TEXTBOOKS IN MATHEMATICS

A MATLAB®
COMPANION
to COMPLEX
VARIABLES

A. David Wunsch
 A MATLAB®
COMPANION
to COMPLEX
VARIABLES
 TEXTBOOKS in MATHEMATICS
Series Editors: Al Boggess and Ken Rosen

PUBLISHED TITLES
ABSTRACT ALGEBRA: AN INTERACTIVE APPROACH, SECOND EDITION
William Paulsen
ABSTRACT ALGEBRA: AN INQUIRY-BASED APPROACH
Jonathan K. Hodge, Steven Schlicker, and Ted Sundstrom
ADVANCED LINEAR ALGEBRA
Hugo Woerdeman
APPLIED ABSTRACT ALGEBRA WITH MAPLE™ AND MATLAB®, THIRD EDITION
Richard Klima, Neil Sigmon, and Ernest Stitzinger
APPLIED DIFFERENTIAL EQUATIONS: THE PRIMARY COURSE
Vladimir Dobrushkin
COMPUTATIONAL MATHEMATICS: MODELS, METHODS, AND ANALYSIS WITH MATLAB® AND MPI,
SECOND EDITION
Robert E. White
DIFFERENTIAL EQUATIONS: THEORY, TECHNIQUE, AND PRACTICE, SECOND EDITION
Steven G. Krantz
DIFFERENTIAL EQUATIONS: THEORY, TECHNIQUE, AND PRACTICE WITH BOUNDARY VALUE PROBLEMS
Steven G. Krantz
DIFFERENTIAL EQUATIONS WITH MATLAB®: EXPLORATION, APPLICATIONS, AND THEORY
Mark A. McKibben and Micah D. Webster
ELEMENTARY NUMBER THEORY
James S. Kraft and Lawrence C. Washington
EXPLORING LINEAR ALGEBRA: LABS AND PROJECTS WITH MATHEMATICA®
Crista Arangala
GRAPHS & DIGRAPHS, SIXTH EDITION
Gary Chartrand, Linda Lesniak, and Ping Zhang
INTRODUCTION TO ABSTRACT ALGEBRA, SECOND EDITION
Jonathan D. H. Smith
PUBLISHED TITLES CONTINUED

INTRODUCTION TO MATHEMATICAL PROOFS: A TRANSITION TO ADVANCED MATHEMATICS, SECOND EDITION

Charles E. Roberts, Jr.
INTRODUCTION TO NUMBER THEORY, SECOND EDITION
Marty Erickson, Anthony Vazzana, and David Garth
LINEAR ALGEBRA, GEOMETRY AND TRANSFORMATION
Bruce Solomon
MATHEMATICAL MODELLING WITH CASE STUDIES: USING MAPLE™ AND MATLAB®, THIRD EDITION
B. Barnes and G. R. Fulford
MATHEMATICS IN GAMES, SPORTS, AND GAMBLING–THE GAMES PEOPLE PLAY, SECOND EDITION
Ronald J. Gould
THE MATHEMATICS OF GAMES: AN INTRODUCTION TO PROBABILITY
David G. Taylor
MEASURE THEORY AND FINE PROPERTIES OF FUNCTIONS, REVISED EDITION
Lawrence C. Evans and Ronald F. Gariepy
NUMERICAL ANALYSIS FOR ENGINEERS: METHODS AND APPLICATIONS, SECOND EDITION
Bilal Ayyub and Richard H. McCuen
ORDINARY DIFFERENTIAL EQUATIONS: AN INTRODUCTION TO THE FUNDAMENTALS
Kenneth B. Howell
RISK ANALYSIS IN ENGINEERING AND ECONOMICS, SECOND EDITION
Bilal M. Ayyub
TRANSFORMATIONAL PLANE GEOMETRY
Ronald N. Umble and Zhigang Han
This page intentionally left blank
 TEXTBOOKS in MATHEMATICS

A MATLAB®
COMPANION
to COMPLEX
VARIABLES

A. David Wunsch
University of Massachusetts Lowell
MATLAB® is a trademark of The MathWorks, Inc. and is used with permission. The MathWorks does not
warrant the accuracy of the text or exercises in this book. This book’s use or discussion of MATLAB® soft-
ware or related products does not constitute endorsement or sponsorship by The MathWorks of a particular
pedagogical approach or particular use of the MATLAB® software.

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

© 2016 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Version Date: 20160315

International Standard Book Number-13: 978-1-4987-5568-9 (eBook - PDF)

This book contains information obtained from authentic and highly regarded sources. Reasonable efforts
have been made to publish reliable data and information, but the author and publisher cannot assume
responsibility for the validity of all materials or the consequences of their use. The authors and publishers
have attempted to trace the copyright holders of all material reproduced in this publication and apologize to
copyright holders if permission to publish in this form has not been obtained. If any copyright material has
not been acknowledged please write and let us know so we may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmit-
ted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented,
including photocopying, microfilming, and recording, or in any information storage or retrieval system,
without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.copyright.
com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and
registration for a variety of users. For organizations that have been granted a photocopy license by the CCC,
a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used
only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com
Contents

Acknowledgments..................................................................................................xi
Introduction.......................................................................................................... xiii
A Note to the Reader.............................................................................................xv
1. Complex Arithmetic........................................................................................1
1.1 The Rectangular Form..........................................................................1
1.1.1 A Caveat on Complex Numbers as Matrix Elements..........3
Exercises.............................................................................................................4
1.2 The Polar Form of Complex Numbers................................................5
Exercises.............................................................................................................8
1.3 Fractional Powers of Complex Numbers............................................9
1.3.1 Using sqrt................................................................................. 11
1.3.2 Reminder: A Warning about Square Roots and
Fractional Powers in MATLAB............................................. 11
1.3.3 Use of roots.............................................................................. 11
1.3.4 A Program to Give You Fractional Roots and
to Do a Check.......................................................................... 14
1.3.5 A Further Caveat with Fractional Powers:
The Plot Function................................................................... 14
Exercises........................................................................................................... 19
1.4 Complex Symbolic Algebra................................................................ 21
1.4.1 Numbers as Symbols..............................................................22
Exercises........................................................................................................... 24
2. Loci and Regions in the Complex Plane and Displaying
Complex Functions........................................................................................ 27
2.1 Meshgrid and Three-Dimensional Plotting.................................... 27
Exercises........................................................................................................... 33
2.2 Two-Dimensional Plots: The Contour Plot.......................................34
Exercises........................................................................................................... 38
2.3 Displaying Regions in the Complex Plane....................................... 39
2.3.1 A Note on pcolor.....................................................................43
Exercises...........................................................................................................44
2.4 Three-Dimensional and Contour Plots for Functions
with Singularities................................................................................. 45
2.4.1 Plots When There Are Isolated Singular Points................. 46
2.4.1.1 Removable Singular Points.................................... 46
2.4.1.2 Pole Singularities..................................................... 48
2.4.1.3 Essential Singularities............................................ 52
2.4.1.4 Branch Cut and Branch Point Singularities:
Three-Dimensional Plots.......................................54

vii
viii Contents

Exercises........................................................................................................... 58
2.5 Contour Plots Affected by Branch Cuts............................................ 60
Exercises........................................................................................................... 66

3. Sequences, Series, Limits, and Integrals.................................................. 69

3.1 Sequences.............................................................................................. 69
Exercises........................................................................................................... 76
3.2 Infinite Series and Their Convergence..............................................77
3.2.1 Using MATLAB to Obtain the Series Coefficients............. 82
3.2.2 Laurent Series..........................................................................84
Exercises...........................................................................................................90
3.3 Integration in the Complex Plane: Part 1, Finite Sums
as Approximations............................................................................... 93
Exercises........................................................................................................... 98
3.4 Integrations in the Complex Plane: Part 2, int and quadgk........ 100
3.4.1 Integration in the Complex Plane with MATLAB........... 104
3.4.2 Waypoints.............................................................................. 105
Exercises......................................................................................................... 107

4. Harmonic Functions, Conformal Mapping,

and Some Applications............................................................................... 113
4.1 Introduction........................................................................................ 113
Exercises......................................................................................................... 125
4.2 The Bilinear Transformation............................................................ 127
4.2.1 The Cross-Ratio..................................................................... 131
Exercises......................................................................................................... 135
4.3 Laplace’s Equation, Harmonic Functions: Voltage,
Temperature, and Fluid Flow........................................................... 138
Exercises......................................................................................................... 150
4.4 Complex Potentials, Cauchy–Riemann Equations,
the Stream Function, and Streamlines............................................ 152
4.4.1 Cauchy–Riemann Equations............................................... 152
Exercises......................................................................................................... 160
4.5 Mapping, Dirichlet, and Neumann Problems
and Line Sources............................................................................ 162
4.5.1 Dirichlet Problems................................................................ 162
4.5.2 A Dirichlet Problem.............................................................. 169
4.5.3 A Neumann Problem........................................................... 173
4.5.4 Line Sources and Conformal Mapping............................. 176
Exercises......................................................................................................... 182

5. Polynomials, Roots, the Principle of the Argument,

and Nyquist Stability.................................................................................. 195
5.1 Introduction........................................................................................ 195
5.2 The Fundamental Theorem of Algebra.......................................... 195
5.2.1 MATLAB roots...................................................................... 196
Contents ix

5.2.2 Use of Zeros........................................................................... 200

5.2.3 Roots versus Factor.............................................................. 201
5.2.4 Derivatives of Polynomials and the Function Roots....... 202
Exercises......................................................................................................... 205
5.3 The Principle of the Argument........................................................ 208
5.3.1 Rouché’s Theorem................................................................. 219
Exercises......................................................................................................... 221
5.4 Nyquist Plots, the Location of Roots, and the Pole-Zero Map.....223
5.4.1 The Use of pzmap................................................................. 230
5.4.2 A Caveat on the Use of pzmap........................................... 231
Exercises......................................................................................................... 233

6. Transforms: Laplace, Fourier, Z, and Hilbert......................................... 235

6.1 Introduction........................................................................................ 235
6.2 The Laplace Transform...................................................................... 235
6.2.1 Branch Points and Branch Cuts in
the Laplace Transform.......................................................... 239
6.2.2 Heaviside and Dirac Delta Functions................................ 240
6.2.3 The Inverse Laplace Transform.......................................... 246
Exercises......................................................................................................... 251
6.3 The Fourier Transform...................................................................... 257
6.3.1 Changing Variables: Fourier to Laplace and Vice Versa...... 266
6.3.2 Convolution of Functions ................................................... 267
Exercises......................................................................................................... 269
6.4 The Z Transform................................................................................ 271
6.4.1 The Z Transform of a Product............................................. 290
6.4.2 Inverse Z Transform of a Product and Convolutions...... 293
Exercises......................................................................................................... 295
6.5 The Hilbert Transform...................................................................... 299
6.5.1 Definition...............................................................................300
Exercises.........................................................................................................306

7. Coda: Fractals and the Mandelbrot Set...................................................309

7.1 Julia Sets.............................................................................................. 323
Appendices to Coda..................................................................................... 326
Appendix A......................................................................................... 326
Appendix B......................................................................................... 328
Appendix C......................................................................................... 329
Exercises......................................................................................................... 330
Index...................................................................................................................... 335
This page intentionally left blank
Acknowledgments

In writing this book, I consulted with a number of very generous people.

One is my brother, Carl Wunsch, formerly of MIT and now at Harvard. He
is a physical oceanographer and knows his fluid dynamics and helped me
with mine. His son, Jared, who is in the mathematics faculty at Northwestern
University, answered some questions about fractals without indicating that
I should have been able to figure these out for myself. He is a good mathema-
tician and a tactful teacher.
Kenneth Falconer, who is in the mathematics department at the University
of St. Andrews, has written a very graceful introduction to the subject of
fractals published by Oxford University Press. He has kindly answered
some of my e-mail inquiries on this subject. And equally generous has been
Professor Bodil Branner of Denmark, the first woman to head the Danish
Mathematical Society, who replied to my e-mails. I can recommend her
essay, The Mandelbrot Set, which appeared in the book Chaos and Fractals
published by the American Mathematical Society.
Michael F. Brown ably assisted me on the last two editions of my book on
complex variables and kindly found time to help me with this text. He is a
good mathematician and a careful reader.
Finally, MathWorks, publisher of MATLAB®, provides a team of consul-
tants to help authors who are writing MATLAB-based books. These people
have responded quickly to my e-mails and have almost always managed to
correct my work and supply helpful advice.

MATLAB® is a registered trademark of The MathWorks, Inc. For product

information, please contact:

The MathWorks, Inc.

3 Apple Hill Drive
Natick, MA 01760-2098, USA
Tel: 508-647-7000
Fax: 508-647-7001
E-mail: [email protected]
Web: www.mathworks.com

xi
This page intentionally left blank
Introduction

I hope the reader will enjoy using this book. It is for someone with at least a
beginner’s knowledge of MATLAB® who is learning the branch of advanced
calculus called “functions of a complex variable.” Complex variable theory
is the calculus of functions dependent on variables that can assume complex
numerical values. I want you to discover that MATLAB is your friend when
you are learning this type of mathematics.
One of the daunting aspects of complex numbers (and variables) is that
simple arithmetic operations that are easily done in one’s head, if they
involve real quantities, become tedious when the numbers are complex. You
can ask a class of third graders for the product of the numbers 3, 4, and 5,
and ­someone with the answer will raise his or her hand in less than 5 sec-
onds. Ask a college student for the product of the numbers (1 + 3i), (1 + 4i),
(1 + 5i) and he or she will, after a little squirming, probably resort to pencil
and paper, although the problem is solvable in one’s head with some serious
­concentration. If you were to ask me this question as I’m typing this sentence,
I would open the MATLAB window on my computer and type in the prod-
uct. Here we see a glimpse of the utility of MATLAB in complex arithmetic.
If you glance at the table of contents, you will see how MATLAB can be
your companion in such staples of complex variable theory as conformal
mapping, infinite series, contour integration, and Laplace and Fourier trans-
forms. Fractals, the most recent interesting topic involving complex vari-
ables, cries out to be treated with a language such as MATLAB, and you
might want to begin this book at its end, The Coda, which is devoted entirely
to this visually intriguing subject. However, I must add that as you progress
through the book, the MATLAB skill required increases gradually, and leap-
ing to the end is not for everyone.
Sometimes while working with MATLAB and complex algebra, you may
be puzzled at what you find. For example, if you ask a class of high school
( )
1/3
students to compute 53 , someone will quickly call out “five.” He or she
will know to multiply the exponents together. Now asking the same class
( )
2 1/2
for ( −1 + i ) , a student will follow the same logic and produce a cor-
rect answer: –1 + i. But another student, using MATLAB, will say 1 – i.
Both answers are correct, but why did MATLAB choose this one? This book
answers such questions and many similar ones.
This textbook does not purport to present MATLAB as a substitute for a
knowledge of the functions of a complex variable any more than MATLAB
can be used as a replacement for an actual understanding of elementary cal-
culus or linear algebra. This is also not a text from which one learns the
elements of MATLAB, although if you already know a little of the language,

xiii
xiv Introduction

it will expand your knowledge. Some books that will get the reader started
in the elements of MATLAB programming are listed as references [1,2].
MATLAB is not without constraints, assumptions, limitations, irritations,
and quirks, and there are subtleties involved in performing the calculus
of complex variable theory with this language that will be made evident
here. Without knowledge of these subtleties, the engineer or scientist who
is attempting to use MATLAB for solutions of practical problems in complex
variable theory suffers the real risk of making major mistakes. This book
should serve as an early warning system about these pitfalls.
This book should be read as a companion to standard texts on functions of
a complex variable. Throughout what follows, we refer to two of the author’s
favorites. Not surprisingly, one volume is his own Complex Variables with
Applications (3rd edition) published by Addison-Wesley in 2005. We refer to
this book as “W” in the text. In most cases, the section numbers referred to
apply to the second edition as well. There is a Spanish translation available
for those who prefer that language: Variable Compleja con Aplicaciones. The ref-
erence book in the Schaum’s outline series, Complex Variables, 2nd edition by
M. Spiegel et al. is an old favorite of mine, and although it is more handbook
than textbook, it is remarkably well done, and I will refer to it with the letter
“S.” Notice that the section numbers that I refer to apply to the second edition
only. Complex Variables and Applications by R. V. Churchill and colleagues is
very well written, but the book has had so many editions that it is difficult to
refer the student to any particular section.

References

1. Hunt, B., Lipsman, R., and Rosenberg, J. A Guide to MATLAB for Beginners and
Experienced Users, 3rd edition. New York: Cambridge University Press, 2014.
2. Hahn, B. and Valentine, D. Essential MATLAB for Engineers and Scientists, 5th
edition. New York: Academic Press, 2013.

A. David Wunsch
University of Massachusetts Lowell
Lowell, Massachusetts
A Note to the Reader

To keep the cost of this book reasonable, all figures are rendered in black and
white and shades of gray. Some of the codes provided here will produce only
these colors, while other programs will produce color plots on your screen
even though the plots are in grayscale in your book. In the solutions manual
I have not avoided color.
MATLAB® is upgraded at least once a year with new releases. Its capa-
bilities change and for the most part are improved. This book is based on
R2015a, the release available through most of the year 2015.
Corrections to this book as well as to the solutions manual will be posted
at the author’s web page hosted by the University of Massachusetts Lowell.
Here is the URL: http://faculty.uml.edu/awunsch/wunsch_complex_­
variables/faculty.htm.
The author invites corrections and comments for his work. An email
address can be found at the above website where you will also find solu-
tions to the odd numbered problems. To assist readers with data entry in
MATLAB, code text can be found for download at https://www.crcpress.
com/A-MatLab-Companion-to-Complex-Variables/Wunsch/9781498755672.
Instructors using my book in the classroom can also receive a copy of a
complete solutions manual, for all problems, if they write to me on their col-
lege stationary at the ECE Department, University of Massachusetts Lowell,
Lowell, MA 01854, USA.

xv
This page intentionally left blank
1
Complex Arithmetic

1.1  The Rectangular Form

A complex number z can be stated in the form

z = x + iy (1.1)

where x and y are real numbers, and there is multiplication between the i and
the y. An equally valid representation is

z = x + yi (1.2)

since complex numbers obey the commutative law of multiplication so that

the order of multiplying y with i is immaterial. Right away we must deal
with an idiosyncrasy of MATLAB®. The statement z = 3 + 4i can be entered
in the MATLAB command window with the result given in the following
example. Note that >> preceding a line of code indicates that the expression
was entered from the keyboard into the command window of MATLAB.

Example 1.1
>> z=3 + 4i
z = 3.0000 + 4.0000i

which is exactly what we hoped for: MATLAB has returned 3.0000 + 4.0000i.
However, entering z = 3 + i4 in the command window will result in an error
message. The value i must appear as the second factor in the multiplication if
i is to be interpreted by MATLAB as a multiplicative factor in z = x + yi. The
practice is only valid if i is preceded by an explicitly stated real (not complex)
number. A symbol cannot be used for that number. This entire convention
can be overlooked provided we employ the MATLAB ✽ for multiplication.
We will adopt that practice throughout this book even though the authors
of MATLAB claim that eliminating ✽ where allowed will speed up calcula-
tions; the advantage is often slight.

1
2 A MATLAB® Companion to Complex Variables

Note that in general, all the arithmetical operations that one does with real
numbers in MATLAB can be carried out for complex numbers, using the
same operators (i.e., the + and − signs for addition and subtraction, the / for
­division, the ^ for raising a number to a power, and as noted, the ✽ for mul-
tiplication). The precise meaning of what the ^ will yield when followed by
a fraction will be treated in section 1.3.
To get the magnitude (or absolute value or modulus) of a complex number,
we apply the operation abs to that number. Thus,

>> abs(3 + 4∗i)

ans = 5

Example 1.2
Multiply 3 + i4 by (1 + 2i) and add i2 to that result. Then find the absolute
value of that quantity.

Solution:
>> (3 + i∗4)∗(1 + 2∗i) + 2∗i
ans = −5.0000 + 12.0000i
>> abs(ans)
ans = 13

Electrical engineers often prefer to use j instead of i, and this notation can
be used automatically in MATLAB. We just use j instead of i and follow the
same conventions as above.
You should not use both i and j in the same code, or you will confuse your-
self and whoever reads what you wrote.

Example 1.3
Use j instead of i, and raise (1 + j3) to the –2 power, divide that result by
2 – j3, and subtract 3 + 4j from that result.

Solution:
>> (1 + 3∗j)^(−2)/(2−j∗3)−(3 + 4∗j)
ans = −2.9985 − 4.0277i

Note that although we used j instead of i, MATLAB returned an answer

employing i.
For some purposes, it is more convenient to put j = –i, and you should
make this statement at the start of your work if that is your preference. (Note
that j2 = –1 as before.) Recall that physicists prefer e –iωt in lieu of the electrical
engineer’s e jωt, where ω is a radian frequency and t is the time.
The real and imaginary parts of a complex number, for example z, are
found from the functions real(z), imag(z), and the conjugate of z is given by
Complex Arithmetic 3

conj(z) as shown by the following sequence of calculations. Note that in the

text material of this book, we will designate the conjugate of a quantity by an
overbar, as in z, which in MATLAB would of course be conj(z).

Example 1.4
>> z=3 + 4∗i
z = 3.0000 + 4.0000i
>> x=real(z)
x=3
>> y=imag(z)
y=4
>> w=conj(z)
w = 3.0000 − 4.0000i

1.1.1  A Caveat on Complex Numbers as Matrix Elements

MATLAB, as its name might suggest, is a computer language for technical
computation that is based on matrices (i.e., rectangular arrays of numbers).
Sometimes the matrix consists of just a horizontal row of numbers, in a
specific order, or a column of numbers in a certain order. Such matrices are
known as vectors and more specifically as row or column vectors. The num-
bers that make up matrices, the elements, can be complex or real numbers,
or symbols. When entering complex numbers as the elements, it is useful to
state them with parentheses surrounding each complex number. In other
words, instead of entering 3 + 4∗i, you should enter (3 + 4∗i). Otherwise,
an inadvertent use of the space bar can result in an error, as shown by the
following where we want to enter a row vector a having two elements, one
being 3 + i and the other 2−i. In the first instance, we will do this correctly
without parentheses, then incorrectly without parentheses, because we have
placed a space where it does not belong, and finally correctly with parenthe-
ses even though there is an unnecessary space:

>> a=[3 + i  2−i]

a = 3.0000 + 1.0000i  2.0000 − 1.0000i
>> a=[3  + i  2−i]
a = 3.0000  0 + 1.0000i  2.0000 − 1.0000i
>> a=[(3  + i)  (2−i)]
a = 3.0000 + 1.0000i  2.0000 − 1.0000i

Comment: Note that in the second instance, the value obtained for a is

  3.0000  0 + 1.0000i  2.0000 − 1.0000i

which is a row vector with three elements, not the desired two elements. The
first element is 3, and the second is 0 + 1.0000i=i. The error was caused by our
having a space after the 3 and before the adjacent + sign.
4 A MATLAB® Companion to Complex Variables

In the third output for a, we got the correct value even though we typed a
space between the 3 and the plus sign following it.
Note that you can always tell the number of elements in a row or column
vector, let us call it a, with the command length(a). If we apply this operation
to the three values of a described above, we get 2, 3, and 2, respectively.

Exercises
1. Using MATLAB, determine
3 − 4i
a. + i2
1 + i2
2+i
b. − 1 − i/2
3 + 5i
i
(1 + i2/3) −
c.
3

3 − 4i
11
 2
 1 + 
d.
i3
5
 2i  1 
e.
 3 + 4i +  5 − 4i  
( ) 
2. The numerical value of the expression (1 + i)37 can be deter-
mined without recourse to MATLAB by the following technique:
( )
18
(1 + i)37 = (1 + i)36 (1 + i) = (1 + i)2 (1 + i). Now complete the calcula-
tion by evaluating (1 + i)2 and noting that i raised to any even power
is easily computed. Check your answer by using MATLAB to find
(1 + i)37 directly.
3. Consider the finite series (N ≥ 0 is an integer) and its sum.
N
1 − z N +1
∑n= 0
zn =
1− z
z ≠ 1, whose derivation is identical to that given

in real calculus for the sum of a finite geometric series.

2 3 10
i  i  i  i
a. Using MATLAB, sum the series 1 + +   +   + ... +  
2  2  2  2
directly by using a loop created from the while or for com-
mands. Use the “long format” of MATLAB as you will need it in
part (b).
b. Verify the answer obtained in (a) by using the closed-form
expression for the sum given above, again using the long format
so that you can notice any small discrepancy.
Complex Arithmetic 5

1.2  The Polar Form of Complex Numbers

Referring to Figure  1.1, we are reminded that a complex number z can be
expressed using the polar coordinates r and θ. Thus,

z = x + iy = r cosθ + ir sinθ = r (cosθ + i sinθ) (1.3)

Sometimes cisθ is used to mean cosθ + i sinθ, but other times, the notation
∠θ is used for the same quantity. Thus,

z = rcisθ = r∠θ (1.4)

Generally, θ is expressed in radians, but one might also use degrees.

Now z = x 2 + y 2 = r, i.e., the polar coordinate r is simply the magnitude
of the complex number z, and we have in MATLAB that r = abs(z). The quan-
tity θ is called the angle or the argument of the complex number z.

4.5

4 z

3.5

2.5
r

1.5

1
θ
0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

FIGURE 1.1
The polar coordinates r and θ for the point z.
6 A MATLAB® Companion to Complex Variables

There are some subtleties with θ since it is multivalued. If θ0 is a possible

value of θ, then so is θ = θ0 + 2kπ, where k = 0, ±1, ±2, …. If we are working in
degrees, the expression becomes θ = θ0 + k360°. Given a complex number in
rectangular form, MATLAB can be used to yield a value of θ in radians from
the function angle(z). But which value will it yield? It yields the principal
value of the argument, the one satisfying −π < θ ≤ π. Note that a few textbooks
use a different definition for the principal value and that the value returned
by MATLAB would not be in conformity with theirs.*

Example 1.5
Find the polar form of the complex number − 3 − i.

Solution:
>> z= −sqrt(3)−i;
r=abs(z)
r = 2.0000
>> theta=angle(z)
theta = −2.6180

Thus, we have shown z = 2∠ −2.6180… in polar form.

A short calculation with MATLAB shows that −2.6180 agrees with
−5π/6 (or −5pi/6 in MATLAB) to as many decimal places as MATLAB
will display in the standard format. Construction of the vector for 
− 3 − i on a piece of paper and recognizing the resulting triangle with
angles 30, 60, and 90 degrees should show where the angle −5π/6 arises
from.

Incidentally, the above rectangular to polar calculation can be done a lit-

tle more easily in MATLAB as [th,r] = car2pol(x,y), which converts complex
numbers from Cartesian to polar form. Thus, using numbers from above,

>> [th,r]=cart2pol(−sqrt(3),−1)
th = −2.6180
r = 2.0000

which agree with what we found.

Example 1.6
Using MATLAB, compute the angles of the complex numbers −1+eps∗i
and −1−eps∗i and compare them by computing their difference. Recall
that in MATLAB, eps is the smallest possible difference between two
floating point numbers. In the version of MATLAB used in this book,
eps ≈ 2.22 × 10−16. If you type the word eps in the MATLAB command
window, you can find out what eps is for your version of MATLAB. We
chose here to work in the “long” format.

* These books take 0 ≤ θ < 2π.

Complex Arithmetic 7

Solution:
>> format long
>> angle1=angle(−1 + i∗eps)
angle1 = 3.141592653589793
>> angle2=angle(−1−i∗eps)
angle2 = −3.14159265358979
>> angle1−angle2
ans = 6.28318530717959
>> 2∗pi
ans = 6.28318530717959

Note that although graphical representation of −1 + i∗eps and −1 − i∗eps

as points in the complex plane will involve neighboring points that are
exceedingly close to one another, these points lie just above and below the
negative real axes. Their principal angles are nearly π and −π, respectively.
The differences in their angles, as computed by MATLAB, are indistin-
guishable from 2π.

Referring to Figure  1.1, we see that θ = arctan(y/x), and we might be

tempted to use the function atan in MATLAB, which yields the arctangent,
in order  to obtain θ, the angle of the complex number z = x + iy. This is
to be avoided; because of the multiplicity of values of the arctangent, the
computer might yield  the wrong value. If, for example, x  =  1 and y =  1,
then atan(1/1) is evaluated by MATLAB to be 0.78539816339745 or π/4, that
is a correct value as a sketch of the vector for this complex number shows.
However, if x = −1, y = −1, we have from MATLAB that atan(−1/−1) is again
0.78539816339745, which is incorrect for the angle of this complex number.
A correct value, as a sketch of the corresponding vector shows, is −3π/4
= −2.35619449019234. In general, it is better to use the angle function of
MATLAB to avoid these situations. However, one might use the function
atan2(x,y) in MATLAB where the real and imaginary parts of the com-
plex number z = x + iy are entered in a particular order into the function.
Additionally, the MATLAB function [th,r] = car2pol(x,y) will always yield the
correct angle th (or θ).
Given a complex number in the polar form (a magnitude r and an angle θ),
we can compute the Cartesian form by asking MATLAB for r cosθ = x and
r sinθ = y. Alternatively, we can use a function M file built into MATLAB
that will automatically do the two calculations. The function is [x,y] =
pol2cart(th,r), where the angle θ in radians, here called th, is entered first
followed by r. Suppose we wish to convert the complex number 2∠π/6 to
rectangular form. We proceed as follows:

>> [x,y] = pol2cart(pi/6,2)

x = 1.73205080756888
y = 1.00000000000000
8 A MATLAB® Companion to Complex Variables

Note that the program does not yield the complex number z = x + iy. It sim-
ply gives the real and imaginary parts of z. To get z, one must take the two
results from above and make that additional computation.

Exercises
1. We know that the polar angle of the product of two complex num-
bers can be expressed as the sum of the angles of each.
( )
a. Multiply (1 + i) by 3 + i using MATLAB. Find the angle of the
product using MATLAB and show that it is the sum of the angles
of each factor.
b. Repeat the above exercise, but use as factors (1 + i) and − 3 + i.
Explain why the angle of each factor, obtained from MATLAB,
when added, does not agree with the angle that MATLAB
yields for the angle of the product. Compute these angles using
MATLAB.
2. When MATLAB is asked for the argument (angle) of a complex num-
ber, it always returns the principal value. Use MATLAB to find the
principal argument of the following numbers where all angles are in
radians except in part (d).
a. 5cis(197)
b. 1/(4∠79)
c. (1 + 2i)17
d. (5cis(53.3°))8 (give answer in degrees)
3. Using the pol2cart function in MATLAB described above, convert
these numbers to rectangular components where the angles below
are in radians:
a. 3cis(−207)
b. 4∠50
4. Write a new function, similar to the pol2cart function in such a
way that if it is given θ in degrees and r, it will convert the polar
form of the number to rectangular form. Call the new function
­pold2cart. Use it to convert these numbers to rectangular form.
Give x and y.
a. 5∠60°
b. 2∠(−320°)
Complex Arithmetic 9

1.3  Fractional Powers of Complex Numbers

First, we make a note on notation. In this book, as in many texts on complex
variable theory, the symbol β is applied only when β is a non-negative real
number and is intended to mean the non-negative square root. Thus, 16
is always 4 and never −4. Similarly, the m β is applied when β has this same
restriction and m is an integer and refers to the mth root of β but only to
the non-negative real value. Thus, 3 64 is 4 and not any of the other two pos-
sible values, which are complex.
Suppose you wish to solve the equation z4 −1 = 0. We have z4 = 1. Notice that
our equation has four solutions: ±1 and ±i. However, if you attempt to solve
the given equation using MATLAB by asking it for z = 11/4, which is written in
MATLAB as 1^(1/4), you will receive just one of the four answers, in this case, 1.
The question arises as to how MATLAB decides which answer to give you
when a number is raised to a fractional power and how we might get the rest.
If z is to be raised to a fractional power n/m where m and n are integers,
with m ≠ 0, and where the fraction n/m cannot be reduced, we convert z to
the polar form z = r∠θ, and as shown in the standard texts, we have m values
given by

( r) cis  θ + 2 kπ  k = 0, 1, 2, ... m − 1 (1.5)

n n
n
z n/m = m
m m

where as we recall, cisϕ = cosϕ + i sinϕ. This result is consistent with the defi-
nition zn/m = (z1/m)n. Notice the order of the operations.
In what follows, we will take m as positive and skip the absolute magnitude
signs around the m. We can do this because it is easily shown that the set of
values of z−n/m is identical to the set of values of zn/(−m). Thus, given the prob-
lem of computing (1 + i)2/(−3), we can compute the set of values of (1 + i)−2/3
and use a positive m.
Equation 1.5 is equivalent to

( r) n   n
n
z n/m = m
cis  θ  cis  2 kπ  k = 0, 1, 2,... (m − 1) (1.6)
m   m

which arises from the identity cis(α + β) = cisαcisβ; it applies to any numbers
α and β. Note that in Equation 1.6, we can generate the same set of values
for zn/m if we let k go through any set of m consecutive integers, for example,
1,2, … m.
The results from Equations 1.5 and 1.6 are m numerically distinct ­values.
When plotted as points in the complex plane, they all lie on a circle having
( )
m
radius n r (i.e., they have identical magnitudes). The angular spacing of
these values is 2π/m. The separation of these points on the circle is uniform.
10 A MATLAB® Companion to Complex Variables

If you ask MATLAB for the computation of a fractional power, for example,
(81i)3/4, you will receive exactly one value, in this case, 10.3325 + 24.9447i, which
in polar form is 27 ∠1.1781…. The angle here is 3π/8 radians. According to
Equations 1.5 and 1.6, there are three other values for (81i)3/4. After some experi-
mentation with MATLAB, you will see how it decides which value of zn/m to
give you.

MATLAB Rule for Fractional Powers The angle θ in Equations 1.5 and 1.6 is
taken as the principal value, and the value of k is set equal to zero. MATLAB returns
the result in Cartesian form.
You should confirm that this convention yields the MATLAB result just
given for (81i)3/4.

Example 1.7
Use Equation 1.5, taking the MATLAB convention for fractional pow-
ers, to determine one value of (−1 − i)2/3, and verify that MATLAB does
indeed yield this value.

Solution:
We have −1 − i = 2 ∠(−3 π / 4), where the principal value of the argument
of (−1 − i) is used.
Using Equation 1.5, with k = 0 we have

( 2 ) ∠  −3 π4 23  =  π
2/3 2
(−1 − i)2/3 =  2 ∠ ( −3π / 4 )  = 3 3
2 ∠  −  = −i 3
2
 2
Proceeding with MATLAB, we have

>> (−1−i)^(2/3)
ans = 0.0000 − 1.2599i

These two answers are numerically identical because the cube root of
2 is 1.2599….
If we compute a value of zn/m and raise it to the m/n power, do we necessar-
ily get back z? That is, does (zn/m)m/n equal z when the inner and outer opera-
tions are done in MATLAB? Because MATLAB will choose just one value of
the n possible values when raising a number to the m/n power, you cannot
say for certain that in MATLAB (zn/m)m/n = z.
( )
3/4
The reader should confirm that in MATLAB we have (−1)4/3 = −i, so the
original z (namely, −1) is not returned. The result comes about because the
principal angle of −1 is π. When MATLAB evaluates (−1)4/3 using this angle,
−1 3
and the rule stated above, the result is −i whose polar representation,
2 2
 π 
using the principal angle, is 1∠  −2  . Raising the preceding to the ¾ power
 3
using the MATLAB rule for fractional powers, we get 1∠  − π  = − i .
 
 2
Complex Arithmetic 11

( )
3/4
For a different problem, MATLAB yields (− i)4/3 = − i. The value of z
is returned. One should be able to predict in advance whether the original
z is obtained if we compute (zn/m)m/n. This matter is explored in problem 2 in
the Exercises.

1.3.1  Using sqrt

If you want to obtain the square root of a number, z, using MATLAB, you
could ask for the computation z^(1/2). The resulting root is the one supplied
by the MATLAB rule given above for fractional powers. Alternatively, we
might ask for sqrt(z), which requires about the same amount of typing. The
same root is obtained as if you were using z^(1/2)—that is, you get rcis(θ/2),
where θ is the principal angle of z. So, for example, we have
>> format short
>> sqrt(−9)
ans = 0 + 3.0000i
We did not get −3i, the other square root.

1.3.2 Reminder: A Warning about Square Roots and

Fractional Powers in MATLAB
If z is a matrix and you ask MATLAB for sqrt(z), then MATLAB will return to
you a matrix whose elements are the square root of each element of z. However,
if you ask MATLAB for z^(1/2), it will return the square root of the matrix z
(provided it is square matrix, like 2 × 2)—that is, a matrix whose square is z.
To get the same matrix as produced by sqrt(z), you will need z.^(1/2), where a
dot follows the z. Similar precautions must be taken where other exponents are
concerned.

