. These are only suggestive, of course, since
one cannot compute with or display infinitely many circles. However, the intuition is theo-
retically sound. (See [1].)
These three scripts provide a quick course in maximal packing. Try it out! Since in
experiments all simply connected complexes K are finite; CirclePack has a single com-
mand max pack (or the droppable icon with the big maize ‘M’) which will automatically
compute a normalized maximal packing PK in the sphere or the unit disc, as appropri-
ate. Maximal packings also exist for non-simply connected complexes; you will consider
multiply-connected cases in Exercise 6.6 and triangulations of the torus in Exercise 6.23.
7. Branching illustrates a further source of control beyond the boundary conditions

we have studied: namely, branching. A circle cv is branched if it is an interior circle whose
neighbors wrap more than once around it. For simple branching, the angle sum at v is 4
and the neighbors wrap twice around it. In general, if the angle sum is 2 n, the neighbors
wrap around n times and the circle is said to have branch order n 1. Investigate the local
geometry at branch circles in the script. Try it out!
8. InputOutput gives you a tour of CirclePack input and output operations. After
you work with CirclePack for a while, you may want to read and write packings, load and
save scripts, create PostScript images and *.jpg screendumps, etc. Try it out!
6.4 Discrete Function Theory

With experience from the previous scripts, you should be ready to investigate discrete
analytic function theory. To see how CirclePack displays functions, repeat Cookie, which
creates the discrete conformal mapping of Figure 6.1.
The script begins with a curve defining a Jordan region . A regular hexagonal packing
H" (i.e., all circles have the same radius ") overlays in the canvas on the right. Using
like a cookie cutter, a packing P of is cut from H" . The associated complex K has
a maximal packing Q D PK in D, which CirclePack computes and displays in the canvas
on the left. We now have, more or less, the situation as pictured in Figure 6.1. This is how
we will represent our functions f W Q ! P in CirclePack: domain on the left, range on
the right.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 328 — #346

i i
Definition 6.4. A discrete analytic function is a map f W Q ! P between circle pack-

ings Q and P that preserves tangency and orientation.
As a practical matter, our discrete analytic functions will have domain and range pack-
ings that share a common complex K (which may vary from one example to another). If v
is a vertex of K with circle Cv in Q, then cv D f .Cv / is the circle for v in P . If hv1 ; v2 ; v3 i
is a positively-oriented face in K, then hCv1 ; Cv2 ; Cv3 i and hcv1 ; cv2 ; cv3 i form mutually
tangent, positively-oriented triples of circles in Q and P , respectively.
Because of Theorem 6.3, we have already generated data for several discrete analytic
functions. In particular, every packing P having a simply connected complex K is the
range of a discrete analytic function: just build its associated maximal packing PK , and
violà, f W PK ! P is a discrete analytic function. You might ask “What is the big deal
about functions?” The function is not about the range (alone); it is about how the domain
changes into the range — it is about the mapping. Which circles grow, which shrink? How
does the boundary behave? Do the circles remain disjoint, or do some come to overlap?
How does the geometry change under f ? Which properties of a packing are preserved?
which ones change?
Subsequent scripts will teach you how CirclePack can aid in generating mappings and
investigating their properties. With domain and range as in Figure 6.1, you can click the
left mouse button on a circle in one packing and both it and its counterpart in the other
packing will be highlighted. Check a few pairs of tangent circles in Q to see that their
counterparts are tangent in P . Also, the circle centered at 0 in Q corresponds to the circle
of P that is also at 0: this is what we mean when we say f .0/ D 0. Clicking the middle
mouse button will highlight corresponding faces of Q and P . CirclePack provides various
buttons for refreshing, zooming, etc., that you should experiment with. Your goal is to be
able to probe the packings to discover properties of the mapping f .
6.4.1 Conformal Maps

The cookie method demonstrated in Cookie is a practical way to create discrete con-
formal mappings, so called because they parallel the conformal (i.e., one-to-one analytic)
mappings of the classical theory.
There is a second approach, modeled on the Schwarz-Christoffel methods seen in Sec-
tion 5.2. The aim is to find a conformal map of the upper half plane or the unit disc to the
region bounded by a given polygonal path .
The classical formula for the unit disc is (5.3). Having such a formula for f suggests
a straightforward computation, but that is not the case. The ˇj are known in advance from
the corner angles of , but the prevertices, the j in the formula, are the points mapping
to those corners, and they are not known. Where does the turning occur? That is the real
challenge with Schwarz-Christoffel. In the discrete version, one confronts this issue head
on. Figure 6.6 illustrates a polygonal packing: the boundary edges of the carrier define
a polygonal curve . The problem, of course, is that in general you start with and a
complex K, not with a packing; you have to find the packing for K whose boundary edges
most closely approximate . See how it goes in the next exercise.
Exercise 6.5. First a little Euclidean geometry. Given a polygonal path , its turning an-
gles reside at the corners: when you travel in the positive direction along and you come
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 329 — #347

i i
qj
bj p
F IGURE 6.6. Packing of a polygonal region.
to a corner, the angle you turn through to start the next edge is by convention taken as
positive if you turn left, negative if you turn right. Compute the algebraic total of all the
turning angles for a simple closed polygonal path .
In Schwarz-Christoffel notation, the j th turning angle is ˇj . We can obtain the bound-
ary angle sum j at the corner by observing (see Figure 6.6) that j D ˇj . Conditions
on the total of turning angles therefore give conditions on the total of all boundary angle
sums: this will constrain your experiments, so you should work this out before jumping
into the script.
DiscreteSC starts with a worked out example for a square. It then challenges you
with a slightly more complicated . You will have to open and read the script to see how
the machinery has been set up and how to manipulate the packing with mouse and key
presses. Try it out! .
Exercise 6.6. The Riemann mapping theorem says that every simply connected proper
subdomain of the plane is conformally equivalent to a particular model simply con-
nected domain, namely, to the open unit disc. This has been extended to certain multiply
connected domains and various model multiply connected domains. For instance, sup-
pose C is a connected open set bounded by n disjoint Jordan curves (a Jordan curve is
a closed curve without self-intersections.). It is known that can be mapped conformally
onto a circle domain, a plane domain bounded by a finite number of circles. In contrast to
the simply connected case, where there are several numerical methods for computing the
classical maps, such as Schwarz-Christoffel, the multiply connected case has few meth-
ods. In the discrete setting, however, maps to circle domains are quite elementary. See
CircleDomains. Try it out!
6.4.2 Basic Function Theory

Let’s look at scripts in which we construct and study additional types of functions. We
start with maps of the unit disc into itself which are important in classical function theory
and convenient for us since we can work in hyperbolic geometry. Schwarz starts with a
somewhat random and small range packing P in D. Incrementally increase the hyperbolic
boundary radii to infinity (in practice, 5 is sufficiently large!) and P morphs into its max-
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 330 — #348

i i
imal packing PK . The map f W PK ! P is a discrete analytic function of D into itself.

We introduce some notation:
Definition 6.7. Let f W Q ! P be a discrete analytic function, where Q and P are
packings with common complex K. The associated ratio function f # W K ! R is defined
by
radius.f .Cv //
f # .v/ D ; v 2 K;
radius.Cv /
where Cv is the circle for v in Q and f .Cv / is the corresponding circle in P .
f # .v/ is the stretch factor at v, the amount that the associated circle is stretched in
going from Q to P . We can now illustrate our first piece of discrete function theory, the
discrete Schwarz lemma.
Theorem 6.8 (Discrete Schwarz Lemma). Let f W PK ! P be a discrete analytic self-
map of the unit disc with f .0/ D 0. Then f # .0/ 1. Equality holds if and only if f is a
rotation; that is, P D PK for some complex with jj D 1.
Exercise 6.9. Run through Schwarz again, but more carefully. Try it out! Can you see
the discrete Schwarz lemma in action? The ratio function plays the role in the discrete
theory that the absolute value of the derivative, jF 0 j, plays in the classical theory, so the
discrete Schwarz lemma precisely parallels its classical counterpart, even up to the equality
statement. (See A.21.)
Exercise 6.10. There is a classical extension of Schwarz’s lemma known as the Dieudonné-
Schwarz lemma. If F W D ! D is analytic and F .0/ D 0, then by Schwarz’s lemma,
jF 0 .0/j 1. According to this extension, there is in fact a constant C; 0 < C < 1, inde-
pendent of F , so that jF 0 .z/j 1 whenever jzj C . Dieudonne sets up experiments so
you can estimate the constant: you manipulate a range packing P in D; with color coding,
CirclePack shows you the circles in PK whose (Euclidean) radii become smaller in P .
How close can these circles come to the origin in PK ? C is larger than 1/2 but quite a bit
smaller than 1. Try it out! (The correct value of C is given at the end of the script.)
The discrete Schwarz lemma is an example of a “monotonicity” result — if you de-
crease the boundary radii of Q to those of P , then the radius of the circle at the origin also
decreases (after repacking). Details are given in [7], but the basic ideas can be seen in:
Exercise 6.11. Consider a mutually tangent triple of circles in Euclidean geometry as
illustrated in Figure 6.7. Using the law of cosines, show that if the radius of circle Cv is
increased, then the angle ˛ at Cv will decrease while the angles ˇ; at the other two circles
will increase.
Exercise 6.12. Using the previous fact, prove:
Theorem 6.13 (Discrete Maximum Principle). Let f W Q ! P be a discrete analytic
function between Euclidean packings Q and P , and suppose Q is locally univalent. Then
the ratio function f # .v/ attains it maximum at a boundary vertex v.
Analogous results hold in hyperbolic geometry and are central to the computation of
packing radii. The failure of monotonicity in spherical geometry may explain the lack of a
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 331 — #349

i i
Cv
a b
F IGURE 6.7. Monotonicity in Euclidean triples.
packing alogorithm there.
Exercise 6.14. Show with a computation in spherical geometry (using the spherical law of
cosines) that the monotonicity of Exercise 6.11 can fail in spherical geometry. What causes
the failure? Try it out!
The analogy between f # and jF 0 j is not perfect. While jF 0 j is zero at branch points of
F , f # can never be zero. Nevertheless, the fact that discrete functions correctly realize the
geometry underlying branching can be seen in the following Exploration.
Exploration 6.15. An important class of classical analytic self-maps of the disc, the finite
Blaschke products, was introduced in Chapter 4. With the explicit product formula shown
in (4.6), a Blaschke product B W D ! D has properties subject to direct computation.
For instance, each factor in the formula accounts for one zero of B, and B extends to a
continuous map on the unit circle, @D, which maps @D n times around itself, where n is
the number of factors in B.
However, there is a more geometric characterization of finite Blaschke products: they
are precisely the proper analytic maps f of D onto itself. “Proper” means that whenever
fzj g is a sequence of points in D converging to the unit circle, so jzj j ! 1, then the images
ff .zj /g also converge to the unit circle, so jf .zj /j ! 1. By orientation, B wraps the
boundary around itself some number n times, and from this the argument principle implies
that B will have n 1 branch points inside the disc. These geometric properties lead you
to discrete finite Blaschke products, as you will see in Blaschke. Try it out! Zooming in
to branch points gives you a chance to demystify their local geometric behavior.
Analytic functions present a fascinating mixture of rigidity within flexibility, which

may be why they have attracted such attention for two centuries now. Though you can
manipulate some features, you also have to live with the consequences. There are two
notions that give some insight into their behavior: harmonic measure and extremal length.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 332 — #350

i i
6.4.3 Harmonic Measure

Let be a Jordan region in the plane (one bounded by a Jordan curve) and let A be a subset
of its boundary, A @. There is a unique harmonic function h defined on which goes
to 1 as z approaches a point of A and goes to 0 as z approaches a point of @nA. (We
suppress technical details about measurability and continuity.) h is known as the harmonic
measure function and written as h.z/ D !.z; A; /. (Notations such as !.z; A; / may
be new to you: this denotes a function whose value depends on three quantities, the point
z 2 , the boundary arc A in @, and the region itself. Hold any of these fixed, and it
becomes a function of the others.)
We also say that !.z; A; / is the harmonic measure of A at z with respect to .
This is called a measure because if you keep z and fixed, then it is a probability
measure on the variable set A. Observe that !.z; A; / 2 Œ0; 1, with !.z; ;; / D 0
and !.z; @; / D 1, and that if A1 ; A2 are disjoint in @, then !.z; A1 [ A2 ; / D
!.z; A1 ; / C !.z; A2 ; /.
There are two nice properties of harmonic measure in the classical setting:
(1) It is sometimes easy to compute: If is the unit disc D and z is the origin, then
!.0; A; D/ D arclength.A/=.2 /. That is, harmonic measure gives the proportion
of the unit circle that A occupies as seen from the origin.
(2) It is a conformal invariant; that is, it is invariant under conformal maps. More pre-
cisely, suppose F W 1 ! 2 is a conformal map (i.e., a conformal homeomor-
phism) between two domains. Under quite weak conditions, it is known that F can
be extended so it also maps the boundary onto the boundary, F W @1 ! @2 .
Given A1 @1, define A2 to be its image A2 D F .A1 / in @2 , and suppose
z1 2 1 and z2 D F .z1 / 2 2 . Then !.z1 ; A1 ; 1/ D !.z2 ; A2 ; 2 /.
If you put these two properties together with the Riemann mapping theorem, you can,
in theory, find the harmonic measure function for any simply connected region. How might
this be useful? Here’s one of many applications. (See Section 3.11.) Suppose a homoge-
neous thin metal plate is cut out in the shape of . Suppose a segment A of the boundary
is held at 100 degrees centigrade, while the remainder of the boundary is held at 0 de-
grees. After the plate has had time to reach thermal equilibrium, so it has a steady state
temperature, what is the temperature at a point z in the interior of the plate? It is evidently
somewhere between 0 and 100, it would approach 100 as z comes closer to A, and 0 as z
gets closer to other parts of the boundary. But what is the temperature at z? The answer is
100 !.z; A; /.
In HarmMeasure we investigate the discrete analogy. I can give the definition now,
since it is motivated directly by property (1). Suppose K triangulates a closed topological
disc, A is a chain of edges in @K, and v 2 K is an interior vertex of K. Let PK be the
maximal packing for K in D that has the circle for v centered at the origin. An edge e 2 A
corresponds to an arc ˛e @D between the centers of the two horocycles forming e.
Definition 6.16. Given K; A; and v as described, the discrete harmonic measure of A at v
with respect to K is X
!.v; A; K/ D arclength.˛e /=.2 /:
e2A
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 333 — #351

i i
So discrete harmonic measure is defined directly on K. The invariance of property (2)

above in the discrete setting is merely due to the definition. In the script, however, you
will see its geometric nature exhibited, much as you would see the temperature distributed
about a plate in the application. Experiments are quite convincing that the same geometry
is at work in both the discrete and the classical settings: in HarmMeasure we discuss
harmonic measure and exit probabilities, comparing the classical and discrete versions for
a region, and we illustrate the extension of domain principle.
6.4.4 Extremal Length