1.3.3  Use of roots

Suppose we want to obtain all m values of zn/m through the use of MATLAB.
A simple method involves the function roots. Let us begin by using roots to
find z1/m, where m is a positive integer. Observe that all m solutions for the
unknown w in the equation wm = z are the m values of z1/m. MATLAB has a
function called roots that will solve the polynomial equation

c1wm + c2wm−1 + c3wm−2 + … + cmw + cm + 1 = 0

and yield all m roots. The procedure is to ask MATLAB for roots(p), where p
is a row vector whose entries, from left to right, are the numerical coefficients
c1, c2, … cm + 1. Note that if we take c1 = 1, c2 = 0, c3 = 0, … cm + 1 = −z in the poly-
nomial equation, we obtain wm −z = 0 whose solutions all satisfy w = z1/m.
12 A MATLAB® Companion to Complex Variables

Suppose for example we wanted all six values of (1 + i)1/6. This is the same
problem as finding all six solutions of the polynomial equation w6 − (1 + i) = 0.
The expression on the left is a polynomial of degree 6. We should see that
this polynomial has coefficients of zero for the terms w5, w4, …, w, while the
coefficient of w0 is −(1 + i). Thus, to solve our polynomial sixth-degree equa-
tion in w, we proceed as follows:

>> coeffs = [1 0 0 0 0 0 −(1 + i)];

>> format short
>> roots(coeffs)
ans =
−1.0504 − 0.1383i
−0.6450 + 0.8405i
−0.4054 − 0.9788i
0.4054 + 0.9788i
0.6450 − 0.8405i
1.0504 + 0.1383i

The above are the six values of (1 + i)1/6. The following provides a check.

>> ans.^6
ans =
1.0000 + 1.0000i
1.0000 + 1.0000i
1.0000 + 1.0000i
1.0000 + 1.0000i
1.0000 + 1.0000i
1.0000 + 1.0000i

There is some round-off error, which is hidden because we are using short
format, but each of these expressions is, to our degree of approximation,
(1 + i). This is correct, as we expect all six values of (1 + i)1/6 when raised to
the sixth power to yield (1 + i).
Suppose we want all seven values of (3 + 4i)5/7. We would proceed as
before, calculating (3 + 4i)1/7. We would then take all seven of these values
and raise them to the fifth power. Here is the procedure:

>> coeffs = zeros(1,8);

%the above takes all the coeffs as zero at first
coeffs(1) = 1;
coeffs(8) = −(3 + 4∗i);
poly_roots = roots(coeffs);
final_answer = poly_roots.^5
Complex Arithmetic 13

The output is
final_answer =
−3.0852 − 0.6691i
0.0342 + 3.1567i
3.0700 − 0.7358i
−1.4005 − 2.8293i
1.3388 − 2.8590i
−2.4467 + 1.9950i
2.4894 + 1.9414i

Note that there are seven answers, as expected. The first line of our code
coeffs=zeros(1,8); establishes a row vector having eight elements. Initially all
eight are zero. Then we set the first element equal to one and the last equal
to –(3 + 4i).
The above technique can be applied to finding all |m| values of zn/m when
m is a negative integer. We compute z−n/|m| and proceed as above, first finding
the |m| values of z1/|m| and then raising them to the –nth power.
Suppose we were to take each of the values of (3 + 4i)5/7 obtained above
and, using MATLAB, raise them all to the 7/5 power. Can we expect in each
case to recover (3 + 4i)? Trying this, we have

>> recover = final_answer.^(7/5)

recover =
−2.8772 + 4.0892i
−2.8772 + 4.0892i
4.7313 − 1.6171i
−4.7782 − 1.4727i
−0.0759 − 4.9994i
−4.7782 − 1.4727i
3.0000 + 4.0000i

Notice that only one of these results is the original 3 + 4i. This is because
there are five possible values of a number raised to the 7/5th power. Just one
( )
7/5
of the above yields ultimately to our having (3 + 4i)5/7 = 3 + 4i.
Keep the following in mind: Suppose using any method at your disposal
you compute all |m| values of zn/m, where n and m are integers, and m ≠ 0.
Now suppose you raise each value to the m/n power using the MATLAB
expression V^(m/n), where V is any one of those m values of zn/m. There is no
guarantee that you will get back z. A simple example will suffice. We know
that (−1)3/2 = (−1)1/2 (−1) = ±i. From MATLAB, we raise both i and −i to the
2/3rds power:

>> i^(2/3)
ans = 0.5000 + 0.8660i
14 A MATLAB® Companion to Complex Variables

>> (−i)^(2/3)
ans = 0.5000 − 0.8660i

In neither case did we recover –1. However, if you computed all values of
i 2/3 and (–i)2/3, you would find in each case that one of the values obtained
is –1.

1.3.4  A Program to Give You Fractional Roots and to Do a Check

Here is a simple MATLAB program that asks you to choose a value of z
and a positive integer m. The program then gives you all values of z1/m.
It  does this by solving the polynomial equation in w : wm – z = 0 using
roots.
We have added some additional checks to our program. Notice that if you
were to factor wm – z into its m factors (w – w1) (w – w2) … (w – wm), then the
­product of the roots w1w2 … wm is seen to equal (–1)m–1z; the product of the
m  ­values of z1/m is thus (–1)m–1z. Similarly, you should convince yourself that
the coefficient of wm–1 obtained when you multiply the m factors is –(w1 + w2 +
w3 + … wm). But since wm – 1 does not appear in wm – z, we conclude that the sum
of the m values of z1/m is zero.
Try the following problem with some numbers:

z = input('the complex value of z to be considered')

m = input('input the positive integer m')
v = zeros(1,m + 1);%creates a row vector length m + 1
%elements are zero.
v(1,1) = 1;% the first coeff is 1, the coeff of w^m
v(1,m + 1) = −z;% the constant term (the last coeff) is −z
%the following solves the equation w^m−z = 0 for w = z^(1/m)
the_roots = roots(v)
prod_roots = prod(the_roots)
%the preceding should equal (−1)^(m−1) times z.
check_prod =(−1)^(m−1)*z
check_roots = sum(the_roots)
%the preceding should be zero to a good approx.

As an exercise, the reader should modify the above code so that the user
is prompted to enter an integer n and a positive integer m where the fraction
n/m is irreducible. The program then computes all m values of zn/m where
again z must be supplied.

1.3.5  A Further Caveat with Fractional Powers: The Plot Function

Suppose x is a row vector whose elements are real numbers. Let w(x) be a
function defined for these values of x. A commonly used line in MATLAB
Complex Arithmetic 15

code is plot(x,w), which will produce a plot in the Cartesian plane of w

­versus x. For example, if x = linspace(0,10,1000) and w = sin(x), the command
plot (x,w) gives you a nice plot of the sine function based on 1000 data points
spaced between 0 and 10.
One must be careful, however. If w is a complex function of the real vari-
able x, then the command plot (x,w) yields a plot of only the real part of w ver-
sus x. This is demonstrated in the following example.

Example 1.8
Using the MATLAB plot function, graph w = x1/3 for the interval −2 ≤ x ≤ 2
and explain your result. Use at least 100 data points.

Solution:
x = linspace (−2,2,101);
w = x.^(1/3);
plot(x,w);grid

which results in Figure 1.2.

1.4

1.2

0.8

0.6

0.4

0.2

0
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 1.2
The plot for Example 1.8.
16 A MATLAB® Companion to Complex Variables

Notice the lack of symmetry about x = 0. In executing the program,

MATLAB issues a warning about the imaginary value of the function
being ignored in the plotting process.
To explain this result, notice that

1 2 
x 1/3 = 3 x ∠  θ + 2 kπ k = 0, 1, 2
3 3 

MATLAB employs the principal value, taking k = 0, and θ = 0 when x > 0.

Thus, for x ≥ 0, we have x 1/3 = 3 x . If x < 0, we have from the principal value

1 
x 1/3 = 3 x ∠  π
3 

because θ = π on the negative real axis. In Cartesian form, the preceding is

x 1/3 = 3 x (cos(π/3)+i sin(π/3)) = 3 x  1 2 +i 3 

2 


13
In our preceding graph above, we have the curve x for x < 0 because
2
only the real part of x1/3 is plotted by MATLAB. This is exactly half the
value obtained for the corresponding positive values of x and is con-
firmed if we study the plot.
To properly graph x1/3 over the given interval, we must employ both the
real and imaginary parts of this function as in the following code:

x = linspace (−2,2,101);
w = x.^(1/3);
plot(x,real(w),'−',x,imag(w),'*')
grid

This results in Figure 1.3, to which we have added some labels.

Finally, as we are on the subject of plotting, the reader should be
reminded that the command plot(w), where w is a row vector having
complex elements, yields a graph in which the imaginary part of w is
plotted against the real part. In other words, w is plotted as points in the
complex plane. Here is a simple example that yields a circle of radius 10
in the complex plane:

>> theta = linspace(0,2∗pi,100);

w = 10∗exp(i∗theta);
plot(w);axis equal

The reader should verify this. Note that axis equal ensures that a cir-
cular plot appears. Without it, MATLAB might use different scales for
the two axes, and the circle would be distorted into an ellipse.
We could have also just plotted 100 stars, like this ∗, for the data points
and not obtained a continuous graph if we used plot(w,’∗’) instead of
plot(w). Here is the result. (See Figure 1.4.)
Complex Arithmetic 17

1.4

Real part
1.2
Imaginary part

0.8

Imaginary part
0.6

0.4

Real part
0.2

0
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 1.3
Plots for x1/3.

–2

–4

–6

–8

–10 –5 0 5 10

FIGURE 1.4
Values of 10eiθ generated by letting θ go through 100 values between 0 and 2π.
18 A MATLAB® Companion to Complex Variables

And while we are on the subject of plot, here is a further peril involv-
ing plot when you are using vectors with complex elements together
with vectors having real elements. Consider the following:

clf
a =[1 + i 5 + 5*i 10 + 10*i];
b =[1 5 10];
plot(a,'k−','linewidth',2);hold on
text(4,4,'the plot of a')
plot(a,'o')
plot(b,'k−','linewidth',2);
plot(b,'o');
text(2.2,5,'the plot of b');grid

We obtain Figure 1.5.
The plot of the vector a shows its imaginary elements plotted against
their corresponding real parts. They all lie on the line y = x, which makes
a 45-degree angle with the real axis.
Now the vector b has only real elements, which are 1 and 5 and 10.
The imaginary part of each element is zero. So, following the logic used
by MATLAB in making the plot for a, we would expect that these points
would occur on the line y = 0 at x = 1 and 5 and 10. We would expect
MATLAB to connect these points with a horizontal line.

10

6
y

The plot of b
5
The plot of a

1
1 2 3 4 5 6 7 8 9 10
x

FIGURE 1.5
Perils of using the plot function.
Complex Arithmetic 19

Instead, MATLAB, seeing that each element in b is real, plots the

value of that element against its index in the row vector b. Thus, the
first element, one, is plotted against its index, one. The second element
has the value 5 and this value is plotted against 2, and finally the third
element is 10 and this is plotted at x = 3, y = 10. The preceding exercise
should serve as a warning and an explanation of seemingly strange
results.

Exercises
1. Using MATLAB, obtain one numerical value for each of the follow-
ing by simply asking for a^b.
a. (−27i)2/3
b. (1 + i)1/15
c. (−1−i)1/9
2. Without using MATLAB, but knowing its convention for produc-
ing fractional roots, predict whether or not the use of MATLAB will
yield these results, and explain your logic:
a. (i1/2)2 = i
b. (i3/2)2/3 = i
c. ((−1)4/3)3/4 = −1
d. ((−1)3/4)4/3 = −1
e. ((−1 + i)5/3)3/5 = −1 + i
f. Now confirm your predictions for parts (a) through (e) by using
MATLAB.
3. Using roots, as discussed, find all values of the following and in
each case plot the values using a ∗ to mark their location.
a. (1 + i)24/5
b. (3 − 4i)5/7
c. (13 + 7i)100/99 (be sure to use the function zeros in this case to cre-
ate the coefficients for roots)
4. Using Equation 1.6, we can sum all the m possible values of zn/m. The
result is
m− 1

( r) cis  θ 
n
∑ cis 2kπ n 
n
S= m
 m  m 
k=0

We can show that this summation S = 0. In other words, the sum

of all possible values of zn/m is zero.
20 A MATLAB® Companion to Complex Variables

Recall De Moivre’s Theorem (cisϕ)k = cis(kϕ), where k is any integer

and ϕ is any number. The result is in all complex variables texts.
Thus, we may rewrite the expression for S as

m− 1

( r) cis  θ 
n
∑ (cis(2πn/m))
n k
S= m
 m  k=0

We can sum this series using the result given in exercise  3,

­section  1.1, for the sum of a geometric series that we rewrite
here as

m− 1

∑ zk =
1 − zm
1− z
z≠1
k=0

Taking z = cis(2πn/m) (where n/m is not an integer), we have that

m− 1
1 − cis(2 πnm/m)
∑ ( cis(2πn/m))
k
= = 0 since cis(2πn) = 1
k=0
1 − cis(2 πn/m)

Using MATLAB, find all the values of (1 + i)11/7 and sum them
together. Note that you will not get exactly zero. This is due to
round-off errors in MATLAB. However, the sum should be many
orders smaller than the magnitude of any one root.
5. Compare Figures 1 and 2 obtained from this program, explaining
differences and similarities:

clf; clear
x = linspace (−3,3,301);
w =(x−1).^(2/5);
figure(1)
plot(x,w);grid;
figure(2)
plot(x, real(w),'−',x,imag(w),'*')
grid;

6. Modify the program given in the text, which computes z1/m so that
when supplied with integers n and a positive m (these numbers have
no common factors), it will compute all m values of zn/m. Try the prob-
lem out by computing (27i)2/3 and (27i)−2/3.
Complex Arithmetic 21

1.4  Complex Symbolic Algebra

One of the more tedious aspects of the algebra of complex numbers is the
manipulation of the symbols in complex expressions, especially the extrac-
tion of the real and imaginary parts following a procedure requiring addi-
tion, multiplication, and division. Suppose, for example, that a, b, c, d, e, f are
all real variables. Consider the equation

1 c + id
w= +
a + ib e + if

if w = u + iv, where u and v are real. Our problem is to find u and v in terms
of a, b, c, d, e, f.
Problems such as these are solved in MATLAB by using the statements
syms or sym. These two statements are not identical in what they do. The
reader might wish to investigate them both in the Help window of MATLAB.
In general, syms is slightly more useful, and we will favor it, but we also use
sym. Briefly, if a and b are symbols, we write

syms a; syms b

or more concisely

syms a b

We can put any other symbols we might be using in the above statement
after a and b. Suppose a is a symbol for a real quantity but b can be complex.
Then our code would contain
syms a real; syms b
However, if both a and b must be real, then we have
syms a b real;

We could have included other symbols that are real after the a and b.
The following program shows how to obtain the real part of the symbolic
expression for u, which is the real part of w given above.

syms a b c d e f real
w = 1/(a + i*b)+(c + i*d)/(e + i*f);
u = real(w)
u = expand(u)
'prettier answer for u'
pretty(u)
22 A MATLAB® Companion to Complex Variables

The output is
u = real((c + d∗i)/(e + f∗i)) + a/(a^2 + b^2)
u = a/(a^2 + b^2) + (c∗e)/(e^2 + f^2) + (d∗f)/(e^2 + f^2)
a ce df
ans = prettier answer for u a 2 + b 2 + e 2 + f 2 + e 2 + f 2

Notice the following: We did not have to declare w, u, or v as symbols as

they are declared equal to a collection of symbolic expressions. We had to
use expand(u) in our code, otherwise we could not have found an explicit
expression that does not involve the real operator. The line of code pretty is
not necessary but does present u as a sum of built-up (as opposed to “shil-
ling”) fractions. Had we used imag instead of real, we would of course have
found the imaginary part of our expression u.
Suppose you wished to substitute the numbers 3 and 7 for a and e, respec-
tively, in our expression for u. This is accomplished by adding this line of
code to the program just given. You may wish to read up on the MATLAB
function subs in the help window.
u = subs(u,[a e],[3 7])

This results in the output:

u = (7∗c)/(f^2 + 49) + 3/(b^2 + 9) + (d∗f)/(f^2 + 49)

where a and e have now been replaced by numbers.

There are some algebraic operations that are sufficiently complicated that
MATLAB in its present version will not simplify them or make the real and
imaginary parts readily accessible. For example, if z = x + iy, then it can be
shown (see W section 1.4, Exercises) that

± x + x2 + y 2 − x ± x2 + y 2
z1/2 = ( x + iy )1/2 = ±i
2 2

However, we cannot write a program like the one given above that will
give this result.

1.4.1  Numbers as Symbols

Sometimes it is useful to have MATLAB treat a number as a symbol and not
convert it to a decimal expression. For example, compare the following two
MATLAB operations:
>> a = sqrt(5)
a = 2.2361
>> a^7
ans = 279.5085
Complex Arithmetic 23

>> b = sym(sqrt(5))
b = 5^(1/2)
>> b^7
ans = 125∗5^(1/2)

In the first instance, we raised the square root of five to the seventh power
and came up with a decimal answer of around 279.50. In the second instance,
we treated the square root of 5 as a symbol. Raising this to the seventh
power we get 125 times the square root of 5. Sometimes this is more conve-
nient. If we now want this result as a decimal we add the line

>> eval(ans)
which again yields
ans = 279.5085

( )
11
Now, consider the following where we find 5 +i 7 :

>> (sqrt(5) + i∗sqrt(7))^11

ans = −8.5407e + 05 − 1.1650e + 05i

We know from the binomial theorem that our final result should be
expressible in terms of the square root of 5 and the square root of 7 and
­integers. But how?

>> a = sym(sqrt(5));
>> b = sym(sqrt(7));
>> z = (a + i∗b)^11
z = (5^(1/2) + 7^(1/2)∗1i)^11
>> expand(z)
ans = −381952∗5^(1/2) − 7^(1/2)∗44032i

This gives the desired result that we can now check against our first
answer.

>> eval(ans)
ans = −8.5407e + 05 − 1.1650e + 05i

which agrees with our first result.

The line of code who placed in your program or command window will
review for you everything that has been defined as a non-numerical symbol
by you or created as a symbol in your program. Thus, using who following
the above MATLAB code yields the following:

Your variables are

a ans 
b z
24 A MATLAB® Companion to Complex Variables

For some purposes, it is best to take pi as a symbol. Compare the following:

>> sin(1000∗pi)
ans = −3.2142e−13
>> sin(1000∗sym(pi))
ans = 0

The first answer is wrong—it results from round-off error. The second one
is correct. By using pi as a symbol, we have MATLAB evaluating sinπ and not
the sine of a numerical approximation to the irrational number.

Exercises
1. Use MATLAB to extract the real and imaginary parts of the expression

1 1 i
w= + +
x iy (( x − 1) + iy )2

where x and y are real. Use the pretty command to make your
answers as easy to read as possible. Use MATLAB to evaluate your
expressions, when x = 1 and y = 1. Note that for these values the
numerical results are easily checked by inspection.
2. Consider the expression

1
z = R + iX =
1 1
+
r1 + ix1 r2 + ix2

where R, X, r1, r2, x1, x2 are all real quantities. This notation is com-
mon in electrical engineering. Write a MATLAB program that will
give explicit expressions for R and X expressed entirely in terms of
real quantities. In addition, have your program determine the values
of R and X if r1 = r2 = 2 and x1 = x2 = 4. Also, consider the case where
x1 = −x2 = 4 and again r1 = r2 = 2.
3. Using MATLAB, find the real and imaginary parts of

1
( x + iy )3

as explicit functions of x and y, assuming that x and y are real.

Complex Arithmetic 25

4. Let
6
z
w = u + iv =  
 z

where z = x + iy. Using MATLAB, obtain explicit real expressions for

u(x, y) and v(x, y). Explain why u2 + v2 = 1, and show using MATLAB
that your results satisfy this.
5. Let w = u + iv = (z + 1/z)8, where z = x + iy. Using MATLAB, obtain
explicit real expressions for u(x, y) and v(x, y). Check your results by
substituting, via MATLAB, the values x = 0, y = 1 in u(x, y) and v(x, y),
and explain how you know that both u and v should be zero.
6. a. Using MATLAB, express the real and imaginary parts of
( )
17
2 + i 3 in terms of integers and 2 and 3.
b. Repeat the above but consider

13
 1 
 2 − i 3 

This page intentionally left blank
2
Loci and Regions in the Complex Plane
and Displaying Complex Functions

2.1  Meshgrid and Three-Dimensional Plotting

In studying the algebra or calculus of real functions of real variables, you
might investigate such equations as y = ex or y = sin x + x2 by graphing y as a
function of x. The matter of graphing for functions of a complex variable is
more complicated.
The following simple example illustrates why. Let w(z) = z2. Since z is com-
plex, we must specify its value with two numbers: the real and imaginary
parts. Typically, w assumes complex values, for example, if z = 1 + 2i, then
w = (1 + 2i)2 = −3 + 4i. Thus, two numerical values are required to specify
values of the dependent variable and two for the dependent variable. A four-
dimensional space, which we cannot visualize, would be required for a plot.
One approach to our plotting functions of a complex variable
w  =  f(z)  =  f(x  +  iy) is to create real functions u(x,y) and v(x,y), the real and
imaginary parts of w(z), so that u(x,y) + iv(x,y) = w = f(z). We study the real
and imaginary parts of f(z) separately and make three-dimensional plots
of u(x,y) and v(x,y). In the case w  =  z2 mentioned above, we have u(x,y) =
Re(z2) = Re(x2 − y2 + i2xy) = x2 − y2 and v(x,y) = Im(z2) = Im(x2 − y2 + i2xy) = 2xy.
Plots of u(x,y) and v(x,y) for functions of a complex variable are typically per-
formed while x and y are restricted to some region of the complex z plane. We
sometimes will also plot f ( z) = u2 ( x , y ) + v 2 ( x , y ) over a region. Recall that a
region is a domain in the complex plane perhaps supplemented by some por-
tion of boundary points of the domain. Remember, too, that a domain is an open
connected set. The reader may wish to review these terms in a standard text.
A region is composed of an infinite number of points in the x−y plane,
and we have no hopes of storing that number of points in our computer in
preparation for plotting u and v as functions of x and y. What we do is create
a mesh consisting of a rectangular grid of lines in the region of interest. The
spacing of the lines is usually but not always uniform. The (x,y) coordinates
at the intersections of these lines provide the values for z = x + iy at which we
will evaluate the real and imaginary parts of f(z).

27
28 A MATLAB® Companion to Complex Variables

To create this mesh, we use the MATLAB® function meshgrid. Meshgrid

accepts two arguments x and y. These are row vectors whose elements are
the x and y coordinates of the points of intersection of the grid that is of
interest. For example, suppose you wished to know the values assumed by
a function at the points in the x−y plane where the vertical lines x = 0, x = 1,
x = 2 intersect the horizontal lines y = 3, y = 5, y = 6, y = 7. Note that these lines
need not have uniform spacing, and in this case, for the horizontal lines, they
do not. One would write in MATLAB
>> x=[0 1 2]; y=[3 5 6 7];
The output of the function meshgrid is then matrices x and y obtained
from the following few lines of code:
>> x=[0 1 2]; y=[3 5 6 7];
>> [x y]=meshgrid(x,y)

 0 1 2 
 
0 1 2
x= 
 0 1 2 
 0 1 2 
 
 3 3 3 
 
5 5 5
y= 
 6 6 6 
 7 7 7 
 

Note that x is no longer a row vector but a matrix, all of whose rows are
identical to one another; these rows are the old row vector x = [0 1 2]. The
number of rows is taken to be the number of elements in the old row vector
y = [3 5 6 7]. One could also say that the number of rows in the new matrix x
is equal to the number of rows in the transpose of the old row vector y.
Observe that y is now a matrix, all of whose columns are identical to one
another; these columns are the vector that is the transpose of the old row
vector y. The number of columns in the new matrix y is taken to be equal
to the number of elements in the row vector x; thus, the dimensions of the
matrices x and y are identical.
Now consider z = x + iy which in MATLAB is
>> format short
>> z=x+i∗y

 0 + 3.0000i 1.0000 + 3.0000i 2.0000 + 3.0000i 

 
0 + 5.0000i 1.0000 + 5.0000i 2.0000 + 5.0000i
z= 
 0 + 6.0000i 1.0000 + 6.0000i 2.0000 + 6.0000i 
 0 + 7.0000i 1.0000 + 7.0000i 2.0000 + 7.0000i 
 
Loci and Regions in the Complex Plane and Displaying Complex Functions 29

The elements of the preceding matrix consist of the complex number

z = x + iy evaluated at the intersections of the grid formed by the lines x = 0,
x = 1, x = 2 with the lines y = 3, y = 5, y = 6, y = 7. As we move across the matrix
from left to right, we encounter the values of x in the order in which we
entered them in our initial row vector, and as we move from top to bottom
we encounter the values of y in the same manner.
At this juncture, we are well positioned to evaluate a function f (z) at those
values of z contained in the preceding matrix. For example, suppose we want
w = z2. We have

>> w=z.^2

 −9.0000 −8.0000 + 6.0000i −5.0000 + 12.0000i 

 
−25.0000 −24.0000 + 10.0000i −21.0000 + 20.0000i
w= 
 −36.0000 −35.0000 + 12.0000i −32.0000 + 24.0000i 
 −49.0000 −48.0000 + 14.0000i −45.0000 + 28.0000i 
 

Note the necessity for the period (the dot) after the z in the first line. This
causes MATLAB to square each element in z. Without the period, the matrix z
would be squared if that were possible.
We can now make three-dimensional plots of the real and imaginary parts
of w as functions of x and y. Calling these u(x,y) and v(x,y), respectively, we
can extract their values from the above matrix as follows:

>> u=real(w)

 −9 −8 −5 
 
−25 −24 −21
u= 
 −36 −35 −32 
 −49 −48 −45 
 

>> v=imag(w)

 0 6 12 
 
0 10 20
v= 
 0 12 24 
 0 14 28 
 

MATLAB allows an abundance of choices for three-dimensional plotting.

A good place to start is mesh(x,y,Z), where x and y are the matrices describ-
ing the rectangular grid, as described above, over which you wish to make
30 A MATLAB® Companion to Complex Variables

the plot in the x−y plane. Z is a real variable that depends on x and y. Do not
confuse it with the complex variable z. For example, Z might be the variable
v just described. We then have, plotting v,

>> mesh(x,y,v);
colormap(gray)
colorbar
xlabel('x');ylabel('y');
title('Imag(z^2)')

which results in the gray shaded Figure 2.1.

The appearance of a surface is created by MATLAB’s connecting adjacent
data points with straight lines. This has been compared to a fishing net with
the knots at the data points. We have included on the right a color bar (by
including the line of code colorbar) that shows how colors (or shades of gray)
are assigned to lines used in the plot. Thus, light lines correspond to the most
positive values of the function, and the darkest lines correspond to the least
positive. You should take a moment to verify that Figure 2.1 is indeed a plot
of Im(z2) = 2xy. For example, where x = 2, y = 7, the height of the plot is seen
to be 28.
Because of my wish to make this an inexpensive book, I have made the
preceding plot in shades of gray. This was ensured by using the line of code
colormap(gray). Had we not included this line of code, MATLAB would revert
to its default of producing a color plot, and the color bar on the right would

25
30

25 20
20

15 15

10

5 10

0
7
6 2 5
5 1.5
y 1
4 x
0.5 0
3 0

FIGURE 2.1
The imaginary part of z2.
Loci and Regions in the Complex Plane and Displaying Complex Functions 31

be in color and would instruct you how to interpret the colors resulting in
the “fishnet.” You should now try obtaining a color plot on your computer.
The subject of color mapping in MATLAB is rather detailed, and the reader
is referred to the topic colormap in the help menu.
The plot given in Figure 2.1 is the view of the surface seen by an observer
whose azimuth position is at −37.5° and whose elevation is 30°. The mean-
ing of these angles is shown in Figure 2.2, which was downloaded from the
MathWorks, Inc., website. These are the default angles that MATLAB uses
in providing three-dimensional plots. Sometimes this is not the best angle
from which to view a three-dimensional surface. If you attempt a plot of
u = Re(z2) = x2 − y2, using the default angles, you will most likely be confused
by the picture thus obtained (try this). Instead, after some experimentation,
we find that azimuth and elevation of 45 and 60 degrees, respectively, give
a better picture. To avoid the default view, enter the line of code view(az,el)
right after the line of code that generates the plot. You must, of course, assign
numerical values to az and el, the desired azimuth and elevation (in degrees)
to use this code.
This better plot is shown in Figure  2.3, where we also show the code
employed. Note that the grid lines used are more numerous than those we
employed in generating Figure 2.1. For x we use 10 values uniformly spaced
on the interval 0 ≤ x ≤ 2, and for y there are 20 values uniformly spaced on
the interval 0 ≤ y ≤ 7. The result is a surface with finer increments in x and y
than would have been obtained from the grid formed for Figure 2.1. Note the
line of code view(45,60) as mentioned.

Viewpoint
y

Center of
plot box x
Elevation

Azimuth

–y

FIGURE 2.2
The default MATLAB viewing angles.
32 A MATLAB® Companion to Complex Variables

–5

–10
50 –15

–20
0
–25

–50 –30
0 8
–35
0.5 6

1 4 –40
x y
1.5 2 –45

2 0

FIGURE 2.3
Real part of z2.

x=linspace(0,2,10);
y=linspace(0,7,20);
[x y]=meshgrid(x,y);
z=x+i*y;
w=z.^2;
u=real(w);
mesh(x,y,u);view(45,60);colormap(gray);colorbar
xlabel('x');ylabel('y')
title ('real part of z^2')

Try running the above code without the statement colormap(gray) to

obtain a nice plot in color.
If you make the increments in x and y very small, the colored lines form-
ing that plot may merge into a continuous colored surface (the fishnet dis-
appears), one involving so many data points that the plot may print very
slowly should you elect to use your printer. If you wish to generate a three-
dimensional plot that looks more like a surface than that sometimes given by
the function mesh, you should use the MATLAB command surf in place of
mesh. Using surf(x,y,u), for example, in lieu of mesh(x,y,u) in the preceding
code, results in Figure 2.4. (Try making one in color.) Another useful plotting
command is meshz, which you will explore in problem 1. This places a refer-
ence plane under the mesh plot of the surface so that you can see how high
the surface rises above its lowest point.
Loci and Regions in the Complex Plane and Displaying Complex Functions 33

–5

–10

50 –15

–20
0
–25

–50 –30
0 8
–35
0.5 6

1 4 –40
x y
1.5 2 –45

2 0

FIGURE 2.4
The real part of z2.

Exercises
1. Consider the function f(z) = z + 1/z. Plot the real and imaginary parts
of this function over the region described by .5 ≤ x ≤ 2 and .5 ≤ y ≤ 2.
Use 50 uniformly spaced increments in x and y. Use the default val-
ues for view. Note that you can generate multiple plots from the
same program by using such commands as figure(1), figure(2), and
so on, placed as code before you call plotting functions such as mesh,
meshz, surf, and so on. Try plots in both color and shades of gray.
a. Plot each surface using the function mesh. Use the default values
for the view.
b. Plot each surface using the MATLAB function meshz.
c. Plot each surface using the MATLAB function surf.
d. Find new values for azimuth and elevation such that you can
improve or make more interesting at least one of the above surfaces.
2. The exponential function f(z) = ez = ex cos y + iex sin y is entire (every-
where analytic) and is the most basic transcendental function in
complex variable theory. Write a single program that will produce
four different plots as described below. The plot should be over the
rectangular region −2 ≤ x ≤ 2, −8 ≤ y ≤ 8. Use 40 uniformly spaced data
points for x and 80 for y and employ one of the plotting functions
34 A MATLAB® Companion to Complex Variables

mentioned in problem 1. Work directly with ez —that is, do not break

it into components u(x,y) and v(x,y). Obtain
a. A plot of Re(ez)
b. A plot of Im(ez)
c. A plot of |ez|
d. A plot of arg(ez)
Suggestion: For part (d) you might not wish to use the default view.
e. We know from any textbook on complex variables that |ez| = ex
and arg(ez) = y + 2kπ, k is any integer. Do your results for parts (c)
and (d) confirm these facts? Because arg(ez) is potentially multi-
valued, how does MATLAB choose a value for arg(ez)?
2
3. We know that the real function of the real variable f ( x) = e − x is the
− z2
well-known bell-shaped curve. In the complex plane, f ( z) = e dis-
plays more complicated behavior. In the following, make grayscale
plots (no color).
2
a. Using mesh, make a three-dimensional plot of e − z over the
region −1 ≤ x ≤ 1, −1 ≤ y ≤ 1. Use a grid in which there are at least
100 values for x and the same number for y.
b. Similarly, plot the real part of f(z). For variety, use meshz.
c. Plot the imaginary part of f(z). For variety, use surf for the plot.
d. The function f(z) has the property that f′(z) = 0 and f″(z) ≠ 0 at
z = 0. A point where the first derivative of an analytic function
vanishes but the second does not is called a saddle point of the
function.* In a region containing the saddle point, the three-
dimensional plots of the real and imaginary parts of the func-
tion should exhibit the shape of a saddle. Is this true of the plots
found in parts (b) and (c)? You may wish to adjust the view func-
tion to verify that saddle shapes have been found.

2.2  Two-Dimensional Plots: The Contour Plot

A useful and popular method for studying the properties of a function of
a complex variable z = x + iy is by means of a set of two-dimensional plots
called contour plots. These are analogous to the plots used by mapmakers to
show us the heights of hills via two-dimensional drawings. A contour plot
of a mountain might show you a curve indicating where the mountain is
1000 feet high, and another plot drawn on the same paper might show you
where it is 3000 feet high.

* More correctly, this is a first-order saddle point. For more information on saddle points, see
any standard text on advanced calculus.
Loci and Regions in the Complex Plane and Displaying Complex Functions 35

If f(z) = u(x,y) + iv(x,y), we study f(z) by generating a set of curves on which

u(x,y) and v(x,y) have prescribed real values. We might also wish to create
curves on which |f(z)| assumes some prescribed values. A disadvantage of
using the three-dimensional plots described in the previous section is that
we must choose our point of view, in the view statement, carefully, if we
are to appreciate the behavior of the function. An inept choice of the point
of view might lead us to overlook some important behavior of the function
being studied, or we might obtain a misleading plot. Because contour plots
are two dimensional, this problem does not arise.
Contour plots are generated by means of the MATLAB command contour.
The reader might wish to study this plotting function by typing help contour
in the MATLAB command window. We will use the contour function with
four input arguments even though it can be called with fewer arguments.
Thus, we will employ contour (x,y,Z,V).
Here x and y are used to describe the rectangular grid over which we seek
to evaluate our functions. These are generated from the meshgrid command
in precisely the same way as was used in such three-dimensional plots as
surf, mesh, and meshz discussed above.
The quantity Z is a real matrix whose elements are specified at the various
values of z = x + iy lying at the intersections in the grid. MATLAB assumes
that these values in Z are samples taken from a function defined everywhere
in a region containing the grid. The symbol V refers to a row vector hav-
ing real elements. Each element in V specifies a value such that MATLAB
will seek to generate, by means of an interpolation process, contour lines on
which the presumed continuous function has that value. Thus, if V=[2 4 6],
MATLAB will attempt to draw contour lines in which the function studied
assumes the values 2 and 4 and 6.
Let us illustrate how to use contour with an example. Suppose we wish
to see the contour lines on which the imaginary part of the function f(z) = z2
assumes the five integer values between –2 and 2 including the end points
(i.e., –2, –1, 0, 1, 2). We will study the function in the region described by
−2.5 ≤ x ≤ 2.5, −2.5 ≤ y ≤ 2.5. The following code will do it.

x=linspace(-2.5,2.5,100);
y=linspace(-2.5,2.5,100);
n=linspace(-2,2,5);
[x,y]=meshgrid(x,y);
z=x+i*y;
w=z.^2;
wi=imag(w);
[c,h]=contour(x,y,wi,n);colormap(gray)
xlabel('x');ylabel('y');title('Contour Lines for Im(z^2)')
grid
clabel(c,h)

This generates Figure 2.5.

36 A MATLAB® Companion to Complex Variables

2.5

–2

2
1.5

–1

1
1

0.5 –2 2
–1 1

0
0 0 0
y

1 –1

0
–0.5 2 –2

–1

–1
–1.5
1
2

–2
–2

–2.5
–2.5 –2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5
x

FIGURE 2.5
Contour lines for Im(z2).

The line of code [c,h]=contour(x,y,wi,n) creates a matrix [c] and a

scalar h containing information about the shape of the contours and the
numerical values assumed on each contour, in this case, by the function
wi  =  Im(z2). The command clabel(c,h) instructs MATLAB to label each
contour with the (constant) numerical value that wi is assuming on that
contour. The numbers in the gap on each contour in Figure  2.5 tell you
the value assumed by Im(z2) on that contour. If you delete the line of code
colormap(gray), you obtain a set of curves of varying colors. You also want
to include the line of code colorbar to help you assign a numerical value to
those colors.
In general, you are better reading off the numbers placed in the gaps in
the curves.
If instead of the line of code [c,h]=contour(x,y,wi,n);colorbar;
we used instead simply contour(x,y,wi,n);colorbar; without creat-
ing c  and h, and if we deleted clabel(c,h), the numerical values would
not appear on the contours. This would simplify the picture, and we could
estimate the values assumed by our function on each contour by the color
coding.
Recall that z2 = x2 − y2 + i2xy so that Re(z2) = x2 − y2 and Im(z2) = 2xy. Thus,
the contours on which Im(z2) = 2xy assumes the values ± 2, ± 1, 0 are the sets
of hyperbolas 2xy = ± 2, 2xy = ± 1, and 2xy = 0, which are readily identified in
Figure 2.5. The case 2xy = 0 simply reduces to the x and y axes, as Figure 2.5
again shows.
Loci and Regions in the Complex Plane and Displaying Complex Functions 37

Suppose we wish to superimpose on Figure 2.5 contours on which Re(z2) =

x − y2 is constant. We add to the preceding code the lines
2

hold on
wr=real(w);
[d,h]= contour(x,y,wr,n);colormap(gray)
clabel(d,h)

which results in Figure 2.6.