Extremal length is a more nuanced measure of conformal shape. Start with a Jordan region
, but this time suppose we are given disjoint closed subarcs A and B of its boundary; the
triple Q D fI A; Bg is known as a conformal quadrilateral.
A classical theorem states that there exists a conformal mapping F W ! R, where
R is a Euclidean rectangle, so that F extends continuously to the boundary and so that the
images F .A/; F .B/ are opposite ends of the boundary of R. F is unique up to Euclidean
similarities. (You may be reminded of the Schwarz-Christoffel situation, as investigated
earlier, but this is different: the boundary points mapping to corners are specified, but the
size of the image rectangle is not set in advance.)
Figure 6.8 illustrates the analogous discrete result you will see in ExtLength. By
a rigid motion, image rectangles can be positioned as shown here, with the images of
A and B on the left and right, respectively. The aspect ratio of R is width over height,
Aspect.R/ D W=H .
W
F IGURE 6.8. Packing for extremal length.
Continuing our discussion of the classical setting, Aspect.R/ is known as the extremal
distance from A to B, relative to , or the extremal length of the quadrilateral Q, denoted
EL.Q/. As with harmonic measure, there are two key properties:
(i) Extremal length is sometimes easy to compute: If is a Euclidean rectangle width
W and height H , and A and B are its ends, then EL.f; A; Bg/ D W=H .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 334 — #352

i i
(ii) Extremal length is a conformal invariant; that is, if Q1 D f1 I A1 ; B1g and Q2 D
f2 I A2; B2 g are conformal quadrilaterals and F W Q1 ! Q2 is a conformal map-
ping such that F .A1 / D A2 and F .B1 / D B2 , then EL.Q1 / DEL.Q2 /
As with harmonic measure, there are physical interpretations. Suppose is a uniform
conducting metal plate in the plane, the boundary segment A is attached to a 1 volt power
supply, the boundary segment B is grounded, and the remaining segments of the boundary
are insulated. Then the current flow between A and B in amperes will be proportional to
EL.f; A; Bg/, the constant depending on the properties of the metal.
Discrete analytic mappings to rectangles having designated boundary vertices as cor-
ners are easy to compute by methods you have already seen in Layout: you specify that
boundary vertices have angle sum except for the four that are to be the corners, and
these you set to =2. ExtLength illustrates the construction of Figure 6.8 and explains
how CirclePack reports the extremal length when you have a rectangle. You are asked to
modify the constructions to answer a question about current flow. You might then test your
intuition with an extremal length analogue of the extension of domain principle we saw
with harmonic functions.
One last script for this section, MiscFtnTheory, illustrates the range of miscella-
neous behaviors you might encounter with discrete analytic functions. They may provide
additional insights you can carry back to the classical setting.
6.5 Function Construction

We have hands-on control of our discrete functions, but this is not the formula-driven and
algebraic control available in the classical setting. In the discrete world, we have limited
ability to add or multiply functions, and essentially no way to consider compositions of
the type studied in Chapter 1. Discrete functions require some geometric hook. We have
seen examples: to get maximal packings in D, we send boundary radii to infinity; to build
Euclidean polygons, we control their boundary angle sums; to create Blaschke products,
we designate branch circles. The next sequence of scripts give additional examples, some
involving techniques, others relying on special circumstances or outright tricks. There is
plenty of room for your creativity here.
6.5.1 Discrete Exponentials

Doyle challenges you to adjust parameters in a family of intriguing circle packings known
as Doyle spirals. With patience you can build a range packing that clearly mimics the
classical exponential function. Try it out! Read the script for guidance, but then experiment
on your own. Examples of good parameter values are provided at the end of the script.
6.5.2 Discrete Rational Functions

A classical rational function has the form F .z/ D P .z/=Q.z/, where P .z/ and Q.z/ are
complex polynomials with no common factors. Projecting stereographically to the sphere
in both domain and range, F may be considered as a map from the Riemann sphere to
itself, F W C ! C, and that will be our view. (See Definition B.11.) If P .z/ D 0, then
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 335 — #353

i i
6.5. Function Construction 335
F .z/ D 0, while if Q.z/ D 0, then F .z/ D 1. The behavior at 1 in the domain depends
on relative degrees: if deg.P / > deg.Q/, then F .1/ D 1, if deg.P / < deg.Q/, then
F .1/ D 0, and if deg.P / D deg.Q/, F .1/ is a non-zero complex number.
The sphere is compact, and F W C ! C is an open continuous map, so key properties
of rational maps (including polynomials) are purely topological: F has a constant valence,
N > 0; that is, for a point p 2 C, F 1 .p/ has N points (counting multiplicities at branch
points, which may include infinity). The valence N satisfies N D maxfdeg.P /, deg.Q/g.
There are precisely 2.N 1/ branch points. As you build examples you should watch for
these properties directly.
Polynomial gives a general method for creating discrete polynomials in the plane.
Treating them as rational maps, however, suggests a broader notion: A map of the form
F D ı P , where P is a polynomial and is a Möbius transformation, is said to be a
polynomial-like rational map. These are characterized by having half of their branching at
a single point, and this is exploited in PolyLike to build examples.
To get a taste for more typical rational maps, Rational illustrates an example con-
structed using Schwarz triangles, with valence seven and with twelve simple branch points.
This is the discrete analogue of the classical rational function
F .z/ D z 2 .z 5 7/=.7z 5 C 1/:
[This formula was found by Edward Crane; the one given in [2] and [7] is not correct.]
Lovely as this discrete example is, it highlights a key capability that is missing in our
discrete theory: we have no algorithm for computing packing labels directly in spherical
geometry. See Exercise 6.25 at the end of the chapter for more on the geometry behind this
Schwarz triangle construction.
6.5.3 Ratio Function Constructions

If f W Q ! P is a discrete analytic function, then the associated local scaling, the ana-
logue of jF 0 j for analytic functions, is given by its ratio function f # . (See Definition 6.7.)
If we wish to build a discrete analogue f for some classical function F , we might begin
from this differential level, mimicking jF 0 j.
RatioBuild builds a discrete analogue for the classical function F .z/ D .1 C z/2 on
the unit disc. You should compare the mapping image the script generates to Figure 4.12,
page 208, created using the applet ComplexTool. Modify the script to build the other ana-
lytic examples illustrated in Chapter 4.
Ratio function constructions also give an approach to entire functions. We illustrate in
Erf a function central to statistics, the error function erf .z/ defined by
Z z
2 2
E.z/ D erf .z/ D p e d :
0
When z is real, z D x, the derivative of E.z/ is the Gaussian distribution, the famous
bell-shaped curve. You may recall that this function has no closed form anti-derivative,
hence the integral representation.
The construction of the discrete error function illustrates several construction issues.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 336 — #354

i i
(1) E is an entire function defined on the whole complex plane. Discrete entire func-
tions would likewise be defined on infinite complexes, as the discrete exponential
we defined earlier on the regular hexagonal packing. Experiments and computations,
however, are restricted to finite complexes. Our approach relies on the method of ex-
haustion: we mimic E using finite packings that take up successively larger, but
finite, portions of C.
(2) E has infinite valence: for every w 2 C, E 1 fwg has infinitely many points. In fact,
its mapping behavior is fascinating to watch. Early in the construction, it may appear
that the image omits two points, w1 ; w2, much as the exponential omits the image
value 0. That, however, would contradict the little Picard theorem,which tells us that
a nonconstant entire function can omit at most one value in the complex plane. Watch
more stages in the construction to learn how our growing function manages to cover
everything. Though not omitted, w1 and w2 are what is known as asymptotic values,
that is, there are paths in the domain whose images under E converge to w1 or w2 .
Can you find such paths in the CirclePack experiments? Try it out!
(3) Does the sequence of packings in the method of exhaustion ultimately converge to an
infinite circle packing? Each stage of construction is finite, so this is a convergence
issue that remains unresolved. The script has both positive and negative indications.
What do you think? If there is a limit packing, this construction would answer in the
negative one of the earliest questions in circle packing:
Question 1. Are the regular hexagonal packings and the Doyle spirals (see Section 6.5.1)
the only locally univalent hexagonal circle packings in the plane?
6.5.4 Harmonic Functions

According to Theorem 4.31, harmonic mappings F on the unit disc may be written as
F .z/ D H.z/ C G.z/ where H and G are analytic in D. While the addition occurring here
is not a geometric feature which can be carried directly by our circles, there is a discrete
fact that captures the local behavior that we will discuss in a moment. The upshot is that
if one just jumps in with discrete functions, the fundamental global phenomena appear to
persist. In this section we start with the mechanics, see some examples parallel to ones in
Chapter 4, then set up a few lines of open experimentation for you.
A run through HGintro shows you how to create discrete harmonic functions. Try it
out! The way CirclePack operates, h and g will be discrete analytic functions on the disc,
so within each example they will share a complex K and have its maximal packing PK as
their common domain. There will be no need to view PK , but in HGintro h is taken as
the identity function, so PK itself appears in p0 as the range of h. (“p0” refers to Pack 0
in CirclePack.) The image of g appears in p1. On execution of h g bar, the sum h C g
is displayed in p2. CirclePack requires that the range packings for h and g be Euclidean,
and a key feature in HGintro is that their boundary radii are identical. The script starts
with g also the identity, so the initial f is boring. However, if you press the letter b while
the cursor is on an interior circle of h, you will cause the circle to become branched for g.
The image of f is immediately more interesting. With additional branch points you will
observe some of the behavior described in Chapter 4. Since the boundary radii are identical
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 337 — #355

i i
6.6. Convergence 337
for h and g, the complex dilatation function w.z/ is a discrete finite Blaschke product (see
Sections 4.6 and 5.7).
The CirclePack mechanics behind h g bar are straightforward: for each vertex, the
center in p0 is added to the conjugate of the center in p1, giving a location that is plotted
in p2. With the centers in place, the edges are drawn in p2 to give the triangulation. In
FtnAddition we display, strictly for purposes of comparison, the analogous addition
h C g (no conjugation bar on g). The result f D h C g in p2 will approximate the
classical sum of these analytic functions. However, there is no sensible way to assign radii
in p2 and there is no packing. So in this algebraic situation, the circles are no longer doing
the work for you.
A last script, HGtrials, sets up various adjustments you can implement with mouse
and keys for more open-ended experiments. You can also add actions of your own devising.
Certain keys will increase boundary radii of g, which introduces you to another feature:
faces of f (in p2) are colored blue if f reverses their orientation. We know from the
classical theory that this can happen only if jG 0 j=jH 0 j > 1 on some part of @D. In the
discrete setting, absolute values of derivatives are replaced by ratio functions; if you return
to the definition of the ratio function and recall that h and g share the same domain packing,
you see that the discrete condition becomes g# = h# > 1 on some vertices of @K. Therefore,
by manipulating the relative sizes of the boundary circles in p0 and p1, we affect the
orientation of f D h C g.
Exercise 6.17. Verify this bit of local geometry, which may account for the properties
of these discrete harmonic functions. Consider Euclidean triangles T and t with vertices
located at fZ1; Z2 ; Z3g and fz1 ; z2 ; z3 g, respectively. Assume that each is positively ori-
ented. Form a new triangle with vertices fw1 ; w2; w3g, where wj D Zj Czj ; j D 1; 2; 3:
(The bar denotes complex conjugation of zj .) The issue is whether is positively oriented.
In general, the answer is “no”, but here is a pretty fact that you should try to prove:
If T and t happen to be formed by mutually tangent triples of circles fC1 ; C2 ; C3 g and
fc1; c2 ; c3g, and if radius.cj /=radius.Cj / 1; j D 1; 2; 3, then is indeed positively
oriented. In other words, the classical boundary condition for preserving orientability of
H C G, namely jG 0 j=jH 0 j < 1, carries over to the discrete case at the local face level.
Use your favorite computer math package to build examples. Show first that can
have reverse orientation in general. Even more is true: show that even if you arrange that
the side lengths of t are individually less than the corresponding side lengths of T , may
still have reverse orientation. Mysteriously, when the side lengths are sums of radii, and
the radii for t are individually less than those for T , then remains positively oriented!
6.6 Convergence
We have introduced the local geometry of circle packings, moved on to defining mappings
between packings, and investigated the emergent global geometric properties. The claim
has been made — with convincing evidence, I hope — that these mappings deserve the
name of discrete analytic function. Can we move beyond mere parallels?
The answer is a resounding yes. Though granularity prevents discrete analytic functions
from actually being analytic functions, with a suitable refinement process for improving
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 338 — #356