By comparing Figure 2.6 with Figure 2.5, we can now see the contour plots
on which Re(z2) assumes constant values. Notice that these curves have right
angle intersections with those on which Im(z2) assumes constant values. This
is a well-known property of an analytic function f(z): the orthogonality, at
their intersections, of the curves on which the real and imaginary parts of
f(z) assume fixed values.* Recall that this orthogonal property need not hold
at a point where f′(z) = 0, and indeed, the intersections are not orthogonal at
the origin in the above. Here f′(z) = 2z is indeed zero.
Suppose you want to make contour plots in order to study some function
u(x,y), but you have no idea of what constant values to choose for each con-
tour. You can let MATLAB make the choices for you. In the command con-
tour (x,y,Z,V), do not enter a vector for V but instead enter a positive integer
(e.g., m), whose value is specified. MATLAB will then provide you with plots

2.5 –
1 01 –2
2 –1 1
–2

2
2

1.5 –2
–1

0
1

1 –1
2

0.5 –2 0 2
–1 1
0

0 0 0
1
y

1 –1
0

2 0 –2
–0.5
2

–1 –1
0
1

–1

–1.5 –2
2
–2

–2
2

–1 –2 0
–1 1
1

–2.5
–2.5 –2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5
x

FIGURE 2.6
Contour lines for real and imag (z2).

* See, for example, W section 2.5 and S Complex Variables, section 3.12.

38 A MATLAB® Companion to Complex Variables

on which u(x,y) assumes m different values of its own choosing. The value
assumed on each contour by u(x,y) chosen by MATLAB can be read off each
plot provided you are using the command clabel(d,h), as described above.
Note that it is possible that you might apparently see more than m plots: for
example, Im(z2) = 2xy = 2 is satisfied by a curve in the third quadrant and one
in the first. Both are branches of the same hyperbola.
Suppose you want the contour on which a function assumes just one
numerical value, let us call it v1. Then in the expression contour (x,y,Z,V),
enter for V the vector [v1 v1]. The matter is explored in problem 2.

Exercises
1. For the region −3 ≤ x ≤ 3, −3 ≤ y ≤ 3, use MATLAB to plot the contours
on which
a. Re(ez) has the values 0 and ±1.
b. Repeat part (a) but use the imaginary part.
c. Plot the results of parts (a) and (b) on the same set of coordinates,
and verify that the intersections of the curves appear orthogonal.
d. Repeat part (a) but have MATLAB automatically choose five
values on which Re(ez) is constant. Compare the plot to the one
obtained in part (a). Do you see any disadvantages to having
MATLAB choose the values?
2. a. For the region −3 ≤ x ≤ 3, −3 ≤ y ≤ 3 use MATLAB to plot the con-
2
tours on which e z has the integer values 2 through 5.
b. Explain why each of these contours is hyperbolic in shape, and
find the equations of these hyperbolas.
2
c. Use MATLAB to plot just the contour on which e z is 1, and
explain why this has degenerated into something that does not
look like a hyperbola.
3. a. For the region −2.5 ≤ x ≤ 2.5, −2.5 ≤ y ≤ 2.5, use MATLAB to plot
the contours on which |cos z − cosh z| assumes the integer val-
ues 0 through 4.
b. You should observe that there is no contour on which |cos z −
cosh z| = 0, although this equation is satisfied at z = 0. Explain
how one could predict in advance the absence of such a contour.
Hint: Recall that the zeros of a nonconstant analytic function
are isolated (i.e., each zero has a deleted neighborhood through-
out which the function is nowhere zero). See, for example,
W section 5.7.
Loci and Regions in the Complex Plane and Displaying Complex Functions 39

4. a. For the region −2.5 ≤ x ≤ 2.5, −2.5 ≤ y ≤ 2.5, use MATLAB to plot
the contours on which Re(cos z − cosh z) assumes the integer
values 0 through 4. On the same set of axes, plot the contours on
which Im(cos z − cosh z) assumes these values.
b. Observe that the two sets of contours obtained in part (a) inter-
sect at right angles except at one point. What point is this?
Explain this in terms of the vanishing of the first derivative of
cos z − cosh z at the point in question. See W section 2.5 and S
section 3.12 for some help.

2.3  Displaying Regions in the Complex Plane

An inequality such as |z − i|< 1 describes a domain in the complex plane. It is
the set of points lying inside a disc having radius 1. The center of the disc is at
z = i. We could, if we wished, make a sketch of this set of points and indicate
the bounding circle |z − i|= 1 is not to be included in the set; sometimes this
is done with a fuzzy or broken line. It is clear in such an elementary problem
that MATLAB is of no particular advantage in sketching out the set.
Suppose, however, that matters are more complicated. The reader might
first wish to review the meaning of the terms union and intersection in a math-
ematics dictionary. For example, imagine we want to sketch the set that is the
intersection of the set of points satisfying the set of inequalities |z − 2| ≤ 3,
|z − 1−i| ≤ 2, Re(z) ≤ Im(z). This can be done with MATLAB and is assigned
as problem 3. Let us try for starters a problem whose solution is known so
that we can confirm that our code works. We look for the intersection of the
sets|z − i| ≤ 2 and |z + i| ≤ 2.
Although not necessary, it is helpful to convince ourselves that all points sat-
isfying either of these inequalities must be no more than three units from the
origin in the z plane. Notice that |z − i| ≥ ||z|− 1|. Thus, if |z| > 3, then |z − i| ≤ 2
cannot be satisfied, and the intersection of the sets |z − i| ≤ 2 and |z + i| ≤ 2
cannot exist if |z| > 3. Therefore, to study the region that is the intersection
of these two sets, with some space left over, we will use meshgrid to create
a tightly spaced grid satisfying −4 ≤ x ≤ 4 and −4 ≤ y ≤ 4. Note that if you are
unsure of how large a region of the complex plane to scrutinize when studying
a mathematically prescribed set, you can if you wish simply make an educated
guess, erring on the side of generosity. You can later shrink the area of interest
to improve your plot. For the present problem, we use the following code:

x=linspace(-4,4,1000);
y=linspace(-4,4,1000);
[x,y]=meshgrid(x,y);
z=x+i*y;
40 A MATLAB® Companion to Complex Variables

w=abs(z-i);
v=abs(z+i);
ww=(w<=2);
vv=(v<=2);
tt=ww&vv;
meshz(x,y,-tt);colormap(gray)
view(2);
hold on
x=linspace(-4,4,9);
y=x;
[x y]=meshgrid(x,y);
plot(x,y,'r');plot(y,x,'r');axis equal
tle('The Intersection of the Sets |z-i|{\leq}2 and |z+i|
ti
{\leq}2')
xlabel('x');ylabel('y')

In the first through third lines, we form a very tightly spaced mesh in the
region −4 ≤ x ≤ 4 and −4 ≤ y ≤ 4, where z = x + iy is evaluated at the 10002 =
one million intersections of the mesh. The variable w = |z − i| is found at
each of these values of z. Similarly, we find v = |z + i| at these same values.
The line of code ww=(w<=2) supplies us with a matrix ww whose elements
are equal to 1, where the inequality |z − i| ≤ 2 is satisfied in the complex
plane. Where the inequality is not satisfied, the matrix has the matrix ele-
ment 0. Similarly, vv=(v<=2) gives us a matrix vv having elements equal to 1
where the inequality |z + i| ≤ 2 is satisfied, and if the inequality is not satis-
fied, we get elements of 0. The logical statement tt=ww&vv yields a matrix
having ones only where both ww and vv have ones. Otherwise, the matrix
element is 0. Thus, tt has ones where the inequalities |z − i| ≤ 2 and |z + i| ≤
2 are both satisfied—that is, at the intersection of the sets specified by these
two inequalities. The code meshz(x,y,-tt);colormap(gray) gives a three-
dimensional plot of this matrix over the complex z plane. The plot is best
viewed from directly above the plane, with a downward view. The MATLAB
statement view(2) accomplishes this. The surface generated by these closely
spaced points through the use of meshz(x,y,-tt) is either 1 unit below the
x−y plane (where the two inequalities are satisfied) or directly on this plane.
The use of the minus sign in –tt in the meshz argument together with the
grayscale colormap place a black mark in the complex plane where both |z
− i| ≤ 2 and |z + i| ≤ 2 are satisfied. Thus, as shown in Figure 2.7, we have a
plot of the intersection of the two given sets.
Had we used the code line meshz(x,y,tt); (without the minus sign), we
would have generated an image in which the region, which is the union of
|z − i| ≤ 2 and |z + i| ≤ 2, would have been displayed in white while the
remainder of the complex plane would be black; this is contrary to common
usage for displaying regions in complex variable texts.
We have not generated the grid of lines that you see in the figure by
using the usual command of grid following the plot. This command would
create a grid based on the data points—one with lines too closely spaced
Loci and Regions in the Complex Plane and Displaying Complex Functions 41

0
y

–1

–2

–3

–4
–5 –4 –3 –2 –1 0 1 2 3 4
x

FIGURE 2.7
The intersection of the sets |z−i| ≤ 2 and |z+i| ≤ 2.

to be discernible. Lines 13 through 16 of the code, which begins with

x=linspace(−4,4,9), instead creates a useful grid of red lines which are
uniformly one unit apart. You can adjust this spacing as you see fit. This
book is printed without colors, so in the above figures the grid is in black. To
generate a black grid on your computer screen, change the “r” to “k” in the
two places it appears near the bottom of the program.
Note that the preceding MATLAB code generates a figure based on 106 data
points, and if you attempt to print this figure you might swamp the buffer
in your printer. The easiest method of printing the MATLAB-generated fig-
ure, without the risk of overwhelming your printer’s buffer with all the data
points, is to first display the figure on the computer screen. If you are using
the Windows operating system, you should then strike the Print Screen key
(while holding down the FN key). This will place a copy of the figure in the
clipboard of your computer. You may transfer this image into either a word
processing file (like Word) or an image processing file (like Photoshop) from
which it is readily printed.
If you are using the Apple operating system, the corresponding technique
involves pressing these keys at the same time: Command Shift 3. This places
the image on your desktop from which it can be moved into a Word or image
processor as described above.
By changing the line of code tt=ww&vv; to tt=ww|vv; we will have
replaced an and operation with an or operation. Now the logical statement
tt=ww|vv; yields a matrix tt having ones only where ww and vv have ones.
42 A MATLAB® Companion to Complex Variables

The resulting program now yields a plot showing the union of the sets|z − i|
≤ 2 and |z + i| ≤ 2. The result is shown in Figure 2.8.
The method presented here has its limitations. If the region being described
is unbounded, then we cannot display the entire region as some uniformly
colored area of the complex plane; we can only hope to show some por-
tion of that region. For example, the union of the sets |z| ≤ 2 and Re(z) > 0 is
unbounded, a fact we should realize before attempting plots like those in the
preceding two figures. We would thus see only a portion of the sought-after
set on the screen. Its unboundedness is something we would have to see in
advance, or we could convince ourselves of this fact by trying grids covering
increasing areas in the complex plane. Another difficulty with the method
presented relates to the boundary points of sets. The set described by |z| ≤ 3
is a closed set and contains all its boundary points, but the set |z| < 3 is an
open set and contains none of its boundary points. However, were we to
apply the MATLAB method described above to displaying both of these sets,
we would generate identical figures. So, one must be careful about boundary
points, and one should find them by studying the inequalities that one has
been given.
We are not limited to codes that find the union or intersection of two sets. In
fact, the number of sets involved is practically unlimited. Suppose we want
to find the intersection of these three sets: |z − 1| ≤ 1, |z − i| ≤ 1, and |z + i| ≤ 1.

0
y

–1

–2

–3

–4
–5 –4 –3 –2 –1 0 1 2 3 4
x

FIGURE 2.8
The union of the sets |z−i| ≤ 2 and |z+i| ≤ 2.
Loci and Regions in the Complex Plane and Displaying Complex Functions 43

Having generated a matrix z of complex numbers like that used to generate

Figure 2.7, we then write code as follows:

w=abs(z-1);
v=abs(z-i);
s=abs(z+i);
ww=(w<=1);
vv=(v<=1);
ss=(s<=1);
tt=ww&vv&ss;

The remainder of the code would be that used to generate Figure  2.7,
beginning with the line meshz(X,Y,-tt);colormap(gray);. Of course, the
title of the plot would be different. A similar technique can be used to find
the union of these sets.

2.3.1  A Note on pcolor

It may have occurred to the reader that the command meshz has been engi-
neered to produce three-dimensional plots and that we are interested in this sec-
tion in two-dimensional plots in the x−y plane. To ensure this two-­dimensional
view, we must include the line of code view(2) to create a plot in the three-
dimensional x−y−z space as seen from directly above as we gaze down on the
x−y plane. There is a MATLAB command pcolor that will generate the plot in
the x−y plane without the additional command of where to view it. The reader
might wish to read up on this command in the help feature of MATLAB.
Basically, we use pcolor (x,y,m) where x and y are vectors used to define the
elements of a matrix locating the x and y coordinates of points in the Cartesian
plane (much as we placed it in meshgrid), and m is a function defined at these
points. Thus, by using pcolor, we eliminate the line of code view(2). Note that
the colormap command is used with pcolor as it was with meshz.
Although we can drop the line view(2), there is an additional line of code that
we must often include. When we invoke pcolor it creates a checkerboard plot in
the x−y plane; the color of each square (or cell) is determined by the numerical
value of the function being plotted. But pcolor places a narrow border around
each square. If there are a great many squares being used (they are densely
populated), the borders can obliterate the colors in the squares. To avoid having
a border around the squares, one must command them not to exist.
Here is how we rewrite the code at the start of this section to use pcolor:

x=linspace(-4,4,1000);
y=linspace(-4,4,1000);
[x,y]=meshgrid(x,y);
z=x+i*y;
w=abs(z-i);
v=abs(z+i);
ww=(w<=2);
44 A MATLAB® Companion to Complex Variables

vv=(v<=2);
tt=ww&vv;
h=pcolor(x,y,-tt);
set(h,'EdgeColor','none');colormap(gray);grid;axis equal
set(gca,'layer','top'); %required if you hope to see the grid
%sets the grid over the plot
ti
tle('The Intersection of the Sets |z-i|{\leq}2 and |z+i|
{\leq}2')
xlabel('x');ylabel('y')

Note that set(h,'EdgeColor','none'); assures that the cells used in the

plot will not have outlines. In the section of this book on fractals, The Coda,
(i.e., Chapter 7) we favor pcolor because we do only plots in the Cartesian plane.

Exercises
1. Write MATLAB code that will display the union of the sets |z − 1| ≤ 1
and |z − i| ≤ 1/2.
2. Write MATLAB code that will display the union of the sets |z − 1|
≤ 1, |z − i| ≤ 1, |z + i| ≤ 1.
3. Write MATLAB code that will display the intersection of these sets:
|z − 2| ≤ 3, |z−1−i| ≤ 2, Re(z) ≤ Im(z).
4. Write MATLAB code that will display the intersection of these sets:
|sin z2|≤ 1, |z| ≤ 1.
5. Using the complex plane and the random number generator in
MATLAB, you can find π, approximately. The method to be described
here is an example of a Monte-Carlo method, and the reader is
encouraged to read the history of this subject in the Wikipedia.
Suppose you draw a square and a circle on a piece of paper. The
circle is of a unit radius and the square has sides of length 2, and
the circle is inscribed inside the square, which means that it will
be tangent to the sides. Suppose you drop some tiny seeds on the
square, at random, in such a way that there is uniform probability of
a seed landing at any point on the square. Thus, the chance of a seed
landing inside the circle is π/4. This is just the ratio of the area of the
circle to the area of the square. We can use this fact to approximately
compute π with MATLAB. Place the center of the circle at z = 1 + i in
the complex plane. The sides of the square are given by the four lines
x = 0, x = 2, y = 0, y = 2.
a. Write a MATLAB code that randomly generates the x coordi-
nates of 1000 points and the y coordinates of 1000 points. Each of
these coordinates must be positive and be less than 2.
Loci and Regions in the Complex Plane and Displaying Complex Functions 45

Hint: Look up the MATLAB documentation for rand. Try

x=2*rand(1,1000); y=2*rand(1,1000); and form the 1000
complex numbers for z = x + iy. Plot these as ∗ in the complex
plane and show, also, the unit circle centered at 1 + i.
b. If a point satisfies |z − 1 − i| < 1, it will be inside the circle. Using
MATLAB, write a line of code that will determine how many
points that you have generated satisfy this condition. Use this
result to find an approximation to π, and compare this with pi
stored by MATLAB.
Hint: The row vector a given by a=abs(z-1-i)< 1; will produce
entries of 1 for any z satisfying the inequality and entries of 0
(zero) where the inequality is not satisfied.
c. Try running the above code repeatedly and notice that your
answers differ. Does your estimate of π improve if you average
these results? Change your code so as to use 10,000 points. Are
your results improved?

2.4 Three-Dimensional and Contour Plots for Functions

with Singularities
Most, but not all, of the interesting functions in complex variable theory
are ones that are analytic, except possibly at some singular points. Some
functions possess isolated singular points; at an isolated singular point the
function is analytic throughout a deleted neighborhood (a punctured disc)
centered at that singular point. The function is not analytic at the center of
the disc. A function with pole singularities exhibits isolated singular points,
1
for example, f ( z) = 2 has poles at z = ±3i and 1. Here we have a
( z + 9)( z − 1)2
1
finite number of poles. On the other hand, the function f ( z) = has an
sin πz
infinite number of poles—there is one at each integer. Another example of an
isolated singular point is that of f(z) = e1/z at z = 0. Such a function is said to
have an isolated essential singularity.* A Laurent expansion of this function
about z = 0 will be found to have an infinite number of terms with z raised
to negative powers.
There are analytic functions whose singularities are not isolated. Then, every
neighborhood of a singularity will contain one or more other singularities.

* There is also such a thing as a function with a nonisolated essential singular point, for exam-
ple, 1/sin(π/z). Observe that every neighborhood of z = 0 will contain another singular point
besides z = 0 at the origin. You cannot find a neighborhood of z = 0 containing just one singu-
lar point. We will not treat such functions here.
46 A MATLAB® Companion to Complex Variables

These singularities most often occur on branch cuts; the branch cut is a means
of creating a single-valued branch of what would otherwise be a multival-
ued function.* An example is f(z) = z1/2, a branch of which we might define by
means of a branch cut lying along the line x = 0, y ≤ 0 together with the fact that
we choose f (1) = −1. Every point on the line x = 0, y < 0 is a singular point of f(z),
because f(z) changes discontinuously as we cross such a point. Moreover, f(z)
fails to be analytic at z = 0, which makes z = 0 a singular point as well.†
Finally, there are isolated singular points that can be removed through
a proper definition of the function at such a point. These singularities are
sin z
known as removable. The most common example is the function f ( z) = .
z
This function is not defined at the origin as you are here dividing 0 by 0. If
one asks MATLAB to evaluate such a function at the origin, an error message
is generated. However, an application of L’Hôpital’s rule shows that the limit
sin z
of this function as z → 0 is 1. If we define f ( z) = z ≠ 0 and f(0) = 1, we
z
have a new function that agrees with the old except at z = 0. Moreover, the
z2 z 4
new function is the sum of the convergent power series 1 − + + ... and
3! 5!
is therefore analytic wherever this series converges—in this case, the whole
complex plane including the origin. We see that by properly defining a func-
tion at an isolated singular point, we can remove its singularity at that point.
The original function is said to have a removable singularity at this isolated
singular point.

2.4.1  Plots When There Are Isolated Singular Points

2.4.1.1  Removable Singular Points
Of the various kinds of singular points described above, the removable
singular point presents the least problem when we are making plots, and
we treat this case first. After identifying mathematically the location of
these points, we take pains in constructing our plots to simply avoid these
points in our evaluations. As the function in question is continuous, we can
evaluate the function as closely as we wish to the singular point as long
as we do not seek to evaluate the function precisely at the singular point.
Alternatively, we might simply redefine the function at the singular point so
as to remove the singularity and let MATLAB work with the redefined func-
tion. The latter course is not usually necessary. Here is one way to obtain a
sin z
­three-dimensional plot of the imaginary part of f ( z) = :
z

* Strictly speaking, a multivalued function is a contradiction in terms because a function has

but one value corresponding to each value of its independent value.
† Note that we might have taken f(1) = 1, which would just mean we are using a different

branch of z1/2.
Loci and Regions in the Complex Plane and Displaying Complex Functions 47

x=linspace(-5,5,101);
y=linspace(-1,1,51);
[x,y]=meshgrid(x,y);
z=x+i*y;
w=sin(z+eps)./(z+eps);
wi=imag(w);
meshz(x,y,wi);colormap gray
grid on
title('The Imaginary Part of sin(z)/z')
xlabel('x');ylabel('y')

The result is presented in Figure 2.9.

Notice that the choices x=linspace(-5,5,101) and y=linspace(-1,1,51)
mean that one element in the row vector for x is at 0, and one element in the
row vector for y is at 0. This causes the complex variable z in our calculation to
sin z
assume, among other values, the value 0. The function w = is undefined
z
at z = 0. Had we used the line of code w=sin(z)./z and run the program,
MATLAB would have generated a warning message that we have divided by
zero, although it is likely that the program will still execute. To sidestep this
problem, we use instead the code w=sin(z+eps)./(z+eps);. The difference
between the functions sin z/z and sin(z + eps)/(z + eps) is numerically insignifi-
cant, and certainly no difference could be perceived in the graphs of the two
functions.
There are other ways to avoid the division by zero. We could take
w=sin(z)./z in the above code but replace z=x+i*y; with z=x+i*y+eps;
In this way, z is never zero in our calculations, although the values used

0.5

–0.5
1
0.5 5
0
0
y –0.5
–1 –5 x

FIGURE 2.9
sin( z)
The imaginary part of .
z
48 A MATLAB® Companion to Complex Variables

for  z would be effectively indistinguishable from those used in the above

program as they differ only by the number eps, which is of the order of 10−16.
A yet more sophisticated ploy is to again take w=sin(z)./z. Returning to
the program that generates Figure  2.9 given above, after z=x+i*y, we place
the additional line z=z+(z==0)*eps. The logical operator (z==0) is zero except
if z is zero. The operator then assumes the value 1. This means that should
the value of z on the right assume the value zero, it will be replaced by the
number eps. Other values of z are unaffected.
Finally, the least sophisticated way to deal with the removable sin-
gularity is to replace the line of code x=linspace(-5,5,101);with
x=linspace(-5,5,100);. In this way, there is no element in the row vector
for x that is equal to zero, which means that no value of z that is zero is in
the preceding program.* Division by zero is avoided. The drawback here
is that we do not evaluate our function as close to the removable singular
point as we do in the preceding methods.

2.4.1.2  Pole Singularities

Recall that if a function f(z) has a pole of order N at the point z0, then this
function has a Laurent expansion valid in a deleted neighborhood of that
point, and the following must be true: lim z→ z0 ( z − z0 )N f ( z) = c− N, where c−N
is neither zero nor infinity. The existence of the Laurent expansion means
1
that in a neighborhood of z0, f(z) has the representation f ( z) = φ( z),
( z − z0 ) N
where ϕ(z) is analytic and nonzero at z0, and ϕ(z0) = c−N. Thus, near z0, the
c− N
function f(z) varies as, and “blows up” as, ( z − z )N . If we are to plot the real
0
or imaginary parts of f(z) or |f(z)|, we must not include z0 as a data point lest
we divide by zero. However, the situation is a bit more complicated than that;
we may run into trouble simply by trying to plot f(z) “close” to z0.
cos z
Consider, for example, the function f ( z) = , which has a simple pole
z
at the origin. Suppose we wish to plot the magnitude of this function in the
complex plane near the pole. We use the following code:

x=linspace(-1,1,200);
y=linspace(-1,1,200);
[x,y]=meshgrid(x,y);
z=x+i*y;
w=cos(z)./(z);
wm=abs(w);
meshz(x,y,wm);colormap gray
grid on

* Recall that in the statement x=linspace(a,b,n) one element in the row vector for x is exactly
halfway between a and b if n is odd.
Loci and Regions in the Complex Plane and Displaying Complex Functions 49

title('The Magnitude of cos(z)/z')

xlabel('x');ylabel('y')

which results in Figure 2.10.

This result is not very useful if we wish to study |f(z)| near the origin.
The problem is that MATLAB finds the largest value of |f(z)| generated by
the code and creates a scale such that this value is the highest shown in the
plot. In the present problem, this maximum occurs where |z| is smallest, at
approximately 0.00502 + i 0.00502, and here |f(z)| is approximately 100. To
accommodate this value, the graph will not show much detailed behavior of
the function except very close to the pole.
If we used x=linspace(-1,1,50);y=linspace(-1,1,50); in place of
x=linspace(-1,1,200);y=linspace(-1,1,200); we would not be evaluating
|f(z)| so close to the pole, and we would obtain the plot shown in Figure 2.11.
Figure  2.11 still makes the presence of the pole apparent while showing
more about the behavior of the function near z = 0.
The order of a pole of a function tells you how rapidly the magnitude of
the function rises to infinity as the pole is approached. Thus, a function will
“blow up” more rapidly near a pole of order 2 than near a pole of order 1. Let
1
us investigate the behavior of f ( z) = near the two poles. The follow-
z( z − 2)2
ing code will generate a plot of |f(z)| over a region:

x=linspace(-1,4,150);
y=linspace(-1,1,150);
[x,y]=meshgrid(x,y);
z=x+i*y;
f=z.*(z-2).^2;

150

100

50

0
1
0.5 1
0 0.5
–0.5 0
y –0.5
–1 –1 x

FIGURE 2.10
cos( z)
The magnitude of .
z
50 A MATLAB® Companion to Complex Variables

40

30

20

10

0
1
0.5 1
0 0.5
0
y –0.5 –0.5
–1 –1 x

FIGURE 2.11
cos( z)
A better plot of the magnitude of .
z
f=1./f;
f=abs(f);meshz(x,y,f);
axis([-.5 2.5 -1 1 0 30]);colormap gray
title('The Magnitude of 1/(z(z-2).^2)')
xlabel('x');ylabel('y')
view(10,15)

The resulting plot is as shown in Figure 2.12.

The values chosen for the elements of x and y in the code ensure that we
never seek to evaluate f(z) at the poles z = 0 and z = 2. There is no division by
zero.
Note especially the line of code: axis([-.5 2.5 -1 1 0 30]).
This establishes the limits for the three axes. Thus, we consider only values
−.5 ≤ x ≤ 2.5, −1 ≤ y ≤ 1, and 0 ≤ z ≤ 30. The last of these is especially impor-
tant. Note that we have used a bold z to distinguish this axis from the com-
plex variable z. If we had not placed a bound on z, the MATLAB program
would have established the z scale by using the very large value assumed
by 1 near the pole of second order at z = 2. This value is sufficiently
z( z − 2)2
large that the remainder of the plot, away from the pole, would look like a
flat surface.
We have elected not to use the default values for the view statement built
into MATLAB. This would produce an azimuth of –37.5 and an elevation of
30 degrees. The values that we have selected of 10 and 15 degrees are some-
what more satisfactory. The reader might wish to experiment with different
values after studying the view command in the MATLAB help folder.
Loci and Regions in the Complex Plane and Displaying Complex Functions 51

30

25

20

15

10

5 1

0
0 y
–0.5 0 0.5 1 –1
1.5 2 2.5
x

FIGURE 2.12
1
The magnitude of .
z( z − 2)2

The proper choice of the arguments in view may be critical when we study
functions with pole singularities. If a function has more than one pole sin-
gularity, it is possible that we might not see all of the poles when we make
a three-dimensional plot of the magnitude of this function if our perspec-
tive in viewing the plot is poorly chosen. For example, if the line of code
in the preceding program is changed from view(10,15) to view(90,15),
the resulting plot will show only the second-order pole at z = 2 because the
simple pole at the origin will be obscured. To avoid these situations, we can
make contour plots of the magnitude of the function under study.
If a function f(z) has a pole of order N at z0, we have in a neighborhood of
φ( z) c− N
z0 that f ( z) = ≈ . The approximation becomes increasingly
( z − z0 ) N ( z − z0 ) N
valid as z approaches z0. Here ϕ(z) is a function that is analytic in a neighbor-
hood of z0, while C−N = ϕ(z0) ≠ 0.
Thus, near a pole, approximately, the magnitude of f(z) is constant on
contours on which |z − z0| is constant (i.e., on circles centered at z0). The
magnitude grows larger as we consider circles of diminishing radius.
Contour  plots of the magnitude of a function near a pole are thus circles
concentric with the pole. At some point, sufficiently far from the pole, these
plots may cease to be circular, because we are beyond the neighborhood
c− N
in which f ( z) ≈ is a fair approximation. The preceding discussion
( z − z0 ) N
should be illuminated by the following example where we plot the contour
52 A MATLAB® Companion to Complex Variables

1
lines on which the magnitude of f ( z) = assumes certain prescribed
z( z − 2)2
values. Note this was the function just studied with a three-dimensional
plot of its magnitude.

clf;clear
x=linspace(-1,3,100);
y=linspace(-1,1,100);
n=[.5 1 2];
[x,y]=meshgrid(x,y);
z=x+i*y;
w=1./(z.*(z-2).^2);
wm=abs(w);
[c,h]=contour(x,y,wm,n);
axis equal
grid
hold on
clabel(c,h);
title('Contour Lines Showing |1/(z(z-2)^2)|')

1
We have plotted the contour lines on which f ( z) = assumes the
z( z − 2)2
values 0.5, 1, and 2. There are lines that are nearly circular around the poles at
z = 0 and z = 2. On these, |f(z)| is 1 and 2, with, as you would expect, the higher
value resulting in contours that are closer to the poles. The value  |f(z)|=.5
does not occur in close proximity to the poles and in fact results in a con-
tour enclosing both poles. Figure 2.13 should be compared with Figure 2.12,
which should make clear why the contours in Figure 2.13 for values 1 and 2
lie closer to the pole at z = 0 than they do for the pole at z = 2.
Note the line of code axis equal. This ensures that equal units of measure-
ment along the x and y axes are represented by equal intervals on the figure
generated by MATLAB.
In this way, a contour will appear as a circle on your computer screen
where appropriate.

2.4.1.3  Essential Singularities

Recall that if a function has an isolated essential singularity at z0, then the
function is analytic in a deleted neighborhood of z0 and has in∞this deleted
neighborhood a Laurent series representation of the form f ( z) = ∑ c (z − z ) .
n =−∞
n 0
n

−1

The portion of the series containing the negative exponents ∑ c (z − z )

n =−∞
n 0
n
is

known as the principal part and has an infinite number of terms; in other
words, we cannot identify any exponent as being the most negative one in
the Laurent series.
Loci and Regions in the Complex Plane and Displaying Complex Functions 53

1.5

1
0.5
0.5 1
0.5

2
2
1

0
0.5

1
–0.5 1

0.5

5
0.
–1

–1.5
–1 –0.5 0 0.5 1 1.5 2 2.5 3

FIGURE 2.13
Contour lines showing 1 .
( z( z − 2)2 )

The behavior of a function near an essential singularity is remarkably

strange. Picard’s theorem, which is proved in advanced courses in complex
variable theory, asserts that in every neighborhood of an essential singular
point, a function assumes every possible (finite) complex value, with possibly
one exception, an infinite number of times.*
To appreciate how strange such a singularity is, consider the function
1 1 1
f ( z) = e 1/z = ... 3 + 2 + + 1, z ≠ 0, which has an essential singularity at
z 3! z 2! z
z = 0. Suppose we approach the origin along the real axis and study the func-
tion. We have f(z) = f(x) = e1/x. Approaching the origin from the right, through
positive values of x, we have lim e 1/x = ∞. Approaching from the left, through
x→0+
negative values of x, we have lim e 1/x = 0. If we approach the origin from the
x→0−
positive imaginary axis, we have lim e − i/y = lim cos y − i sin y, which has no
y →0+ y →0+
limit, although the magnitude of this expression is fixed at the value one as y
shrinks to zero. Approaching the origin from the negative y axis similarly
has no limit, although we are looking at an expression whose magnitude is
also one. Because of the difficulties involved in making three-dimensional
plots  of functions near their essential singularities, we will avoid such
ventures.

* See, for example, Conway, John, Functions of One Complex Variable I (2nd ed.). Springer, 1978.
54 A MATLAB® Companion to Complex Variables

2.4.1.4  Branch Cut and Branch Point Singularities: Three-Dimensional Plots

Functions such as log z, zβ (β is not an integer), and sin−1 z are not, strictly
speaking, functions, because without our having further information they
appear to be multivalued. The log of 1, for example, might be 0 or i2π or in
fact i2kπ where k is any integer, while z1/2 evaluated at −1 could be ±i. We
establish analytic branches of these functions, which are single valued, by
performing two tasks: we choose a branch cut to describe the function and
assign at some point in the complex plane, not on the cut, a specific numeri-
cal value for our branch. This value cannot be chosen haphazardly but must
be chosen from the collection of permissible values of the original multi-
valued function. The branch cut can be thought of as cutting, or removing,
all the points along its length from the complex plane. This, together with
the numerical value just mentioned, creates a function that is analytic in a
domain, which is the complex plane with points removed by the branch cut.
Some branches require more than one branch cut for their complete speci-
fication. Not every multivalued function of a complex variable can be used
to create an analytic single-valued branch; for example, one could create a
single-valued function from log z , but it is nowhere analytic. If all possible
branch cuts used to create an analytic function have a point in common,
we refer to that as a branch point of the analytic functions being created. For
example, log z has a branch point at z = 0 since all branch cuts for all pos-
sible branches of the log must go through the origin. Similarly, all branches
of (z − 1)1/2 pass through z = 1, which means that z = 1 is a branch point of
this function.
An alternative interpretation of a branch cut is that it is a line, which when
crossed, causes the function defined by the cut to change discontinuously.
The function in question is analytic but not on the cut.
If we ask MATLAB for the numerical value of a function such as log z or
tan−1 z, which, as they stand, are multivalued, how does it decide which value
to produce? Moreover, if we were to ask MATLAB for the three-dimensional
plots of the real or the imaginary parts or the magnitude of such a function,
what branch would it choose to plot? Unfortunately, the Help command in
MATLAB does not provide much help here. As an illustration, try typing
Help sqrt. These questions are best resolved through experimentation with
MATLAB. Let us tackle the matter of sqrt(z). We use the following code to
get the imaginary part of this function:

x=linspace(-1,1,50);
y=linspace(-1,1,50);
[X,Y]=meshgrid(x,y);
z=X+i*Y;
w=sqrt(z);
wi=imag(w);
meshz(X,Y,wi);colormap gray
grid on;view(45,30)
Loci and Regions in the Complex Plane and Displaying Complex Functions 55

title('The Imaginary Part of the Square Root of z')

xlabel('x');ylabel('y')

which results in Figure 2.14.

Notice that the function plotted changes discontinuously along the line y = 0,
x ≤ 0. This line must be the branch cut for the function sqrt(z). Had we asked
MATLAB for a plot of the real part of z1/2, we would have obtained a plot with
no discontinuity. In hunting for discontinuities associated with branch cuts,
we should make plots for both the real and imaginary portions of the function
in question. A discontinuity in one or both is an indicator of a branch cut.
If we ask MATLAB for the numerical value of say sqrt(1), we get 1 as a
result. Thus, it would appear that z1/2 is computed in MATLAB by means
of a branch cut along the negative real axis and that the numerical values
1

( )
1/2 Logz
obtained are consistent with the choice z1/2 = e Logz = e 2 , where Log z is
the principal branch of the logarithm, which is defined by a branch cut along
the negative real axis and includes z = 0 as shown in Figure 2.15.
For the principal branch of the log, we have Log z = Log|z| + i arg z, where
−π < arg z < π. Actually, MATLAB uses not the principal branch of the log but
the principal value defined by Log z = Log|z| + i arg z, where −π < arg z ≤ π.
There is a subtle distinction between the principal branch and the principal
value. Notice that < π (in the principal branch) has been replaced by ≤π (in
the principal value). In this way, we can ask MATLAB to compute, for exam-
ple, square roots of negative real numbers; such numbers of course have
1 1 i

( )
1/2 Log z Log z + arg z
arguments (angles) of π in the expression z1/2 = e Log z = e2 = e2 2 .
We cannot seek the square root of negative real numbers if we are using the

1.5
1
0.5
0
–0.5
–1
–1.5
–1 1
–0.5 0.5
0 0
0.5 –0.5
x y
1 –1

FIGURE 2.14
The imaginary part of the square root of z.
56 A MATLAB® Companion to Complex Variables

0
y

–2

–4

–6

–10 –8 –6 –4 –2 0 2 4 6 8 10
x

FIGURE 2.15
Branch cut for Log(z) is y = 0, −∞ ≤ x = 0.

principal branch, because these numbers have been removed (by a branch
cut) from the plane in which the principal branch of z1/2 is defined. The func-
tion z1/2 is analytic in that cut plane.
Using the principal value from MATLAB, we have
>> format short
>> sqrt(−1)
ans = 0 + 1.0000i
1 1 i π
Log ( −1) Log ( −1) + arg( −1) i
since Log(−1) = i π and e 2 = e2 2 = e 2 = i.
MATLAB will also tell you that the square root of 0 is zero by using the
limit of the real part of Log z as z → 0+. The function produced by MATLAB
for z1/2 is actually not analytic anywhere on the line y = 0, x ≤ 0, even though
MATLAB yields numerical values for the function for points on this line.
The difficulty is that if you move from a point on the line downward to a
point below the line, the value of z1/2 changes discontinuously. Consider the
following two results:
>> sqrt(−3)
ans = 0 + 1.73205080756888i
>> sqrt(−3−i*eps)
ans = 0.00000000000000 − 1.73205080756888i
Moving from −3 to its very near neighbor −3−i*eps produces results for the
square root that are so different that you feel that you are falling off a cliff.
Indeed, such a cliff is shown in Figure 2.14.
Loci and Regions in the Complex Plane and Displaying Complex Functions 57

If we ask MATLAB to compute the logarithm of z ≠ 0, it will, as expected,

produce the principal value. Asking for the log of zero results in a warning
and the result of minus infinity. In general, MATLAB computes zβ, where β
is not an integer, by means of eβLog z, where the principal value is used as the
capital L indicates. Error messages may arise if z = 0. If you ask MATLAB for
0^(1/2), it will produce the value zero, but asking for 0^(1/2 + i) will inconsis-
tently give you the result NaN meaning “not a number.” An expression like
(z − z0)β is computed as eβLog ( z − z0 ). The function is discontinuous if we cross
the branch cut for Log (z − z0)—that is, a semi-infinite line beginning at z = z0
parallel to the real axis and extending to the left.
If we ask MATLAB to evaluate an expression such as f(z) = (az + b)β, where
β is any number that is not an integer, then MATLAB uses f(z) = eβLog(az + b)
provided (az + b) ≠ 0. Because the principal value of the log used here has
a branch cut where (az + b) is negative real or zero, f(z), as computed by
MATLAB, will exhibit such a cut, and a three-dimensional plot of either or
both the real and imaginary parts of the function will exhibit a discontinuity
there. The matter is explored in problems 9 and 12.
It is more challenging to consider how even more complicated functions
such as f(z) = (z2 − 1)1/2 = (z − 1)1/2 (z + 1)1/2 and f(z) = sin−1 z = −i log(iz +
(1 − z2)1/2) are treated by MATLAB. We could try to figure out how MATLAB
chooses cuts for such functions by making three-dimensional plots of the
real and imaginary parts. Let us study f(z) = sin−1 z = −i log(iz + (1−z2)1/2). The
multivalued function (1 − z2)1/2 can be defined by a variety of branch cuts.*
Some experimentation involving MATLAB computation of numerical values
of this function at selected values of z, or three-dimensional plots of the real
and imaginary parts of the function (see problem 10), shows that the branch
used by MATLAB is defined by cuts along the lines y = 0, x ≥ 1 and y = 0,
x ≤ −1. The function is chosen to be 1 when z = 0. MATLAB evaluates the
logarithm using the principal value. However, no branch cut is needed for
this log. The function Log z is discontinuous along the line where Im(z) = 0,
Re(z) ≤ 0. Therefore, using the function −i log(iz + (1−z2)1/2), we must look
to see where the argument iz + (1 − z2)1/2 is zero or negative real. One can
show almost immediately that it can never be zero with the branch of the
square root we are using, and a slightly more complicated argument shows
that it can never be negative real. Thus, the branch of (1 − z2)1/2 that MATLAB
chooses, and the decision to use the principal value of the log, is sufficient to
define the branch of the inverse sine produced by MATLAB.
A plot of the imaginary part of sin−1 z is given in Figure 2.16 for the region
−2 ≤ x ≤ 2, −2 ≤ y ≤ 2. Note the discontinuities as we cross the branch cuts
y = 0, x ≥ 1 and y = 0, x ≤ 1. The real part exhibits no such discontinuity. Note,
however, that in problem 11 where you study asinh(z), it is the real part of the
function that is discontinuous across the branch cut.

* The reader may wish to consult section 3.8 of W for a discussion of branch cuts of some simi-
lar functions.
58 A MATLAB® Companion to Complex Variables

–1

–2
–2
–1 1
0 0
1 –1 y
x 2 –2

FIGURE 2.16
The imaginary part of arcsin z.

Exercises
ez − 1
1. a. Show that the function f ( z) = has a removable singularity
z
at z = 0 through the use of either L’Hôpital’s rule or a Maclaurin
series expansion of the numerator. Be sure to state how f(0)
should be defined in order to remove the singularity.
b. Make a plot of the surface describing |f(z)| over the region
−2  ≤  x  ≤ 2, −2 ≤ y ≤ 2; use at least 1000 data points. Your code
should be written in such a way that you never divide by zero
(i.e., MATLAB should not give you a warning message).
c. Repeat part (b) but use the real part of the function f(z).
d. Repeat part (b) but use the imaginary part of f(z).
sin( z − e)
2. a. Show that the function f ( z) = has a removable singu-
Log( z) − 1
larity at z = e through the use of L’Hôpital’s rule. Be sure to state
how f(e) should be defined in order to remove the singularity.
b. Make a plot of the surface describing |f(z)| over the region .5 ≤ x
≤ 4, −2 ≤ y ≤ 2; use at least 1000 data points. Your code should be
written in such a way that you never divide by zero (i.e., MATLAB
should not give you a warning message). You should provide a
view statement in your code such that you can view the height of
your surface at the removable singular point.
Loci and Regions in the Complex Plane and Displaying Complex Functions 59

c. Repeat part (b) but use the real part of the function f(z).
d. Repeat part (b) but use the imaginary part of f(z).
1
3. Make a plot of the magnitude of f ( z) = . Choose the limits on
z( z − i)2
your axes and the arguments in your view statement such that you
can clearly see the behavior near both poles. Your plot should dem-
onstrate that the surface rises more steeply near the pole of order 2
than near the pole of order 1.
z
4. a. Where are the two poles of f ( z) = that are nearest the ori-
sin πz
gin, and what is their order? Where is the removable singularity?
Prove that it is removable.
b. Make a plot of the magnitude of this function that clearly shows
the behavior near these poles, but do not include any other poles
in your plot. Make certain that your program does not seek to
divide by zero.
In the following three problems, be sure to use axes of equal scales (the axis equal
command in MATLAB does this) so that contour plots near the poles will appear
nearly as circles if you are sufficiently close to the poles.
1
5. Make a plot of the contours on which the magnitude of f ( z) =
z( z 2 + 1)
assumes the values 0.5, 1, and 5.
Suggestion: Use the region described by |x| ≤ 2, |y| ≤ 2. Do the con-
tours encircle the individual poles? Explain.
1
6. Make a plot of the contours on which the magnitude of f ( z) =
sin( πz)
assumes the values 0.5, 1, and 3. Use a region that includes only the
poles at 0 and ± 1. Which contours encircle the poles?
7. a. Identify the location of the pole of 1 , and state its order.
( Log z − 1)2
b. Make contour plots on which the magnitude of this function
assumes the values 1, 2, and 10. A suggested region to use satis-
fies 0 ≤ x ≤ 10, −5 ≤ y ≤ 5. Place an asterisk * at the location of the
pole, on your plot. Comment on how well these contours appear
to be circles surrounding and concentric with the pole.
8. Make three-dimensional plots of both the real and imaginary parts
of f(z) = Log z for the region −2 ≤ x ≤ 2, −2 ≤ y ≤ 2. The plot of the real
part should clearly show the discontinuous behavior of Log |z| at
z = 0. The plot of the imaginary part should show the discontinuous
nature of Log z at the branch cut for this function.
9. Consider the function f(z) = (iz − 1)1/3. Where do you expect the
branch cut for this function, as created by MATLAB, to lie? Verify
60 A MATLAB® Companion to Complex Variables

the branch cut that MATLAB uses in this case by making three-
dimensional plots of the real and imaginary plots of this function
over the region −2 ≤ x ≤ 2, −2 ≤ y ≤ 2.
10. a. Consider the function f(z) = (1 − z2)1/2. Establish the branch cut
that MATLAB uses in evaluating this function by making three-
dimensional plots of the real and imaginary plots of this function
over the region −2 ≤ x ≤ 2, −.25 ≤ y ≤ .25. For view a good choice is
view(−30,70). What value does MATLAB produce at z = 0?
b. Repeat part (a) but choose f(z) = (z2 − 1)1/2. Notice that the branch
cuts are now different from those in part (a). Where are they?
What value does MATLAB choose for f(0)?
11. Consider the function f(z) = sinh−1(z) = log(z + (z2 + 1)1/2), the inverse
hyperbolic sine of z. Establish the branch cut that MATLAB uses
in evaluating this function by making three-dimensional plots of
the real and imaginary plots of the function sinh−1z over the region
−2  ≤  x ≤ 2, −2 ≤ y ≤ 2. Explain why no branch cut arises from the
logarithm, and explain what branch of the log is used by MATLAB.
12. Consider the function f(z) = zi. Establish the branch cut that MATLAB
uses in evaluating this function by making three-dimensional plots
of the real and imaginary plots of this function over the region
−2 ≤ x ≤ 2, −2 ≤ y ≤ 2. What relationship does this have to the branch
cut for Log z?
Suggestion: Try view(−125,30) to view your plot. What numerical val-
ues does MATLAB assign to 1i and ii?

2.5  Contour Plots Affected by Branch Cuts

This section is in some sense a continuation of the previous one, but here
we focus on how to create contour plots when the function being studied
exhibits a branch cut in the domain in which we are seeking the plot. To see
how this might arise, suppose we seek a contour plot of the imaginary part
of Log z. We proceed with this code:
x=linspace(−5,5,1000);
y=x;
[x y]=meshgrid(x,y);
z=x+i*y;
w=log(z);
wm=imag(w);
[g h]=contour(x,y,wm);
clabel(g,h);grid

whose output is as shown in the plot presented in Figure 2.17.

Loci and Regions in the Complex Plane and Displaying Complex Functions 61

1
4

1
3

1
0 2 –3 –2 3 2 –0 0
1
–3
–1

–2

–3
–2

–4
–1
–5
–5 –4 –3 –2 –1 0 1 2 3 4 5

FIGURE 2.17
Contour lines for the imaginary part of Log z.

Notice the jumble of contour lines and numbers near the negative real axis.
The imaginary part of Log z is the principal value of the angle of z (i.e., the
principal value of arg(z)). The curves on which this angle assumes constant
values are simply rays extending out from the origin. The command contour
should produce rays spaced at equal angles. This is not the case in the above.
The difficulty arises from the branch cut for Log z, which extends out from the
origin along the negative real axis. Just above the cut we have arg z = π, and just
below the cut arg z = −π. In constructing the contour lines, MATLAB errone-
ously assumes that Im(Log z) assumes all possible real values lying between
−π and π in any domain containing the negative real axis. It does not allow for
this function having a step-type discontinuity. It is possible that later versions
of MATLAB will sense the discontinuity and that what we will describe here
is not necessary. As noted, we are using here MATLAB release 2015a.
To prevent what just happened, before we invoke the contour command,
we must insert a line of code asserting that if z lies in a narrow band contain-
ing the negative real axis, then z “is not a number.” This is done below and
on the next page:
x=linspace(-5,5,1000);
y=x;
[x y]=meshgrid(x,y);
z=x+i*y;
62 A MATLAB® Companion to Complex Variables

1
4

2
3

1
0 3 0
–3
–1

–2

–3
–2

–4
–1
–5
–5 –4 –3 –2 –1 0 1 2 3 4 5

FIGURE 2.18
An improved version of the preceding plot.

w=log(z);
wm=imag(w);
del=.01;
wm(abs(y)<del&x<=0)=nan;
[g h]=contour(x,y,wm);
clabel(g,h);grid

Notice the lines of code del=.01; and wm(abs(y)<del&x<=0)=nan;

This indicates that if z lies in a strip of width 2* del straddling the negative
real axis, that imag(w) is not to be regarded as a number.
The output of the program, which displays the desired result, is as shown
in Figure 2.18.
Some experimentation can be required in choosing del. If del is chosen
too small (e.g., del = .001), then the same problem arises as when we did not
exclude the real axis from our domain of consideration. You will once again
get Figure 2.17. However, if you choose del too large (e.g., del = .1), you will
find that some portion of the desired contours are missing. You should try
both these values to see the results.

Example 2.1
Using the ideas presented above, obtain a satisfactory plot of the contour
( z − 1)
lines of the imaginary part of w = Log . Let MATLAB choose the
( z + 1)
lines, and then change the program so that it plots exactly six lines.
Loci and Regions in the Complex Plane and Displaying Complex Functions 63

Solution:
This function has a branch cut where (z + 1)/(z − 1) is zero or negative
real. Some study (which you should do now) reveals that this is only
where y = 0, −1 ≤ x ≤ 1. We remove from consideration a strip containing
points along abs(y)<del, −1 ≤ x ≤ 1. We try del =.01. The following code
is used:
clf
x=linspace(-5,5,1000);
y=x;
[x y]=meshgrid(x,y);
figure(1)
z=x+i*y;
w=log((z-1)./(z+1));
wm=imag(w);
del=.01;
wm(abs(y)<del&-1<=x<=1)=nan;
[g h]=contour (x,y,wm,6);
clabel(g,h);grid
title('contour lines for Im log((z-1)/(z+1))')
xlabel('x');ylabel('y');

the output is as shown in Figure 2.19.

We have elected to use six contours as is indicated by the line
[g h]=contour(x,y,wm,6);
The reader is invited to play with the previous program to see the
effects of changing del, the number of contours plotted, as well as the
parameters in linspace.

4 0.44451

2
1
445

1 335
1.3
0.4

2.2225
0
y

–1.
–1 333
1

5
–0.4445

–2

–3

–4 1
445
–0.4
–5
–5 –4 –3 –2 –1 0 1 2 3 4 5
x

FIGURE 2.19
 ( z − 1) 
Contour lines for Im log  .
 z + 1 
64 A MATLAB® Companion to Complex Variables

One should not think that the branch cut can create problems only in
the imaginary part of an analytic function. Here is an instance where the
branch cut appears in both real and imaginary parts.

Example 2.2
Obtain the contour lines for both the real and imaginary parts of the
function f(z) = u(x,y) + iv(x,y) = zi, where the principal branch of the func-
tion is used.

Solution:

u( x , y ) + iv( x , y ) = z i = e i(Log z ) = e i(Log z + i arg( z ))

= e − arg( z ) cos ( Log z ) + ie − arg( z ) sin ( Log z )

Both the real and imaginary parts of the above are discontinuous as
we cross the negative real axis because of the discontinuity in the prin-
cipal value of arg(z). Both real and imaginary parts are discontinuous as
|z|→ 0 because of the singularity in Log|z|. Thus, in computing the con-
tour lines for both the real and imaginary parts of this function, we must
again exclude points from a strip that extends outward from the origin
along the negative real axis. We use the same strip as we used in creat-
ing Figure 2.18. We seek the contours on which the levels of 0 through 5
(integer values) are achieved. The following code will do the job. If you
run this program, the plot will be in color.

%example 2 sec 2_5

clf
x=linspace(-5,5,1000);
y=x;
[x y]=meshgrid(x,y);
z=x+i*y;
w=z.^i;
del=.01
levels=(0:5);%the values 0,1...5 for the contours
wr=real(w);wm=imag(w);
wr(abs(y)<del&-1<=x<=1)=nan;%accounts for branch cut
figure(1)
[g, h]=contour(x,y,wr,levels);
clabel(g,h);grid
title('contour lines for Re z^i')
xlabel('x');ylabel('y');
wm(abs(y)<del&-1<=x<=1)=nan;%accounts for branch cut
figure(2)
[G, H]=contour(x,y,wm,levels);
clabel(G,H);grid
xlabel('x');ylabel('y');
title('contour lines for Im z^i')

whose output is as shown in Figures 2.20 and 2.21.

Loci and Regions in the Complex Plane and Displaying Complex Functions 65

5 0

4 0

2
0

0
y

1 1

2 54
3
2

–1 5

0
4
–2 1
3
–3
2
0 1
–4
0
–5
–5 –4 –3 –2 –1 0 1 2 3 4 5
x

FIGURE 2.20
Contour lines for Re zi.

1
0

0
y

1
3
12 4
0

–1
5

–2
2
–3

–4 3

–5
–5 –4 –3 –2 –1 0 1 2 3 4 5
x

FIGURE 2.21
Contour lines for Im zi.
66 A MATLAB® Companion to Complex Variables

If you were to make both sets of plots on the same set of axes, you would
find that the sets of curves intersect at right angles. This is a well-known
property of analytic functions (see W section 2.5 and S section 3.12). If
you were to delete the lines of code wr(abs(y)<del&-1<=x<=1)=nan;
wm(abs(y)<del&-1<=x<=1)=nan, you would find that the curves form
a jumble along the negative real axis due to the discontinuity caused by
the branch cut described above.
The near absence of contour lines in the upper half plane in the two
figures can be remedied if we seek some contours having numerical val-
ues between 0 and 1.

Exercises
1. a. Consider f(z) = Log(z2 + 1). Obtain contour line plots for the real
and imaginary parts of this function. Before doing so, determine
the location of the branch cut that makes a domain of analyticity
for this function, and in your code eliminate points on this cut
through the use of NaN. Do this for both plots. Use a grid like
that in Example 2.1.
b. Try out your program but do not include the use of NaN, and see
what happens to your plots. Are both affected?
2. a. Consider f(z) = sin(Log z). This function is obviously discontinu-
ous as we cross the branch cut y = 0, x ≤ 0. Explain why the real
part is not discontinuous.
Hint: Recall sin(a + ib) = sin a cosh b + i cos a sinh b. Obtain contour
plots of the real part of this function, noting that you need not
use NaN. Consider a grid filling |x| ≤ 2, |y| ≤ 2 and let MATLAB
choose the values on the contours. Explain why the unit circle is
a contour on which this function is equal to 0.
b. Using the same domain and procedure as above, plot contours
for the imaginary part of the function but insert NaN where
needed. Try leaving out NaN and see what happens.
3. a. Consider f(z) = Log(sin(z)). Explain why to define this function in
the complex plane you need an infinite number of branch cuts.
Prove that these cuts are along the lines where any one of the fol-
lowing is true:
y = 0, sin x ≤ 0, or the set of lines where cos x = 0, sin x < 0,
−∞ < y < ∞.
b. Write a MATLAB plot to generate two separate contour program
of the real and imaginary parts of f(z).
Loci and Regions in the Complex Plane and Displaying Complex Functions 67

Suggestion: Consider the space −6 < x < 6, and −2 < y < 2, and
let the contour function choose the values on the contours. You
might also wish to make both plots on the same set of axes to
verify that they intersect at right angles. Be sure to use NaNs
to avoid evaluating the imaginary part of f(z) at or near the
branch cuts. Explain why this is not necessary for the contour
lines for the real part.
4. a. Consider f(z) = zz, where we use the principal branch. Explain
why both the real and imaginary parts of this function are dis-
continuous in any domain containing any portion of y = 0, x ≤ 0.
b. Obtain a set of contour plots for both the real and imaginary
parts of this function. A suggested range for both x and y is that
their absolute value be less than or equal to 2. Use NaNs where
required. Let the contour function choose the values assumed by
the function.
c. Obtain a contour plot showing only where the imaginary part
of this function assumes the value 1. A suggested range for both
x and y is that their absolute value be less than or equal to 5.
Notice that one of the contours appears coincident with a portion
of the positive real axis. This would seem to be wrong, because
if z = x is positive real, then zz = xx must be real and cannot have
an imaginary part equal to one. Can you explain this? Think in
terms of asymptotic behavior.
This page intentionally left blank
3
Sequences, Series, Limits, and Integrals

Complex variable theory is mostly about the properties of analytic f­unctions

and the integration of these functions. Among the most salient features of
­analytic functions is the fact that they are the sums of convergent ­infinite series.
Thus, we must have some understanding of infinite series to c­ omprehend ana-
lytic functions. We must also know what it means to say that a series is con-
vergent, and here we rely on the fundamental concepts of sequences and their
limits. Integrals in the complex plane, like the real integrals of elementary cal-
culus, are the limits of sums, and here also we are concerned with limits.
MATLAB® can be a useful tool in determining how rapidly a sequence
of functions approaches its limit—if indeed it has one. Additionally, a finite
portion of an infinite series—say the first N terms—can be a useful device
for approximating not only an analytic function but also an integral, and a
numerical comparison employing MATLAB can be useful to investigate the
validity of approximations. Infinite series, as a means of approximating func-
tions, do have their limitations, and sometimes infinite products are prefer-
able; we do not have the space to explore that subject here, and the reader is
referred to Chapter 9 of W.

3.1 Sequences
In complex variable theory, we deal with both infinite and finite sequences,
the former being more common. A finite sequence of complex numbers is
simply a list z1, z2, z3, ..., zn of n numbers, where n is any positive integer.
Sometimes we begin the list with z0 instead of z1 so that the list has n + 1
­elements instead of the n just shown. If n is allowed to grow without bound,
we have an infinite sequence. An example of a finite sequence might be i, i2,
i3, i4, i5. Thus, the nth member or element in the sequence is in. If we were to
let n here run beyond the number 5 and increase all the way to infinity, we
would have an infinite sequence.
Sometimes we have sequences of functions. Each element in the sequence
is defined by a function whose value typically depends on the index that
locates that function in the list defining the sequence. Here is an infinite
sequence of functions: sin z, sin 2z, sin 3z, ..., whose nth element is sin nz.
Here  n is allowed to grow without bound. Restricting the size of n would
result in a finite sequence.

69
70 A MATLAB® Companion to Complex Variables

The most useful infinite sequences have a limit. This has a precise
­mathematical meaning. Suppose the sequence is of the form

p1(z), p2(z), p3(z), ... pn(z), pn+1(z), ... (3.1)

where the terms in the sequence might be functions of z, like sin nz, n = 1,
2, …, or simply constants like in.
The sequence has a limit (or “converges to the limit”) P(z) if the following
is true.
Given any positive number ε, there exists a positive integer N such that

|P(z) – pn(z)|< ε for all n > N (3.2)

The preceding says simply that if we go to the Nth term in the sequence,
then the magnitude of the difference between all terms beyond the Nth
(to the right of the Nth) and P(z) will be less than any preassigned positive
number ε that you choose, no matter how small. N almost always depends
on the value chosen for ε, and the smaller you make ε the larger must be N.
If the terms pn(z) are not constants but are functions of z, then almost invari-
ably N will depend on z as well as ε. Loosely speaking, if a sequence has a
limit P(z), then its terms must cluster closer and closer to P(z), in the complex
plane, as you move to the right in Equation 3.1. A sequence might have a
limit for some values of z but not others. For the latter, it would be futile to
try to find P(z). The reader might wish to see W section 5.2 and S sections
2.15 through 2.17 for more information on sequences.

Example 3.1
Let us consider the simple sequence

1 + 1/(1 + i), 1 + 1/(1 + i)2, 1 + 1/(1 + i)3, ...

1
whose nth term is pn ( z) = 1 + . We use MATLAB to plot the ­elements
(1 + i)n
of this sequence.

Solution:
It should be obvious that the limit of this sequence is P = 1, since

 1  1
lim  1 + n
= 1 + lim =1
n→∞  (1 + i)  n →∞ (1 + i)n

1
This follows from the well-known fact that lim n = 0 if |w|>1. We can
n→∞ w
take w = 1 + i.
Sequences, Series, Limits, and Integrals 71

The reader should verify that we can take N in Equation 3.2 as any
1
Log
integer N satisfying N > ε . Thus, if we use MATLAB to plot the
Log(1 + i)
elements pn(z) of this sequence in the complex plane, they should cluster
more and more closely around the point 1 + i0. We prove this with the
following code and Figure 3.1:

N=15;
p=ones(1,N);rp=real(p);ip=imag(p);
for n=1:N
p(n)=1+1/(1+i)^n;
rp(n)=real(p(n));
ip(n)=imag(p(n));
b=num2str(n)
plot(rp(n),ip(n));
text(rp(n),ip(n),b,'FontSize',6);hold on
pause(.25)
n
p(n)
end
grid
plot(p,'−')
title('The First 15 Terms in the Sequence 1+1/(1+i)^n')
hold off

0.2

5 6
0.1
7

4 13
0 12 14 8
11 15
10 9
–0.1

–0.2

3
–0.3

–0.4

2 1
–0.5
0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5

FIGURE 3.1
1+ 1
The first 15 terms in the sequence .
(1 + i)n
72 A MATLAB® Companion to Complex Variables

The line of code

p=ones(1,N);rp=real(p);ip=imag(p);
is not, strictly speaking, necessary. It tells the MATLAB compiler to set
aside space for vectors, p, rp, and ip having N elements. By doing this
before the FOR loop is executed, we save computation time. Although
time is not an issue in this program, it is a good habit to create space in
the beginning for any array that is to be generated.
Note in the code that the statement pause(.25) places a pause of
­one-quarter second between the plotting of each point on the figure.
In this way one can follow visually the generation of each point. It should
be evident that the points are clustering increasingly close to the limit of
the sequence at z = 1.
The statement b=num2str(n) converts the number n to a ­character b.
This number can be placed at the point in the complex plane
­corresponding to nth element in the sequence. We have arranged to do
this with a small-sized font. This is accomplished with
text(pr(n),pi(n), b,'FontSize',6);hold on

Here the number 6 is the font size. Using a larger integer than 6 here
will allow the values of n to be printed larger, but there will be increased
crowding of these numbers near the limit of the sequence at 1.

Some sequences are generated through an iterative scheme. Each new

element is obtained by putting the previous element or elements through a
defined process, as presented in Example 3.2.

Example 3.2
()
 ii 
i 
( ii ) 
Investigate the sequence i, i , i , i i 
, …. Thus, the nth element pn
is  obtained by taking the previous element pn–1 and using it for the
­exponent of i so that pn = i pn−1, where n ≥ 2. We will agree to take p1 = i.
We  use the p ­ rincipal value in each of these calculations, so that, for
example,

( )
i i = e iLog ( i ) = e iiπ/2 = e − π/2

while,

i − π/2 − π/2 − π/2

)( iπ/2)
i( i ) = i( e )
= e(e ) Log ( i )
= e(e

Solution:
Using MATLAB we can make a plot of the elements of this sequence in
the complex plane. Here is the code and the resulting Figure 3.2 when
we use the first 50 terms:
clear
clf
Sequences, Series, Limits, and Integrals 73

N=50;
p=ones(1,N);rp=real(p); ip=imag(p);
p(1)=i;
for n=2:N
p(n)=i^p(n−1);
rp(n)=real(p(n));
ip(n)=imag(p(n));
b=num2str(n)
plot(rp(n),ip(n),'*');
pause(.3);% gives a .3 sec delay in plotting
hold on
% text ( rp(n) , ip(n) , b , 'FontSize' , 6 );hold on
n
p(n)
end
plot(p(1),'*');
grid
title('Sequence from p_n=i^{p_{n−1}}')
hold off

Note that if you remove the % sign at the start of

% text( rp(n),ip(n), b,'FontSize', 6 );hold on

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

FIGURE 3.2
Sequence from pn = i np− 1.
74 A MATLAB® Companion to Complex Variables

you will place at each point on the plot the index n corresponding to
each point in the plot. If you do this, you should eliminate the line
plot(rp(n),ip(n),'*');. The numbers for n will be hard to read on the
plot near the limit of the sequence, where they become crowded.
Why are there three spiral arms? Does the sequence in q ­ uestion
have a limit, and what it is it? This problem has been studied in
two papers by Greg Packer, Steve Abbott, and Steve Roberts.* The
limit is close to 0.4383 + i0.3606. We can obtain the limit by means of
MATLAB.
Notice that lim pn − 1 = lim pn = p, assuming the limit p exists. Now
n→∞ n→∞
observe that lim pn = lim i pn−1. The left side of the preceding is this p. Let
n→∞ n→∞
us assume that we can swap the limit and exponentiation process of the
right side.† We thus have the equation

p = ip (3.3)

which must be satisfied by the limit p. We can solve this equation in the
complex plane using MATLAB. Let p = x + iy. Using this in Equation 3.3,
we have

x + iy = ix+iy = e(iπ/2)(x+iy) = e –πy/2 e(iπx/2)

Let us equate corresponding parts from both sides of this equation:

x = e –yπ/2 cos(πx/2) (3.4)

y = e –yπ/2 sin(πx/2) (3.5)

Dividing the second of these by the first and making a little rearrange-
ment, we have

y = x tan(πx/2) (3.6)

We use this in Equation 3.4 to get

x = e –(πx/2)(tan πx/2) cos(πx/2)

We can solve this equation by using the MATLAB function fzero to

find the solutions (if any) of

x − e–(πx/2) tan πx/2 cos(πx/2) = 0 (3.7)

* Greg Packer and Steve Abbott, “Complex Power Iteration,” The Mathematical Gazette 81 (492,
Nov. 1997): 431–434.
† The justification for this swap lies with the “Continuous Function Theorem for Sequences.”

See, for example, George B. Thomas, Ross L. Finney, Maurice D. Weir, and Frank R. Giordano,
Thomas’ Calculus, 10th ed. (Reading, MA: Addison-Wesley, 2000), 614.
Sequences, Series, Limits, and Integrals 75

The code follows:

% considers i^p=p solution
format long
myfunc='exp(−pi/2*x.*tan(pi*x./2)).*cos(pi*x./2)−x'
[x,value]=fzero(myfunc,[0 10])
y=x*tan(pi*x/2);
z=x+i*y
check=i^(z)
The reader may wish to review the function fzero in the MATLAB
help  file. Notice that the argument [0 2] that we have used means that
we are seeking a solution of Equation 3.7 in the interval 0 < x <  2. We
were guided by the approximate location of the limit shown in Figure 3.2.
The output for x is 0.438282936727032. The line of code y=x*tan(pi*x/2)
uses Equation 3.6 and yields y = 0.360592471871385. Observe that these
values of x and y are consistent with the coordinates of the limit point that
we can approximately visualize from Figure  3.2. Our final line of code
computes check= i^(z). According to Equation 3.3, if z is the limit of our
sequence, then the value of check should be the same as z. The agreement
to within the 16 digits displayed by MATLAB is perfect.

A relatively recent use of sequences having complex elements, and

MATLAB, is in the generation of fractal sets. These are sets of points in the
complex plane describing a shape that is not one dimensional (like a straight
line) or two dimensional (like a disc). These shapes have fractional dimen-
sionality. At the end of this book you will find a Coda devoted to fractals and
there find references that will explain what is meant by a fractional dimen-
sion, a topic not treated in this book.
In preparation for dealing with fractals, and one particular kind, the
Mandelbrot Set, we introduce here the concept of divergence to infinity. Suppose
we are given the sequence in Equation 3.1. The sequence diverges to infin-
ity if the following is satisfied: if you are given any positive number ρ there
exists a number N such that
|pn(z)|> ρ for all n > N (3.8)

where typically N depends on ρ and z.

Thus, if we consider terms in the sequence to the right of the Nth term, we
find that the magnitude of all of them will exceed any a priori number cho-
sen in advance. Here is a sequence that diverges to ­infinity ez, e2z, e3z, …, enz,
e(n+1)z, …, where z = x + iy and we take Re(z) > 0.
Recall that |enz| = enx and that this expression grows with increasing n.
To satisfy Equation 3.8, we require that enx > ρ for n > N. Hence, by taking
logs, we see that if x > 0, we can choose N as any integer greater than log ρ .
x
A sequence might diverge but not to infinity—an example being the
sequence whose nth element is in. A more interesting example is the sequence
p(n) = sin(nπ/10)zn, where n = 1, 2, 3, …. If |z|≥1, this sequence diverges, but not
to infinity. The matter is considered in p ­ roblem 3.
76 A MATLAB® Companion to Complex Variables

Exercises
n
 x
1. All textbooks on calculus have the formula e x = lim  1 +  , where
n→∞  n
x is any real number. By studying the derivation of this formula, we
find that it applies equally well to the case where z is any number,
not just reals.* Thus, we can say that ez is the limit of the sequence
(1 + z), (1 + z/2)2, (1 + z/3)3, ..., (1 + z/n)n, …. Take z = (1 + i) and make
a plot similar to that in Figure 3.2 in which you display the first 100
terms of this sequence. Also display in this plot the value of e1+i.
2. a. Consider the sequence in which each term is obtained by cub-
ing the preceding term and squaring this same term, summing
these and adding them to i/2. Thus, p(n) = [p(n – 1)]3 + [p(n – 1)]2 +
i/2. Taking p(1) = 0, obtain a plot like Figure 3.1, which generates
a line ­connecting the first 1,000 terms in the sequence.
Suggestion: Use the xlim and/or ylim functions of MATLAB to
show the portion of the x–y plane near what is the apparent limit
of this sequence. From your plot or from the value of p(1000),
estimate the limit of this sequence.
b. Repeat part (a) but take p(1) = .9∗i. Compare p(1000) arising from
each case. Do you suspect that both sequences are heading for
the same limit?
c. Investigate what happens if you take p(1) = i. Leave off the xlim
and ylim. Does the sequence appear to have a limit?
d. No matter what you take for p(1), if this sequence has a limit, let us
call it p, show that it must satisfy the equation p3 + p 2 – p + i/2 = 0.
Find the roots of this equation by using the MATLAB function
roots. Which of the three roots is close to the limit you estimated
in parts (a) and (b)? How good is the agreement between the result
obtained here for the limit and the result for parts (a), (b), and (c)?
3. Consider the sequence p(n) = sin(nπ/10)zn, where n = 1, 2, ....
a. Let z = 1 + i/10. Obtain a plot comparable to Figure 3.2 showing
the first 100 terms of this sequence.
Suggestion: Put a pause statement in your code giving a pause
of around 0.5 second as each point is plotted so that you can
see how the points are generated. What terms in your sequence
have  the smallest absolute magnitude? Prove mathematically
that this sequence cannot diverge to infinity.

* See, for example, K. Knopp, Infinite Sequences and Series, English edition (Mineola, NY:
Dover, 1956): 153–154.
Sequences, Series, Limits, and Integrals 77

b. Let p(n) = sin(0.1 + nπ/10)zn. Using the same value of z, obtain a

plot like the one in part (a). Notice the similarity in the two plots.
Prove mathematically that the sequence in this case does diverge
to infinity. Write your MATLAB program in such a way that it
will tell you the term with the smallest magnitude and the value
of n used in generating that term.
Hint: See the MATLAB function [Y,I] = min(p).
4. a. In Example 3.2 we considered the sequence in which p(n) = ip(n–1),
where n ≥ 2 and p(1) = i. Let us consider here something of the
reverse: p(n) = [p(n – 1)]i, where n ≥ 2 and p(1) = i. Compute and
plot the first million terms of this sequence, and connect the
points with a solid line in the order in which they are gener-
ated. Do not use the pause statement employed in Example 3.2
for obvious reasons. It is not necessary to separate the real and
imaginary parts of p. Just use plot(p) for your plot. Using the tic
and toc functions figure out how much time MATLAB took to
solve this. On the author’s laptop computer, purchased in 2014,
this required about 0.2 seconds.
b. Explain why the plot is so simple (compare it to Example 3.2) and
why the sequence diverges. Does it diverge to infinity?

3.2  Infinite Series and Their Convergence

The topic of infinite series is central to complex variable theory. Infinite
series are not just a useful tool in numerical computation—they have a close
connection to the functions we use in complex variables: analytic functions.
An analytic function can be represented in a domain by the infinite series
called a Taylor series: a power series convergent in a domain has a sum that
is an analytic function. Laurent series play a special role in the evaluation of
contour integrals, and we study them here, too.
For our purposes, an infinite series is an expression of the form
∞ ∞

∑ u (z), or  with different indexing, ∑ u (z). Here the u (z), are usually,
n= 0
n
n=1
n n

but not always, functions of z and almost always depend on the index n. With

∞ ∞

∑
e inz
the second indexing, examples of such series are
∞ n= 0
n + 1
and ∑
n2 ( z + i ) ,
n

∑
n=1
1
while a series with no z such as also qualifies as infinite. The most
n=1
n2
useful series converge. Loosely speaking, it means that if you look at the first
N

N terms in the summation as in ∑ u (z) = u (z) + u (z) + ... + u

n= 1
n 1 2 N ( z), we  get
78 A MATLAB® Companion to Complex Variables

closer  and closer to some value (which depends on z) as N gets b

­ igger
and bigger. Convergence has a precise mathematical meaning: the series
∞ ∞

∑
n=1
un ( z)  converges to the function S(z), which is written ∑ u (z) = S(z), if
n=1
n

the  sequence formed by taking the Nth partial sums (where N = 1, 2, …),
namely, u1(z), u1(z) + u2(z), u1(z) + u2(z) + ... + uN(z) converges to S(z).
∞

A Taylor series, which has the form ∑ c (z − z )

n=1
n 0
n−1
, typically converges to

a sum S(z) when z is limited to some neighborhood of the point z0.