i i
resolution they can be made to approximate analytic functions. Circle packing began with
exactly this approximation issue with the celebrated Rodin-Sullivan theorem explained
below. Let’s see it in action first.
DiscreteRMT picks up from our earlier script Cookie. It builds a range packing P
using cookie to cut a region from a regular hexagonal packing. The maximal packing
Q having the same combinatorics becomes the domain. A standard CirclePack normaliza-
tion puts the circle for the ˛ vertex at the origin in both Q and P , and a vertex on the
imaginary axis in each. The map f W Q ! P is then a discrete conformal map.
At issue: How well do the discrete mappings approximate the classical conformal map-
ping F W D ! guaranteed by the Riemann mapping theorem? This script provides op-
tions to change the radii in the hexagonal overlay. Making those circles smaller — hence
using more of them — is an example of refinement. The effectiveness of such refinements
is the content of the foundational theorem by Burt Rodin and Dennis Sullivan, paraphrased
here.
Theorem 6.18 (Rodin and Sullivan, [6]). Let be a bounded simply connected domain
in the plane and for each " > 0 let f" be the discrete conformal mapping created as in
the script, based on a regular hexagonal packing with circles of radius ". Let F W D !
be the classical conformal mapping with the same normalization. Then the functions f"
converge to F uniformly on compact subsets of D as " ! 0. Moreover, the ratio functions
f # converge uniformly on compact subsets of D to the modulus of the classical derivative,
jF 0 j.
The script illustrates the quality of approximation in two ways. One is self explanatory,
imposing a spoke-and-wheel grid on the domain disc and having the discrete analytic func-
tion carry it forward to . This result is compared in Figure 6.9 with the analogous result
computed using Toby Driscoll’s Schwarz-Christoffel software ([5]). (These markings are
related to the flow lines employed in Chapter 3.)
A second test of quality is illustrated in the script by using color-coding. Treat both
Q and P as Euclidean packings and pay attention to their carriers. We can think of f W
Q ! P as a more familiar type of function; namely, as a piecewise affine function
F IGURE 6.9. Grid lines from two approximations of a conformal mapping: the left was computed
using Schwarz-Christoffel and the right using CirclePack.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 339 — #357

i i
6.6. Convergence 339
f W carr.Q/ ! carr.P / mapping each face of carr.Q/ affinely onto the corresponding
face of carr.P /, with the vertices (circle centers) mapped to corresponding vertices. The
map may be defined using barycentric coordinates: if T1 is a face of carr.Q/ and t1 the
corresponding face of carr.P /, then for appropriate real values a; b; c; d; x0; y0, f has the
form
f .x; y/ D .ax C by C x0 ; cx C dy C y0 /; .x; y/ 2 T1 :
p
This is quasiconformal on T1 with dilatation given by k D w C w 2 1, where w D
.a2 C b 2 C c 2 C d 2 /=.2.ad bc//. Note that k 1; moreover, f is conformal (analytic)
on T1 if and only if k D 1, that is, if and only if T1 and t1 are similar triangles. The larger
k becomes, the further f is from being conformal. The experiments in DiscreteRMT
color-code the faces in the range based on k. Faces are white when k D 1 and become
more red as k gets larger up to some cutoff that you can set. Faces with dilatation larger
than the cutoff are blue. In DiscreteRMT, commands have been set up to display both
grids and the color coding: run the script with increasing refinement levels to see what you
can say about the affects on k.
The convergence of discrete objects to their classical counterparts seen in the Rodin-
Sullivan theorem holds more generally, indeed, in essentially all the settings we have dis-
cussed in this chapter. First, a few cautionary words on increasing the complexity:
(a) Computational penalty: repacking and display times can increase dramatically.
(b) Roundoff error: errors accumulate, interfering with layout, display, and other opera-
tions.
(c) Display crowding: objects and indices are harder to see, harder to pick out with the
mouse, and display updates are slower.
Nevertheless, CirclePack can handle packings with hundreds of thousands of circles, so
give the following scripts a try, noting that a few ill-advised mouse clicks can push you
into computational neverland — but of course, you can always kill CirclePack and restart!
Exercise 6.19. RandomRefine illustrates the use of random packings in place of reg-
ular hexagonal packings. You can create your own simply connected region and then
apply the commands developed in the script to create and investigate the resulting discrete
conformal mappings. Refinement is accomplished by increasing the number of circles in
the random packing.
Exercise 6.20. MultiRefine illustrates maps between multiply connected regions, as
in CircleDomains. If you want to modify the region , note that each component of
@ requires its own PATH file.
Exercise 6.21. Run ErfRefine to pursue hex refined packings for the error function
encountered in Erf (see [7, 14.4]). One method for judging how closely the results mimic
the classical erf .z/ is to compare the images of the unit circle under the discrete map and
the classical map. Methods are given in the script to display the former. However, you
should use a math package to generate the latter; the script explains how to read and
display a path given as a list of x; y coordinates.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 340 — #358

i i
Exercise 6.22. This exercise sets a simple task: estimate the complex number B.3i=4/
for a classical finite Blaschke product B. Here is what you know about B: B.0/ D 0 and
B.1/ D 1; B has three zeros (counting multiplicities); and B branches (i.e., B 0 vanishes)
at z1 D .1 C i /=2 and z2 D 1=2. It is known that appropriate discrete finite Blaschke
products bn will converge uniformly on compact subset of D to B (an extension of the
Rodin-Sullivan theorem to branched packings). BlaschkeRefine allows you to create
discrete functions bn (at various refinement levels) based solely on this type of data. You
will have to tailor the construction and normalizations to match this particular B. Then you
are asked to apply two methods for estimating B.3i=4/: (1) Use the values bn .3i=4/ for
the bn you constructed. (2) Locate the zeros of the bn , use those to build a classical finite
Blaschke product B b close to B, then compute B.3i=4/
b in a math package. The point is
that classical Blaschke products are best constructed from their zeros, while their discrete
analogues are best constructed from their branch points.
Nearly every script from Sections 6.4 and 6.5 involves packings that can be refined
as we discussed, and invariably the discrete objects will converge under the refinement to
their classical counterparts. The reader should pick a topic and create a script that imple-
ments appropriate refinement: discrete exponentials, discrete rational maps, discrete ratio
function constructions, discrete harmonic functions — each project will have its distinct
character.
6.7 Wrapup
I hope you have enjoyed this chapter. Perhaps it has given you a new way to look at and
think about analytic functions. As discrete theories go, circle packing is arguably unique in
the breadth, depth, and fidelity with which it captures its subject. The essential phenomena
appear with even the simplest of packings, and you seem to glimpse the heart of analyt-
icity. As complexity increases, the fidelity only grows, to the point that you can reliably
investigate — perhaps even discover new phenomena — in the classical theory. So it is
a full-featured quantum theory, classical in the limit, with experimental and visualization
capabilities to boot.
This chapter may also have introduced you to a topic of interest in its own right. There
is a lot of room for growth in circle packing: packings on surfaces, packings of random tri-
angulations, graph embedding, and a growing list of applications. Wherever circle packing
goes, there, too, will be complex analysis!

Here are three additional exercises to stretch your understanding of circle packing. The first
challenge in each is understanding the mechanisms being illustrated in the scripts. After
that you may see new directions for experiments, and you are encouraged to set your own
path of discovery.
Exercise 6.23. Seeing the connection between triangulations and circle packings may have
piqued your curiosity about more general triangulations. Until now, most of our complexes
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 341 — #359

i i
have been simply connected, but Torus is different: K triangulates a torus. A torus is
topologically equivalent to the surface of an inner tube, so you would rightly guess that we
cannot realize a circle packing for K on the plane or on the sphere. Yet the script presents
you with the “packing” P shown on the left in Figure 6.10.
F IGURE 6.10. Circle packing of a torus and its universal covering with color-coded side pairings.
Interpretation of P rests on the color coded side-pairings: The 32 shaded circles form
the packing, but additional tangencies exist on the torus. Since we cannot realize them
in the plane, we create ghost circles (the unshaded ones on the left): the ghost circles
centered along one colored edge are identified with corresponding circles along the other
edge sharing that color. You may be able to see the matchings in the image either from the
combinatorics or the radii, but the script should help you explore. The image in the right of
Figure 6.10 presents this in another fashion; translated copies of the packing tile the plane.
It is up to you to come to terms with the representation for circle packings of tori. It may
help to create your own triangulated torus and insert it into the script data section in place
of the initial packing. Encoding of complexes is described in the Formats tab of the Help
Frame of CirclePack. Let me offer one specific challenge: create a combinatorial torus K
having just seven vertices — the smallest number possible for a proper triangulation (that
is, one in which two faces are disjoint, share a single vertex, or share an edge and its end
vertices). This may be a little harder than you expect.
Exercise 6.24. An experimental capability of CirclePack that we touched on briefly in
Patterns involves the adjoin operation for triangulations. Observe that when two
simply connected triangulations K1 and K2 have the same number of boundary vertices,
they can be pasted together along their boundaries. The result is a new triangulation K of
the sphere. A packing P for K in C then provides simultaneous packings for K1 and K2
and gives an embedding for their common boundary curve, the so-called welding curve, on
the sphere.
The basics are demonstrated with Owl in OwlWelding. Interest lies in how the shape
of the welding curve is determined by our choices in attaching K1 and K2 . There is a
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 342 — #360

i i
connection with conformality related to a classical topic called conformal welding (see
[10]). Learn about the double and adjoin operations, and let the experiments begin.
Exercise 6.25. The CirclePack algorithms for computing circle packings work in Eu-
clidean and hyperbolic geometry. Unfortunately, there is (as yet) no general algorithm
that works directly in spherical geometry, and the spherical packings in previous scripts
have been stereographic projections from the plane or disc. All is not lost, however. At
least a smattering of examples can be created by hand using basic spherical geometry in
conjunction with combinatorial symmetries. Figure 6.11 illustrates an example with the
combinatorics of the soccer ball: the left side is the maximal packing, while the right is
a branched packing having the same combinatorics. The 12 shaded circles are the branch
circles. Investigate this further in SoccerPack. Try it out! .
Q P
F IGURE 6.11. Circle packings with soccer ball combinatorics.
Both packings in Figure 6.11 are constructed using the geometry of Schwarz triangles.
The univalent packing Q shows the symmetry of the regular dodecahedron: twelve sym-
metrically distributed circles each with five neighbors, the remaining twenty circles each
with six neighbors. (This packing can also be created by standard methods in CirclePack;
see MaxPackSph). The packing P shares the same combinatorics, but all degree five cir-
cles are branch circles, so P defines a 7-fold covering of the sphere. The map f W Q ! P
is thus a discrete rational function with twelve simple branch points and valence seven.
Both Q and P can be created using the spherical triangles of Figure 6.12, two of
several triangles known as Schwarz triangles. The Schwarz h2; 3; 5i triangle t (i.e., one
having angles =2; =3; and =5) is pictured in Figure 6.12(a). A second triangle, the
t T
F IGURE 6.12. Schwarz triangles t and T .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 343 — #361

i i
h2; 3; 5=2i Schwarz triangle T , is shown in (b) (with a different scale). Dashed lines on T
show how it is built from seven copies of t. A pattern of two circular arcs is imprinted on
t in (a) and the analogous pattern is imprinted on T in (b).
A Schwarz triangle can be reflected across any of its edges to form a copy of itself with
reverse orientation. Repeated reflections will tile the sphere; for t this involves 120 non-
overlapping copies. Carrying along the imprinted arcs during the reflection process gives us
a univalent circle packing with 32 circles, our maximal packing Q. The identical reflection
process, starting with T , gives 120 copies which tile the sphere seven layers deep and
whose imprinted arcs give our packing P . In any copy of T , it takes ten reflections around
the vertex with angle 2=5 to close up, and this is what happens at each of the twelve
branch points of P . The script SoccerPack lets you check a branch circle visually to see
the five (huge) neighboring circles wrap twice around it.
Schwarz triangles were first used to generate branched circle packings of the sphere in
[2] (see Appendix H of [7]). However, the example of Figure 6.13 and other equally beau-
tiful examples were created by Samantha Corvino, one of my REU (Research Experience
for Undergraduates) students.
F IGURE 6.13. A branched packing due to Samantha Corvino.
There are a handful of additional Schwarz triangles and other options for circular mark-
ings, so the challenge to you is to create more examples. For each, you will need to encode
the combinatorics K for CirclePack, which can then generate the associated maximal pack-
ing. However, CirclePack cannot find the radii and centers for the branched packing. For
this you will need to apply some spherical trigonometry, record the results in a packing file
along with K, and have CirclePack display them. To get you started, I’ve put another of
Samantha’s examples with more details in Tri432. Try it out!
6.9 Bibliography
[1] Alan F. Beardon and Kenneth Stephenson, Circle packings in different geometries,
Tohoku Math. J., 43 (1991), 27–36.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 344 — #362

i i
[2] Philip L. Bowers and Kenneth Stephenson, A branched Andreev-Thurston Theorem