Replacing S(z) by the more familiar function f(z), we know from complex
variable theory that if f(z) is analytic at the point z0, we can expand this
function in a Taylor series—that is, a series that converges to f(z) in a circle
described by |z0 – z0|< r, where r is the distance from z0 to the singularity
of f(z) lying closest to z0. We call z0 the center of expansion. Thus, with a slight
­reindexing, we write
∞

f ( z) = ∑ c (z − z )
n= 0
n 0
n
(3.9)

where
f n ( z0 )
cn = (3.10)
n!
The above two statements give us Taylor’s theorem.
Taylor’s theorem with Remainder is as follows:
N −1

f ( z) = ∑ c (z − z )
n= 0
n 0
n
+ RN (3.11)

where the same values for cn are used as in Equation (3.9). R N is known as the
remainder in the series and is given by this integral:

( z − z0 ) N f ( z′ )
RN =
2 πi ∫ ( z′ − z )
C
0
N
( z ′ − z)
dz′

where the integration is done in the complex plane, using the variable z′, around
the circle |z′ – z0|= b, where b is a positive real number less than r given above.
Thus, the circle excludes any singular points of f(z′). It is not hard to show that
as n → ∞, the remainder vanishes. If we delete the ­remainder in Equation 3.11,
we may approximate f(z) with a series containing N terms as follows:
N −1

f ( z) ≈ ∑ c (z − z )
n= 0
n 0
n
= c0 ( z − z0 )0 + c1 ( z − z0 )1 + ... + c N − 1 ( z − z0 )N − 1 (3.12)

MATLAB provides a convenient way of comparing an analytic function

with its Taylor series approximation.
Sequences, Series, Limits, and Integrals 79

Example 3.3
∞

Let us look at the series e =

z
∑ c z , the Taylor series expansion about
n= 0
n
n

the origin. Since ez is an entire function, this expansion is valid within

1
a circle of infinite radius centered at the origin. Here cn = . Using the
n!
approximation in Equation 3.12, we have

z2 z3 zN −1
ez ≈ 1 + z + + + ... + (3.13)
2! 3! ( N − 1)!

Setting z = 2 + 2i, we will use MATLAB to compute e(2+2i) and ­compare

this with the preceding series approximation where we consider
N = 1, 2, …, 11 terms. Thus, we are ultimately using the 10th power in
our final approximation. The code to do this is

Nmax=11;
a=0:Nmax−1;
q=1./factorial(a);
b=(2+2*i).^a;
SN=q.*b;
format long
S=(cumsum(SN)).';% this sums all the elements in the row
%vector SN
N=[1:Nmax]';
disp('number of terms series approximation')
[N S]
disp('MATLAB computes exp(2+2i)')
exp(2+2*i)

The output is
number of terms series approximation

ans =
 1.
000000000000000 + 0.000000000000000i 1.000000000000000 +
0.000000000000000i
 2.
000000000000000 + 0.000000000000000i 3.000000000000000 +
2.000000000000000i
 3.
000000000000000 + 0.000000000000000i 3.000000000000000 +
6.000000000000000i
 4.
000000000000000 + 0.000000000000000i 0.333333333333333 +
8.666666666666666i
 5.
000000000000000 + 0.000000000000000i −2.333333333333333 +
8.666666666666666i
 6.
000000000000000 + 0.000000000000000i −3.400000000000000 +
7.600000000000000i
 7.
000000000000000 + 0.000000000000000i −3.400000000000000 +
6.888888888888888i
 8.
000000000000000 + 0.000000000000000i −3.196825396825396 +
6.685714285714285i
 9.
000000000000000 + 0.000000000000000i −3.095238095238095 +
6.685714285714285i
80 A MATLAB® Companion to Complex Variables

10.
000000000000000 + 0.000000000000000i −3.072663139329805 +
6.708289241622574i
000000000000000 + 0.000000000000000i −3.072663139329805 +
11.
6.717319223985890i

MATLAB computes exp(2+2i)

ans = −3.074932320639359 + 6.718849697428250i

The left-hand column shows the number of terms we have elected to use
in our series approximation (Equation 3.11), while the adjacent c­ olumns give
the real and imaginary parts of the resulting sum. At the bottom of the out-
put, we see what value MATLAB gives for e2+2i. By the time we use a series
with 11 terms, the disparity between the ­approximate and “exact” value
is tiny.
The command S=(cumsum(SN)).'; in the above code may be ­unfamiliar.
If  the argument of cumsum is a vector, either row or column, the result-
ing vector is a vector that has elements that are the cumulative sum of the
­elements in the argument. For example, from MATLAB:

cumsum([2 7 9])
ans = 2 9 18

Thus, S=(cumsum(SN)).'; will sum the elements in the row vector SN;,
next the ′ (prime) converts this to a column vector, and the . (dot) before the
prime prevents this operation from taking the conjugate of these complex
elements.
The representation of ez by a polynomial with N terms as in Equation 3.13
may seem puzzling. According to the fundamental theorem of algebra, such
a polynomial has N – 1 roots (i.e., vanishes at N – 1 ­locations in the complex
plane). Thus, the more terms used in the series, the greater is the number of
these points. However, ez is known to be nonzero throughout the complex
plane. (See W section 5.7 and S problem 2.59.) Thus, the series might appear
to become increasingly not valid as N increases. This paradox is studied in
problem 3.

Example 3.4
Here we look at a series that is valid only within a domain having a finite
radius. ∞

Consider
1
∑
=
cosh z n = 0
cn z n, where we are using a Maclaurin expan-

sion (i.e., a Taylor series valid in a neighborhood containing the ori-

gin). The singularities of the function on the left side lying closest to
z0 = 0 are those where cosh z vanishes—that is, at z = ± i π . Thus, the
π 2
series is not valid outside the domain z < . Let us obtain the first three
2
Sequences, Series, Limits, and Integrals 81

nonzero coefficients in the Taylor expansion. We could use Equation 3.10

1
where f ( z) = , but the differentiation quickly becomes tedious.
cosh z
Instead one might do a long division as follows:

1 1 z 2 5z 4
= 2 4 ≈ 1− + (3.14)
cosh z z z 2 24
+1+ + ...
2! 4!
The reader should fill in the details of this calculation. Note that on
the above right,

1 1 d2 1 −1 1 d 4 1 5
=1, 2 = , =
cosh z z = 0 2! dz cosh z z= 0 2 4! dz 4 cosh z z = 0 24

The derivatives of odd order of 1 all vanish, z = 0. Although

cosh z
the business of obtaining the coefficients in this Taylor expansion can be
tedious, we will see in a moment that MATLAB makes it simpler.
The approximation sign is used in Equation 3.14 because we have
used only the first three terms in the Maclaurin series. If we had used
an i­ nfinite number of terms, we would still require the caveat |z| < π / 2.
Let us compare the left side of the Equation 3.14 with the right side when
z is real and confined to the real axis (where both sides are real). We make
the MATLAB plot as shown in Figure 3.3.

10

Three-term series approx.

1/cosh(z), y = 0

–5
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 3.3
1
and its series approximation.
cosh( z)
82 A MATLAB® Companion to Complex Variables

This plot arises from this code:

clf
x=linspace(−2,2,100);
z=x+i*0;
w=1./cosh(z);
u=real(w);
v=imag(w);
plot(x,u)
grid
hold on
ser=1−z.^2/2+5*z.^4/4;
plot(x,ser,'−−r','LineWidth',2)
ylim([−5,10]);
xlabel('x');ylabel('1/cosh(z) and its series approximation')

The series we use is a three-term approximation to an infinite series

π
that is valid only for z < . If we confine z to the real axis, this r­ estriction
2
π π
becomes − < x < , y = 0 even though 1/cosh(z) is analytic at the end-
2 2
π
points of this interval z = ± . Notice from our figure that the three-term
2
1
series is a fair approximation to in the interval –.5 ≤ x ≤ .5, y = 0.
cosh z
Had we used more terms, we could have broadened this interval. Notice,
−π π 1
too, that for x < or x > , y = 0 the series in no way a­ pproximates ,
2 2 cosh z
and the situation could not be improved by increasing the n ­ umber of
terms.

3.2.1  Using MATLAB to Obtain the Series Coefficients

Had we attempted to obtain the first 10 coefficients in the above series by the
process of long division that we just used, we would face a tedious problem.
Just as forbidding would be our getting these coefficients via Equation 3.10
1
and successive differentiation of . Here MATLAB comes to our rescue.
cosh z
MATLAB can be used to differentiate any differentiable function as many
times as is practically necessary. We could thus obtain any derivative of
1/cosh z through the MATLAB function diff. Here is an example of our
obtaining the fourth derivative of 1/cosh z. We are working in the Command
window of MATLAB. You must have the Symbolic Math Toolbox installed
to accomplish this. You might wish to study the commands syms and diff
in MATLAB help.
>> syms z a
a=diff(1/cosh(z),4)
a = 5/cosh(z) − (28∗sinh(z)^2)/cosh(z)^3 + (24∗sinh(z)^4)/cosh(z)^5
Sequences, Series, Limits, and Integrals 83

Had we wanted say the 20th derivative, we would have put a 20 instead
of a 4 in the diff expression. Note that the line of code syms z is required to
establish z as a symbolic variable.
Suppose we wished to evaluate the preceding answer at z = 0 and
divide that result by 4! in order to obtain the coefficient of z 4 in the Taylor
­expansion of 1/cosh(z) about z = 0. To do this, we create a function handle
and an ­anonymous function in MATLAB, topics the reader might wish to
review.

syms z b
b(z)=@(z)diff(1/cosh(z),4)
the_coeff=b(0)/factorial(4)

1 d4 1
The output is identical to the 5/24 = 4
we used in our series
above. 4! dz cosh z z= 0
There is an even easier way to obtain the Taylor series of an analytic func-
tion by means of MATLAB. Suppose we want the first three terms in the
1
Taylor expansion of about z = i. We use the MATLAB function taylor,
cosh z
in the Symbolic Math Toolbox, from the Command window as follows:

>> syms z T
>> T=taylor(1/cosh(z),z,i,’order’,3)

The output is

T = 1
/cosh(i) − (sin(1)*(z − i)*i)/cosh(i)^2 − ((sin(1)^2/
cosh(i)^2 + 1/2)*(z − i)^2)​
/cosh(i)

The series contains the first three terms: (z – i)0, (z – i)1, (z – i)2.

The reader should investigate the function taylor in MATLAB help. Notice
that had we wanted a different number of terms, we would replace 3 above
with that number. The reader should convince himself or herself that the
Taylor expansion of 1 about i is valid within a circle in the z plane
cosh z
­centered at i and having a radius π − 1.
2
Suppose we want to have the above series as a function that depends on z.
And we want to get the numerical value of this three-term series at .1+.1∗i.
The following code will accomplish this:

syms z b
b(z)=@(z)taylor(1/cosh(z),z,i,'order',3)
b(.1+.1*i)
eval(b(.1+.1*i))
84 A MATLAB® Companion to Complex Variables

The output is

b(z) = 1
/cosh(i) − (sin(1)*(z − i)*i)/cosh(i)^2 − ((sin(1)^2/
cosh(i)^2 + 1/2)* (z − i)^2)/cosh(i)
ans = 
((sin(1)^2/cosh(i)^2 + 1/2)*(4/5 + (9*i)/50))/cosh(i) +
(sin(1)*(−9/10 − i/10))/cosh(i)^2 + 1/cosh(i)
ans = 3.5883 + 0.6864i

3.2.2  Laurent Series

If a function f(z) is analytic in a ring-shaped domain described by
r1 < |z–z0| < r2, where z0 is the center of the ring and r1 and r2 are the inner
and outer radii, then f(z) can be written
∞
as the sum of a Laurent series when
z is in this ring, and we have f ( z) = ∑ c (z − z ) . The series on the right is
n =−∞
n 0
n

said to be the Laurent series expansion of the function on the left. If a func-
tion is analytic in a deleted neighborhood of z0 described by 0 < |z–z0| < r2,
then a Laurent expansion of this function is also possible, and it is valid
in this “punctured disc.” The coefficients of the series in either case can be
expressed as contour integrations around closed contours in either the ring
or the ­punctured disc (deleted neighborhood) centered at z0. The reader
might wish to consult a book on complex variables to recall what these inte-
grals are (e.g., W section 5.6 and S section 6.9).
Ordinarily we do not obtain the coefficients by means of these integrals
but instead use the method of partial fractions or changes of variable within
2
a known Taylor series. For example, e 1/z is analytic in a deleted neighbor-
w2 w3
hood of z = 0. Recall that ew has the Taylor expansion 1 + w + + + ...
2! 3!
valid for all finite w. We can replace w with 1/z in the preceding (as long
2

2 1 1 1
as  z  ≠ 0) and obtain the Laurent expansion e 1/z = 1 + 2 + + + ...,
z 2! z 4 3! z6
which is valid for all z ≠ 0.
Can we use MATLAB to obtain the coefficients in a Laurent series expan-
sion much as we used it to obtain the coefficients in a Taylor series? The
answer is no… and yes—there is no function laurent like taylor in the release
2015a. However, there is a quirk in the taylor function that allows it to yield
Laurent expansions of functions.

Example 3.5
Use MATLAB to obtain the first few terms in the Laurent expansion of
1
f ( z) = valid in a deleted neighborhood of z = 0.
sin z
Solution:
We notice that sin z = 0 at z = 0, z = ±π, …. An expansion of the ­function
is thus possible in the deleted neighborhood of the origin. This domain
Sequences, Series, Limits, and Integrals 85

is described by 0 < |z|< π. If you ask MATLAB for a Taylor series expan-

sion of 1/sin z about z = 0, you will get an error message—MATLAB
recognizes that the expansion is not possible because a function must
be analytic at its center of expansion z0 for an expansion to be available.
However, look at

z z
g( z) = zf ( z) = =
sin z z3 z5
z − + − ...
3! 5!

Here we have used the Maclaurin series of sin z. If z ≠ 0, we may divide

the right side of the preceding to obtain

z 1
=
sin z z2 z4
1− + − ...
3! 5!

from which it becomes clear that if we define g(0) = 1, we have removed

the singularity of g(z) at the origin since

1
z2 z4
1− + − ...
3! 5!

is the reciprocal of a convergent power series (i.e., the reciprocal of an

analytic function that is nonzero at the origin).
Let us ask MATLAB for the first six terms in the Maclaurin series of
g(z) with its singularity removed:

syms z b
b(z)=@(z)taylor(z/sin(z),z,0,'order',6)

The answer is

ans = (7*z^4)/360 + z^2/6 + 1

Notice that MATLAB realizes that z/sin(z) has a removable singular-

ity at z = 0. Observe also that MATLAB writes the series in descending
powers of z. In our example above where we expanded 1/cosh z about
z = i, the result was in ascending powers of (z − i). A quirk of MATLAB
(as of release 2015a) is that the function taylor will yield the Taylor series
in ascending powers of (z − z0) if the expansion is not about z0 = 0. For an
expansion about zero, we obtain descending powers.

In the preceding expansion, we asked for six terms, but we seem to have
gotten only three. Note that z/sin(z) is an even function of z. In our series, the
first term is simply 1, the second is 0, the third contains the second power
of  z, the fourth vanishes, the fifth is the fourth power, while the sixth is
absent. The terms with odd exponents of z vanish.
86 A MATLAB® Companion to Complex Variables

The infinite series of which we have obtained only the first three nonzero
terms converges for |z|< π to the function z when z ≠ 0 and to 1 when
sin z
z = 0. Dividing this series by z, we obtain the desired Laurent expansion:

1 1 z z3
= + +7 + ... 0< z <π
sin z z 6 360

where we have placed the terms in their usual order of ascending powers.
The function 1 is analytic in the ring-shaped domain π < |z| < 2π, and
sin z
a Laurent expansion of the function should be possible there. We can use the
previously obtained Laurent series, valid for 0 < |z| < π, to get us partway to
the solution. The details can be found in W section 6.3 problem 41.
There are other ways to use the Taylor series function in MATLAB to obtain
Laurent series. This is especially true when a function is analytic at infinity.
Recall that a function f(z) is analytic at infinity if the function F(w) = f(1/w) is
1
analytic in the w plane at w = 0. For example, f (z) = is analytic at
( z − 1)( z − 3)
1
z = ∞ since F(w) = is analytic at w = 0 (verify this), while f(z) = z
(w −1 − 1)(w −1 − 3)
is not analytic at z = ∞ since 1/w is not analytic at w = 0.

Example 3.6
z
Expand f ( z) = in a Laurent series in powers of z, valid
( z − 1)( z + 3)
for |z|>3.

Solution:
We verify that this function is analytic in the finite plane except at z = 1
and z = −3. Thus, if we are working with series having powers of z, a
Taylor series is available for 0 ≤  |z| < 1. A Laurent series can be found
for 1 <  |z| < 3, and another for |z| > 3. The MATLAB function taylor
will produce the Taylor series. Now

w −1 w
F(w) = f (1/z) = f ( z) = =
(w − 1)(w −1 + 3) (1 − w)(1 + 3w)
−1

is analytic at w = 0. Surprisingly, taylor in MATLAB will generate a

Laurent series in powers of z. The series will be valid in the unbounded
domain containing infinity, in this case, |z| > 3, as follows:
>> syms z c
c(z) = @(z) taylor(z./((z − 1). ∗ (z + 3)),’ExpansionPoint’,inf,’Order’,10)
c(z) = 1 /z − 2/z^2 + 7/z^3 − 20/z^4 + 61/z^5 − 182/z^6 + 547/z^7
− 1640/z^8 + 4921/z^9
Sequences, Series, Limits, and Integrals 87

For our Laurent series, valid at z  = ∞, we chose inf as the center of

expansion. We have elected to obtain 10 terms, but only the odd powers
1 … 9 are nonzero.

We see that the function taylor is in a sense a misnomer. If we apply taylor

to a function of z that is analytic at infinity, we obtain not a Taylor expansion but
a Laurent expansion containing powers of 1 , where n ≥ 0 is an integer. Here
is another example: zn

>> c(z) = @(z)taylor(exp(1./z),’ExpansionPoint’,inf,’Order’,5)
c(z) = 1/z + 1/(2∗z^2) + 1/(6∗z^3) + 1/(24∗z^4) + 1
The above is the Laurent expansion (the first five terms) of e1/z. Notice the
rather strange ordering of the terms: 1/z∞ is the first term, and 1/z0 is the last
term. The actual Laurent expansion is ∑ n! z
1 1 . MATLAB does not yield the
n
n= 0

Laurent expansion of, for example, ze1/z, because this function has a simple
pole at infinity, as the reader should verify with the substitution w = 1/z.
z
In Example  3.6, we mentioned that the function f ( z) = has
( z − 1)( z + 3)
a Laurent expansion in the domain 1 < |z| < 3. The methods we have been
using with MATLAB will not produce this series.
However, MATLAB can still help us to obtain the series. We express f(z) in
the form

α β
f ( z) = +
z+3 z−1

The first fraction is then expanded in a Taylor series in powers of z, valid

for |z| < 3, while the second is expanded in a Laurent series in powers of z,
valid for |z| > 1. The two series are added together to give us the desired
Laurent series.
Although we should know how to obtain α and β in the above from the
methods of elementary calculus, MATLAB can do it for us. At this point,
you might wish to study the MATLAB function residue in the help folder.
We proceed as follows:

>> [L M N] = residue([1 0], [1 2 −3])

 0.7500 
L= 
 0.2500 
 −3.0000 
M= 
 1.0000 
N = []
88 A MATLAB® Companion to Complex Variables

Here is what was done. The denominator of our fractional expression

is (z  –  1)(z + 3) = z2 + 2z  –  3. This is a polynomial of second order whose
­coefficients, in descending powers, are given by in the row vector [1 2 −3].
The  numerator of our fractional expression is the polynomial z whose
­coefficients we supply to MATLAB are in the form [1 0]. Again we go from
highest to lowest power, with the zero arising because there is no constant
(z0) term in the numerator. The 1 is simply the coefficient of z in the numera-
tor. Now L gives the ­coefficients in the partial fractions, while M supplies
the location of the poles (i.e., information to make the denominators). Thus,
the first fraction is .75 , while the second is .25 . The vector N here is empty.
z+3 z−1
It arises in nonempty form if we are asked to reduce to partial fractions the
ratio of two polynomials in which the degree of the numerator equals or
exceeds that of the denominator, as is discussed later. The reader should
check our answer—that is, verify that

.75 .25 z
+ =
z + 3 z − 1 ( z − 1)( z + 3)

Now we expand the first fraction in a Taylor series and the second
in  a  Laurent series, each in powers of z, and we add the two series, as
follows:

syms z a b
a(z)=@(z)taylor(.75/((z+3)),'ExpansionPoint',0,'Order',4);
b(z)=@(z)taylor(.25/(z−1),'expansionpoint',inf,'order',4);
a(z)+b(z)

We chose to use four terms in each series.

The output is

ans = 1/(4∗z) − z/12 + 1/(4∗z^2) + z^2/36 + 1/(4∗z^3) − z^3/108 + 1/4

Note the unorthodox grouping of the terms in a + b. Ordinarily, we would

first state the terms with negative exponents of z (i.e., those in the series for b),
while those in the series for a would be given second but in the order oppo-
site to that given here.
Suppose the problem had been to expand

z3 + 2 z2 + i
f ( z) =
( z − 1)( z + 3)

in a Laurent series valid for |z| > 3. We again seek a partial fraction decom-
position, but note that we are dealing here with an improper rational
Sequences, Series, Limits, and Integrals 89

expression—the numerator is of degree 3 and the denominator of degree 2.

We proceed as follows:
>> [L M N]=residue([1 2 0 i],[1 2 −3])

 2.2500 − 0.2500i 
L= 
 0.7500 + 0.2500i 
 −3.0000 
M= 
 1.0000 
N = [1 0]
Note that N is not empty.
The preceding means that the partial fraction decomposition of f(z) is

2.25 − .25i .75 + .25i

f ( z) = + +z
z+3 z−1

The denominators of the two fractions are in the elements of M, while

the corresponding numerators come from L. The vector N supplies the
polynomial that must be added to the fractions—a consequence of f(z)
being an improper fraction. There are two elements in N that tell us that
the polynomial has two terms, the first containing z and the second z 0.
The c­oefficients are 1 and 0, respectively. We now expand each of the
two ­f ractions in Laurent series. We seek four terms in each series, for brev-
ity. We do not seek to expand z, because it is already a one-term Taylor
series.

syms z a b
a
 (z)=@(z)taylor((2.25−.25*i)/(z+3),'ExpansionPoint',inf,​
'Order',4)
b(z)=@(z)taylor((.75+.25*i)/(z−1),'expansionpoint',inf,'order',4)

a(z)+b(z)

The output of this program is

ans = 3/z + (−6 + i)/z^2 + (21 − 2∗i)/z^3

Our final answer, the desired Laurent series for |z| > 3, requires our add-
ing the term z to the above. Thus, we have

z3 + 2 z2 + i
f ( z) = = z + 3/z + (−6 + i)/z ^ 2 + (21 − 2∗i)/z ^ 3 + 
( z − 1)( z + 3)

for |z| > 3.
90 A MATLAB® Companion to Complex Variables

Exercises
1. Perform a calculation similar to Example 3.3 but compute sin(3 + 4i)
using the Maclaurin series expansion of sin z. Compare the sum of
series having 1, 2, ... 10 nonzero terms to the MATLAB determination
z3 z5
of sin(3 + 4i). Recall that sin z = z − + − ...
3! 5!
How many terms would you need in your series to get agreement
between the MATLAB value of sin(3 + 4i), and the series approxima-
tion, that is within 1%?
1
2. Consider the series = 1 + z + z 2 + z 3 + ... . This series is u
­ niformly
1− z
convergent in the disc |z| ≤ r if r < 1. If necessary, review the concept
of uniform convergence in either W section 5.3 or S section 6.4. Because
the series is uniformly convergent, we may integrate it term by term
along a contour connecting z = 0 with z = w in the complex plane. We
require that |w| < 1 and that the contour lies in the above disc. Thus,
w w w w
dz w2 w3

∫
0
1− z
=
∫
0
∫
0
∫
dz + z dz + z 2 dz + ... = w +
0
2
+
3
+ ... w < 1

w
dz 1
∫ 1 − z = − Log(1 − z)
w
But notice also that = Log . Thus,
0
(1 − w)
0

1 w2 w3
Log =w+ + + ... |w| < 1
1– w 2 3
i
a. Take w = in the preceding equation. By taking the real and
3
imaginary parts of this equation, show that

3 −1 1 1
Log = + − + ...
2 2 × 3 4 × 9 6 × 27
and
π 1 1 1 1  1 1 1 
= − + − ... = 1− + − − ... 
6 3 3× 3 ×3 5× 3 ×9 3  3 × 3 5 × 9 7 × 27 

b. The last equation provides a method of computing π provided

you know the value of 3 . Multiply both sides of the ­preceding
equation by 6, and compute π by using different finite series
ranging from 1 to 10 terms, total. Use Example 3.3 as your model,
and compare the result obtained from each series with the value
that MATLAB chooses for π.
Sequences, Series, Limits, and Integrals 91

3. An apparent paradox. According to the fundamental theorem of

algebra (see W section  4.6 and S section 5.2 and problem 5.10), the
algebraic equation a0 + a1z + a2z2 + … +  anzn  =  0, where the a’s are
constants, n is a positive integer, and an ≠ 0 has exactly n roots in the
complex plane. Now suppose we approximate ez by the first n + 1
terms in its Maclaurin series:

z2 zn
ez ≈ 1 + z + + ... +
2! n!

It is apparent that the more terms we use in this Maclaurin

polynomial approximation, the more places in the complex plane
where the approximation has the value zero. However, ez ≠ 0 in the
finite complex plane. Thus, it appears that the more terms we use in
the Maclaurin expansion, the greater is the number of locations in
the complex plane where the series cannot well approximate ez. It can
be proved that this paradox can be resolved by showing that the roots
1
of a0 + a1 z + a2 z 2 + ... + an z n = 0, ak = , move to infinity as n → ∞. We
k!
­confirm this numerically.
a. Write a MATLAB program that will find the roots of
z2 zn
1+ z + + ... + for n = 1, 2, …, 25. These roots should be plot-
2! n!
ted as points in the complex plane and labeled with the number
n from the corresponding equation. Observe that the roots do
move farther from the origin with increasing n. Use the MATLAB
function roots. You might want to open help roots in MATLAB
to refresh your memory. Note that when entering the coefficients
of the polynomial whose roots you are seeking, that numbers
must be entered for the highest power first, and then descend in
order.
b. Alter the preceding program so that it will find the root, for each
value of n, that lies closest to the origin in the complex plane. Your
new program should plot the magnitudes of these roots versus n.
1 1 1
4. a. Show that Log z = ( z − 1) − ( z − 1)2 + ( z − 1)3 − ( z − 1)4 +
2 3 4
1
( z − 1)5 − ... by direct use of Equations 3.9 and 3.10. State the cen-
5
ter and radius of the domain in which the series converges to the
function on the left.
b. Determine the first five coefficients by using the MATLAB func-
tion diff in the Symbolic Math Toolbox. Recall that the MATLAB
function log is the principal value.
c. Determine the series in (a) by using the MATLAB function taylor.
92 A MATLAB® Companion to Complex Variables

d. Assume z = x (is real). Obtain a plot comparable to Figure 3.3

in which you compute Log(x–1) and compare it with a five-
term Taylor expansion (the first five nonzero terms) about
z = 1. Consider the interval 0  <  x  <  3. Comment on where
the  ­agreement is reasonably good and where it is not, and
be sure to  include ­discussion where the infinite Taylor series
is valid.
 a+b 
e. Derive the formula tan −1 ( a) + tan −1 (b) = tan −1  , where a and b
 1 − ab 
are real. This is easily done from the product (1  +  ia)(1 + ib) =
1 – ab + i(a + b) and the fact that the argument (or angle) of a p ­ roduct
of complex factors is the sum of the arguments of each factor. Show
from your result that tan–1(1/3) + tan–1 (1/2) = π/4.
f. From the preceding, we have 4[tan–1(1/3)  +  tan–1(1/2)]  =  π.
Let z = 1  +  iy be the result of part (a). Derive the series
y3 y5
tan −1 ( y ) = y − + − .... State the restrictions on the real vari-
3 5
able y for this series to be valid.
g. Prove that

 1 1 1 1 1 1 1 1 (−1)n− 1  1 1  
π = 4  +  −  3 + 3  +  5 + 5  + ... +  2 n− 1 + 2 n− 1  ... 

 3 2  
3 3 2  
5 3 2  2n − 1 3 2 

Now, taking Example 3.3 as your model, use the preceding to compute π.

Compare the result obtained from each finite series (n = 1, 2, …) with the
value that MATLAB produces for π. Is this series more useful for computing
π than the one in problem 2(b)?
5. a. Consider the principal branch of the function

1
f ( z) =
(1 − z)1/2

Where in the complex plane is this an analytic function?

Where is this Maclaurin series expansion

∑c z
n= 0
n
n
=
1
(1 − z)1/2

valid? Give the center of the domain and its radius. Through the
use of Equation 3.10, find a formula for cn in terms of n.
b. Using the MATLAB function diff, show that the first six terms
in the Maclaurin series as given by the formula you derived in
Sequences, Series, Limits, and Integrals 93

(a) agree with those obtained with having MATLAB generate the
derivatives.
c. Confirm the results in (b) by using the MATLAB function
­taylor. Convert the expression found to an anonymous function
of z using the @(z) notation.
d. Show that you can use your series to find 2 by suitable choice
of z. Using this value of z in the above derived function handle,
find what values you get for 2 by using the series having six
terms found in (c) and compare this result with the MATLAB-
generated sqrt(2). Use double precision. Now repeat this but use
50 terms. Are the results better with 50 terms?

3.3 Integration in the Complex Plane: Part 1,

Finite Sums as Approximations
One reason that we have been paying so much attention to sums and limits is
that these concepts provide the basis for understanding the most important
and common operation in complex variable theory: contour integration. To
perform a contour integration, we need a piecewise smooth curve (a contour)
and a function, f(z), to integrate along the contour. This function is often,
but not always, analytic. The integral of f(z) along the contour is—like all
integrals—defined as the limit of a sum. In the case of a contour integral, we
have this definition. The reader should refer to Figure 3.4.
We are here integrating the function f(z) along the arc C in the complex
plane. We integrate from point A to point B. The arc is divided into n sub-
arcs. The endpoints of these arcs are (x1, y1), (x2, y2), ..., (xn+1, yn+1). Somewhere
on each arc we choose just one point. The coordinates of these points are,
respectively, (x1,y1), (x2,y2), ..., (xn,yn), and for each of these points there are cor-
responding complex numbers z1 = x1 + iy1, z2 = x2 + iy2, ..., zn = xn + iyn. Each
arc is subtended by a vector chord creating complex numbers. Here, Δz1  =
Δx1 + iΔy1, Δz2 = Δx2 + iΔy2, and so on. The vectors are shown in the figure.
Thus, Δxk = Xk+1–Xk and Δyk = Yk+1–Yk, where k = 1, 2, ..., n. We now define
the integral
B n

∫ f ( z) dz = lim
n→∞ ∑ f (z )∆z
k =1
k k (3.15)
A

where all Δzk → 0 as n → ∞. Note that the complex number A = X1 + iY1 and
B = XN+1 + iYN+1. We must think of the summation on the above right as arising
from summing the following n products: the function f(z) evaluated at a point
on C multiplied by the complex number for the neighboring vector chord.
94 A MATLAB® Companion to Complex Variables

y
Xn+1, Yn+1
B

Δyn Zn

Xn , Yn

etc.

Δz
2 z2

X1 , Y1 Δz1
A X2 , Y2
z1
Δx1 x

FIGURE 3.4
Defining a contour integral.

If we do not pass to the limit on the right, but instead use some finite value
of n, we obtain only an approximation of the integral

B n

∫ f (z) dz ≈∑ f (z )∆z
A k =1
k k (3.16)

As n increases, the sum on the right ultimately becomes a better approxi-

mation to the integral on the left, although we cannot argue that the improve-
ment is monotonic. For a particular f(z) and C, we cannot say for certain that,
for example, by going from n = 10 to n = 11 the sum on the right must be
a better approximation to the integral. Note that the result on the right in
Equation 3.16 depends to some extent on where we have placed each zk on the
corresponding subarc. However, in the limit as n → ∞ and all Δzk → 0, this
placement does not matter.
Let us investigate a particular integral whose value is known.

Example 3.7
1

Consider
∫ cos z dz, where the integral is along the arc C in the first
i
quadrant y = 1 – x2.
Sequences, Series, Limits, and Integrals 95

Approximate this integral by sums like that on the right side of

Equation 3.16, where n = 3, and then compare the sum with the value of
the integral obtained by using the antiderivative of cos z.

Solution:
We create the subarcs, arbitrarily, by making each x-axis projection of
Δzk, namely, Δxk, to be of identical length. Figure 3.5 shows what h ­ appens
with three such subdivisions.
Since x advances from 0 to 1 as we integrate, we have when n = 3 that
X1 = 0, X2 = 1/3, X3 = 2/3, X4 = 1. The corresponding values of the Y
­coordinates are obtained from the equation of C—that is, y = 1–x2. Thus,
Y1 = 1, Y2 = 8/9, Y3 = 5/9, Y4 = 0.
Having found the endpoints of each subarc, we have that Δz1 = 1/3 – i/9,
Δz2 = 1/3 – i/3, Δz3 = 1/3 – i5/9 as some study of the figure should confirm.
Notice that the sum of these three quantities is simply 1 – i which is a
vector connecting the start of integration at (1,0) with the fi ­ nish (0,1). This
will always be the case with the Δzk quantities—their sum will equal a
vector going from the start to the finish of integration.
The quantities zk must now be chosen. Recall that they must each reside
somewhere on the corresponding subarc. We choose arbitrarily to place
them in such a way that each xk lies exactly between the two values of Xk
defining the ends of the arc. The corresponding value of yk is found from
the equation of the contour C. Thus, yk = 1 – xk2. We now have x1 = 1/6,
y1 = 35/36, x2 = 1/2, y2 = 3/4, x3 = 5/6, y3 = 11/36. Our corresponding three-
term sum on the right in Equation 3.16 is

1.2

X1, Y1
1 z1
X2, Y2
Δz1
0.8
z2

Δz2
0.6
y

X3, Y3

0.4
Δz3 z3

0.2

X4, Y4
0
0 0.5 1 1.5
x

FIGURE 3.5
The arc C and vector chords.
96 A MATLAB® Companion to Complex Variables

∑ f (z )∆z
k =1
k k = cos(1/ 6 + i35 / 36)(1/ 3 − i / 9) + cos(1/ 2 + i3 / 4)(1/ 3 − i / 3)

+ cos(5 / 6 + i11/ 36)(1/ 3 − i5 / 9)

which is easily evaluated with MATLAB to be 0.8302 − 1.2061i.

Using complex variable theory, one may evaluate the integral on the
left in Equation 3.17
1

∫ cos z dz
i

exactly (see W section 4.4 and

z
S section 4.13).
2

Recall the following. Let

∫ f (z) dz be an integral taken along a con-
z1
tour C connecting z1 with z2. Assume that C lies in a simply connected
domain (one with no holes) and that in this domain we have found an
analytic function F(z) such that dF/dz = f(z). Then the value of our integral
is given simply by
z2

∫ f (z) dz = F(z ) − F(z )

z1
2 1

F(z) is called the antiderivative of f(z); if f(z) is not analytic in a domain

containing the contour of integration, then we cannot find an analytic
F(z) as it does not exist.
The function cos z is entire and is the derivative of another entire func-
tion sin z. The path of integration is immaterial, and we have that

∫ cos z dz = sin 1 − sin i
i

which we evaluate, to four decimal places, with MATLAB, as 0.8415 −

1.1752i. Thus, with only a three-term series, we got a fair approximation
to the “exact” value of the integral.

The following MATLAB program redoes the preceding example by means

of an n term series on the right in Equation 3.16, where n ranges from 1 to 10
values.

Example 3.8
Repeat Example 3.7 but use series that range from 1 to 10 terms.

Solution:
nmax=input('max number of terms in series')
for(n=1:nmax)
Sequences, Series, Limits, and Integrals 97

m=n+1;% number of points needed on curve to define the n

% subarcs
X=linspace(0,1,m);% this gives
%the values of X needed
Y=1−X.^2;%gives values of Y on curve
delta_x=(X(2)−X(1))/2;
x=X(1:n)+delta_x;%the n values of x coordinate where we...
% evaluate f(z)on curve (near middle of chord)
y=1−x.^2; %the values of y corresponding to the above
z=x+i*y; % the complex values of z needed on the curve
fz=cos(z);%the n values of the function on the curve
yd=Y(2:m)−Y(1:m−1);% the vertical component of the vector
% chord
xd=2*delta_x;%the horizontal component of the vector chord
zdchord=xd+i*yd;% the vector chord complex representation
s=fz.*zdchord; %the product of the complex vector chord...
% and the function cos z evaluated nearby on the curve
series_sum(n)=sum(s);% the sum of those products
end
number_of_terms=[1:n]';
format long
approx_value=series_sum.';
disp('number of terms in series series sum')
disp([number_of_terms approx_value])
disp('sin(1)−sin(i)')
disp(sin(1)−sin(i))

The output of this program is

max number of terms in series10

nmax = 10

number of terms in series series sum

1.000000000000000 0.741951831691765 − 1.530431115914932i

2.000000000000000 0.817101149113634 − 1.246464676977509i
3.000000000000000 0.830203869938418 − 1.206054884648359i
4.000000000000000 0.835034307863057 − 1.192415447540212i
5.000000000000000 0.837320720853477 − 1.186178843527661i
6.000000000000000 0.838576965669872 − 1.182810093770062i
7.000000000000000 0.839339429392264 − 1.180785102772347i
8.000000000000000 0.839836339025017 − 1.179473274644714i
9.000000000000000 0.840177956756501 − 1.178574997871836i
10.000000000000000 0.840422785323087 − 1.177933013899162i

sin(1)−sin(i)
0.841470984807897 − 1.175201193643801i

Notice that the agreement between the above “exact” value and the value
obtained by the series approximating the integral is surprisingly good.
With only a five-term series, the disparity in both real and ­imaginary parts is
less than 1%, and agreement continues to improve with the use of more terms.
98 A MATLAB® Companion to Complex Variables

As a check, we should observe that the results obtained from the three-term
series agree with that obtained in the previous example.
The reader might wonder what the point of the preceding exercise is if we
can evaluate the integral so easily by using the antiderivative, which quickly
yields the exact result sin(1) − sin(i). Of course, there is no point in deter-
mining the integral numerically in the way we just have. However, had the
integrand been a function that is not analytic, for example, f(z) = cos|z| or
f ( z) = cos z, we cannot find an antiderivative, and a numerical determination
is required.
The function f ( z) = cos z is analytic except at z = 0 and is therefore
z
­analytic throughout a simply connected domain containing the path of inte-
gration like the one just employed. However, an antiderivative F(z) cannot be
found for this function in terms of the elementary functions of calculus, and
a numerical integration is necessary unless one might have access to numeri-
cal tables (in this case, they do exist) yielding this integral.
MATLAB provides several functions that will perform n ­umerical
­integrations, among them are quad, quadl, and quadgk. The last is among
the most recent in MATLAB and of the most use to us here, because under
certain circumstances, it permits integration between complex limits, as well
as limits that can be infinite. It is so valuable that the next section is devoted
to quadgk.

Exercises
1+ i

1. a. Approximate the integral

∫ sin z dz along the contour y = x by
0
2

using a three-term series in the expression ∑ f ( z )∆z . Choose

k k
k =1
the vector chords Δzk as they were done in Example 3.7 (i.e., they
have identical projections on the real axis). Also, choose the
xk values to be identical to the value of x in the middle of each
chord, as was done in Example 3.7.
b. Explain why the exact value of the integral is 1 – cos(1 + i), and
compare this with the value found in (a).
c. Write a MATLAB program that will approximate the integral
by n terms where n varies between 1 and 10. Use the same
criteria for Δzk (uniform x-axis projections) as was done in
Example 3.7 and the same criteria for choosing xk as was done
in part (a).
Sequences, Series, Limits, and Integrals 99

2. a. Consider the integral

∫ sinh z dz taken along the contour y = sin x.
0
Perform this by means of a numerical integration like that done
in Examples 3.7 and 3.8. Use Δzk that have identical projections
along the x-axis. Use n vector chords to approximate the contour,
where n varies from 1 to 10. Show your results for all 10 values.
Now evaluate the integral exactly by using the antiderivative of
sinh z. Are the results of your numerical integration real num-
bers? Explain why, ideally, they should be real.
b. Use the above technique to approximate the integral by a sum
with 100 terms. How different is this result from the one obtained
with 10 terms? Consider especially the imaginary part.
i

3. Consider the integral

∫ Log z dz performed in the half plane Re(z) ≥ 0
−i
along the half circle satisfying x2 + y2 = 1. Perform this by means of a
numerical integration like that in Examples 3.7 and 3.8 but here use
Δzk that have identical projections along the y-axis. Again use n vec-
tor chords to approximate the contour where n has only the values
10, 100, and 1,000, in other words, n increases by factors of 10. The
integrand is analytic in the given half plane and has an antideriva-
tive that you can find in a table of integrals if you have forgotten it.
Using this function, find the exact value of the integral and compare
it with the “exact” value.
2
sin z
4. a. Consider the integral
∫ 0
z
dz performed in the complex plane

along y = x(2 – x). Perform the integration numerically by fol-

lowing the technique in Examples 3.7 and 3.8, which uses values
for Δzk having identical projected lengths along the x-axis. If you
evaluate sin z/z at those values of z whose x coordinates lie at
the middle of each chord approximating the curve, then you will
never have to ask MATLAB for the undefined sin (0)/0.
Use series having 10, 100, and 1,000 terms and state your
answers in double precision.
b. The integrand is not analytic at z = 0 because it is undefined
there. However, the integrand has a removable singularity at this
sin z
point. Note that lim z = 1 as can be verified with L’Hôpital’s
z→ 0
rule. We can remove the singularity in the integrand by ­defining
the integrand as 1 when z = 0. This creates an integrand that is an
entire function. The reader may wish to review “removable sin-
gularities” in W section 6.2 or S section 3.11. The integrand is thus
100 A MATLAB® Companion to Complex Variables

an entire function, and we can deform the contour of integration

into a straight line connecting z = 0 with z = 2. You may wish to
review “deformation of contours” or “path independence” in W
section 4.3 or S section 4.15. Thus, an integral having the same
2
sin x
value as the one in (a) is
∫
0
x
dx on this line. This kind of inte-

gral turns up frequently in antenna engineering and optics where

one encounters the function called the “sine integral” defined as
x′
sin x
Si ( x′) =
∫
0
x
dx. If you Google the words “sine integral, Si, and

high accuracy,” you will come to a website that gives values of

this integral for different real values of x’ to a high degree of
accuracy. Thus, the results in part (a) should approximate Si(2).
The site yields Si(2) = 1.60541297680269484858. Compare this with
the result obtained by using the series in part (a) with 1 million
terms.

3.4  Integrations in the Complex Plane: Part 2, int and quadgk

MATLAB provides us with a wealth of methods for integration. We confine
ourselves here to two of the most useful tools: int and quadgk. The first of
these, int, will do symbolic integration. Ask it for the indefinite integral of cos x
and it will return sin x. It will also perform definite integrals. If you ask it to
integrate cos πx from 1/13 to 1/11, it will produce (sin(pi/11) − sin(pi/13))/pi.
If  possible, it does not produce decimal values for the results. However, if
you ask int for the value of an integral which it cannot obtain by evaluating
an antiderivative at the two limits, it can give you a numerical result based
on numerical integration. You will also receive a warning that an “Explicit
Integral” could not be found. To get the numerical result, you must use the
command vpa as described below. The command double can also be used but
is less reliable and will give the answer to exactly 16 decimal places. With vpa
you have more flexibility and reliability.
The tool quadgk is only for definite integrals, which it evaluates numeri-
cally. The output of this function is a decimal number. The great utility of
quadgk for us is that it will perform a contour integration in the complex
plane provided that the contour is composed of straight lines. This last
restriction might seem like a handicap, but we know that if our integrand is
analytic, we can usually deform a curved contour into a straight line or into a
contour composed of straight lines. In the case of both int (when used for def-
inite integrals) and quagk, we can have one or both limits of integration set to
Sequences, Series, Limits, and Integrals 101

infinity which we designate inf. The number inf is the real number, say x, as
it increases without bound through positive values, inf = lim
x→∞
x, while –inf is a
real number x as it decreases without bound through negative ­values –inf =
lim x. One should note that like quadgk, int will do an integration in the
x →−∞
complex plane, but the contour must be a single straight line and cannot be
composed of a sequence of straight lines. For example, quadgk can integrate
all the way around or part of the way around a triangle, while int would
integrate only along one side.
At this point the reader should study both methods of integration, int
and quadgk, by going to the help feature of MATLAB. In the remainder
of this section, we see how int and quadgk can be used to verify results
obtained with residue calculus or in some cases perform integrations that
are possible with complex variable theory but require tedious mathematical
manipulations.

Example 3.9
∞
x 2 dx
Find I =
∫
0
x4 + 9
by using residue calculus and by using both int and

quadgk.

Solution:
a. With residues:
Let us recall a theorem from complex variable theory (see W
­section 6.5 and
∞
S problems 7.15–7.17). Suppose we must evaluate
P( x) , where P and Q are polynomials in the
the integral
−∞
∫
Q( x)
dx

variable x.
Assume that the highest power of x in Q exceeds that in P
by two or more, in other words, the degree of Q is at least two
bigger than P, and we also agree that Q(x) ≠ 0 for all real x. Then,

∞
 P( z) 
∫ Q(x)dx = 2πi∑ Res  Q(z)  @ all the poles in the upper half plane (3.17)
P( x)

−∞

We will use uhp to abbreviate “upper half plane,” and Res of

course means “residue.”
In words: the integral above is 2πi times the sum of the resi-
dues of the function P(z)/Q(z) at the poles of this function in the
uhp. If we wish, we may instead sum the residues of the poles in
the lower half plane. However, should we make that choice, we
would need to put a minus sign in front of the two on the right
side of the preceding equation.
We were given an integration for I that has limits from 0 to
infinity. Because of the even symmetry in the integrand, we
102 A MATLAB® Companion to Complex Variables

∞ 1
x 2 dx
have that I =
∫x
−∞
2
+94
. Thus, in Equation 3.17, we take P(z) = z2/2

and Q(z) = z4 + 9. The equation Q(z) = z4 + 9 = 0 has roots where

z = (−9)1/4 = 3 e ± iπ/4 and 3 e ± i 3 π/4. Two of these roots of Q(z) lie
in the upper half plane: z = (−9)1/4 = 3 e iπ/4 and 3 e i 3 π/4. These
roots result in P(z)/Q(z) having simple poles since all four roots
of Q(z) are distinct (nonrepeated). Recall (see W section 6.3) that
if f(z) = g(z)/h(z) has a simple pole at a point z0, where g(z0) ≠ 0,
then the residue of f(z) is

g ( z) g ( z0 )
Res @ z0 = (3.18)
h( z) h′( z0 )

This is one of the handiest formulas in complex variable

theory. Thus, if we take P(z) = g(z) and Q(z) = h(z), the residues
of P(z)/Q(z) at z = (−9)1/4 = 3 e iπ/4 and 3 e i 3 π/4 are, respectively,
1 2
z /(4 z 3 ) = 1/(8 z) evaluated at these two points. The residues
2
1 − iπ/4 1 − i 3 π/4
are e and e . Summing these and applying
8 3 8 3
Equation 3.17, we have finally

2 πi π
I=
8 3
[ cos(π/4) − i sin(π/4) + cos(−3π/4) + i sin(−3π/4)] = 2 6
We note that the answer must be real, and it is.
b. To integrate with int, we use this MATLAB code:

syms x;
I=int((x.^2)./(x.^4+9),0,inf)

The output is
I = (pi*6^(1/2))/12
π
Observe that this is the same as π 6 / 12 = , the result
obtained by residues. 2 6
Notice the importance of the line of code:
syms x;

This tells MATLAB that x is a symbolic variable.

c. To integrate with quadgk, we use this code:

f=@(x) (x.^2)./(x.^4+9)
format long
I=quadgk(f,0,inf)

The output is
I = 0.641274915080932
Sequences, Series, Limits, and Integrals 103

In the preceding code, the line format long is optional. It gave us an

output with 16 decimal places that we can use to compare with the out-
put of part (b). If you use MATLAB to convert (pi*6^(1/2))/12 to a deci-
mal expression and use long format, you will obtain the same result as in
part (c). To make the conversion, use the line of code

double ((pi*6^(1/2))/12) or vpa ((pi*6^(1/2))/12, 16).

The number 16 here tells vpa to yield an expression with 16 decimal

places. You can use other integers here.

Some comments:
Obviously, quadgk and int each have their uses. If you want to obtain
an immediate numerical expression for use in other calculations,
then quadgk would be your best bet. If you want an expression with
numerical symbols to place inside an engineering or physics formula,
you might prefer int. Of course, you can do the integration in the pre-
ceding example with residue calculus. However, if you were given the
1
x 2 dx , you could not quickly find the
­problem to find, for example,
0
∫
x4 + 9
answer with r­ esidues. You might resort to an integral table or perhaps
just use quadgk or int depending on your needs. The code would be
like that used in Example  3.9 except that the upper limit of integra-
tion would no longer be inf but would be 1. There are symbolic prob-
lems  in  integration that cannot be done with int (i.e., it cannot do a
closed-form expression for the result). See problem 2 that follows for
example.
∞
x 2 dx
A problem such as I =
∫
0
x4 + x2 + 3 + i
can in principle be done with

residue calculus, but you would be faced with the messy job of locating
the roots of z4 + z2 + (3 + i) in the complex plane. Here you are much
better off with either int or quadgk, depending on your goal. A similar
integral is considered in exercise problem 5.
3

∫
1/2
A problem like e i(sinh z ) dz will pose a difficulty for int as the ­following
0
code shows:

syms z
I=int(exp(i∗sqrt(sinh(z))),0,3)

Which yields the following warning:

Warning: Explicit integral could not be found.
However, an additional line of code:

vpa(I,10)
yields this answer, in numerical form, correct to 10 decimal places.
ans = 2.15702077∗i + 0.1021467801
104 A MATLAB® Companion to Complex Variables

3.4.1  Integration in the Complex Plane with MATLAB

The great utility of quadgk, as we have mentioned, lies in its ability to do
integrations in the complex plane along contours composed of straight lines.
If we ask MATLAB for quadgk( fun, a, b) where fun is expressed as a function
handle using the α notation, and a and b are numbers that can be complex,
then the integral is evaluated numerically along a straight line connecting a
and b in the complex plane. Here is an example. Notice that when using the
anonymous function, we do not have to tell MATLAB that the argument of
α is a symbol. This is assumed. But if you forget and use syms, the program
will still run.

Example 3.10
(1+ i )

Find I1 =
∫e
(1− i )
iz
dz by using quadgk, and verify the answer with ­complex

variable theory.

Solution:
The required code is simply as follows:

syms z
f=@(z) exp(i*z)
format long
I=quadgk(f,1−i,1+i)
check_answer=(exp(i*(1+i))−exp(i*(1−i)))/i

Whose output is
I = −1.977795411525730 + 1.269927829569473i
check_answer = −1.977795411525731 + 1.269927829569473i

Notice the “check answer” portion of the output. We recall that since
eiz is an analytic function, we may simply evaluate the integral using the
e iz e i(1+ i) − e i(1− i)
antiderivative of this function. Thus, I = |11+− ii = . Observe that
i i
the agreement between the two methods is nearly perfect. Had the given
integrand been e iz, which is not analytic, we could not have checked the
results of our numerical integration so easily, as this function does not have
an antiderivative.

A note is appropriate here about the use of infinity (inf) in the limits of
quadgk if we are integrating in the complex plane. One can use this as one of
the limits provided that the other limit is real. One can also use - inf if the other
limit is real. You can use –inf and inf together. We cannot really use infinity
as a limit unless we stick to the real axis, although complex integrands are
permissible when integrated on the real axis.
Sequences, Series, Limits, and Integrals 105

3.4.2 Waypoints
Using quadgk, we can integrate along a contour composed of straight
lines. The waypoints vector, placed in quadgk, tells quadgk where the end-
points of these straight lines lie. For example, suppose we wish to inte-
grate cos z/z from 1 + i to −3i using this path of successive straight lines:
from 1 + i to 2i and then along the line from 2i to −1 + 2i, and from there
to −1 − i, and finally from −1 − i to −3i. (See the solid lines with arrows in
Figure 3.6.) We write code in which waypoints is a row vector whose ele-
ments are the points where the contour abandons its old straight line path
and starts on a new one. We do not include in waypoints the starting points
and endpoints in the numerical integration. These points still appear as
the limits of integration. The numerical entries in waypoints are the points
on the path where the tangent of the overall contour of integration is dis-
continuous; we enter these points as complex numbers in waypoints from
left to right in the order in which they are encountered on the path. If you
elect to use waypoints, you may not use infinity (i.e., inf or −inf) in the
waypoints vector or in either the starting point or endpoint of the integra-
tion if any of the elements in waypoints is not real, or if any of the limits
of integration are not real.
Here is the required code to solve the problem described above:
format long
syms z
f=@(z)cos(z)./z;
IA=quadgk(f,1+i,−3i,'waypoints',[2i,−1+2i,−1−i])

y
2

x
0
0

–1

–2

–3
–3 –2 –1 0 1 2 3

FIGURE 3.6
An illustration of “Waypoints.”
106 A MATLAB® Companion to Complex Variables

The output is

IA = 4.078219914209672 + 4.425139846864735i

Notice (refer to the figure and the solid lines) that the integration begins
at 1 + i and ends at –3i and that the waypoints are 2i, –1 + 2i, and –1 – i.
We can check this result. If we integrated along a straight-line path
connecting 1 + i with –3i by using I=quadgk(f,1+i,−3i,), we would not
obtain the above value for IA. This is because the operation we are doing
depends on the path of integration. The function cos z/z has a singular-
ity (a simple pole) at z = 0 but is otherwise analytic. We cannot deform
the ­integration along the original path (the solid line in Figure  3.6) into
the ­dotted-line path without forcing the contour to pass through the
singularity.
Suppose, however, we were to take the result IA from the above and add it
to the integral of cos z/z taken along the straight line going from –3i to 1 + i.
cos z
The sum of these two integrals would be the value of the integral
∫ z
dz
taken in the positive (counterclockwise) direction around the closed contour
that is the polygon shown in Figure 3.6.
We can do this integration with residues, because cos z/z has a simple pole
cos z
at z = 0, which is enclosed. The result should be 2 πi Res[ @ z = 0], where
z
Res is an abbreviation for residue. Using Equation 3.18, we have that the resi-
due is cos 0 = 1. So the result of our integration around the closed contour in
Figure 3.6 is 2πi.
Let us add our integral IA obtained above from quadgk to the integral of
cos z/z taken along the dotted line in Figure 3.6. We call the latter integral IB.
The sum of these should equal 2πi and would provide a check on the value
of IA. Here is our code:

format long
syms z
f=@(z)cos(z)./z;
IA=quadgk(f,1+i,−3i,'waypoints',[2i,−1+2i,−1−i])
IB=quadgk(f,−3i,1+i)
total_is= IA +IB
two_pi_i=2*pi*i

Whose output is
IA = 4.078219914209672 + 4.425139846864735i

IB = −4.078219914209674 + 1.858045460314853i

total_is = −0.000000000000002 + 6.283185307179587i

two_pi_i = 0 + 6.283185307179586i
Sequences, Series, Limits, and Integrals 107

Observe that the agreement between the sum of these two integrals and
the desired 2πi is excellent, and any disparity is due to round-off errors.
Rather than integrate around the quadrilateral by combining two inte-
grals, we can perform the integration with a single quadgk statement. We
take the beginning and final limits of integration as 1 + i and use the remain-
ing corners as waypoints. The code is

format long
syms z
f=@(z)cos(z)./z;
IC=quadgk(f,1+i,1+i,'waypoints',[2i,−1−i,−3i])

which yields
IC = −0.000000000000020 + 6.283185307179271i

This agrees with the total sum of the two prior integrals, IA plus IB, up to
13 decimal places. We would have been happier if the real part above had
turned out to be zero, but rounding errors prevented this.

Exercises
1

∫
1. Evaluate e i sin x cos x dx by the following three methods and ­compare
0
the results
a. By using the antiderivative of the integrand
b. By using int
c. By using quadgk
1

2. Evaluate
∫e 0
i sin x
dx. Note that, unlike the previous problem, an

a­ ntiderivative of the integrand does not exist in terms of standard

functions. Determine this integral by means of
a. int
and
b. quadgk and compare the results.
∞
dx
3. Evaluate the integral I =
∫x
−∞
4
−i
by these methods:

a. Using residue calculus

b. Through the use of int (obtain the result both in symbols and in
a decimal expression)
108 A MATLAB® Companion to Complex Variables

c. Through the use of quadgk.

d. Verify that the three answers above are numerically the same (to
within round-off errors).
∞
dx
4. Using int, evaluate the integral I ( a) =
−∞
+ a2 ∫x
, where a ≠ 0 is an
2

­arbitrary complex number. If necessary, read the documentation for

int to see how to evaluate an integral containing a symbolic param-
eter, which in this case is a. Briefly, you must mention that parameter
in the syms statement. You might wish to make your answer look
more attractive by using the command pretty. Now take a = ρeiϕ,
where ϕ is the principal angle, –π < ϕ ≤ π and ρ is positive. Compare
your result with that obtained by residues. What values of a must be
avoided if we are to avoid singularities on the path of the integra-
tion? How is this reflected in your MATLAB result?
∞
dx
5. Evaluate I =
∫x
0
4
+ x2 + i
by using the following:

a. The function int. Convert your symbolic result to a decimal one

using both vpa and double. How different are they? The preced-
ing integration can be done using residues, but it is tedious (as can
be seen by your answer, in symbols, to this part), and you would
have to find the zeros of z4 + z2 + i in the complex plane and then
find the residues at each of these poles. We do this in part (c).
b. The function quadgk. Compare your results to those in part (a).
c. If we are determined to do this problem with residues, MATLAB
can assist us.
First, find the location of all the zeros of z4 + z2 + i by using
MATLAB’s roots. Explain why all the poles of 1/(z4 + z2 + i)
∑
1
are simple. Observe that the integral is I = πi Res 4
z + z2 + i
@ poles in uhp. Prove that the residue at a pole is 1/(4z3 + 2z) eval-
uated at the pole. Using MATLAB, find the residues at all the
poles and sum those that arise in the upper half plane. Compare
your final answer with those obtained in parts (a) and (b).
Hint: If vrt is the column vector containing the roots of the above
polynomial, then uu=imag(vrt)>0 is a column vector having
ones where the corresponding roots are in the uhp and zeros
where they are in the lower half plane (lhp).
6. MATLAB can compute residues. Suppose you have a rational expres-
sion consisting of two polynomials in z:

bn z n + bn− 1 z n− 1 + ... + b0
f ( z) =
am z m + am− 1 z m− 1 + ... + a0
Sequences, Series, Limits, and Integrals 109

Then MATLAB can find the location of the poles of this expression
as well as the residue at each pole. At this point, you should read the
documentation for residue in the help folder. Briefly, you employ the
statement

[r,p,k]=residue(b,a)

where b is the row vector consisting of the coefficients of the poly-

nomial in the numerator. These are entered in order of descend-
ing powers, just as is done in using roots. Be sure to enter a zero
if a term is missing. For example, if the numerator is 4z3+2z+i,
then b = [4 0 2 i]. In a similar way, a is a row vector describing the
denominator.
The outputs of residue are the vector r, which is the residue of
the expressions at the poles, and the vector p, which is the loca-
tion of the poles (stated in the same order as the residues are
given). If m > n, then the vector k has no entries. If m ≤ n, then k
gives the coefficients of a polynomial of degree n−m. The poly-
nomial is the remainder after f(z) is resolved into partial frac-
tions. The coefficients are given in the usual order, just described.
Notice that ­r esidue works only for rational expressions (ratios of
polynomials).
3 z 2 + 2 iz + (1 + i)
a. For the function f ( z) = , use MATLAB to find all
5 z 4 + iz 3 + i
the poles and state the residue at each pole.
∞
3 x 2 + 2 ix + (1 + i) . Remember to
b. Using these residues, find
∫
−∞
5 x 4 + ix 3 + i
dx

use only the residues in the uhp or, if you prefer, only the lhp.
c. Check your answer to (b) by using int, being sure to convert
your answer to a decimal form to facilitate comparison with the
answer from (b).
7. Recall from complex variable theory that if we are given the integral
∞
P( x)
∫ Q(x) e
−∞
iνx
dx

where v > 0 and the degree of Q(x) exceeds that of P(x) by 1 or more,
∞

then
P( x)
∫ Q(x) e iνx
dx = 2 πi ∑ Res QP((zz)) e iνz
@ all poles in the uhp .
−∞

If v < 0, we would use the poles in the lower half plane and replace
2 with –2.
110 A MATLAB® Companion to Complex Variables

a. Using residues, find

∞
ei2 x

∫
−∞
x
( x − 1)2 + 1
dx

b. Evaluate this integral using int in MATLAB.

c. Evaluate the same integral using quadgk. Notice that the answer
you have given is erroneous as can be seen by comparing your
answer with the result of part (a). The function quadgk some-
times fails to evaluate an integral where the integrand is oscil-
latory, dies out slowly, and one or both limits are infinite. Try
integrating between –1,000 and 1,000 and see if this helps. Notice
that a difficulty did not arise in part (b) because int uses the anti-
derivative to evaluate the integral.
d. Using the result of (a) or (b), find

∞
cos 2 x

∫ x (x − 1)
−∞
2
+1
dx

and

∞
sin 2 x

∫ x (x − 1)
−∞
2
+1
dx

You might wish to review W section 6.6 or S section 7.4.

1+ i

8. a. Using quadgk, perform the integration

∫ cosh zdz
1− i
along the

straight-line segment connecting the two limits.

b. Integrate these same two points by using the ­broken-line path
going from 1 – i to –1 – i then to –1 + i and from there to 1+i. Use
waypoints. Compare your answer to (a).
c. Check your answer by using the antiderivative of the integrand
and compare all three results.
9. a. Repeat part (a) above but use as the integrand cosh z/z.
b. Repeat part (b) above but use the new integrand.
c. Explain why in parts (a) and (b) you obtained differing results.
d. Reconcile the difference between the results by using a residue.
Sequences, Series, Limits, and Integrals 111

10. a. Consider
cos( πz)
∫ sin(πz) dz
where the integral is done on the contour |z|=3/2. Explain why we
can get the same result if we integrate instead in the positive direc-
3 3
tion around the square having corners at z = ± and z = ± i.
2 2
b. Perform the given integration by using residues.
c. Check your answer by using quadgk and the square contour.
Employ waypoints. Are the results the same?
d. By using quadgk, do the integration
cos( π z )
∫
sin( πz)
dz

around the square. Explain why we cannot do this problem with

residue calculus.
i

∫
11. Compute Log z dz using quadgk and waypoints. Use the following
−i
two different paths:
a. Straight lines connecting –i, 1, and i.
b. Straight lines connecting –i, –1, and i.
c. Discuss the fact that the answers to (a) and (b) are not the same.
Note that you are using the principal value of the Log.
d. Derive both of the answers analytically by using the antideriva-
tive of the log.
Hint: For part (b), notice that the imaginary part of log changes
discontinuously as you cross the branch cut at z = –1. Thus, you
should break your integration into two integrals: one from –i
to –1-i∗eps (at the bottom of the cut) and the other from –1 to i.
Notice that using principal values in MATLAB log(–1–i∗eps) ≈ –iπ
and it takes log(–1)=iπ.
12. a. Using quadgk and waypoints, find
sin z
I=
∫ sinh z (z 2
+ 1)( z + 10i)
dz

taken around the circle |z–1|= 4.

Hint: Deform your contour into a box that encloses the same
poles as the circle.
b. Check your answer to (a) by using residue calculus.
This page intentionally left blank
4
Harmonic Functions, Conformal
Mapping, and Some Applications

Harmonic functions and conformal mapping are two of the richest and
most interesting topics in complex variable theory. It is our good fortune
that harmonic functions, which are always the real and imaginary parts
of analytic functions, can be used to describe two-dimensional configura-
tions of fluid flow, heat transfer, and electrical fields. Historically, confor-
mal mapping achieved importance as a means of solving two-dimensional
problems in these branches of engineering and physics. Although numeri-
cal methods on modern computers have rendered this application of less
importance, conformal mapping still provides us with solutions to certain
canonical problems that can be used to verify the correctness of computer-
generated solutions. This is analogous to the practice of learning to perform
integrations in e­ lementary calculus courses even though, for example, the
MATLAB® Symbolic Mathematics Toolbox will perform them for you. One
must know what kinds of answers to expect from a computer and to have a
means of checking them.
In Chapter  2, we learned some graphical techniques for studying
functions of a complex variable. These included three-dimensional
plots as well as the use of contour lines. Conformal mapping provides
another means of visualizing the properties of a function of a complex
­variable. In this case, we work in just two dimensions and use two sepa-
rate planes.

4.1 Introduction
Let z = x + iy and w = u + iv be complex variables. If w = f(z) describes a func-
tion of the complex variable z and is defined at the point z0, we can say that
this function maps the point z0 into the point w0 where w0 = f(z0).
We can display this fact using two planes—the z and w planes. The
points z 0 = x0 + iy 0 and w 0 = u 0 + iv 0 might be plotted as * or some other
symbol. We say that w 0 is the image of z 0 under the given transformation
w = f(z).

113
114 A MATLAB® Companion to Complex Variables

We might also say that z0 is the image of w0 under an inverse t­ ransformation.

Suppose, for example, that z = g(w) yields z when w is known. Here g(w) is
said to be the inverse transformation of f(z). We have z0 = g(w0). We assume
in the preceding discussion that we are dealing with well-behaved, single-
valued functions.
For example, suppose w = f(z) = log  z and z0 = 1 + i so that its image is
π π
w0 = Log(1 + i) = Log 2 + i ≈ .3466 + i = u0 + iv0. Incidentally, z = g(w) = ew
4 4
and z0 = 1 + i = e w0 = e Log 2 +iπ/4.
We display the mapping with the code:

clf
z=(1+i);
w=log(z);
figure(1);
plot(z,'*');grid on;xlabel('x');ylabel('y');
%the above plots the point 1+i in z plane,
% and labels the axes.
text(1.02,1.02,'z_0')
% above labels the above point, we use 1.02, 1.02
%and not 1,1, so as to keep the letter zo off the star *
text(1,1.6,'\fontsize {12} The z Plane')
figure(2);
plot(w,'*');grid on;xlabel('u');ylabel('v')
%above plots log(1+i) in w plane.
text(real(w)+.05,imag(w)+.05,'w_0');
%note the use of .05 in the above to keep
%the wo off the *
%labels the above point
text(0,1.7,'\fontsize{12} The w Plane')

The output is as shown in Figure 4.1.

Suppose we wish to map a large number of points from one plane to
another. We might designate them as A, B, C … (we obviously have 26 of
them if we need that), and their images as A, B, C …. We resort here to a
bold italic font. Alternatively, we might simply use a, b, c or some other such
variation in lieu of A, B, C.
To understand how to effectively use letters, realize that we may convert
any letter of the alphabet, regarded as a character, to a unique numerical
value using a convention called the ASCII standard representation. The capi-
tal letters A, B, C, …, Z have ASCII numbers 65, 66, 67, …, 90. Similarly, the
lowercase letters a, b, c, …, z have ASCII numbers 97, 98, 99, …, 122.
We may manipulate characters as if they were numbers by operating on
the corresponding number in the ASCII representation. For example, we
may begin with the letter A in a string and produce the letter D by adding
Harmonic Functions, Conformal Mapping, and Some Applications 115

1.8

1.6

1.4

1.2
z0
1
y

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x
(a)

1.5

1
w0
v

0.5

–0.5
–1 –0.5 0 0.5 1 1.5
u
(b)

FIGURE 4.1
(a) The point for 1 + i and (b) the principal value of log(1 + i).

3 to the ASCII number for A (which is 65) and then converting the result
(which is 68) back to a character which will be D. We convert from the ASCII
number to the actual letter of the alphabet by use of the operation char.
Here is how we might convert a letter A to letter D. Recall the use of single
quotes to make a character, ‘…’.
116 A MATLAB® Companion to Complex Variables

>> uu=‘A’
uu = A
>> vv=uu+3
vv = 68
>> char(vv)
ans = D
Here is how we might go from the lowercase letter z to the lower case u.
Recall that there are four letters between u and z and that the ASCII number
for u is smaller than that for z.
>> tt=‘z’
tt = z
>> LL=tt−5
LL = 117
>> char(LL)
ans = u
Now consider the 11 points on the line x = 1 given by z = 1, 1 ± 1i, 1 ± 2i, …
1 ± 5i. We will identify them with the letters A … K (starting with 1 – 5i and
proceed upward along x = 1 to 1 + 5i). Using the mapping w = log z, we will
plot not only the given values of z but their images in the w plane and label
them as A … K.
This is accomplished with the following code:
clf
y=linspace(−5,5,11);
x=1;
z=1+i*y;
w=log(z);
figure(1)
plot(real(z),imag(z),'.');grid on;
%plots a dot for each value used in
%z plane
xlabel('x');ylabel('y');title('z plane');grid on
figure (2)
plot(real(w),imag(w), '.');axis equal;
%plots a dot at each value of logz
xlabel('u');ylabel('v');title('w plane')
for n=1:length(y)
letter=char(n+64);
%the above creates the letters A, B,...
figure(1);
text(1.02,imag(z(n)),letter,'FontSize',10);
% the above plots the letters A,B,
%note that using 1.02 instead of 1 in the above
%keeps the text off the dot in the plot.
%a similar thing is done for figure(2)
figure(2)
Harmonic Functions, Conformal Mapping, and Some Applications 117

te
xt(real(w(n))+.02,imag(w(n)),letter,'FontSize',10, ...
'FontAngle',...
'italic','FontWeight','bold');
% the above plots in the w plane the
% italicized bold A, B, C...
% which are the images of the points in the z plane
hold on
end;grid on

This yields the output as shown in Figures 4.2 and 4.3.

Every point that we consider in the z plane is marked with a dot and a
letter A through K. Its image in the w plane is also marked with a dot and a
corresponding letter A through K in boldface italic.
If we use plot(real(z),imag(z)) in place of plot(real(z),imag(z),'.'),
we obtain a continuous line connecting the points we have chosen to map.
A similar change can be made to plot(real(w),imag(w),'.') to obtain a
continuous line.
Sometimes, instead of wanting to map a selection of discrete points from
one plane to another, we wish to map a curve, generally one that is smooth
or piecewise smooth. Since a curve contains an infinite number of points,

5 K

4 J

3 I

2 H

1 G

0 F
y

–1 E

–2 D

–3 C

–4 B

–5 A
0 0.2 0.4 0.5 0.8 1 1.2 1.4 1.6 1.8 2
x

FIGURE 4.2
Some points for values of 1 + iy.
118 A MATLAB® Companion to Complex Variables

J K
I
H
1

0.5

0 F
v

–0.5

–1
D
C
B A
–0.5 0 0.5 1 1.5 2 2.5
u

FIGURE 4.3
Principal value of the logarithm for numbers in preceding figure.

we cannot map them all. The usual procedure is to map a sufficient num-
ber of points from the curve in the z plane into the w plane—the number
of points is large enough so that if MATLAB connects them together the
resulting curves in either plane resemble the usually smooth ones that we
are considering. If in doubt, overestimate the number of points you think
you might need. The only drawback will be increased required time for the
computation, and this is often not a consideration.
Let us consider two intersecting curves and again map them with the
transformation w = log z. These will be the circle |z| = 2 and the line seg-
ment x = Re(z) = 1, −1 ≤ y ≤ 1.
As a first attempt, we proceed as follows with this code:

clf
y=linspace(−3,3,101);
x=1;
z1=1+i*y;
w1=log(z1);
th=linspace(−pi,pi,100);
z2=2*exp(i*th); % yields 100 points on circle of radius 2
w2=log(z2);
Harmonic Functions, Conformal Mapping, and Some Applications 119

figure(1)
plot(real(z1),imag(z1),'linewidth',2);hold on
plot(real(z2),imag(z2),'linewidth',1);grid on;axis equal
xlabel('x');ylabel('y');title('z plane');grid on
figure (2)
plot(real(w1),imag(w1),'linewidth',2);hold on
plot(real(w2),imag(w2),'linewidth',1);
grid on;axis equal
xlabel('u');ylabel('v');title('w plane')
y=linspace(−3,3,6);th=linspace(−.9*pi,pi,6);

The output is as shown in Figures 4.4 and 4.5.