for circle packings of the sphere, Proc. London Math. Soc., 73 (1996), 185–215.
[3] Charles R. Collins and Kenneth Stephenson, A circle packing algorithm, Computa-
tional Geometry: Theory and Applications 25 (2003), 233–256.
[4] Tobin A. Driscoll and Lloyd N. Trefethen, Schwarz-Christoffel Mapping, Camb. Univ.
Press, New York, 2002.
[5] Tobin Driscoll, Schwarz-Christoffel Toolbox (for Matlab),
http://www.math.udel.edu/driscoll/software/SC/.
[6] Burt Rodin and Dennis Sullivan, The convergence of circle packings to the Riemann
mapping, J. Differential Geometry 26 (1987), 349–360.
[7] Kenneth Stephenson, Introduction to Circle Packing: the Theory of Discrete Analytic
Functions, Camb. Univ. Press, New York, 2005.
[8] Kenneth Stephenson, Circle packing: A mathematical tale, Notices Amer. Math. Soc.
50 (2003), no. 11, 1376–1388.
[9] William Thurston, The finite Riemann mapping theorem, Invited talk, An International
Symposium at Purdue University in celebration of de Branges’ proof of the Bieberbach
conjecture, March 1985 (unpublished).
[10] G. Brock Williams, Discrete conformal welding, Indiana Univ. Math. J., 53 (2004),
765–804.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 345 — #363

i i
Background
A
This appendix summarizes some major topics and theorems from a standard undergradu-
ate complex analysis course. Since it is a review, we do not prove results; for more details
you can look to reference texts such as [1] and [3], and occasionally to the more advanced
text [2]. A few advanced topics are also summarized here, including the Riemann map-
ping theorem, the open mapping theorem, and Schwarz’s lemma. The extended notion of
analyticity on the Riemann sphere C is addressed separately in Appendix B.
A.1 Functions of a Complex Variable as Mappings

p
Let R denote the real numbers, i the imaginary number 1, and C D fx Ciy W x; y 2 Rg
the complex numbers. (We employ standard notations which the reader can find in the
Index of Notation on page 365.)
A complex-valued function f defined on a set C is denoted f W ! C. We
often denote the domain set of f by domain.f /, and express z D x C iy D .x; y/ and
f .z/ D f .x; y/ with the implicit understanding that x D Re.z/ and y D Im.z/. The term
“mapping”, synonymous with “function”, is often used in geometric settings such as the
complex plane, and tempting as it may be to treat f as mapping two real variables to two
real variables, it is far better to work with it as a complex function of a single complex
variable. The set f ./ D ff .z/ W z 2 g is called the range of f . If A C, we denote
the inverse image (or preimage) of A under f by f 1 .A/ D fz 2 W f .z/ 2 Ag.
A.1.1 Linear Functions

We can examine the linear mappings f .z/ D azCb, for constants a; b 2 C, by considering
the simpler functions h.z/ D az and g.z/ D z C b.
Writing z D r e i and a D jaje i ˛ in polar form, we can express the image point as
h.z/ D az D r jaje i.C˛/ . Hence, h maps z to the point h.z/, which geometrically is
a stretching or contraction of the modulus by jaj and a rotation about the origin by the
angle ˛. Follow this with the action of g.z/ D z C b, which translates a point by jbj
units in the direction of the vector b. The result is the geometric action of the composition
f .z/ D az C b D g.h.z//. See Figure A.1 for an example.
A.1.2 Power Functions

Let f .z/ D z n for fixed n 2 N, where N denotes the natural numbers. Writing z D r e i in
polar form, we see that f .z/ D .r e i /n D r n e i n ; that is, under the action of f the point
345
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 346 — #364

i i
346 Appendix A. Background
5 Rotate by Arg(a) = p/4.

Translate by b = i.
2.5 Stretch by factor

of |a| = 2.
5 2.5 2.5 5
F IGURE A.1. An illustration of the geometric action of f .z/ D az C b for a D 2ei =4 and b D i .
z has its modulus raised to the nth power and its argument multiplied by n. See Figure A.2
for an example.
2 2
1 1
z a z3
–2 –1 1 2 –2 –1 1 2
–1 –1
–2 –2
F IGURE A.2. An illustration of the geometric action of z 7! z 3 .
Roots of unity
For a positive integer n, we call any of the n solutions to the equation z n D 1 an nth
root of unity. These have the form !k D e 2k i=n for k D 0; : : : ; n 1, and written as
1; !1 ; !12 ; : : : ; !1n 1 , we see they are equally spaced points on the unit circle jzj D 1.
A.1.3 The Exponential Function

The definition of the exponential function arises naturally out of Euler’s formula
e i D cos C i sin .
Definition A.1. If z D x C iy, then
e z D e x .cos y C i sin y/:
The definition has two important consequences:

(a) The exponential function maps C onto Cnf0g.
(b) The exponential function is periodic with period 2 i .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 347 — #365

i i
A.1. Functions of a Complex Variable as Mappings 347
A.1.4 The Logarithm Function

The logarithm function provides an inverse of the exponential function. However, since the
exponential function is not one-to-one, special consideration must be taken to understand
in what sense the logarithm is an inverse. We begin by first understanding arg.z/.
Definition A.2. For a complex number z ¤ 0, the argument of z, denoted arg.z/, is the set
of all 2 R such that z D jzje i . The principal value of the argument, denoted Arg.z/, is
the unique such angle with < .
Definition A.3. The logarithm function log.z/ is the multiple-valued function
log.r e i / D ln.r / C i;
where ln.r / denotes the real-valued logarithm. We also write
log.z/ D ln jzj C i arg.z/:
Since log.z/ is multiple valued (because arg.z/ is multiple valued), it is not a func-
tion in the traditional sense. To make it a function, we can use the principal value of the
argument, Arg.z/.
Definition A.4. The principal value of the logarithm is Log.z/ where
Log.z/ D ln jzj C i Arg(z):
It has domain set Cnf0g and range < Im.w/ , but fails to be continuous at any
point of the negative real axis.
Technology Note: We use Log.z/ and Arg.z/ to denote the principal values of the loga-
rithm and argument functions, but the applet ComplexTool uses log.z/ and arg.z/ to express
them.
Remark A.5. While the principal branch of the logarithm is the one most widely used,
it is possible to define other branches of the logarithm function that are continuous, for
example, along the negative real axis. Any branch of the logarithm must involve a branch
cut, a line of discontinuities running from infinity to 0. More information is in [1] and [3].
A.1.5 Trigonometric Functions

The trigonometric functions of a complex number z are defined in terms of the exponential
function.
Definition A.6. Given a complex number z, we define
e iz e iz
e iz C e iz
sin z D and cos z D :
2i 2
We can show that the image of the infinite vertical strip =2 Re z =2 under
either map is the entire plane, so these trigonometric functions are unbounded in C.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 348 — #366

i i
A.2 Continuity and Analyticity in C

Definition A.7. Let C and consider a complex-valued function f W ! C.
(a) We say that f is continuous at z0 2 if limz!z0 f .z/ D f .z0 /; i.e., for each " > 0
there exists ı > 0 such that jf .z/ f .z0 /j < " whenever z 2 and jz z0 j < ı. We
say f is continuous on a set U if it is continuous at each point of U .
f .z/ f .z0 /
(b) We say that f is differentiable at z0 when its derivative f 0 .z0 / D limz!z0 z z0
exists.
(c) When a function is differentiable at all points of an open set , it is said to be analytic
on . (Some authors refer to such a function as holomorphic in .)
(d) The term domain refers to an open connected subset of the plane (or of the Riemann
sphere), since these are the natural domain sets for analytic functions.
A.2.1 Cauchy-Riemann Equations

Writing f .z/ D f .x; y/ D u.x; y/ C iv.x; y/ with u.x; y/ D Re f .x; y/ and v.x; y/ D
Im f .x; y/, we have the partial derivatives
@f @u @v @f @u @v
D Ci and D Ci : (A.1)
@x @x @x @y @y @y
When f is differentiable at z0 , we have f 0 .z0 / D @f .z / D i @f

@x 0
.z /, which by
@y 0
equating real and imaginary parts yields the Cauchy-Riemann equations
@u @v @u @v
D and D : (A.2)
@x @y @y @x
Theorem A.8. A function f .z/ D u.x; y/ C iv.x; y/ is analytic in an open set if

and only if the first partial derivatives ux ; uy ; vx ; vy exist, are continuous, and satisfy the
Cauchy-Riemann equations.
Using the Cauchy-Riemann equations (A.2), we can show that if f .z/ D u.x; y/ C
iv.x; y/ is analytic in an open set , then the component functions u and v are harmonic
in , that is, uxx C uyy D 0 and vxx C vyy D 0.
A.2.2 Conformal Mappings

A function f W ! C is called univalent if it is one-to-one; i.e., if z1 ¤ z2 in ,
then f .z1 / ¤ f .z2 /. A function that is both analytic and univalent on an open set is
said to be conformal there. Geometrically, conformal means that the function is locally
angle-preserving, preserving both the magnitude and sense of the angle between intersect-
ing curves. You can get a feel for this by graphing a univalent analytic function with the
ComplexTool applet, using either a rectangular or circular grid, and zooming in on points
in the range to see that the right angles where grid lines intersect are preserved. An analytic
map f is locally univalent at z0 2 when its restriction to some neighborhood of z0 is
univalent. As discussed in Section A.6.1, this is the case if and only if f 0 .z0 / ¤ 0.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 349 — #367

i i
A.3. Contour Integration 349
A.3 Contour Integration

Contour integration is a form of line integration and involves expressions of the form
R
C f .z/ dz. Here C is a curve in the plane and f is a complex function defined on C .
A.3.1 Planar Contours

A path in the plane is defined as a continuous map W Œa; b ! C where Œa; b is a
parameter interval on the real line. It suffices to consider piecewise differentiable functions
, and since the parameterization is generally irrelevant in computations, the term contour
(also, curve) will be used in place of path for the trace C (i.e., the image of ).
Definition A.9. Suppose W Œa; b ! C defines a piecewise smooth contour C in the
plane.
(a) Points z0 D .a/ and z1 D .b/ are known as the initial and terminal points of C ; the
orientation of the contour is from z0 to z1 .
(b) C is a closed contour if its initial and terminal points coincide.
(c) C is a simple closed contour if it is a closed contour that does not cross itself. It is
positively oriented if the inside is on the left as you travel around the contour.
See Appendix B.3 for information regarding topology on both the plane and the Rie-
mann sphere.
A.3.2 Examples
The value of a contour integral on C is independent of the continuous function which is
used to parameterize C . In practice, of course, some parameterization W Œa; b ! C must
be chosen, then the integral is calculated as
Z Z b
f .z/ dz D f . .t// 0 .t/ dt: (A.3)
C t Da
H
When C is closed and positively oriented, the special integral sign is often used to
emphasize that fact.
Here are two examples.
(a) The line segment C from z0 to z1 can be parameterized by .t/ D .1 t/z0 C tz1 ,
where 0 t 1. Thus,
Z Z 1
f .z/ dz D f ..1 t/z0 C tz1 / .z1 z0 / dt:
C 0
(b) The contour C defined as wrapping once counterclockwise around the circle of radius
R centered at a can be parameterized as .t/ D a C R e i t , where 0 t 2 . Hence,
I Z 2
f .z/ dz D f .a C Re i t / Ri e i t dt:
C 0
Though the value of a contour integral on C does not depend on the parameterization , it
would seem to depend on the geometry of C and on its endpoints if C is not closed. In the
study of analytic functions, however, the following results are central.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 350 — #368

i i
A.3.3 Four Integral Theorems

Theorem A.10 (Cauchy’s Theorem). Let f .z/ be analytic everywhere on and inside a
simple closed contour C . Then I
f .z/ dz D 0:
C
One corollary is that if f .z/ is analytic on and inside the region between two contours
R
C1 and C2 which share the same initial and same terminal points, then C1 f .z/ dz D
R
C2 f .z/ dz. Another is that if f is analytic on and in betweenR
positively-oriented simple
R
closed contours K1 and K2 with K1 interior to K2 , then K1 f .z/ dz D K2 f .z/ dz.
These are both aspects of the principle of deformation of contours.
Theorem A.11 (Cauchy’s Integral Formula). Let f .z/ be analytic everywhere on and
inside a simple closed positively-oriented contour C . Let a be a point inside C . Then
I
f .z/
dz D 2 i f .a/:
C z a
Theorem A.12 (Cauchy’s Generalized Integral Formula). Let f .z/ be analytic everywhere
on and inside a simple closed positively-oriented contour C . Let a be a point inside C .
Then, for n 2 N,
I
f .z/ 2 i f .n/ .a/
dz D :
C .z a/nC1 nŠ
If C is not a simple closed contour, then these integral theorems do not apply. Fortu-
nately, we may still be able to avoid direct use of (A.3) by an application of the fundamen-
tal theorem of calculus.
Theorem A.13. Let C be a contour from z0 to z1 . Suppose f has an antiderivative in
some open set containing C ; that is, there exists a complex function F W ! C so that
R
F 0 .z/ D f .z/ for all z 2 . Then C f .z/ dz D F .z1 / F .z0 /.
A.4 Taylor Series and Laurent Series

A.4.1 Taylor Series
The theory of Taylor series carries over from the theory in real variables. However, it is
even better here, as the following theorem shows:
Theorem A.14. Let f .z/ be an analytic function in a domain and let z0 2 . Then
(a) f .z/ can be represented by a power series which converges in the disk jz z0 j < R,
where R is the distance from z0 to the nearest singularity of f .z/. If f .z/ is entire,
i.e., is analytic on all of C, then R D 1 and the power series converges everywhere
in the plane.
(b) The series representing f .z/ is its Taylor series; namely,
1
X f .n/ .z0 /
f .z/ D an .z z0 /n ; where an D :
nD0
nŠ
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 351 — #369

i i
A.4. Taylor Series and Laurent Series 351
Here f .n/ .z0 / denotes the nth derivative of f at z0 . The an are also given by the
contour integrals
I
1 f .z/
an D dz;
2 i C .z z0 /1Cn
where the contour C is any positively-oriented circle centered at z0 of radius r < R.
(c) The Taylor series is the unique power series representing f at z0 .
Power series can be added, subtracted, multiplied and divided. They can also be differ-
entiated and integrated term by term to obtain new series. Of particular note is the fact that
the derivative of an analytic function is itself analytic.
A.4.2 Laurent Series