The line x = 1 in the z plane is the heavier line in Figure  4.4, and the
­circle |z| = 2 is the thinner line in the same figure. The image of x = 1 is the
heavier line in Figure 4.5, while the image of the circle is the heavier line in
the same figure. To see the effect of using too few data points for your map-
ping, try changing the line of code above from y=linspace(−3,3,101); to
y=linspace(−3,3,11);. You will get a jagged result.
What is lacking in the above two figures is any indication of how points
from one curve are mapped into another. We can remedy this by labeling
some points on each of the two curves in Figure 4.4, using the letters A, B, C, …

0
y

–1

–2

–3
–3 –2 –1 0 1 2 3
x

FIGURE 4.4
Circle |z| = 2 and Line Re(z) = 1.
120 A MATLAB® Companion to Complex Variables

0
v

–1

–2

–3
–3 –2 –1 0 1 2 3 4
u

FIGURE 4.5
Images of the preceding curves when mapped by Log z.

and a, b, c, … while indicating their image points with bold italic versions of
these letters. We add on this code to the above code. The additional lines
­provide an idea of how each curve is deformed by the mapping process:
y=linspace(−3,3,6);th=linspace(−.98*pi,.98*pi,6);
z1=1+i*y; z2=2*exp(i*th);
%we will use fewer data points for the letters
%on the curves than were needed to make the curves
for n=1:length(y)
letter1=char(n+64);
% above creates the letters A, B,...
letter2=char(n+96);
%above creates letters a,b, ...
figure(1);
text(1,imag(z1(n)),letter1,'FontSize',10);
%plots the letters A,B,
% in the z plane on x=1;
text(real(z2(n)),imag(z2(n)), letter2,'FontSize',10)
% 
above plots the letters a,b, on the circle in
% the z plane
w1(n)=log(z1(n));
w2(n)=log(z2(n));
figure(2);
Harmonic Functions, Conformal Mapping, and Some Applications 121

tex
t(real(w1(n)),imag(w1(n)),letter1,'FontSize',10, 'FontAngle',
...'italic','FontWeight','bold');
% the above puts in the w plane the italicized bold A,B,..
% which are the images of the points A,B,.. in the z plane
t(real(w2(n)),imag(w2(n)),letter2,'FontSize',10,'FontAngle',
tex
...'italic','FontWeight','bold');
%the above puts in the w plane the italicized bold
%a,b..which are the
%images of a,b,c in the z plane
hold on
end;

With these new lines appended to the preceding program, the output is as
shown in Figures 4.6 and 4.7.
Notice the line of code: th=linspace(−.98∗pi,.98∗pi,6);. We do not set limits
on th (the angle used to locate points of the circle in Figure 4.6) from –pi to pi.
Otherwise in Figure 4.6, the points marked a and f would lie on top of one
another because of the line of code z2=2*exp(i*th), which would produce
two identical values for z2;.
We see how the vertical line x = 1 in Figure 4.6 is bent into the bow shape in
Figure 4.7 that is symmetrical about the real axis. We also see how the circle
of radius 2 is deformed into a line segment. It is not hard to show (try this)

3
F

2
e E

d
1
D

f
0
y

C
–1
c

B
–2
b

A
–3
–3 –2 –1 0 1 2 3
x

FIGURE 4.6
Circle |z| = 2 and Line Re(z) = 1.
122 A MATLAB® Companion to Complex Variables

3
f

2
e

E F
1
D d

0
v

C c
–1
B A

b
–2

a
–3
–3 –2 –1 0 1 2 3 4
u

FIGURE 4.7
Images of the preceding curves when mapped by Log z.

that the segment lies along the line u = Log 2, −π < v ≤ π, because MATLAB
uses the principal value of the log.
Consider the intersection between the circle and the straight line in
Figure  4.6 near the point marked E. The reader should show that the
­intersection is at x = 1, y = 3. With a little high school geometry, he or she
should see that the acute angle between the tangent to the circle and the ver-
tical line is 60°. See Figure 4.8 (detail).
If you study Figure  4.9, you will find that the intersection between the
images of the curves shown in the detail appears to also be about 60°.
It is in fact exactly 60°; notice that in each case the curve with the lowercase
letters is displaced 60° counterclockwise from the one with uppercase let-
ters. What we have here is an illustration of the conformal property in complex
variable theory (see W section 8.2, S section 8.4): if w = f(z) is analytic in a
domain D, and if C1 and C2 are two curves intersecting at a point z0 in D, then
the images of these two curves in the w plane will intersect in such a way
that their angle of intersection (at w0 = f(z0)) will be the same as that between
C1 and C2, provided that f′(z0) ≠ 0. Not only is the magnitude of the angle
preserved, but so is its sense. For example, if we must rotate C1 clockwise to
stretch it onto C2 (using the smallest possible angle), then this must be true of
their respective images. We now see where the word conformal comes from
Harmonic Functions, Conformal Mapping, and Some Applications 123

60°
e E

FIGURE 4.8
Detail of an intersection.

60 F
E

FIGURE 4.9
Detail of the images of the intersection for curves in preceding figure.

in “conformal mapping”: the angular intersection of intersecting curves in

one plane conforms in magnitude and direction to the angular intersection
of their images in the other. To repeat, conformal mapping assumes that
an analytic function is used for the mapping and that its first derivative is
nonvanishing at a point of intersection where angle preservation must be
preserved.
A nonconstant analytic function will always map a domain into a domain
(see W section 8.3). Suppose all the points in a domain D in the z plane are
mapped into the w plane by means of an analytic function w = f(z). Recall that
because the function is analytic, it is single valued. The result of the mapping
is a domain D′ in the w plane. If z1 and z2 are points in D, and the equation
f(z1) = f(z2) is only satisfied in D if z1 = z2, then we say that the mapping from
D to D′ is one-to-one. Examples of mappings that are and are not one-to-one
are given in Exercises 2 and 3.
In Figure 4.6, there is a domain whose boundaries are defined by the line
segment BCDE and the arc containing the letters c and d. With the aid of
Photoshop™, we show this domain heavily shaded in Figure 4.10. We can see
how this domain is mapped into the w plane by studying how its boundary
points are mapped into the w plane. Studying Figures 4.6 and 4.7 leads us to
conclude that the dark-shaped area is mapped into the dark domain shown
in Figure 4.11.
124 A MATLAB® Companion to Complex Variables

F
3
60°
2 E
e
d
1

0
y

–1 c

–2 B
b
–3 A

–4

–5
–6 –4 –2 0 2 4 6
x

FIGURE 4.10
A domain in the z plane.

3
f

2 e
60°

1 E F
d

0
v

–1 B

A
b
–2

a
–3
–3 –2 –1 0 1 2 3 4
u

FIGURE 4.11
The image of the domain in the preceding figure under the transformation w = Log z.
Harmonic Functions, Conformal Mapping, and Some Applications 125

For confirmation, note that the point z = 3/2, which lies in the domain
shown in Figure 4.10, has an image lying at w = Log(z) ≈ .4055, which is seen
to lie in the domain depicted in Figure 4.11.

Exercises
1. a. Consider the unit circle |z| = 1. Construct this circle in the z plane
by choosing 100 points uniformly spaced around the ­circle, plac-
ing them in the z plane, and having MATLAB connect them with
a smooth curve. Be sure to use axes of equal scales to see the cir-
cle. Suppose you had used only the nine points at arg(z) = 0, 45,
90, 135, … 315, 360 degrees, what kind of curve would MATLAB
give you, and how well would it approximate a circle?
b. Using the 100 points employed above, map the circle into the
w plane using the transformation w = ez.
c. Place the nine points described in part (a) on the circle generated
with 100 points in part (a). Label these points with the letters
A, B, C, … .
d. Using MATLAB, plot the images of these points with the letters
(italic) A, B, C, … on the figure in part (b).
2. Consider the rectangular-shaped domain in the z plane whose
boundary points are on the four straight-line segments: y = 0,
−1 ≤ x ≤ 1; y = π, −1 ≤ x ≤ 1; x = ±1, 0 ≤ y ≤ π.
a. Using w = ez, map these four line segments into the w plane by
means of MATLAB. Do  this by mapping 100 points from each
segment. Use a heavy line width (e.g., 3) for the boundary in each
plane.
Using MATLAB, place a different capital letter of the alpha-
bet at each of the four corners of the domain in the z plane, and
place one at the midpoint of each line segment. Put correspond-
ing bold italic letters at the images of these points in the w plane.
b. The interior points of the above rectangle form a domain. How is
it mapped into the w plane by the given transformation?
c. Explain why the mapping from the z plane to the w plane is one-
to-one—that is, if we choose any point in the domain found in
part (b), for example, f(z1), it has exactly one image in the domain
given in the z plane (i.e., there is only one possible value for z1).
d. Suppose we change the rectangle in the z plane to y = 0, −1 ≤ x ≤ 1;
y = 2π, −1 ≤ x ≤ 1; x = ±1, 0 ≤ y ≤ 2π, and use the same mapping.
126 A MATLAB® Companion to Complex Variables

Is the mapping again one-to-one? Into what domain is the inte-

rior of the rectangle mapped?
e. Suppose we replace 2π by 3π in the above. Is the mapping again
one-to-one?
3. a. Consider the domain in the z plane whose boundary is the circle
|z  −  1  −  i| = 1. The domain is the interior of this circle. Using
MATLAB, produce a plot of the image of this boundary under
the transformation w  =  z2. Use at least 100 points to make this
circle. Place at least four points uniformly spaced on the circle,
designating them with capital letters of the alphabet, and put
corresponding lowercase letters at the images of these points in
the w plane.
b. Is there a one-to-one mapping between the domain described
above and the domain created by the given transformation?
Explain.
c. If we change the circle to |z − 1 − i| = 2, is the mapping one-to-
one? Explain.
4. Consider the bounded domain in the z plane whose boundaries are
the lines y = 0, .1 ≤ x ≤ 1; x = 0, .1 ≤ y ≤ 1; |z| = .1, 0 ≤ arg z ≤ π/2; |z| = 1,
0 ≤ arg z ≤ π/2.
This domain is mapped with the transformation w = zi, where
the principal value is used.
a. Using MATLAB, plot the boundaries of the domain in the z plane
and the boundaries of the image in the w plane. In the z plane,
place letters A, B, C… at these points: z = .1, .5, 1, eiπ/4, i, .5i, .1i,
.1 eiπ/4, and put corresponding italicized letters at the images of
these points in the w plane.
b. Prove that the mapping between the two domains is one-to-one.
5. A domain in the z plane consists of all the points lying between
two squares with a common center. The centers of both squares
lie at  z  =  1  + i. The outer square has sides parallel to the x−y
axes and has side length 2. Two sides are on the axes. The inner
square has sides parallel to the same axes and has a side length of
one half.
a. Using the transformation w = z3, map these two squares into the
w plane with MATLAB. Place letters A, B … H on each of the
eight corners and place bold or italic versions of these same let-
ters at their images in the w plane. We suggest using a small font
size (e.g., 6) in the w plane.
b. Study the mapping in part (a). Where is it not conformal, and why?
c. Is the mapping of the domain from the z plane to the w plane
one-to-one? Explain.
Harmonic Functions, Conformal Mapping, and Some Applications 127

4.2  The Bilinear Transformation

One of the most commonly used conformal mappings in engineering is the
bilinear transformation given by

az + b
w( z) = (4.1)
cz + d

(See W section 8.4 and S section 8.10.)

a b
Here a, b, c, d are constants. Notice that if = , this would reduce to w
c d
being  a constant and so the mapping is not conformal. The same would
be true if we solve the above for z(w), the inverse transformation. Then
z  becomes a constant. To avoid these possibilities, we require that ad ≠ bc.
You should easily prove that this inequality guarantees that w′(z) ≠ 0 for all
finite z (i.e., the mapping is everywhere conformal).
Among the simpler special cases of bilinear transformations are such
z−i
examples as w = 2 iz, w = 1/ z, w = . The inverse of a bilinear transfor-
z+i
mation is a bilinear transformation (i.e., if you were to solve Equation 4.1 for
z, you would obtain z in terms of a bilinear expression in w). Although this is
a high school level exercise, we perform it here so as to demonstrate the solve
function in MATLAB.
The program is

z= solve('w=(a*z+b)/(c*z+d)','z')
'a pretty answer for z='
pretty (z)

The output is

z = −(b − d*w)/(a − c*w)

a pretty answer for z=

b − dw
− (4.2)
a − cw
which is obviously also a bilinear transformation. Note that if you run the
above program, you will receive a warning message (not shown here) that
the result is valid only if ad ≠ bc.
You may avoid this warning by using, if you wish,
solve('w=(a*z+b)/(c*z+d)','z','IgnoreAnalyticConstraints',true)

to the right of the equals sign in the first line of code in lieu of

solve('w=(a*z+b)/(c*z+d)','z').
128 A MATLAB® Companion to Complex Variables

The most useful property of the bilinear transformation is this: If an

i­nfinite straight line in the z plane is mapped by means of the transforma-
tion (Equation 4.1), its image in the w plane is found to be either an infinite
straight line or a circle. If a circle in the z plane is similarly mapped, the result
in the w plane will be either a circle or an infinite straight line. This is proved
in all standard texts (see, e.g., W section 8.4). By an infinite straight line, we
mean one along which we can travel in two opposite directions and arrive
at infinity. Thus, x + y = 1 is infinite, but x + y = 1, y > 0 is not infinite but
could be called semi-infinite. We will use the term “straight line” to mean an
infinite straight line.
Since the inverse of a bilinear transformation is still a bilinear transforma-
tion, it means that a circle or straight line in the w plane, obtained by map-
ping some curve in the z plane by means of a bilinear transformation, must
have arisen from a straight line or circle in the z plane.
If the coefficients in a bilinear transformation a, b, c, d are real in
Equation  4.1, then it has the property that w( z ) = w( z). This means that
points that are conjugates of each other in the z plane will have image
points in the w plane that are conjugates of one another under this map-
ping. Thus, a circle whose center is on the real axis in the complex z plane
will have as its image in the w plane either another circle with center on the
real axis or an infinite line parallel to the imaginary axis. An infinite line
parallel to the imaginary axis in the z plane will be mapped into either a
circle in the w plane, with center on the real axis, or an infinite line parallel
to the imaginary axis.

Example 4.1
z−1
Let us verify some mappings using the transformation w = and
z+i
MATLAB. We will map, by means of this transformation, these two
circles: |z| = 1 and |z − 1 − i| = 2.

Solution:
Before beginning, notice that as we approach the point z = −i, the func-
tion w(z) → ∞. Now z = −i is on the circle |z| = 1. Thus, at least one
point of the circle has an image at infinity, which means that the image
of the entire circle must be an infinite straight line. Using MATLAB to
plot points from the z to the w plane, we must be careful to avoid z = −i.
Otherwise, we will be dividing by zero and will receive an error mes-
sage. Observe, too, that the given circle passes through z = 1, where w = 0.
Thus, the given line must pass through the origin in the w plane.
The circle |z − 1 − i| = 2 does not pass through z = −i. Thus, w is never
infinite on this locus, and the image of the circle must be another circle.
The code is
th=2*pi*linspace (0,1,1000);
z=exp(i*th);
figure (1);
Harmonic Functions, Conformal Mapping, and Some Applications 129

plot(real(z),imag(z));axis equal;grid on
xlabel('x');ylabel('y');title 'Unit Circle z plane'
w=(z−1)./(z+i);
figure (2)
plot(real(w),imag(w));grid on;axis equal;
xlabel('u');ylabel('v');
z=1+i+2*z;
figure(3)
plot(real(z),imag(z));axis equal;grid on
xlabel('x');ylabel('y');title 'Circle |z−1−i|=2'
w=(z−1)./(z+i);
figure(4);
plot(real(w),imag(w));grid on;axis equal;
xlabel('u');ylabel('v');

The resulting output is shown in Figures 4.12 through 4.15.

There are some things to note here. We have used 1,000 data points on
the two circles in the z plane in order to map these circles into the w plane.
This may seem excessive, but in the case of the transformation applied to the
z−1
circle |z − 1 − i| = 2, it is not. This is because the mapping function w =
z+i
changes rapidly as we approach z = –i. If we expect to obtain a smooth curve
from the mapping, we must obtain enough image points in the w plane to
yield a circle.

0.8

0.6

0.4

0.2

0
y

–0.2

–0.4

–0.6

–0.8

–1 –0.5 0 0.5 1
x

FIGURE 4.12
Unit circle, z plane.
130 A MATLAB® Companion to Complex Variables

600

500

400

300

200
v

100

–100

–200
–300 –200 –100 0 100 200 300 400 500 600 700
u

FIGURE 4.13
( z − 1)
Image in the w plane of the unit circle for w = .
( z + i)

2.5

1.5

1
y

0.5

–0.5

–1.5 –1 –0.5 0 0.5 1 1.5 2 2.5 3 3.5

x

FIGURE 4.14
Circle |z – 1 – i| = 2.
Harmonic Functions, Conformal Mapping, and Some Applications 131

3.5

2.5

1.5

1
v

0.5

–0.5

–1

–1.5

–5 –4 –3 –2 –1 0 1
u

FIGURE 4.15
( z − 1)
Image in the w plane of the circle |z – 1 – i| = 2 for w = .
( z + i)

z−1
We noted that w = goes to infinity at –i and that the unit circle passes
z+i
through this point. On the unit circle, we have z = exp(iθ) 0 ≤ θ ≤ 2π. By not
using θ = 3π/2, we can avoid evaluating w at this point. This is satisfied by
the line of code th=2*pi*linspace(0,1,1000);. In fact, we can replace 1,000
with any even integer and avoid the troublesome point, as the reader should
verify.

4.2.1  The Cross-Ratio

Let the points z1, z2, z3, z4 lie in the complex plane, and let their images, cre-
ated by means of a bilinear transformation, be w1, w2, w3, w4. Then the “cross-
ratio” of these points is preserved (see W section  8.4 and S section  8.10),
which means

( z1 − z2 ) ( z3 − z4 ) ( w1 − w2 ) ( w3 − w4 )
=
( z1 − z4 ) ( z3 − z2 ) ( w1 − w4 ) ( w3 − w2 )

Let us take one of these points, say z4, to be the arbitrary variable z having
image w. Substituting these values in the preceding, we have
132 A MATLAB® Companion to Complex Variables

( z1 − z2 ) ( z3 − z ) ( w1 − w2 ) ( w3 − w )
= (4.3)
( z1 − z ) ( z3 − z2 ) ( w1 − w ) ( w3 − w2 )
The preceding can be solved for either z(w) or w(z) as we wish.
Notice that the values z1, z2, z3 must differ from each other, and the same
goes for their corresponding images w1, w2, w3. (Recall that the bilinear
transformation is one-to-one.) If one of the quantities on the left, say z3, is
infinite, we replace the quotient of the two terms (in this case (z3 − z)/(z3 − z2))
involving this quantity by the number 1. The same technique is used on the
right if there is an infinite quantity there. We can write a MATLAB pro-
gram in which one supplies z1, z2, z3 as well as the desired images w1, w2, w3.
The output from the program can be either w(z) or z(w), or both. The matter
of what  to do when a quantity is infinite is taken up in problem 2 in the
exercises.

Example 4.2
Write a MATLAB program in which you enter the finite values z1, z2, z3
and w1, w2, w3 and receive back the bilinear transformation w(z).
Solution:
The following code will work:
w=solve('(w1−w2)*(w3−w)/((w1−w)*(w3−w2))=(z1−z2)*(z3−z)/
((z1−z)*(z3−z2))','w');
zvalues= input(' z1 z2 z3 as row vector elements')
z1=zvalues(1);
z2=zvalues(2);
z3=zvalues(3);
wvalues= input(' w1 w2 w3 as row vector elements')
w1=wvalues(1);
w2=wvalues(2);
w3=wvalues(3);
% the following line substitutes the numerical values into
the formula %for w, in place of the symbols
w=s
ubs(w,{'w1','w2','w3','z1','z2','z3'},{w1,w2,w3,z1,
z2,z3});
simplify(w)
factor(w)

Some things to note include that the line of code using solve will
yield the symbolic solution for w. This employs the symbols for the six
values z1, z2…w1, w2…. The line containing the function subs places the
numerical values in these symbols. The function factor will help us rec-
ognize that we have a bilinear transformation. Notice, however, that
simplify does not always live up to its name. If you attempt simplify
((z − i)/(z2 + 1)), MATLAB will not factor the denominator, divide the com-
mon factor, and give you back 1/(z + i). Thus, the fact that such a func-
tion is a bilinear transformation is not evident. You must be prepared
for such outcomes. Sometimes factor can be helpful in these situations.
Notice also that the function solve results in the output of the program
Harmonic Functions, Conformal Mapping, and Some Applications 133

warning you that we cannot find w(z) if the points to be mapped in either
plane are not numerically distinct. The warning can be eliminated by
adding 'IgnoreAnalyticConstraints', true in the solve expression
as discussed above.

Let us try out the preceding program. Taking z1 = 0, z2 = 2, z3 = 1 + i with

1 2i
corresponding images w1 = −1, w2 = 1/ 3, w3 = + , we have finally for w
5 5
the output

ans = (z − 1)/(z + 1)
ans = [z − 1, 1/(z + 1)]

These two answers are consistent; the second simply factors the first.
You can check this formula by installing the previous values of zn and veri-
fying that the desired values of wn are obtained.
Because this bilinear transformation contains only real coefficients, we
know that it will map a circle, in the z plane, with center on the x axis into
either a circle with center on the u axis or into an infinite line perpendicular
to that axis. The circle with center at z = 1 having radius 1 passes through the
given points 0, 2, 1 + i. The image of this circle in the w plane passes through
1 2i
the three points w1 = −1, w2 = 1/3, w3 = + and has center at the average of
5 5
–1 and 1/3 (i.e., at w = –1/3). Suppose we want to see how the transformation
thus obtained maps the circle |z − 2i| = 1. We add on these lines of code to
the preceding:
th=linspace (0,2*pi,100);
z=
2*i+exp(i*th);% creates the points on the circle, center at 2i
w=subs(w,{'z'},{z});% puts the values we need into w
figure (1)
plot(real(w),imag(w));axis equal

For the values of z1, z2 … w1, w2 … specified above, the output of this would
be a circle.
Because a nonconstant analytic function maps a domain into a domain, it
should be possible to find a bilinear transformation that maps the real axis
in the z plane into the unit circle in the w plane in such a way that the image
of the upper half of the z plane is the interior of that circle. Notice that we can
rewrite Equation 4.1 as

a  z + b/a 
w ( z) = (4.4)
c  z + d/c 

where we must assume that a and c ≠ 0. If c were zero in Equation 4.1, we could
not argue that z = ∞ has an image at a finite point in the w plane. The limit
of the right side in the above expression as z → ∞ is a/c. As we want this to
134 A MATLAB® Companion to Complex Variables

a
lie on the unit circle in the w plane, we can take = e it, where t is any real
c
number. Suppose z = x is real. We want the image for such a point to have
magnitude 1.
Then we obtain from Equation 4.4

 x + b/a   x + b/a 
w( z) = e it   =  =1
 x + d/c   x + d/c 

Now we have |x + b/a| = |x + d/c| that can only be satisfied if b/a = d/c or

b/a = ( d/c ). We must reject the former as unworkable (see  Equation  4.4), as
the transformation reduces to a constant. Thus, with the latter, and p = –b/a,
we have for the conformal mapping that transforms the real axis into the
circle |w| = 1,
 z− p
w ( z ) = e it  t is real
 z − p 

or
 z− p
w ( z) = m  m = 1 (4.5)
 z − p 

How do we know that the space above the axis is mapped into the interior of
the circle? If we take p as any point in the z plane satisfying Im(p) > 0, we see
from the above that it is mapped into the origin of the w plane—the center of
the circle. Because domains are mapped into domains by the above, we con-
clude that the transformation in Equation 4.5, with Im(p) > 0, maps the upper
half plane into the domain bounded by the unit circle and the ­boundary
y = 0 into the unit circle.

Example 4.3
Write a MATLAB program that will accomplish the following. It is to
produce a bilinear transformation that will transform the real axis in
the z plane into the unit circle of the w plane. The program will request
a particular point in the upper half plane that will be mapped into the
origin in the w plane. Finally, the program will request a particular point
on the real axis z that is to have a specified image on the unit circle in
the w plane.

Solution:
The following program will work:

% the line below solves for w(z)

clear
m=solve('w=m*(x-p)/(x-conj(p))','m');
Harmonic Functions, Conformal Mapping, and Some Applications 135

p=input('point in upper z plane mapped to w =0');

x=
input('point on real axis mapped to a place on unit circle');
w=input('image of above pt. on unit circle');
m=subs(m, {'w','x','p'},{w,x,p})
'above requires |m|=1'
syms w z;
w=m*(z-p)/(z-conj(p))

Here we try the program out. We will ask that z = i be mapped into
π
i
w = 0 and that z = 1 be mapped into w = e 4 , which, as required, is on the
unit circle.
Here are the inputs, outputs, and the result:

point in upper z plane mapped to w = 0  i

point on real axis mapped to a place on unit circle  1
image of above pt. on unit circle exp(i∗pi/4)
m = 2^(1/2)∗(−1/2 + i/2)
above requires |m|=1
The output is
w = (2^(1/2)∗(z − i)∗(−1/2 + i/2))/(z + i)

Exercises
1. a. In Example  4.1, using MATLAB, we mapped the unit ­ circle
|z|  =  1 into the w plane by means of the transformation
z−1
w= . We obtained what appears to be the infinite line v = u.
z+i
Prove mathematically that this line is indeed obtained.
Hint: In the given expression for w(z), multiply numerator and
denominator by the conjugate of the denominator. We can gener-
ate the given circle in the x–y plane with z = eiθ, where 0 ≤ θ ≤ 2π.
Substitute this expression for z in the equation just obtained, and
show that if w = u + iv, then u and v are such that v = u which is
the equation of a straight line through the origin in the w plane.
It should be obvious that the line passes through infinity when
z goes through –i.
b. Using MATLAB, plot in the w plane the image of the circle
|z − 1 − i| = 1 under the transformation w = 1/z. The original cir-
cle has center at z = 1 + i. Does its image have center at 1/(1 + i)?
Where is its center? Figure this out analytically and confirm it
from your plot.
136 A MATLAB® Companion to Complex Variables

This is something to think about: If a bilinear transformation

maps a circle into a circle, does it always map the center of the
first circle into the center of the second one?
2. a. Consider the cross-ratio in which the variable points z and w are
images of each other but the point z3 lies at infinity, the other
points being finite. In this limit, the cross-ratio Equation 4.3
becomes

( z1 − z2 ) ( w1 − w2 ) ( w3 − w )
=

( z1 − z ) ( w1 − w ) ( w3 − w2 )

Write a MATLAB program in which you supply at a prompt

the five numerical values w1, w2, w3, z1, z2. The program should
then solve for w as a function of z, using the solve function.
Use the function simplify to simplify your answer.
b. Use the program you just obtained to find one transforma-
tion that maps the infinite line x + y = 1 onto the unit circle.
Explain why there is more than one transformation that will
do this.
3. a. A fixed point of a mapping is one in which some value of z,
say zj, has the same numerical value as its image wj. Thus, a
az + b
fixed point for a bilinear transformation must satisfy z = .
cz + d
Write a MATLAB program in which you supply at a prompt
the numerical values of a, b c, d, and you will get back the fixed
points. You should use the function solve. Note that in general
there are at most two fixed points. When will there be no fixed
points? When will there be one fixed point? When will there
be an infinity of fixed points? Use your program to determine
( z − i)
the two fixed points of w ( z ) = . Note that if your answer
z+1
is two values of z, include the line of code eval(z) to see their
numerical values.
b. Check your numerical answer to part (a) by showing that for the
two values found, we have w(z) = z.
4. A bilinear mapping of a bilinear mapping is still bilinear. This is
­easily proved. Let

az + b ez + f
f ( z) = and f1 ( z ) =
cz + d gz + h

which are each bilinear. Then
Harmonic Functions, Conformal Mapping, and Some Applications 137

az + b
a
f ( f1 ( z ) ) = cz + d
+b
=
( )
a 2 + bc z + bd + ab
+ d ( ac + dc ) z + bc + d
az + b 2
c
cz + d

which is also a bilinear transformation.

Obviously, f(f(z)) and f(f(f(z))), and so on, are bilinear transfor-
mations.
a. Write a MATLAB program in which you enter at a prompt the
four numerical values a, b, c, d. Now let f0 = f(z) and f1 = f(f(z)),
f2 = f(f(f(z))), and so on. The program should produce as output
f k(z) for k = 1, 2, …, 10. Test your program with these functions:

z+1 z+i
f ( z) = and also f ( z) =
z−1 z − 2i

b. Explain how your results show for all k ≥ 1 that with

z+i
f ( z) = , we have f k(z) can be only three possible functions,
z−i
even though  we have investigated only the first 10 functions.
What are these functions?
c. Use your program to find how many possible functions are
z−1
generated if you begin with f ( z) =
­ ? What are these
z+1
functions?
5. a. Consider Equation 4.5. How should this equation be modi-
fied, if at all, to map the real z axis into the unit circle of the w
plane with the lower half of the z plane being mapped onto the
interior?
b. Consider Equation 4.5. How should this equation be modified to
do the following: the imaginary axis in the z plane gets mapped
into the unit circle in the w plane while the right half of the
z  plane is mapped onto the interior of that circle? Write a pro-
gram comparable to that in Example 4.3 that will do the follow-
ing: the program requests a point in the right half of the z plane
that is to be mapped to the origin of the w plane. The program
requests a point on the imaginary axis in the z plane that is to be
mapped into a point, also specified by the user, on the unit circle
in the w plane.
c. Use the above program to find the bilinear transformation that
maps the imaginary axis into the unit circle, and with the point
138 A MATLAB® Companion to Complex Variables

z = 1 + i being mapped to the origin. The point z = 2i is to be

mapped into w = –1.
d. Using the above transformation, map the line segments y = x,
0 ≤ x ≤ 10 and y = −x, 0 ≤ x ≤ 10 into the w plane. Use a heavier
line for the first mapping so that you can distinguish the two
image curves. At what angle should these images intersect,
and why?

4.3 Laplace’s Equation, Harmonic Functions:

Voltage, Temperature, and Fluid Flow
Let f(z) = u(x, y) + iv(x, y) be a function that is analytic in some domain D.
Then it is easily proved (see W section 2.5 or S section 3.4) that both real and
imaginary parts satisfy Laplace’s equation:

∂2 Ω ∂2 Ω
+ = 0 (4.6)
∂x2 ∂y 2

in this domain—that is, we may put either u(x, y) or v(x, y) in place of Ω

in the above and find that the equation is true. For example, f(z)  = z2  =
x2 − y2 + i2xy is an entire function, and it is readily seen that both u(x, y) =
x2 − y2 and v  =  2xy will satisfy Equation 4.6 in any domain. Functions
­satisfying Laplace’s equation in a domain are said to be harmonic func-
tions.  A ­function not satisfying Equation 4.6 throughout some domain
cannot be either the  real or imaginary part of an analytic function in a
domain.
If a function Ω(x, y) does satisfy Equation 4.6, then there exists an analytic
function of the form f1(z) = Ω(x, y) + iv(x, y). Given Ω(x, y), there are pro-
cedures for finding v(x, y) (see W section 2.5 and S problem 3.7). We show
how MATLAB can help do this in the next section where we learn how
knowledge of v(x, y) as well as f 1(z) might be useful. We say that v(x,  y) is
the h­ armonic ­conjugate of Ω(x, y). Note that there also exists another analytic
­function of the form f2(z) = u(x, y) + iΩ(x, y).
That both the real and imaginary parts of an analytic function are har-
monic has enormous utility for the solution of two-dimensional physical prob-
lems in electrostatics, heat transfer, and fluid mechanics. In these problems,
the sources of electric fields, heat, and fluid flow are infinitely long structures
in a direction perpendicular to the x−y plane. The properties of these sources
do not vary along the dimension perpendicular to the plane, and no physical
property is assumed to change in time.
Harmonic Functions, Conformal Mapping, and Some Applications 139

The space surrounding these sources is assumed to be filled with a mate-

rial that will do at least one of the following: support an electric field, permit
the conduction of heat, and sustain a flowing fluid. The material is of infinite
extent in any direction perpendicular to the complex plane. The complex
plane is assumed to show a cross section of the material and is perpendicu-
lar to any sources that are present. The behavior of all physical quantities is
assumed to be identical in all planes parallel to the complex plane.
For an example of two-dimensional heat conduction, imagine that all of
space is filled with a material that conducts heat. An infinitely long, hot
tube is maintained at a fixed elevated temperature in this material. The
tube passes perpendicularly through the complex plane. Then the tem-
perature in this material depends only on the variables x and y, and we
would expect the temperature in the material to fall as we move farther
from the tube.
We can complicate the heat flow configuration described above, adding
other tubes, each with its own temperature, or ones in which the cross sec-
tions of the tubes are not necessarily circular.
Further complications, in various physical situations, might involve the use
of a material surrounding the sources that is bounded except in directions
normal to the x−y plane. We might require, depending on the physics, that
temperature, fluid flow, or electric fields satisfy certain prescribed ­conditions
on the boundaries of the material.
All of these three kinds of physical situations are analyzed by means
of harmonic functions to which we give the symbol ϕ(x, y). In the case of
electrostatics, this harmonic function is called the electrostatic potential, or
sometimes the voltage, in fluid mechanics it is known as the velocity poten-
tial, while for heat conduction it is called simply the temperature. The use
of the potential in fluid problems comes with three caveats: like water, the
fluid cannot be compressed nor are there present in the fluid vortices (or
­whirlpools)—the fluid is irrotational. This term is explained in all standard
books on fluid mechanics. Also, there is assumed to be no internal friction
among the moving fluid particles.
From ϕ(x, y), we can, in electrostatics, obtain the components of the electric
field Ex(x, y) and Ey(x, y). These components give the x and y directed com-
ponents of force on a unit (1 Coulomb) test charge placed at (x, y). In fluid
mechanics, we can obtain from ϕ(x, y) the components of the fluid velocity
vector Vx(x, y) and Vy(x, y) describing the motion of a particle of fluid at (x, y).
In heat conduction, we can use ϕ(x, y) to get the components of the heat flux
density vector Qx(x, y) and Qy(x, y) at a point. The vector gives the direction
of heat flow and the rate of flow at (x, y).
The equations involved are as follows:
For electrostatics,

− ∂φ − ∂φ
Ex ( x , y ) = Ey ( x , y ) = (4.7)
∂x ∂y
140 A MATLAB® Companion to Complex Variables

For fluid flow,

∂φ ∂φ
Vx ( x , y ) = Vy ( x , y ) = (4.8)
∂x ∂y
For heat conduction,

− k ∂φ − k ∂φ
Qx ( x , y ) = Qy ( x , y ) = (4.9)
∂x ∂y
Here k is a constant peculiar to the heat-conducting medium and is called
its thermal conductivity. If you are familiar with the concept of the gradient or
the del operator, ∇, you see that

E = −∇ϕ = −grad ϕ (4.10)

V = ∇ϕ = grad ϕ (4.11)

Q = −k∇ϕ = −k grad ϕ (4.12)

The vector E, for example, which has components Ex(x, y) and Ey(x,  y),
is “minus the gradient of ϕ”; it is the vector whose components are
− ∂φ − ∂φ
and . Note that some authors would put minus signs after the
∂x ∂y
equals signs in Equations 4.8 and 4.11. There is no unanimity here.
MATLAB is useful to us here in several ways. First, given a function ϕ(x, y),
how can we tell if it is harmonic? For a simple function, say ϕ(x, y) = x2 − y2,
it should be evident that Equation 4.6 is satisfied with ϕ = Ω, and by contrast,
ϕ(x, y) = x2 + y2 does not satisfy the equation and is not harmonic. Verify
this. Thus, the first of these can serve as a voltage, velocity potential, or tem-
perature function while the second cannot. Suppose we are given a more
( )
complicated function φ ( x , y ) = e e cos y cos e x sin y . It is not obvious if this is
x

harmonic, although the question is readily resolved with the following code:

syms x y
g=exp(exp(x)*cos(y))*cos(exp(x)*sin(y));
A=diff(g,x,2)+diff(g,y,2)

The output is:

A = 0

Notice that the third line of code:

A=diff(g,x,2)+diff(g,y,2)

sums together the second x and second y derivatives of the function

( )
φ ( x , y ) = e e cos y cos e x sin y . The result is found to be zero, which proves that
x
Harmonic Functions, Conformal Mapping, and Some Applications 141

this function is harmonic in any domain of the complex plane. This function z
was constructed by a choice of ϕ(x, y) as the real part of the function e( e ),
which must of course be harmonic. The reader should compute the real and
imaginary parts of this function.
Suppose we proceed as above and try to determine whether the function
−2 xy
φ ( x, y ) =
(x )
2 2
+ y2

is harmonic. The code

syms x y
gg=(-2*x*y)/(x^2+y^2)^2;
B=diff(gg,x,2)+diff(gg,y,2)

has as its output

B = (48∗x∗y)/(x^2 + y^2)^3 − (48∗x^3∗y)/(x^2 + y^2)^4 − (48∗x∗y^3)/(x^2 + y^2)^4

Since, unlike in the previous example, we do not simply get zero, we might
conclude that this is not a harmonic function. Wrong. The additional line of
code, which simplifies the above, B=simplify(B) has as its output B =  0
which confirms that we have a harmonic function, one that the reader should
1
verify is the imaginary part of f ( z) = . We had to simplify our result
( x + iy )2
to see this. We should, to be safe, try to simplify our result in such calcula-
tions, to determine if we have a harmonic function. However, before con-
−2 xy
cluding that the function φ ( x , y ) = 2 is harmonic everywhere, take
(
x + y2
2
)
note of the fact that it is undefined at z = 0 and therefore is not differentiable
at the origin. MATLAB ignores this. Thus, the correct conclusion is that this
function is harmonic in any domain not containing the origin.
In the preceding, we began with a real function ϕ(x, y) and used MATLAB
to determine whether the function is harmonic. Alternatively, we might begin
with an analytic function f(z) and extract its real or imaginary part.  Each
must be harmonic in any domain where f(z) is analytic. Often the problem of
extracting u(x, y) + iv(x, y) from f(z) is simple. For example,
x y
f ( z) = 1/ z = 1/ ( x + iy ) = 2 2
−i 2
x +y x + y2
whose real and imaginary parts

x −y
and
x + y2
2
x + y2
2

are harmonic in any domain not containing the origin.
142 A MATLAB® Companion to Complex Variables

But suppose we have a more complicated function like f ( z ) = e cos( z ).