Laurent series are a generalization of Taylor series. Suppose f is analytic throughout an
annular domain D fz W R1 < jz z0 j < R2 g. Then f has a unique representation as a
Laurent series
X1 X1
bn
f .z/ D an .z z0 /n C ; z 2 : (A.4)
nD0 nD1
.z z0 /n
Both the values R1 D 0 and R2 D C1 are allowed. The coefficients an are as given in
the integral representation above and the bn by
I
1 f .z/
bn D dz;
2 i C .z z0 /1 n
where the contour C is any positively-oriented circle centered at z0 of radius r; R1 < r <
R2 .
A.4.3 Isolated Singularities

A function f that is analytic on .z; "/ n fz0 g, that is, on a deleted " neighborhood of z0 ,
can be expressed by its Laurent series as in (A.4). This leads to three types of singularities:
Definition A.15.
(a) When the coefficients bn D 0 for every n D 1; 2; : : : , then f .z/ is said to have a
P
removable singularity at z0 . Then f .z/ D 1nD0 an .z z0 /n , and setting f .z0 / D a0
will make f analytic in the full neighborhood .z; "/ (we have removed the singular-
ity).
(b) When there exists N 2 N such that bN ¤ 0 and bn D 0 for all n > N , then f has a
pole of order N at z0 . We can express f in a factored form f .z/ D .z z0 / N h.z/
where h.z/ D bN C bN 1 .z z0 / C is analytic on .z; "/ and h.z0 / D bN ¤ 0.
(c) When infinitely many bn are nonzero, we say f has an essential singularity at z0 .
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 352 — #370

i i
A.4.4 Residue Theorem

The Laurent expansion is very useful in complex integration. When f .z/ is expressed by
a Laurent series as in (A.4) on a punctured disk (R1 D 0) with center z0 , then the residue
of f at z0 is the coefficient of .z 1z0 / in the Laurent expansion, that is, Res.f; z0 / D b1 .
Theorem A.16 (Residue Theorem). Let C be a simple closed positively-oriented contour
in C. If a function f is analytic on and inside C except for a finite number of isolated
singularities zk , k D 1; : : : ; n, inside C , then
I n
X
f .z/ dz D 2 i Res.f; zk /:
C
kD1
Definition A.17. A function f is meromorphic in a domain if it is analytic at every

point of except possibly at poles.
A.5 Key Theorems

A.5.1 Maximum Modulus Theorem
Two key results about the maximum modulus of an analytic function are the following.
Theorem A.18 (Maximum Modulus Theorem). If a function f is analytic and not con-
stant in a domain C, then jf .z/j has no maximum value in . That is, there is no
point z0 2 such that jf .z/j jf .z0 /j for all z 2 .
Corollary A.19. Suppose that a function f is continuous on a closed and bounded region
E C and is analytic and non constant in the interior of E. Then the maximum value of
jf .z/j in E, which is always attained, occurs on the boundary of E and not in the interior.
A.5.2 Argument Principle

The argument principle lets us count the number of zeros and poles inside a contour.
Theorem A.20 (Argument Principle). Let C be a positively-oriented simple closed con-
tour. Suppose f is meromorphic on an open set which contains C and the open set
bounded by C . Suppose f has no poles or zeros on C . Then
I
1 f 0 .z/
dz D N0 Np ;
2 i C f .z/
where N0 is the number of zeros of f inside C and Np is the number of poles of f inside
C . In determining N0 and Np , zeros and poles are counted according to their orders (or
multiplicities).
A.5.3 Schwarz’s Lemma

We have the following geometric result regarding functions on the open unit disk D.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 353 — #371

i i
A.6. More Advanced Results 353
Theorem A.21 (Schwarz’s Lemma). Suppose f W D ! D is analytic and f .0/ D 0. Then

jf 0 .0/j 1 and jf .z/j jzj for all z 2 D. If jf 0 .0/j D 1 or if jf .z/j D jzj for any
z 2 D n f0g, then f is of the form f .z/ D e i z for some 2 R (i.e., f is a rotation).
A.6 More Advanced Results

A.6.1 Local Properties of Analytic Maps
The existence of power series representations of analytic maps is a powerful tool for under-
standing their local, and sometimes global, behavior. If an analytic function f is defined
in a domain C, then for any z0 2 , Theorem A.14 tells us that
f .z/ D a0 C a1 .z z0 / C a2 .z z0 /2 C : : : (A.5)
on any Euclidean disk .z0 ; r / D fz 2 C W jz z0 j < r g which is contained in .

Suppose for the moment that f 0 .z0 / D a1 ¤ 0. Then for z very close to z0 we see that
the linear contribution a0 C a1 .z z0 / dominates the rest of the series in (A.5). Omitting
the details, we can say that f .z/ A1 .z/ D a0 C a1.z z0 / for z 2 .z0 ; "/ when " > 0
is small. Since A1 .z/ is a linear map, with well understood properties, we can expect that
f .z/ will have the same properties as A1 .z/ when z 2 .z0 ; "/. In particular, for every
w near a0 D f .z0 /, there is exactly one value z1 near z0 that maps to w. This holds for
A1 , and for f . Thus f is locally one-to-one (also called locally univalent) at z0 when
f 0 .z0 / ¤ 0. See Figure A.3.
w
z1 A1
z0 a0
D(z0, e)
D(a0, |a1|e)
F IGURE A.3. An illustration of the map A1 .z/ D a0 C a1 .z z0 / mapping .z0 ; "/ onto
.a0 ; ja1 j"/ in a one-to-one fashion.
Similarly, we can use approximations to understand the local behavior of f .z/ when
a1 D 0, i.e., f 0 .z0 / D 0. If f is not constant,
f .z/ D a0 C ak .z z0 /k C : : : (A.6)
where ak is the first non-zero coefficient (other than possibly a0 ) in (A.5). Then for z
very close to z0 , the terms a0 C ak .z z0 /k dominate the rest of the series. Omitting the
details, we can say that f .z/ Ak .z/ D a0 C ak .z z0 /k for z 2 .z0 ; "/ when " > 0
is small. Note that Ak is a composition h3 ı h2 ı h1 of the simple maps h1 .z/ D z z0
(translation), h2 .z/ D z k (power function), and h3 .z/ D a0 C ak z (linear function). As
such, Ak .z/ has well understood properties, and so we can expect that f .z/ will have the
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 354 — #372

i i
z1 zk
Ak w
z0 a0
z2 D(a0, |ak|ek)
D(z0, e)
F IGURE A.4. An illustration of the map Ak .z/ D a0 C ak .z z0 /k for k D 3 mapping .z0 ; "/
onto .a0 ; jak j"k / in a k-to-one fashion.
same properties when z 2 .z0 ; "/. In particular, for every w near a0 D f .z0 /, there
are k values z1 ; : : : ; zk symmetrically arranged near z0 that map to w. This holds for Ak ,
and for f , though since we have only an approximation the k values may not be exactly
symmetrically arranged about z0 . See Figure A.4. Thus we say that f is locally k-to-one at
z0 when f 0 .z0 / D f 00 .z0 / D D f .k 1/ .z0 / D 0, but f .k/ .z0 / ¤ 0. We also describe
this situation by saying that z0 maps to f .z0 / with degree (or order or multiplicity or
valency) k.
When z0 maps to f .z0 / with multiplicity k, we can rewrite (A.6) as f .z/ D a0 C .z
z0 /k Œak C akC1 .z z0 / C : : : . Because ak C akC1 .z z0 / C : : : determines an analytic
map on D, we have the following.
Lemma A.22. Let f be a function analytic at z0 with multiplicity k. Then there is a map h
that is analytic at z0 such that h.z0 / ¤ 0 and f .z/ D f .z0 /C .z z0 /k h.z/. In particular,
if f .z0 / D 0, that is, if f has a zero of order k at z0 , then f .z/ D .z z0 /k h.z/.
Open Mapping Theorem

A fact not usually introduced in standard undergraduate complex variables courses is that
analytic maps are open maps, which we define more carefully below. Some chapters in
this text use this and other such results, and so we present them here. Proofs are in more
advanced texts such as [2, p. 344-348].
Consider a non-constant function f that is analytic at z0 . Section A.6.1 shows that,
whether or not f 0 .z0 / D 0, the image of a small neighborhood of z0 covers a small neigh-
borhood of f .z0 / D a0 . This is enough to assert the following.
Theorem A.23 (Open Mapping Theorem). If f is a function on a domain C that is
analytic and non-constant, then the range f ./ is an open set.
Since open maps are those for which the image of an open set is an open set, non-
constant analytic maps are open maps. An application is the following.
Theorem A.24 (Inverse Function Theorem). Suppose that C is a domain and that
f W ! C is a univalent analytic map. Then the inverse function f 1 W f ./ ! is
analytic.
A use for the inverse function theorem is that it shows that analytic locally one-to-one
functions have local analytic inverse functions. Let f be an analytic map defined on a
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 355 — #373

i i
A.7. Bibliography 355
domain , where f need not be one-to-one on all of . If f 0 .z0 / ¤ 0 for some z0 2 ,

then there is a small disk D D .z0 ; "/ on which f is one-to-one. Hence by the inverse
function theorem, there is a map g W f .D/ ! D, the local inverse of f , defined on the
open set f .D/ that is analytic and satisfies .g ı f /.z/ D z for all z 2 D.
Riemann Mapping Theorem

A very important theorem in complex analysis states that all simply connected domains
(other than the whole plane) are conformally equivalent to the unit disk:
Theorem A.25 (Riemann Mapping Theorem). Suppose that is a simply connected do-
main in the complex plane, ¤ C, and that z0 2 . Then there exists a unique conformal
mapping f of onto the unit disk D satisfying f .z0 / D 0 and f 0 .z0 / > 0.
A proof is in [2, p. 420].
A.7 Bibliography
[1] Ruel V. Churchill and James Ward Brown, Complex Variables and Applications.
[2] Bruce P. Palka, An Introduction to Complex Function Theory: Undergraduate Texts
in Mathematics. Springer-Verlag, New York, 1991.
[3] Edward B. Saff and Arthur David Snider, Fundamentals of Complex Analysis with
Applications to Engineering, Science, and Mathematics. Prentice Hall, New Jersey,
third edition, 2003.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 356 — #374

i i
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 357 — #375

i i
The Riemann Sphere

B
This appendix describes how “infinity” is incorporated in complex analysis via the Rie-
mann sphere and how complex arithmetic extends slightly beyond the complex plane. As
points z in C move arbitrarily far from the origin we informally say that they go to infinity.
To make that precise, we adjoin a point denoted 1 to C to get the Riemann sphere (or
extended complex plane), denoted here as C D C [ f1g.
We extend the concepts of continuity and analyticity from C to C by using stereo-
graphic projection, the intrinsic spherical metric , and the corresponding topology. We
even give an answer to the division-by-zero problem. We do not prove results or even give
a full exposition; our goal is to provide enough background so the reader can grasp the
material in chapters that require it. More details can be found in the reference texts [1], [3],
and the more advanced text [2].
B.1 Stereographic Projection and Spherical Geometry

We model˚ the Riemann2 sphere C by identifying
points in C with points in the unit sphere
2 2 3
S D .x1 ; x2; x3 / W x1 C x2 C x3 D 1 R through stereographic projection. See
Figure B.1. Label N D .0; 0; 1/ the north pole on S. Then for any z D x C iy D .x; y; 0/
in C, consider the line in R3 between z and N . The line intersects S at two points, N
and a second point we call Z D .x1; x2 ; x3/. We then associate the points z and Z. For-
mally, we define a map W C ! S given by .z/ D Z. Omitting the details (which
can
be found in [2, p.351]), the formula for the map is Z D .z/ D .x C iy/ D
2x 2y jzj2 1
; ;
jzj2 C1 jzj2 C1 jzj2 C1
. However, we rarely use the formula since it is the idea and the
picture that provide the necessary understanding.
x3
N
S
ZZ
z x2
C
x1
F IGURE B.1. Stereographic projection with shaded northern hemisphere.
357
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 358 — #376

i i
358 Appendix B. The Riemann Sphere
The only point on S that is not associated with a complex number is the north pole N
and the association between C and S n fN g is bijective (that is, both one-to-one and onto).
We now decide how to associate a point with N .
If jzj is large (i.e., z is very far from the origin), then Z 2 S is close to N . It is therefore
natural to say that a point associated with N must be infinitely far from the origin. It is for
this reason that we adjoin 1 to C and extend the definition of by defining .1/ D N .
We now have an identification, i.e., a bijection, between all of C D C [ f1g and all of S.
We can think of C in two ways: as the complex plane with an added point 1 infinitely far
from the origin, or as the sphere S with N representing 1. Both views are useful.
Stereographic projection identifies the origin 0 2 C with the south pole .0; 0; 1/ in
S, the unit disk D D fz W jzj < 1g with the southern hemisphere, the unit circle jzj D 1
with the equator, and the real axis with a great circle through the two poles and the points
.1/ D .1; 0; 0/ and . 1/ D . 1; 0; 0/. In the plane there are many ways a point can
go to infinity. On the real axis, for example, it is normal to say that a point moving off
arbitrarily far to the left is going to 1 and one moving to the right, to C1. On S,
however, these point come together at N . There is only one 1 on S.
As an exercise you are asked to match the curves in C with their projections onto C
given in Figure B.2.
–5 5
–5
F IGURE B.2. Curves in C projected to C; 1 is the point N on top of the sphere.
B.2 The Spherical Metric

The standard way we measure the distance between two points z and w in C is by jz wj,
the Euclidean metric on C. Is there a natural way to define the distance between points on
C using the sphere model? The answer is yes. We use the natural spherical metric on S
and transfer that back to C. Here’s how.
The spherical distance between two points Z and W on S is defined to be the arclength
of the shortest path on S that connects Z and W , which is the shorter arc of the great
circle that runs through Z and W . We denote this distance by d.Z; W /. For example, the
distance between .1; 0; 0/ and N D .0; 0; 1/ is =2, a quarter of the circumference of a
great circle with radius 1.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 359 — #377

i i
B.3. Topology in C and C 359
We can now transfer the metric from S to C using the map W C ! S by defining
the distance between two points z; w 2 C to be .z; w/ D d.Z; W / D d..z/; .w//.
In other words, project z and w to the sphere S and measure the distance there. The new
metric (distance function) on C does not treat 1 as special; the point 1 plays the same
role as any other point of C.
There is a formula for .z; w/. We do not need it, but a few observations about and
a few examples may be helpful. The largest possible spherical distance is , and for r 2
.0; /, 4 .a; r / D fz 2 C W .z; a/ < r g is the spherical disc of radius r and center a.For
example, 4 .0; =2/ is the southern hemisphere of the Riemann sphere and corresponds to
the unit disc D, while 4 .1; =2/ is the northern hemisphere and corresponds to C n D.
In general, spherical discs are spherical caps of the sphere cut off by planes in R3 . As
suggested in Figure B.2, circles in C project under to circles in the metric (though
the center in C does not generally project to the center in C). Straight lines in C, all of
which are regarded as passing through 1, project to circles in C passing through 1. The
extended real axis R D R [ f1g, for instance, projects to a great circle through 1, as
does every line in C through the origin. In general, a spherical disc D corresponds under
to a halfplane of C if 1 2 @D, to the outside of a Euclidean circle if 1 2 D, and to the
inside of a Euclidean circle otherwise.
B.3 Topology in C and C

With the spherical metric in hand, we can now define the corresponding topological con-
cepts on C. We begin by defining the interior, closure, and boundary of a set E C as
follows: The interior of E, denoted Int(E), is the set of points z 2 E for which there exists
r > 0 so that .z; r / E. The closure of E, denoted E or closure.E/, is the set that
contains all points z 2 C such that for any r > 0 we have 4 .z; r / \ E ¤ ;. The bound-
ary of E, denoted @E, is the set that contains all points z 2 C such that for any r > 0
we have .z; r / \ E ¤ ; and .z; r / \ .C n E/ ¤ ;: Hence, @E D E \ C n E and
E D E [ @E.
Example B.1. Readers should convince themselves that Int.D/ D D, that @D is the unit
circle jzj D 1 , and that D D fz 2 C W jzj 1g. In general, @4.a; r / D C.a; r /, where
C.a; r / D fz 2 C W jz aj D r g is the Euclidean circle of radius r > 0 and center a 2 C.
Likewise, for a 2 C and r 2 .0; /, @4 .a; r / D C .a; r /, the spherical circle. What
happens when r D ? Show that fz 2 C W .a; z/ > r g is a spherical disc and give its
center.
Exercise B.2. Identify the interior, closure, and boundary of the following subsets of C.
No proofs are required. Try it out!
(a) A D ; (b) B D fz 2 C W jzj 1g
(c) C D fz 2 C W jzj D 1g (d) D D fz 2 C W jzj > 1g
(e) E D R (f) F D C n R
(g) G D fx C iy 2 C W x; y 2 Qg, where Q is the set of rational numbers in R
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 360 — #378

i i
Definition B.3.
(a) A set A C is called open in C if Int.A/ D A, i.e., if for each point z 2 A, there
exists r > 0 such that 4 .z; r / A.
(b) A set A C is called open in C if for each point z 2 A, there exists r > 0 such that
4.z; r / A.
Definition B.4.
(a) A set A C is called closed in C if its complement C n A is open in C.
(b) A set A C is called closed in C if its complement C n A is open in C.
Remark B.5. Often we refer to open or closed sets without making explicit whether we
are working in C or in C. Context will in general make the meaning clear. These facts may
also help. Take a moment to convince yourself of their validity:
(a) A set A C is open in C if and only if it is open in C.

(b) A set E C is bounded if there exists some R > 0 such that jzj R for all z 2 E.
A bounded subset E of C is closed in C if and only if it is closed in C.
(c) There exist sets B C which are closed in C but not closed in C. For example, R or
C itself.
An open set A C is called connected if given any two points z; w 2 A there exists a
polygonal line in A that connects z to w. A polygonal line, is the union of a finite number
of line segments joined end to end. If U is an open set and z0 2 U , then the set U.z0 / of
all points z 2 U such that there is a polygonal line in U which connects z0 to z is called
the (connected) component of U containing z0 . Two facts regarding an open set U can be
confirmed, proofs are in [2]:
(a) Components U.z/ and U.w/ are equal exactly when there is a polygonal path in U
which connects z to w.
(b) An open set U is equal to the union of its components, i.e., U D [z2U U.z/.
A domain is a non-empty open connected set in C. (The domain set of a function is

not necessarily a domain in this sense.) A domain D C is simply connected if every
closed curve in D can be shrunk continuously to a point while remaining in D. Informally,
this means that D has no “holes” in it. For example, the unit disc D is a simply connected
domain, while the set D n f0g is not. A Jordan curve in C is a simple closed curve, that
is, one which does not intersect itself. The Jordan curve theorem tells us that removing
a Jordan curve from the Riemann sphere leaves two simply connected domains. The
domain inside is the one on its left as you traverse , and this is known as a Jordan
domain. Observe that D is a Jordan domain, whereas the slit disc, D n Œ0; 1/, though simply
connected, is not a Jordan domain.
A neighborhood of a point z 2 C is an open set that contains z. A deleted neighbor-
hood of z is a set U n fzg where U is a neighborhood of z. When a neighborhood U is of
the form .z; "/ or .z; "/ we refer to it as an "-neighborhood of z.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 361 — #379

i i
B.4. Continuity on the Riemann Sphere 361
B.3.1 Convergence in the Riemann Sphere

With the spherical metric on C and the notion of "-neighborhood, we can now define
convergence of sequences in C:
Definition B.6. A sequence of points fzn g1
nD1 in C is said to converge to z 2 C if
.zn ; z/ ! 0 as n ! 1, in which case we write zn ! z.
The statement zn ! z when z 2 C (i.e., z ¤ 1) conforms to the standard notion of
convergence in C. The next proposition records this fact and more that relate the Euclidean
metric in C to the spherical metric on C.
Proposition B.7. For points zn ; z; and w in C, we have:
(a) .zn ; z/ ! 0 if and only if jzn zj ! 0.
(b) jzj > jwj if and only if .z; 1/ < .w; 1/.
(c) zn ! 1, jzn j ! C1, and .zn ; 1/ ! 0 are equivalent.
(d) zn ! 0, jzn j ! 0, 1=jzn j ! C1, and 1=zn ! 1 are equivalent.
B.4 Continuity on the Riemann Sphere

Let be a subset of C and consider an extended complex-valued function f W ! C.
We write limz!z0 f .z/ D L and we say the function f approaches L as z approaches
z0 , if for each " > 0 there exists ı > 0 such that .f .z/; L/ < " whenever z 2 and
0 < .z; z0 / < ı. We say that f is continuous at z0 if z0 2 and limz!z0 f .z/ D f .z0 /.
We say f is continuous on a set U in its domain set if it is continuous at each point
of U . From Proposition B.7, this notion of continuity conforms with our usual notion of
continuity in C (see Section A.2).
Example B.8. We consider the map W z 7! 1=z and its relation to the division-by-zero
problem. Using complex arithmetic, we can see that is a continuous function from C to
itself, except at z D 0 and z D 1 where it is not defined. Use Proposition B.7 to show
that .0/ D 1 is the only way to extend the definition of so that it becomes continuous
at 0. Thus, in this context 1=0 D 1. Show as well that .1/ D 0 is the only way to make
continuous at 1, meaning 1=1 D 0.
Example B.9. Mappings from C to itself can also be treated geometrically by identifying
C with the sphere S. Consider the x1 -axis of S in R3, which pierces S at the points .1; 0; 0/
and . 1; 0; 0/, corresponding to 1 and 1 in C. What is the effect of rotating S by 180ı D
radians about this axis? Observe that the rotation carries S one-to-one onto itself, fixes
each of the points 1 and 1, carries the equator (the unit circle) onto itself, carries the real
axis (a great circle) onto itself, and interchanges the north and the south poles. Readers
should convince themselves that this rotation is precisely the map W z 7! 1=z.
You can use W z 7! 1=z for addressing continuity (and later analyticity) on the sphere
by the judicious interchange of 1 and 0.
Theorem B.10. Let z0 ; w0 2 C and let f .z/ be a complex-valued function defined in a
deleted neighborhood of z0 . Then,
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 362 — #380

i i
1
(a) lim f .z/ D 1 if and only if lim D 0.
z!z0 z!z0 f .z/

1
(b) lim f .z/ D w0 if and only if lim f D w0 .
z!1 z!0 z
1
(c) lim f .z/ D 1 if and only if lim D 0.
z!1 z!0 f .1=z/
By pre- or post-composing f with the rotation W z 7! 1=z, the above theorem
converts each statement on the left (involving 1) to the corresponding statement on the
right (which avoids 1). By considering the topology of the Riemann sphere C near 1
and the continuity of , the details of the proof can easily be checked (and they can also be
found in [1, p. 51]).
The simplest and most useful class of maps that are continuous on all of C is the class
of rational maps.
Definition B.11. A quotient of two polynomials is called a rational function.
P.z/
Let f .z/ D Q.z/ be a rational function in reduced form (i.e., polynomials P .z/ and
Q.z/ have no common zeros). Although this formula for f is defined only for complex
values where Q is not zero, we can regard f (as a mapping into C) as being defined
and continuous on all of C. If Q.a/ D 0, we set f .a/ D 1. We also set f .1/ D
limz!1 f .z/. We leave it to the reader to check that this gives f the desired continuity
properties. We illustrate this in the next examples.
2
Example B.12. Let f .z/ D z 2 C 5i and g.z/ D z3z2 C2z5 . Then f .1/ D 1; g.1/ D
3; g. 2/ D 1, and g.0/ D 1. Defining the maps this way, we see that f and g are
continuous at every point of C. For instance, f .1/ D limz!1 f .z/ D 1 because
1 1 z2
lim D lim D lim D 0:
z!0 f .1=z/ z!0 .1=z/2 C 5i z!0 1 C 5iz 2
The reader should use Theorem B.10 to check the continuity of g at 2; 0; and 1.
Example B.13. Special among rational maps are the Möbius transformations. These are of
form M.z/ D CAzCB
zCD
where A; B; C; and D are complex numbers satisfying AD BC 6D
0. The inversion map z ! 1=z discussed earlier is one example. Show that in general
M. D=C / D 1 and M.1/ D A=C and explain the interpretations when C D 0. For
more on Möbius transformations, see Sections B.1 and 4.2.
Example B.14. In contrast to the case for rational functions, the exponential function
f .z/ D e z cannot be defined at 1 to make it continuous there. You can see this by
noting that on the real line limx!C1 e x D 1, but limx! 1 e x D 0, which means that
limz!1 e z does not exist.
B.5 Analyticity on the Riemann Sphere

As the rotation z 7! 1=z was used in Theorem B.10 to explain the notion of continuity on
the Riemann sphere C, so too can we use it to extend our notion of analyticity.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 363 — #381

i i
B.5. Analyticity on the Riemann Sphere 363
Definition B.15. Let C be an open set and suppose f W ! C is continuous:

1
(a) If z0 2 \ C and f .z0 / D 1, then f is analytic at z0 if f .z/ is analytic at z0 .
(b) If 1 2 and f .1/ ¤ 1, then f is analytic at 1 if f . 1z / is analytic at 0.

1
(c) If 1 2 and f .1/ D 1, then f is analytic at 1 if is analytic at 0.
f .1
z/
One way to summarize the above is to say that a map is analytic in the extended sense if
after moving each instance of 1 to 0 (by pre- or post-composing with the map z 7! 1=z),
we get a map that is analytic in the usual sense.
Exercise B.16. Show that z 7! sin z1 is analytic at 1, but z 7! sin z is not.
Remark B.17. When z0 2 C is a pole of f of order k, then by Definition A.15 we know
that we can write f .z/ D .z z0 / k h.z/ where h is analytic at z0 (in the usual sense of
Section A.2) and h.z0 / ¤ 0. Thus, f 1.z/ D .z z0 /k = h.z/ is analytic at z0 , which means
we can say f is analytic at z0 by Definition B.15(a), in the extended sense. Since the word
“analytic” is used in this text in the usual sense of Section A.2 and in the extended sense,
the reader must always be careful to use context to decide in which sense it is being used.
Since the context is usually clear, no confusion should arise.
A map f with an isolated singularity at z0 2 C (see Definition A.15) is analytic in the
usual sense for a removable singularity and analytic in the extended sense for a pole. Thus,
only if z0 is an essential singularity does f fail to be analytic in any sense.
1 z2
Example B.18. Let f .z/ D 1Cz 2
. Since f . 1z / D 1
D is analytic at 0 (and
1C. 1
z/
2 z 2 C1
has the value 0 there), we can say that f is analytic at 1 (and f .1/ D 0). Similarly,
since 1=f .z/ D 1 C z 2 is analytic at ˙i , f is analytic at the poles ˙i , a fact stated more
generally in Remark B.17.
1
Example B.19. Let f .z/ D 3z z
so f .1/ D 1. Since
1 1 z
k.z/ D D 1 D
f . 1z / 3. z / z 3 z2
1
is analytic at zero we say that f .z/ is analytic at 1. Because f 0 .z/ D 3 C z2
, f 0 .1/ D
2
3. We also have k 0 .z/ D .33Cz z 2 /2
yielding k 0 .0/ D 1=3 D 1=f 0 .1/. The relationship
between k 0 .0/ and f 0 .1/ is an important and general property:
Lemma B.20. Suppose f is analytic at 1 with f .1/ D 1. If k is defined by k.z/ D
1 0 0 0 0
1 , then f .1/ D 1= k .0/. If k .0/ D 0, then by our conventions f .1/ D 1.
f .z /
Proof. Assuming f is analytic at 1 means k is analytic at 0; also, f .1/ D 1 means

k.0/ D 0. We let N denote the multiplicity of the zero of k at 0. There exists r > 0 such
that k is analytic on .0; r /. Therefore, f .z/ D 11 is analytic on fz 2 C W jzj > 1=r g.
k. z /
Hence f may be represented by a Laurent series
1
X 1
X
f .z/ D an z n C bn z n
nD0 nD1
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 364 — #382

i i
on fz 2 C W jzj > 1=r g. Since

X 1 1
X
1 1 n
Df D an z C bn z n
k.z/ z nD0 nD1
has a pole at 0 of order N , we have aN ¤ 0 and an D 0 for all n > N . Hence f .z/ D
z N h.z/ where h.z/ D C aN 1 z 1 C aN .
Let g.z/ D h. 1z / D aN C aN 1 z C . Then
1 1 zN
k.z/ D D D :
f . 1z / z 1
N h. /
z
g.z/
N zN 1
g.z/ g0 .z/z N
Thus k 0 .z/ D and so
Œg.z/2

0 1=aN when N D 1
k .0/ D
0 when N > 1:
Because h0 .z/ D aN 1z
2
, as z ! 1 we have

aN when N D 1
f 0 .z/ D N z N 1
h.z/ C z N h0 .z/ !
1 when N > 1:
Thus f 0 .1/ D 1= k 0 .0/ as claimed.
B.6 Bibliography
[1] Ruel V. Churchill and James Ward Brown, Complex Variables and Applications,
[2] Bruce P. Palka, An Introduction to Complex Function Theory: Undergraduate Texts in
Mathematics, Springer-Verlag, New York, 1991.
[3] Edward B. Saff and Arthur David Snider, Fundamentals of Complex Analysis with
Applications to Engineering, Science, and Mathematics, Prentice Hall, New Jersey,
third edition, 2003.
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 365 — #383

i i
Index
Index of Notation Index of Terms
N D the natural numbers, 345 analytic
Q D the rational numbers, 359 Bieberbach conjecture, 213
R D the real numbers, 345 convolution, 252
R D R [ f1g D the extended real numbers, 359 Koebe 41 -theorem, 213
C D fx C iy W x; y 2 Rg D the complex num- Koebe function, 212
bers, 345 right half-plane map, 211
C D C [ f1g D the extended complex plane or analytic dilatation, 291
Riemann sphere, 357 analytic function
complex, 348
D D the open unit disk, 352
discrete, 328
4.a; r/, open Euclidean disk, 353
real, 10
.z; w/ D spherical metric on C, 359
anamorphosis, 198
4 .a; r/, open spherical disk, 359
anticonformal, 127
C.a; r/ D Euclidean circle, 359 Appell F1
C .a; r/ D spherical circle, 359 function, 283
E D closure.E/ D closure of a set E in C, 359 integral representation, 283
@E D boundary of a set E in C, 359 applets
Int(E) = interior of a set E in C, 359 Complex Function Iterator, 3
Complex Newton Method, 3
Re.z/ D the real part of z, 345 ComplexTool, xvi, 272
Im.z/ D the imaginary part of z, 345 Cubic Polynomial Complex Newton Method,
arg.z/ D the argument of z, 347 3
Arg.z/ D the principal argument of z, 347 DiffGeomTool, 88, 90
log.z/, 347 export, xvii
Log.z/, 347 FlowTool, 161
cos z, 347 Global Complex Iteration for Polynomials, 3
sin z, 347 LinComboTool, 198, 247
domain.f / D domain set of f , 345 Mandelbrot Set Builder, 3
f ./, 345 MinSurfTool, 89, 113
f 1 .A/, 345 Parameter Plane and Julia Set, 3
Res.f; z0 / D residue of f at z0 , 352 PolyTool, 272, 289
Real Function Iterator, 3
Ag .w/, 8 Real Newton Method, 3
F .g/, 25 ShearTool, 198, 224
J.g/, 25 StarTool, 272, 303
K.g/, 27 arcsine, 273
Kn , 33 argument, 291, 347
M , 38 principal value, 347
argument principle, 352
S, 210
analytic functions, 292, 293
SH , 216
O harmonic functions, 295, 299, 310
SH , 216
aspect ratio, 274, 333
2 F1 , 277 associated surfaces, 112
365
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 366 — #384

i i
366 Index
asymptotic line, 134 common boundary condition, 16

asymptotic value, 59, 336 compact topological disk (ctd), 42
Complex Function Iterator Applet, 3
basin of attraction, 8 Complex Newton Method Applet, 3
immediate, 35 complex potential
beta function, 279 uniform interval source, 184
bivalent, 297 complex potential function, 172
Blaschke factor, 305 ComplexTool, xvi, 272
Blaschke product, 236, 304, 305, 307, 309, 311, component, 360
331 condition A, 241
discrete, 331, 337 conformal invariant, 332, 334
boundary behavior, 300 conformal mapping, 33, 174, 348
boundary of a set, 15 discrete, 328
bounded set, 360 conformal parametrization, 108
branch circle, 327, 334, 342 conjugate minimal surface, 111
branch point, 331 conjugation
discrete, 327, 331 global, 12, 81
local, 13
Cantor bouquet, 54 connected set, 17, 58, 360
finger, 56 connectedness locus
thickness, 56 of z d C c, 52
hair, 57 of z 2 C c, 42
endpoint, 57 continuous, 348
open, 58 contour, 349
Cantor dust sets, 42 contour integration, 349
Catalan surface, 107 deformation principle, 350
catenoid, 91 contraction lemma, 55
Cauchy’s generalized integral formula, 350 convex, 55, 264
Cauchy’s integral formula, 294, 350 convex corner, 273
Cauchy’s theorem, 285, 350 convex in the direction ei' , 221
Cauchy-Riemann equations, 165, 348 critical orbit
chain rule, 298 of cez , 59
Chaos (James Gleick), 52 of c cos z, 61
chaotic, 2, 16, 25 of c sin z, 61
chaotic set, see Julia set of z 2 C c, 37
CHD, 221 of Newton’s method of a cubic polynomial, 19
circle packing, 322, 323 critical point, 35
angle sum, 324 Cubic Polynomial Complex Newton Method Ap-
branched, 327 plet, 3
carrier, 322, 323 curl, 165
complex, 323 curvature, 97
flower, 323 curvature line, 134
label, 323 cycle, 30
locally univalent, 323 multiplier, 30
maximal, 326 cylinder, 111
packing label, 324
refinement, 338 deformation of contours principle, 294, 350
univalent, 323 degree
CirclePack, 321 vertex, 323
closed contour, 349 degree (local), 35, 354
closed set, 360 dendrite, 37
cluster set, 300 density of hyperbolicity conjecture, 39
CMC surface, 155 devil’s colosseum, 65
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 367 — #385

i i
Index 367
devil’s staircase, 64 super attracting, 10

Dieudonné-Schwarz lemma, 330 flow
differentiable, 348 around a corner, 175
DiffGeomTool, 88, 90 around a cylinder, 176
dilatation, 142, 220, 304, 337, 339 in a channel, 179
analytic, 291 inside a disk, 182
star mappings, 311, 313 FlowTool, 161
dipole, 192 fractal, 15
dipole moment, 194 Frenet frame, 96
disconnected, 42 function
divergence, 165 analytic, see analytic function
domain, 272, 348, 360 conformal, see conformal mapping
domain set, 7, 345 entire, see entire function
Doyle spiral, 334, 336 exponential, see exponential function
dynamics harmonic, see harmonic function
complex, 1 logarithm, see logarithm function
discrete iterative, 2 meromorphic, see meromorphic function
of c cos z, 60 fundamental form
of c sin z, 60 first, 100
of cez , 53 second, 100
of z 2 C c, 26 fundamental theorem of algebra, 4
of a Möbius map, 80 fundamental theorem of calculus, 350
electric field, 165 gamma function, 279

elliptic integral, 274 Gauss map, 126
end of a minimal surface, 132 Global Complex Iteration for Polynomials Ap-
Enneper’s surface, 89 plet, 3
with 2k C 2 leaves, 132 global dynamics, 6
entire function, 350
discrete, 336 Hadamard, Jacques, 7
epicycloid, 210 half-plane function, xvii
error function, 335, 339 harmonic, 284, 348
discrete, 335 Bieberbach conjecture, 229
essential singularity, 351, 363 convolution, 255
Euler representation of hypergeometric, 279 Koebe 16 -conjecture, 229
Euler’s formula, 346 Koebe function, 227
exponential function, xvii, 346 Riemann mapping theorem , 230
discrete, 334 right half-plane map, 230
extension of domain principle, 333 square map, 231
exterior angle, 272 harmonic conjugate, 168
extremal function, 212 harmonic extension, 290, 297, 300, 309
extremal length, 333 harmonic function, 117, 166, 214, 336
canonical decomposition, 215, 291, 336
Farey addition, 47 discrete, 336
Fatou set, 25, 78 harmonic measure, 332
Fatou, Pierre, 25 helicoid, 98
Feynman, Richard P., 2 helix, 98
first fundamental form, 100 Henneberg surface, 107
fixed point, 8 holomorphic, see analytic
attracting, 9 horocycle, 325
classification of, 23 hyperbolic component of Int(M ), 33
multiplier, 23 center, 33
repelling, 20 hypercycloid, 217
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 368 — #386

i i
368 Index
hypergeometric function, 277 antennae, 49

Euler representation, 279 order of junction point, 49
integral representation, 279 spokes, 49
baby, 45, 61
immediate basin of attraction density of, 61
cycle, 36 closed, 39
fixed point, 35 conjugate bulb, 46
inner function, 236 hyperbolic component of Int(M ), 33
interval source, 183 center, 33
inverse function theorem, 354 main cardioid, 28, 32, 45
inverse image, 345 no holes, 71
isothermal, 108 Mandelbrot Set Builder Applet, 3
Mandelbrot set image, 40
Jordan curve, 329, 360 baby, 41
Jordan domain, 360 elephants, 41
Julia set, 24, 79 galaxy, 41
filled-in, 27 mapping problem, 314
no holes, 44 Mathematica code, 181, 182
Julia set image maximum
Cantor bouquet, 54 modulus theorem, 352
dendrite, 37 principle
dragon, 41 discrete, 330
rabbit, 41 mean curvature, 99
Sierpinski curve, 63
meridians, 96
star cluster, 41
meromorphic function, 120, 352
Julia, Gaston, 24
method of exhaustion, 336
jump discontinuity, 299
method of images, 191
metric, 359
Koebe-Andreev-Thurston theorem, 326
minimal graph, 105
lakes of Wada, 17 minimal surface, 104
Laplace’s equation, 166 n-noid, 124
Laurent series, 351 associated surfaces, 112
Lewy’s theorem, 220, 291 asymptotic line, 134
LinComboTool, 198, 247 Catalan surface, 107
linear combinations, 240 catenoid, 91
linear fractional transformation, see Möbius trans- conjugate, 111
formation curvature line, 134
linear mappings, 345 end, 132
linearizing, 13 Enneper’s surface, 89
local inverse, 355 Enneper’s surface with 2k C 2 leaves, 132
locally k-to-one, 354 helicoid, 98
locally one-to-one, see locally univalent Henneberg surface, 107
locally univalent, 219, 348, 353 Richmond’s surface, 156
logarithm function, xvii, 347 Scherk’s doubly periodic surface, 92
principal value, 347 Scherk’s singly periodic surface, 107
Scherk’s surface with n leaves, 124
Möbius strip, 94 wavy plane, 135
Möbius transformation, 12, 80, 205, 272, 325, with a planar end and an Enneper end, 132
326, 362 with a planar end and two catenoidal ends, 134
Mandelbrot set, 38 MinSurfTool, 89, 113
p=q bulb, 46 Misiurewicz point, 37
root, 46 Montel’s theorem, 79
p=q limb, 49 moving frame, 96
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 369 — #387

i i
Index 369
multiplicity, see degree (local), see order polygon, 277, 280, 288
multiplier map on K1 , 28 simple, 273
multiplier map theorem, 33 polygonal line, 360
multipole, 194 polynomial
discrete, 335
n-gon, 277, 280, 288 PolyTool, 272, 289
n-noid, 124 potential function, 166
neighborhood, 360 complex, 172
Newton map, 5 real, 166
Newton’s method power functions, 345
attracting property, 10 pre-periodic, 37
circle of convergence, 11 prevertices, 273, 328
complex, 10 principal directions, 99
of z 3 1, 14 principle of superposition, 170
of family of cubic polynomials, 17 proper map, 331
of quadratic polynomial, 13
radius of contraction, 68 quasiconformal dilatation, 339
radius of convergence, 11, 67
real, 4 r-star, 301–303
non-convex corner, 273 Rado-Kneser-Choquet theorem, 296, 299
non-orientable surface, 94 random
normal curvature, 99 basin of attraction, 64
normalize, 209 dynamics, 64
Fatou set, 65
open mapping theorem, 354 Julia set, 66
open maps, 354 orbit, 64
open set, 360 range of f , 345
orbit, 5, 7 ratio function, 330, 335
order, 354 rational function, 362
of branch point, 327 discrete, 334
of pole, 351 Real Function Iterator Applet, 3
of zero, 291, 354 Real Newton Method Applet, 3
rectangle (conformal map onto), 273
parallels, 96 removable singularity, 351, 363
parameter residue, 352
bifurcation, 34 residue theorem, 352
stable, 28, 33 Richmond’s surface, 156
unstable, 34 Riemann mapping theorem, 175, 209, 329, 355
parameter curve, 93 Riemann sphere, 127, 325
Parameter Plane and Julia Set Applet, 3 root of function, 3
parametrization, 89 roots of unity, 346
periodic point, 30 Rouché’s theorem, 4, 70
Picard’s theorem, 53, 336
piecewise affine, 338 Scherk’s surface
planar harmonic mapping, 139 doubly periodic, 92
Pochhammer symbol, 277 singly periodic, 107
Poincaré metric, 325 with n leaves, 124
point sink, 170 schlicht, 210
point source, 170 Schwarz triangle, 335, 342
Poisson integral formula, 285 Schwarz’s lemma, 353
Poisson kernel, 286 discrete, 330
pole, 351, 363 Schwarz-Christoffel
Polya vector field, 168 discrete, 328
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 370 — #388

i i
370 Index
half-plane formula, 273 stable set, see Fatou set

map onto bowtie, 281 star, 282, 301–303
map onto rectangle, 273 star mappings, 301
map onto regular n-gon, 277, 280 StarTool, 272, 303
map onto square, 278, 280 steady state temperature, 186, 332
map onto star, 282 step function, 290
unit disk formula, 276 stereographic projection, 325, 357
second fundamental form, 100 stream function, 173
self-similarity, 44 streamlines, 173
sense-preserving, 219, 291 Sumi, Hiroki, 65
sensitive dependence, 25 surface of revolution, 92
sharp inequality, 213
shear construction, 221, 288 tangent plane, 93
ShearTool, 198, 224 Taylor series, 350
shifted factorial, 277 temperature, 287
Sierpinski curve, 63 torus, 91
similar matrices, 12 discrete, 340
simple closed contour, 349 totally disconnected, 42
simply connected, 360 transcendental entire, 53
sine, xvii, 273 triangle (conformal map onto), 288, 300
singular inner function, 236 turning angle, 328
singularity, 351
essential, 351 unit normal, 94
pole, 351 unit speed curve, 97
removable, 351 univalent, 208, 348
sink, 170, 176
at infinity, 178 valency, see degree (local), see order
source, 170, 176 vector field, 164
at infinity, 178 conservative, 166
dipole, 192 electric field, 165
effective strength, 176, 191 incompressible, 165
interval, 183 irrotational, 165
multipole, 194 magnetic field, 164
non-uniform density, 185 Polya, 168
not on boundary, 190
strength of, 171 wavy plane, 135
spherical disc, 359 winding number, 292, 293
square (conformal map onto), 280
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 371 — #389

i i
About the Authors

Michael Brilleslyper was raised in Southern California. He earned a bachelor’s degree in
applied mathematics and a master’s degree in mathematics from Arizona State University.
He obtained his Ph.D. from the University of Arizona in 1994, under the direction of Doug
Pickrell. He spent the first six years of his career at Arizona State working extensively with
their introductory and calculus courses. For two of those years he served as coordinator
for their First Year Mathematics Program. In 2000, he came to the United States Air Force
Academy in Colorado Springs, CO. At the Academy, Dr. Brilleslyper enjoys teaching at
all levels of the curriculum. He has worked on numerous curricular development projects
involving writing, technology, and fundamental skills. He is extremely active in the MAA,
having been in the second cohort of Project NExT fellows. More recently he has served
as Rocky Mountain section chairman, he has twice been a program chair, he has served
as chair of the professional development committee, and he currently serves as Governor
of the Rocky Mountain section. He lives in Colorado Springs with his wife and their two
daughters.
Michael Dorff is a professor of mathematics at Brigham Young University. He was born in

Duluth, Minnesota and grew up in southern California. After teaching high school for four
years, he earned an MS degree at the Univ. of New Hampshire and in 1997 a PhD from
the Univ. of Kentucky in complex analysis. He was a professor at the Univ. of Missouri-
Rolla before accepting a position in 2000 at BYU. He has published over 20 research
papers and has given talks at over 100 different conferences, universities, and colleges. He
founded the BYU mathematics REU and in 2007 he founded CURM, the national Center of
Undergraduate Research in Mathematics, which promotes, trains, and supports professors
across the U.S. in doing research with undergraduate students. He is a member of the MAA,
AMS, CUR, and Project NExT, and has served in many positions including governor of the
MAA Intermountain section, chair of several MAA committees, member of the Executive
Board of CUR, and member of the editorial boards of the journals American Math Monthly
and Involve. He was a Fulbright Scholar in Poland and has received numerous teaching
awards including the MAA’s Deborah and Franklin Tepper Haimo Award in 2010. He is
married with 5 daughters. His interests include reading (Dostoyevsky and Dickens through
Stegner and Saramago), traveling (invite him to visit you!), running (even at 3 am on the
streets in Utah), music (classical, Norah Jones), and soccer.
Jane McDougall received her PhD from Northwestern University in 1996, where she stud-
ied functions of one complex variable. She has been a member of the Department of Math-
371
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 372 — #390

i i
372 About the Authors
ematics and Computer Science at Colorado College since 1997, during which time her
research interests have grown into geometric function theory and harmonic mappings.
Jim Rolf was born in Nashville, TN but moved to his adopted home state of Texas at
age two. He earned his Bachelor of Science degree in mathematics in his home town of
Waco, TX while attending Baylor University. This was followed by a Master of Divinity
degree from Southwestern Baptist Theological Seminary in Fort Worth, TX. In 1997 he
completed his doctoral work in numerical analysis at Duke University under the direction
of William K. Allard. After three years on the faculty at West Point, Jim moved to Colorado
Springs, CO and the math department at the United States Air Force Academy. In the fall of
2012, Jim will join the math department at Yale University. In addition to his professional
interests, Jim enjoys cycling, gardening, and woodworking.
Lisbeth Drews Schaubroeck was born in the Netherlands, but spent most of her childhood
in Geneseo, IL. She earned her bachelor’s degree in Mathematics and Mathematics Edu-
cation at Wartburg College in Waverly, IA. In 1998, she completed her doctoral work in
complex analysis at the University of North Carolina at Chapel Hill under the direction of
John Pfaltzgraff. Currently, Beth is a professor at the United States Air Force Academy in
Colorado Springs, CO. There she enjoys teaching all levels of undergraduate mathematics
to future Air Force officers. She is active in the Rocky Mountain Section of the MAA and
has been on the awards committee and the student activities coordinator. Beth especially
enjoys her work in faculty development in a wide range of venues. She mentors Air Force
officers who are new to teaching at the Air Force Academy, new PhDs through Project
NExT, and seasoned faculty from civilian universities and community colleges through an
NSF-funded PREP workshop. She lives in Colorado Springs with her husband and two
sons.
Richard L. Stankewitz was born in Royal Oak, Michigan. He earned a bachelor’s degree
with distinction in mathematics from the University of Michigan at Ann Arbor in 1991. He
obtained a Ph.D. in 1998 under the direction of his advisor Aimo Hinkkanen at the Univer-
sity of Illinois at Urbana in the field of complex dynamics. Dr. Stankewitz held positions at
Texas A&M University at College Station and Penn State University at Erie before coming
to Ball State University in 2002, where in 2010 he attained his current position of Profes-
sor of Mathematics. He is a member of the American Mathematical Society and has a long
time interest in supporting efforts to encourage undergraduate research in mathematics. In
addition to publishing over 15 research articles, he has given talks in over 50 national and
international conferences, universities, and colleges.
Ken Stephenson was born in South Haven, Michigan, in 1945. He received a BS in math-
ematics at Michigan and an MS in mathematics at Wisconsin before serving 3 years as
a Naval Officer. He returned to his studies at Wisconsin, receiving his PhD under Walter
Rudin in 1976. From Wisconsin he joined the faculty at the University of Tennessee, where
he continues his research and teaching.
Complex function theory and the associated conformal geometry have remained Dr.
Stephenson’s core mathematical interest, but in 1985 a fascinating talk by Bill Thurston on
i i
i i
i i
“ECA˙Book” — 2012/8/28 — 13:58 — page 373 — #391

i i
About the Authors 373
“circle packing” profoundly changed his outlook. He began contributing to the develop-
ment of circle packing and its associated “discrete” conformal geometry through publica-
tions and software, culminating in his 2005 book Introduction to Circle Packing: the The-
ory of Discrete Analytic Functions, Cambridge University Press, and his software package
“CirclePack”. This is a new type of experimentally driven mathematics, and his principal
goal now is promoting circle packing and its applications.
Dr. Stephenson has been supported throughout his years of research by grants from the
National Science Foundation, the Tennessee Science Alliance, and currently the Simons
Foundation. He has spoken on circle packing at numerous national and international con-
ferences and has held visiting positions at the University of Hawaii, the Open University
(England), the University of Cambridge, Florida State University, the Technical University
of Berlin, the Free University of Berlin, Oak Ridge National Laboratory, and the Mathe-
matical Sciences Center (Tsinghua University, Beijing). He is a member of the American
Mathematical Society and the Mathematical Association of America, and is a Fellow of
AAAS (American Association for the Advancement of Science).
i i
i i

Explorations in Complex Analysis

Uploaded by

Copyright:

Available Formats

Explorations in Complex Analysis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Explorations in Complex Analysis

Uploaded by

Copyright:

Available Formats

MAA-Complex Analysis back_Layout 1 9/7/12 2:13 PM Page 1

Explorations in Complex Analysis

9 780 883 85 355 9

Classroom Resource Materials

Classroom Resource Materials

“ECA˙Book” — 2012/8/28 — 13:58 — page i — #1

“ECA˙Book” — 2012/8/28 — 13:58 — page ii — #2

Library of Congress Catalog Card Number 2012943816

“ECA˙Book” — 2012/8/28 — 13:58 — page iii — #3

Explorations in Complex Analysis

Published and Distributed by

“ECA˙Book” — 2012/8/28 — 13:58 — page iv — #4

Council on Publications and Communications

“ECA˙Book” — 2012/8/28 — 13:58 — page v — #5

CLASSROOM RESOURCE MATERIALS

101 Careers in Mathematics, 2nd edition edited by Andrew Sterrett

“ECA˙Book” — 2012/8/28 — 13:58 — page vi — #6

Proofs Without Words, Roger B. Nelsen

MAA Service Center

“ECA˙Book” — 2012/9/5 — 14:07 — page vii — #7

1 Complex Dynamics: Chaos, Fractals, the Mandelbrot Set, and More 1

2 Soap Films, Differential Geometry, and Minimal Surfaces 85

“ECA˙Book” — 2012/9/5 — 14:07 — page viii — #8

3 Applications to Flow Problems 161

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

4 Anamorphosis, Mapping Problems, and Harmonic Univalent Functions 197

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

5 Mappings to Polygonal Domains 271

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271

“ECA˙Book” — 2012/9/5 — 14:07 — page ix — #9

5.9 The Dilatation for Star Mappings . . . . . . . . . . . . . . . . . . . . . . 311

6 Circle Packing 317

6.1 First Impressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318

B The Riemann Sphere 357

About the Authors 371

“ECA˙Book” — 2012/8/28 — 13:58 — page x — #10

“ECA˙Book” — 2012/8/28 — 13:58 — page xi — #11

“ECA˙Book” — 2012/8/28 — 13:58 — page xii — #12

“ECA˙Book” — 2012/8/28 — 13:58 — page xiii — #13

“ECA˙Book” — 2012/8/28 — 13:58 — page xiv — #14

Chapter 1: Complex Dynamics As a text that hopes to inspire students by showing

Chapter 2: Soap Films, Differential Geometry, and Minimal Surfaces

“ECA˙Book” — 2012/8/28 — 13:58 — page xv — #15

Chapter 3: Applications to Flow Problems I am particularly indebted to Harrison Pot-

Chapter 5: Mappings to Polygonal Domains We gratefully acknowledge the numerous

“ECA˙Book” — 2012/8/28 — 13:58 — page xvi — #16

Using Java Applets

F IGURE 0.1. The applet ComplexTool

“ECA˙Book” — 2012/8/28 — 13:58 — page xvii — #17

Using Java Applets xvii

“ECA˙Book” — 2012/8/28 — 13:58 — page xviii — #18

Using Links in the Electronic Book

“ECA˙Book” — 2012/9/5 — 14:07 — page 1 — #19

Complex Dynamics: Chaos, Fractals,

“ECA˙Book” — 2012/8/28 — 13:58 — page 2 — #20

“ECA˙Book” — 2012/8/28 — 13:58 — page 3 — #21

1.2. Newton’s Method 3

How to Use this Chapter

1.2 Newton’s Method

that jf n .z/ aj ˇ n jz aj ˇ n " ! 0 whenever z 2 with jz aj < ". Thus

simpler map g. Iterates are related as such, gn D . ı f ı 1 / ı . ı f ı 1 / ı ı

D .f p /0 .w0 / D f 0 .w0 /f 0 .w1 / : : : f 0 .wp 1/ D .f p /0 .wj /; j D 0; : : : ; p 1; (1.1)