2

This is an entire function whose real and imaginary parts are harmonic
everywhere. The following code shows how to extract the real and imagi-
nary parts u(x, y) and v(x, y) and confirms in addition that these parts are
harmonic:

%getting real part and imaginary part

clear
syms x y real
z=x+i*y;
w=exp(cos(z^2))
u=real(w)
v=imag(w)
%check on harmonic behaviour
u_check=diff(u,x,2)+diff(u,y,2);
v_check=diff(v,x,2)+diff(v,y,2);
u_check=simplify(u_check)
v_check=simplify(v_check)

The output is

w = exp(cos((x + y∗i)^2))
u = exp(cosh(2∗x∗y)∗cos(x^2 − y^2))​∗cos(sinh(2∗x∗y)∗sin(x^2 − y^2))
v = −exp(cosh(2∗x∗y)∗cos(x^2 − y^2))​∗sin(sinh(2∗x∗y)∗sin(x^2 − y^2))
u_check = 0
v_check = 0

Notice an important line of code:

syms x y real

This tells MATLAB that x and y are real. Without this line, we could not use
the real(w) and imag(w) operations to extract the real and imaginary parts of
w because MATLAB would assume that x and y are complex.
We have chosen to check that the real part of w, which we call
wr(x,  y), is harmonic through the inclusion of the line of code
u_check​ =diff(u,x,2)+diff(u,y,2).
The output of zero, which is the sum of the second x and y derivatives
of wr(x, y), confirms that we have a function that is harmonic everywhere.
A similar check is made on the imaginary part of w.
Incidentally, the functions u(x, y) and v(x, y) obtained from the preceding
code might be more comfortably written as

(cosh(2 xy )cos( x − y )) cos

( sinh ( 2xy ) sin ( x ))
2 2
u ( x, y ) = e 2
− y2
Harmonic Functions, Conformal Mapping, and Some Applications 143

and

(cosh(2 xy ) cos( x − y )) sin

( sinh ( 2xy ) sin ( x ))
2 2
v ( x, y ) = − e 2
− y2

Sometimes we might find that having supplied MATLAB with a function

f(z), we discover that the function is sufficiently complicated that MATLAB
cannot determine explicit expressions for the real and imaginary parts. The
following example will suffice:

clear
syms x y real
syms z
z=x+i*y
A0=sin(z)
A0r=real(A0)
A1=sin(1/z)
A1r=real(A1)
A2=sin(z*sin(1/z))
A2r=real(A2)
A3=sin(z^2)
A3r=real(A3)

Whose output is
z = x + y∗i
A0 = sin(x + y∗i)
A0r = cosh(y)∗sin(x)
A1 = sin(1/(x + y∗i))
A1r = cosh(y/(x^2 + y^2))∗sin(x/(x^2 + y^2))
A2 = sin(sin(1/(x + y∗i))∗(x + y∗i))
A2r = real(sin(sin(1/(x + y∗i))∗(x + y∗i)))
A3 = sin((x + y∗i)^2)
A3r = cosh(2∗x∗y)∗sin(x^2 − y^2)

Refer to A2r above. Note that MATLAB was unable to present the real part
of sin(z sin(1/z)) as an explicit real function of x and y. In the case of the other
functions sin(z), sin(1/z), sin(z2), we were successful.
To deal with the case of sin(z sin(1/z)), we can break the operation into
steps for MATLAB, first obtaining the real and imaginary parts of sin(1/z)
and entering that result into the argument of sin(z sin(1/z)). It will be found
that the following will produce an explicit function for the real part of
sin(z sin(1/z)):
clear
syms x y real
144 A MATLAB® Companion to Complex Variables

syms z
z=x+i*y
A1=sin(1/z)
A1r=real(A1);
A1i=imag(A1);
A2=sin(z*(A1r+i*A1i))
A2r=real(A2)

The output is

A2r = c osh(x∗cos(x/(x^2 + y^2))∗sinh(y/(x^2 + y^2)) −  y∗cosh​(y/(x^2 +

y^2))∗sin(x/(x^2 +  y^2)))∗sin(x∗cosh(y/(x^2 + y^2))∗sin(x/(x^2 + y^2))
+ y∗cos(x/(x^2 + y^2))∗sinh​(y/(x^2 + y^2)))

Using MATLAB, you can readily show that the preceding is a harmonic
function in any domain not containing the origin.
In Section 2.2 we saw how the properties of a real function of the two real
variables, like ϕ(x, y), can be studied through a set of contour plots. In electric-
ity, if the harmonic function ϕ(x, y) describes the potential in space, then such
contour plots are the loci on which the voltage is constant. These are the traces
in the x−y plane of the surfaces of constant voltage, often referred to as equipo-
tential surfaces. Similarly, if a temperature is given by ϕ(x, y), then the contour
plots of ϕ(x, y) are isotherms, or surfaces of constant temperature. Finally, in
fluid mechanics, we can also plot and speak of equipotential surfaces.
To review this subject, consider for example an infinitely long, uniformly
charged, filament passing perpendicularly through the complex plane at the
point (x0, y0). The electric potential (voltage) created in the surrounding mate-
rial is given by

1
φ( x , y ) = C Log (4.13)
( x − x0 ) + ( y − y 0 )
2 2

Here C is a real constant that depends on the density of electric charge

along the filament as well as the electrical properties of the surrounding,
nonconducting, medium.* If z = x + iy and z0 = x0 + iy0, we have also that

ϕ(x, y) = C Log(1/|z − z0|) = −C Log(|z − z0|) = −C Re[Log(z − z0)] (4.14)

One should use MATLAB (as was done earlier) to quickly verify that ϕ(x, y)
is a harmonic function in any domain not containing (x0, y0).

* In this case, C = ρL/(2πε0), where ρL is the charge per unit length on the filament and ε0 is
a constant found in handbooks called “the permittivity of free space.”
Harmonic Functions, Conformal Mapping, and Some Applications 145

Electrical engineers like to model electric power lines as two charged

­ laments separated by some distance. Each filament carries an equal and
fi
opposite charge. There is an electric field surrounding the filaments.

Example 4.4
An electrical transmission line is being approximated as two oppo-
sitely charged filaments. The positive filament passes through z = 1
while the negative one passes through z = –1. Each filament is perpen-
dicular to the complex plane. The potentials created by the filaments
can be obtained from Equations 4.13 and 4.14. We take z 0 = 1 or z 0 = −1
as needed. Thus,

φ1 ( x , y ) = − Log ( x − 1)2 + y 2 = − Log|z − 1|= − Re ( Log ( z − 1))

and

φ2 ( x , y ) = Log ( x + 1)2 + ( y ) = Log|z + 1|= Re ( Log ( z + 1))

2

These arise from the charges passing through z = 1 and z = –1,

r­espectively. The constant C has been set equal to 1. The combined
potential is ϕ(x, y) = Re(−Log(z  −  1) + Log(z  +  1)). We can plot the
­equipotentials for this function using the familiar MATLAB function
contour as follows:

x=linspace(−4,4,100);
y=linspace(−4,4,100);
n=linspace(−2,2,5);
[x,y]=meshgrid(x,y);
z=x+i*y;
w=−log(z−1)+log(z+1);
wr=real(w);
[c,h]= contour(x,y,wr,n);colorbar;
xlabel('x');ylabel('y');
title('Contour Lines for Log|z+1|−Log|z−1|')
grid
clabel(c,h);axis([−2.5 2.5 −2.5 2.5])

This generates Figure 4.16.

Figure 4.16 is in black and white (gray scale). If you run the above code,
you will obtain a more attractive color figure and a color bar.
As mentioned in our presentation of Equations 4.7 through 4.9, the gra-
dient of the potential ϕ(x, y) is useful. In the present case, it can give us
the components of the electric field, but it can also yield temperature and
146 A MATLAB® Companion to Complex Variables

2.5

0
2

1.5

1
–1 1
0.5
–2

2
0
y

–0.5
–1 0 1 1
–1

–1.5

–2

–2.5
–2.5 –2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5
x

FIGURE 4.16
Contour lines for log|z + 1| – log|z – 1|.

the heat flux density in other situations. The electric field is the vector
­having components given in Equations 4.7 and 4.10.
MATLAB provides a convenient graphical means for visualizing the
gradient vector. This involves using the MATLAB functions gradient
and quiver, and the reader should read the documentation for these in
the help folder. Let the function ϕ(x, y) be approximated by numerical
values at the intersections of a rectangular grid, perhaps one gener-
ated by meshgrid. The MATLAB function gradient uses the differences
between values of ϕ(x, y) at adjacent points to compute numerical values
of the components of the gradient at the same grid points. For example,
at some point with coordinates xj, yk,

( ) (
∂φ φ x j+1 , y k − φ x j , y k
≈
)
∂x ∆x

which approximates the x directed component of the gradient of ϕ.

Unless told otherwise, MATLAB takes the spatial distance between the
values used to compute such an expression as unity so that ∆x = 1, which
∂φ
means is approximated as ϕ(xj+1, yk) − ϕ(xj, yk). The reader should keep
∂x
in mind that we are obtaining a numerical approximation, perhaps a
crude one, by using this function.
Harmonic Functions, Conformal Mapping, and Some Applications 147

More often, we simply want a graphical picture of the gradient of

ϕ(x, y), not numerical values. This is accomplished by our using, in addi-
tion to the above, the MATLAB function quiver. The entries giving the
components of the gradient at each intersection of the grid in the x−y
plane are used by quiver to generate pictorial arrows signifying the gra-
dient. These arrows are placed on a plot at the intersections of the grid.
The length of the arrow is proportional to the strength of the gradient
vector, while the arrow’s direction is the direction of the gradient vector.
Following is a simple example.

Example 4.5
Let ϕ(x, y) = x2 − y2 = Re(z2), a harmonic function. The following code
illustrates the use of gradient and quiver to show us a picture of the gra-
dient of the function. You might wish to study quiver in the help folder.

clf
v=linspace(−3, 3, 100);
[x,y]=meshgrid(v);
f=(x.^2−y.^2);
figure (1)
n=linspace(−2,2,5);
[c,h]=contour(x,y,f,n);axis equal
clabel(c,h);colorbar
hold on
%following uses coarse spacing for gradient
%which is desirable
v=linspace(−3, 3, 10);
[x,y]=meshgrid(v);
f=x.^2−y.^2;
[gx,gy]=gradient(f);
quiver(x,y,gx,gy);
title('contour lines and gradient for x^2−y^2')

The output is as shown in Figure 4.17.

There are several matters to notice here. The gradient vector at each
location is normal to the contour of constant ϕ passing nearest to that
location. The vector points toward the nearest other contour on which
ϕ has a more positive numerical value than on the first. The length of
the vector is inversely proportional to the spacing of adjacent contours,
which is because the narrower the spacing, the more rapidly is ϕ chang-
ing as we move normal to the contours.
Another important matter is the code. We use
v=linspace(−4, 4, 100)
[x,y]=meshgrid(v);

to plot the contours of constant ϕ. However, to plot the gradient, we

employ
v=linspace(−4, 4, 10)
[x,y]=meshgrid(v);

The 100 was replaced by 10.

148 A MATLAB® Companion to Complex Variables

–2 1
–
1
2

0
2
–2

–1
1

2
1
0
0

0
1

–1
2

–1

–2
–2

2
0

1
–1 –2

0
–3
–3 –2 –1 0 1 2 3

FIGURE 4.17
Contour lines and gradient for x2 – y2.

Notice that we are using a coarser grid to plot the gradient (­compare 100
versus 10). Had we used as fine a grid to plot the gradient as was done
with the surfaces of constant ϕ, we would have obtained a plot from
quiver so densely filled with arrows as to be unintelligible. The reader
should verify this for him- or herself. The plot shown above is in black
and white, but the code will generate a more attractive color picture.

To find the numerical values of the gradient of ϕ, we are best served by using
the formula for grad  ϕ rather than the line of code  [gx,gy]=gradient(f),
which would yield a result based on an approximate evaluation of the deriva-
tives with respect to x and y. In the present problem, the gradient is a vector

having components
(
∂ x2 − y 2 ) = 2 x and ∂( x 2
− y2 )
= −2 y. These are readily
∂x ∂y
evaluated on any grid of points by means of MATLAB. Observe that both
components of the vector vanish at the origin, which explains the absence of a
vector in the above figure at this point.

Example 4.6
Using quiver, plot the negative of the gradient, for the equipotentials
shown in Figure 4.16. We are using the negative of the gradient because
the electric field is generally the vector with components −∇ϕ, where
ϕ(x, y) is the electrostatic voltage.
Harmonic Functions, Conformal Mapping, and Some Applications 149

The following code will work. We simply add these lines to those used
in Example 4.4:
hold
v=linspace(−2.5, 2.5, 10);
[x,y]=meshgrid(v);
z=x+i*y;
w=−log(z−1)+log(z+1);
wr=real(w);
[gx,gy]=gradient(−wr);
quiver(x,y,gx,gy,.75);

This generates the output shown in Figure 4.18.

Notice the line of code:
quiver(x,y,gx,gy,.75);

The .75 scales down the length of the arrows to three quarters of their
normal length. Without this number, the arrows are longer and crowd
each other in the figure. Observe that the vectors representing the elec-
tric field emanate out from the positive charge on the right and head
toward the negative one on the left, as is to be expected. The arrows
strike the equipotentials at right angles.
Your plot will be in color and will have a color bar. We give here only
a gray-scale version.

2.5
0

1.5

1
–1

0.5
–2

0
y

–0.5
–1

–1

–1.5

–2

–2.5
–2 –1 0 1 2
x

FIGURE 4.18
Electric field for –log|z – 1| + log|z + 1|.
150 A MATLAB® Companion to Complex Variables

Notice also that we use [gx,gy]=gradient(−wr);. We do not apply

−gradient(wr), which would result in an error message—an idiosyn-
crasy of MATLAB, which I am told will be remedied.

Figure 4.18 has interpretations beyond electrostatics. Imagine the filament

on the right (passing through x = 1, y = 0) as a source of heat and the one
on the left (passing through x = –1, y = 0) as a sink for heat—a place where
heat is absorbed. The arrows indicate the direction of heat flow through a
heat-conducting medium surrounding the filaments, while the length of the
arrows is a measure of the intensity of that flow.
A similar interpretation is available in fluid mechanics: because of the
absence of the minus sign in Equation 4.11 (compare with its neighboring
equations), the filament on the left emits fluid uniformly in all directions and
fills some unbounded container of liquid, while the filament on the right is a
sink for this flow—it receives all the fluid sent by the filament on the left. The
arrows above are opposite to the direction of fluid flow, while their length is
proportional to the density of flow.

Exercises
1. Using MATLAB, determine whether the following functions are har-
monic. If a function is harmonic in general except at certain points,
state where these points are located.
a. cos(x2 + y2) cos(2xy)
b. cos(x2 − y2) cos(2xy)
c. Log(sinh2 x + sin2 y)
d. Log(cosh2 x + sin2 y)
1
e. e
− tan −1( y/x )
( ) 
sin  Log x 2 + y 2  , where the principal value of the
2 
arctan is used. In MATLAB you will have to use atan(y/x) and
not atan2(y,x), as MATLAB will not do symbolic differentiations
involving the latter function. As the two functions differ only by
a constant, their derivatives are identical.
2. With z = x + iy, express the following functions in the form
f(z) = u(x, y) + iv(x, y), finding u(x, y) and v(x, y) by means of MATLAB,
and use MATLAB to show that both u(x, y) and v(x, y) satisfy Laplace’s
equation in a domain. State where the equation is not satisfied.
a. cos(1/z)
b. cos(cos z)
c. 1/(z − i)5
Harmonic Functions, Conformal Mapping, and Some Applications 151

d. (z + z−1)5
e. sinh(sin z2)
3. An electrostatic potential in space is being created by two identi-
cal positively charged filaments. One filament passes through z = 1,
while the other passes through z = –1. Each filament is perpendicu-
lar to the complex plane. The potentials created by the filaments are
from Equations 4.13 and 4.14.

φ1 ( x , y ) = − Log ( x − 1)2 + y 2 = − Log|z − 1|

and

φ2 ( x , y ) = − Log ( x + 1)2 + ( y )
2
= − Log|z + 1|

where the constant C has been set equal to 1. The combined potential
is given by

φ ( x , y ) = − Log ( x − 1)2 + y 2 − Log ( x + 1)2 + y 2

a. Plot the equipotentials created by this pair.
Suggestion: Use the equipotentials –2, –1, 0, 1, 2. Use a grid with
1002 points satisfying |x| ≤ 3, |y| ≤ 3.
b. On the above figure, plot the negative of the gradient of the
potential by using grad and quiver.
Suggestion: Use a much coarser grid than that used in part (a), oth-
erwise you will have too many arrows to be visible. You may also
wish to shorten the length of the arrows. Make sure the arrows
(which are proportional to the electric field) are normal to the
nearest equipotentials and point from higher to lower potentials.
4. A strip of electric charge is of width 2. The density of charge on the
strip is uniform. The strip, of infinite length, is perpendicular to the
x−y plane and occupies the space y = 0, –1 < x < 1. It can be shown
that the electrostatic potential created by the strip is of the form
C Re[(z – 1) log(z – 1) – (z + 1) log(z + 1)], where C is a real constant,
and z = x + iy.
The value of the real constant C is directly proportional to the den-
sity of electric charge on the strip.
a. Taking C = 1, plot the equipotentials created by this charged strip.
Suggested values for the equipotentials are ϕ = –4, –3, –2, –1, 0.
Use a grid like that described in problem 3(a).
b. In the above figure, plot the negative of the gradient of the
­potential. Use the suggestions in problem 3(b).
152 A MATLAB® Companion to Complex Variables

4.4 Complex Potentials, Cauchy–Riemann Equations,

the Stream Function, and Streamlines
In the preceding section, we discussed a harmonic function ϕ(x, y) which
can be interpreted as giving the voltage, temperature, or velocity potential in
various two-dimensional configurations. Suppose this to be the real part of
the analytic function Φ(x, y) or Φ(z). Thus, we put

Φ(x, y) = ϕ(x, y) + iψ(x, y) (4.15)

Here ψ(x, y) is a real function that is known as the stream function because
of its connections to fluid mechanics, although the term is used in electricity
and heat transfer as well. Thus, ψ and ϕ are conjugate functions. The former
can be as useful as the latter, as we see later. The function Φ(x, y) is called the
complex electric potential (in electrostatics) or complex temperature (in the case of
heat flow) and the complex velocity potential (for fluid flow). Suppose we know
either ϕ or ψ. Can we use one to find the other?

4.4.1  Cauchy–Riemann Equations

Recall that all standard textbooks on complex variable theory treat his
s­ ubject. (See W section  2.5 and S section  3.5.) In familiar notation, we are
given the analytic function f(z) = u(x, y) + iv(x, y), where

∂u/∂x = ∂v/∂y (4.16)

∂u/∂y = −∂v/∂x (4.17)

As a simple example, to refresh the reader’s memory, we consider the func-

tion u(x.y) = y + x2 − y2 + x. Putting Ω(x, y) = u(x, y) in Equation 4.6, we readily
see that this function is harmonic. We now seek v(x, y) such that f(z) = u(x, y) +
iv(x, y) is analytic. We have from Equations 4.16 and 4.17 that

2x + 1 = ∂v/∂y (4.18)

−2y + 1 = −∂v/∂x (4.19)

We integrate Equation 4.18 with respect to the variable y and Equation 4.19
with respect to the variable x. We also swap the locations of the minus sign
in Equation 19 and get Equations 4.20 and 21:

v = 2xy + y + c1(x)+ C2 (4.20)

v = 2xy − x + c2(y)+ C4 (4.21)

Harmonic Functions, Conformal Mapping, and Some Applications 153

Here c1(x) is a function that can depend on x but not on y, while c2(y) is a func-
tion that can depend on y but not on x. C2 and C4 are true constants. One should
verify the above results by substituting back into Equations 4.18 and 4.19.
Now Equations 4.20 and 4.21 should both be valid expressions for v. The
­difference between them must be identically zero. Subtracting the second
from the first gives us the expression we call difference:

difference = y + x + c1(x) − c2(y) + C2 − C4

Setting difference equal to zero, we see that we must take c1(x) = −x, c2(y) = y
as well as C2 = C4.
The number C2 = C 4 is simply a true constant, independent of x and y,
and its value can only be found if the numerical value of v(x, y) is known at
some value of (x, y). Thus, using our derived values in either Equation 4.20
or Equation 4.21 gives

v = 2xy + y − x + C2

Here we accomplish the same thing with MATLAB:

clear all
syms y x v ux uy v1(y) v2(x) c1(x) c2(y) difference;
u=y+x^2−y^2+x;
ux=diff(u,x);
uy=diff(u,y);
%below solves du/dx=dv/dy
v1=dsolve(ux==diff(v1),y)+c1(x);
%below solves du/dy=−dv/dx
v2=dsolve(uy==−diff(v2),x)+c2(y);
v1=simplify(v1)
v2=simplify(v2)
difference=v1−v2;
difference=simplify(difference)

The reader might wish to study the MATLAB help for diff and dsolve.
Notice that the lines of code ux=diff(u,x) and uy=diff(u,y) ­compute
∂u ∂u
and . The next two lines of code, which employ dsolve, solve
∂x ∂y
∂u ∂ v ∂u − ∂ v
= and = . We add on “constants” c1 and c2, which can depend
∂x ∂y ∂y ∂x
on x and y, respectively. We derive two expressions for v(x, y), which we call
v1 and v2, respectively. We form the difference of these expressions. The
­output of the program is

v1 = C2 + y + c1(x) + 2*x*y
v2 = C4 − x + c2(y) + 2*x*y
difference = C2 − C4 + x + y + c1(x) − c2(y)
154 A MATLAB® Companion to Complex Variables

Studying the above line, we see that if the difference in our two ­expressions
for v is to be zero throughout a region of the x–y plane, then c1(x)=-x
and c2(y)=y. We may install these in the expression for either v1 or v2.
The constant C2=C4 remains undetermined. We have just found the value
of v = 2xy + y − x + C2 to within a constant. In this particular problem, C2
and C4 are the numerical values of v(0, 0). In the exercises, you will do some
MATLAB problems similar to the above where the computer program can
save us some labor.
A program similar to the above can be written if we are supplied with the
imaginary part of an analytic function and are seeking the real part. This is
left to the exercises.
Recall that in textbooks in complex variable theory, the Cauchy–Riemann
equations are derived by equating two equally valid expressions for the
derivative of a function of a complex variable. Thus, with f(z) = u(x, y) + iv(x, y),
∂u ∂ v ∂v ∂u ∂v
we have both f ′ ( z ) = +i and f ′( z) = − i . We can eliminate in
∂x ∂x ∂y ∂y ∂x
the first of these by means of a Cauchy–Riemann equation (Equation 4.17)
∂u ∂u
and obtain f ′ ( z ) = − i . If we take Φ(z) = ϕ(x, y) + iψ(x, y) as our analytic
∂x ∂y
∂φ ∂φ
function, we have Φ′ ( z ) = − i , and finally
∂x ∂y

∂φ ∂φ
Φ′ ( z ) = +i (4.22)
∂x ∂y

We have noted that if ϕ(x, y) is an electrostatic potential, then the x and

y components of the electric field in space are given, respectively, by (see
∂φ ∂φ
Equation 4.7) Ex ( x , y ) = − and Ey ( x , y ) = − . Thus, the negative of the
∂x ∂y
expression in Equation 4.22 is a complex quantity whose real and imaginary
parts are, respectively, the x and y components of the electric field.
We have

Ex ( x , y ) + iEy ( x , y ) = − Φ′ ( z ) (4.23)

which should show the utility of the complex potential in electrostatics.

In a similar way, we may use the complex velocity potential of fluid
mechanics to obtain a complex number whose real and imaginary parts
are the vector components of the fluid velocity:

Vx ( x , y ) + iVy ( x , y ) = Φ′( z) (4.24)

For this complex potential Φ(z), the real part ϕ(x, y) is the velocity potential
describing fluid flow.
Harmonic Functions, Conformal Mapping, and Some Applications 155

Finally, the complex temperature of heat flow has this property:

Qx ( x , y ) + iQy ( x , y ) = − k Φ′ ( z ) (4.25)

The real and imaginary parts on the left are the components of the heat
flux density vector, and k is the thermal conductivity of the material in use.
Of what use is the stream function ψ(x, y)—the imaginary part of the
­potential—in each of these three cases? Recall that for the analytic function
Φ(z) = ϕ(x, y) + iψ(x, y), if we generate a set of curves ϕ(x, y) = α1, α2, …, and
another set of curves ψ(x, y) = β1, β2, … (where the α and β values are real con-
stants), the two sets of curves will meet at right angles—they are orthogonal.
The orthogonality property can break down at a point where Φ′(z) = 0. Recall
also that in electrostatics the electric field at each point in space is at right
angles to the curve of constant potential (the curve of constant ϕ) passing
through that point. Thus, at each point in space, the electric field vector must
be tangent to the curve of constant ψ passing through that point. The state-
ment breaks down only at those places where the electric field vanishes—
here we cannot assign it a direction.
In a similar fashion, in fluid mechanics, the curves along which the
stream function maintains constant values, the “streamlines,” are the
curves that are tangent to the direction of fluid flow. The curves do not
by themselves show the actual sense of flow (i.e., the direction of motion
along the s­ treamline)—this must be computed from either Equation 4.24 or
Equation 4.8. However, we can frequently tell the direction along the curve
by a simple physical argument. Finally, in heat conduction problems, the
streamlines—the curves of constant ψ—are the curves that are tangent to
the direction of heat flow.
In fluid mechanics, we require that any impermeable boundary be coinci-
dent with a streamline. Let us look at the following case.
The analytic function Φ(z) = z, z = x + iy is not a very interesting complex
velocity potential. We have from Equation 4.24 that Φ ′ ( z ) = 1 = Vx + iVy so
that Vx = 1 and Vy = 0. Thus, fluid is moving uniformly in the positive x
direction with uniform velocity. The stream function ψ(x, y) = y and so the
streamlines are simply the lines on which y assumes fixed real values, for
example, y = 0, ±1, ±2, …. Suppose we add to this analytic function the ana-
lytic function 1/z. Thus, Φ(z) = z + 1/z, where we exclude the origin. Notice
that as z → ∞, Φ(z) → z, which means that the flow is uniform at infinity.
Let us ask MATLAB to generate the streamlines for this potential. This code
will do that:

clf;clear
x=linspace(−6,6,200);
y=linspace (−6,6,200);
n=linspace (−2,2,9);
[x,y]=meshgrid(x,y);
156 A MATLAB® Companion to Complex Variables

z=x+i*y;
w=z+1./z;
wi=imag(w);
[c,h]= contour(x,y,wi,n);colorbar;
axis equal
xlabel('x');ylabel('y'); grid
clabel(c,h);

This output is as shown in Figure 4.19.

The arrows that appear in Figure 4.19 were not generated by the code but
were added later as will be explained. Do not be alarmed that we cannot read
the numbers inside the unit circle—they are not needed.
It appears that the contour on which ψ = 0 is satisfied both on the unit
circle and on the line y = 0, −∞ < x < ∞. This can be confirmed as follows:

Φ( z) = z + 1/z = x + iy + 1/( x + iy ) = x + iy +
( x − iy )
x2 + y 2

6 2

1.5
4

1
2 2
2 2 2 2
1.5 1.5
1.5 1 1.5 1.5 0.5
1
1 0.5 1 1
0.5 –1
0.5 0.5 0.5
–0 1.5
–

–2
0
.5

0 0 0 0 1.5 0 0 0 0
y

0.5
0
12

–0.5 –0.5 –0.5

–0.5
–1 –0.5 –1 –1
–1
–1.5 –1 –1.5 –1.5 –0.5
–1.5 –1.5
–2 –2 –2 –2
–2 –2
–1

–4
–1.5

–6 –2
–6 –4 –2 0 2 4 6
x

FIGURE 4.19
1
Streamlines for z + .
z
Harmonic Functions, Conformal Mapping, and Some Applications 157

y
Thus, ψ ( x , y ) = Im Φ ( x , y ) = y − . If we equate the preceding
x + y2 ( 2
)
to zero, we find that the equation is satisfied by either y = 0 or x2 + y2 = 1
(which is |z| = 1) as the figure suggests. Thus, the unit circle is a streamline
of this given complex potential.
The complex velocity for this flow is found from Equation 4.24. With the
given potential, we have Vx + iVy = 1 − 1/z 2. We see that as z → ∞, Vx → 1 and
Vy → 0. There is a flow of fluid that moves from left to right. It begins as a
uniform flow and ends the same way. Knowing the direction of the flow, we
added arrows to some of the streamlines.
The contour |z| = 1 is a streamline, which means that flow is everywhere
tangent to this circle. We may place a rigid barrier in coincidence with this
circle and not affect the flow of fluid in and outside this barrier. Of course,
no fluid will get inside the barrier. Thus, we may interpret the streamlines
outside the unit circle in Figure  4.19 as showing the flow around what
might be an infinitely long wooden log of unit radius that has been placed
inside a uniform stream of infinite extent. Under these circumstances, the
streamlines shown inside the unit circle in this same figure have no physical
meaning, and the fact that we cannot read the numbers on these lines is of
no consequence. If you want to know the actual fluid velocity at any point
­outside the log, you can use

Vx + iVy = 1 − 1/z = 1 − 2 ( x + iy )2 (4.26)

(x )
2 2
+ y2
to establish the components Vx and Vy.
If both components of the velocity were to vanish at a point, then we could
not assign a direction to the fluid velocity vector. We cannot say that it is tangent
to the streamline. In such cases, the tangent to the streamline is not defined.
Studying Equation 4.26, we see that if z 2 = 1, which means z = ±1, the fluid
velocity is zero. Studying Figure 4.19, we see that indeed we cannot establish
the tangent to the streamline as this point—the streamline on y = 0 branches off
into two different directions. In fluid mechanics, such points where the velocity
vanishes are called “stagnation points” as the fluid is nonmoving (stagnant).
Let us turn our attention to electrostatics and create a new potential by
taking the one from the previous example and multiplying it by i. The code
is essentially the same as above except that we replace w=z+1./z; with
w=i*(z+1./z);.
We have also generated a plot of a unit circle, by means of the lines of code
added at the end of the program:

hold on
axis([−2 2 −2 2])
th=linspace(0,2*pi,1000);
z=exp(i*th);
plot(z,'linewidth',2)
158 A MATLAB® Companion to Complex Variables

We have added arrows to the streamlines after the plot has been formed.
Figure 4.20 presents the result.
Our complex potential is now

Φ ( z ) = φ ( x , y ) + iψ ( x , y ) = iz + i/z = − y + ix + i/( x + iy ) = − y + ix +
( y + ix ) (4.27)
x2 + y 2

If |z| >> 1, we have approximately that Φ(z) ≈ iz = −y + ix. Thus, ϕ(x, y) =

− ∂φ − ∂φ
Re(Φ(z)) ≈ −y. Recalling that Ex = and Ey = , we realize that, as we
∂x ∂y
move far from the origin, the electric field asymptotically tends to Ex = 0 and
Ey = 1. The field is uniform and pointing in the positive y direction. In this
way, we were able to assign arrows to the streamlines in Figure 4.20.
From Equation 4.27, we have that the electrostatic potential
y
φ ( x , y ) = Re Φ ( z ) = − y + 2 . Observe that on the unit circle, x 2 + y2 = 1,
x + y2
we have that ϕ(x, y) = 0. Thus, the unit circle is a locus of constant poten-
tial, and we would expect that the streamlines strike this circle at right
angles. Figure 4.20 would appear to confirm this except at z = ±1. The fail-
ure of orthogonality can be explained if we notice that Φ′(z) = i(1 − 1/z2)

1.5

1
–0.5 –1 1.5

0.5
0

0.5
–2

2
5 1
–

1.
–2

0
y

2
2

–0.5

–1
–1.5

–0.5

1.5
0.5
–1
–2

–1.5
1

2
0

2
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 4.20
A plot of a unit circle and the streamlines of arrows.
Harmonic Functions, Conformal Mapping, and Some Applications 159

is zero if z = ±1. Recall that Ex ( x , y ) + iEy ( x , y ) = − Φ′ ( z ). Thus, the electric

field vanishes at z = ±1, and we cannot assign it a direction or find a tan-
gent to the streamlines at these points.
If an electric field strikes a perfect electrical conductor, the intersection
must be orthogonal. We can thus insert a perfectly conducting metal tube
of unit radius into the configuration of Figure 4.20. The axis is perpendicu-
lar to the z plane and passes through the origin. We may regard this as a
metal tube placed into an electric field that is uniform and parallel to the y
axis at infinity. The field will not penetrate the tube, the field inside is zero,
and we should ignore the streamlines inside the unit circle that appear in
the plot of Figure 4.20. The streamlines outside the tube show the direction
of the electric field. The numerical values of the field can be found from
Ex ( x , y ) + iEy ( x , y ) = − Φ′ ( z ).
Of what use is the numerical value of the stream function? Consider a
piecewise smooth, closed curve C such as that in Figure 4.21. Suppose Φ(z) =
ϕ(x, y) + iψ(x, y) is a complex analytic potential describing a configuration in
heat flow, fluid mechanics, or electrostatics. Let C′ be a segment of C with α
and β the values of z at the extremes of the segment. We encounter α before
β in moving counterclockwise. Let

Δψ = ψ(α) − ψ(β) (4.28)

If α and β were to lie on the same streamline, this expression would

be zero, because ψ is constant along a streamline. Now it can be shown

1.5

β
1
C´ α
0.5
Closed contour C
0

–0.5

–1

–1.5

–2
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2

FIGURE 4.21
A contour for studying ψ.
160 A MATLAB® Companion to Complex Variables

(see  W  chapter  8 and its appendix) that in heat conduction, the expression
H = kΔψ gives the flux of heat flow (the rate at which energy is transported
by heat conduction) across a surface of unit length (perpendicular to the x–y
plane) whose edge is this segment C′. Positive flow is from the interior to
the exterior of the domain whose boundary is C. Recall that k is the heat
conductivity of the medium in use. In fluid mechanics, the expression
G = −Δψ = ψ(β) − ψ(α) is the rate at which fluid moves across the same surface
(assumed permeable) from interior to exterior. Finally, in electrostatics, the
formula F = εΔψ gives the flow of electric flux lines across the surface. Here,
ε is a constant, called the permittivity, that describes the electrical properties
of the medium in which the electric field exists. Note that all of the preceding
statements involving Equation 4.28 presume that ψ varies continuously as we
move along the surface in question.
Let us look at the configuration of Figure  4.19. If the cylinder were not
present, then the flow of fluid would be uniform and given by the potential
Φ(z) = x + iy. How much fluid flows through a rectangular surface like the
one described above whose edge is the line segment x = 1, −1 ≤ y ≤ 1? Now
following the notation of Equation 4.27, we have α = (1, −1) and β = (1, 1).
Since ψ = y, we have Δψ = −1 − 1 = −2 . The rate of flow of fluid is G = −Δψ = 2.
This might be measured in such units as liters per second or gallons per
minute, depending on the system in use. If the cylinder of unit radius shown
in Figure 4.19 is now placed in the fluid, so as to impede the flow, we have
y
shown that ψ ( x , y ) = Im Φ ( x , y ) = y − 2 . Evaluating ψ at points at α
( )
x + y2
and β, we have –½ and ½, respectively. Thus, G = −Δψ = 1. The flow through
the surface has been reduced from 2 to 1 by this impediment.

Exercises
Solve the following problems by means of suitable MATLAB programs. All
of the functions given are harmonic.
1. If the real part of an analytic function is sin(x2/2 − y2/2 + x) cosh(xy + y),
find the imaginary part. Evaluate any arbitrary constant in your answer
by assuming that the sought-after function is zero at the origin.
2. If the real part of an analytic function is y+ arctan(y/(x − 1)), find the
imaginary part. Evaluate any constant by assuming that the func-
tion is zero at z = 0.
x2 − y 2
3. If the real part of an analytic function is x 2 − y 2 + 2 + 2 , find
( )
x2 + y 2
the imaginary part. Evaluate any constant by assuming that the
function is zero at z = 1.
Harmonic Functions, Conformal Mapping, and Some Applications 161

4. If the imaginary part of an analytic function is

1   y 

2 ( ) 
sin  log x 2 + y 2 + y + 1 cosh  arctan   − x  ,
   x 
find the real part to within an additive constant.
5. If the imaginary part of an analytic function is

e ( )
arctan y /x
( )
cos Log x 2 + y 2 + 1 + 2 y − x

find the real part. Assume that the real part equals 1 at z = 1.
6. In the previous section, Example 4.5, we used gradient and quiver
in MATLAB to create arrows pointing in the direction of the gradi-
ent of the function u(x, y) = x2 − y2. The gradient at each point in
space must be normal to the surface, on which u is constant, passing
through that point. The gradient must therefore be tangent to the
streamline passing through that point. Determine the stream func-
tion for this function u(x, y) and plot the streamlines on the same
figure as the vectors for the gradient. Verify that the arrows from
quiver are along the streamlines.
7. Suppose fluid motion is caused by a flow that is uniform at infin-
ity and in the x direction. Suppose that there are present, normal to
the complex plane, a line source of fluid and a corresponding sink
of identical strength. The source and sink penetrate the x axis. The
streamlines for this overall arrangement form interesting shapes.
One in particular forms an oval shape known as a Rankine Oval.
Consider the complex velocity potential given by Φ(z) = ϕ(x, y) +
iψ(x, y) = V∞ z − C Log(z − 1) + C Log(z + 1). Here V∞ is the velocity
of the fluid at infinity, while C (a positive real) is the strength of the
fluid source at z = –1. There is a corresponding sink at z = 1.
a. Take V∞ = 1 and C = 1. Using MATLAB, plot the streamlines of
this flow. A suggestion for a nice plot: let the values of ψ on the
plot be the nine integers satisfying –4 ≤ ψ ≤ 4.
Consider |x| ≤ 5 and |y| ≤ 5.
b. Study your plot. What is special about the curve on which ψ = 0?
This is called a Rankine Oval. Other such ovals are obtained
by adjusting the separation of source and sink, the strengths of
these, C, and the value of V∞.
c. Consider a rectangular surface that freely allows the passage of
fluid through itself. The surface is normal to the x axis and is
described by x = 2, 0 ≤ y ≤ h, where h is real. The surface is one
unit long in the direction normal to the z plane. Write a MATLAB
program, using the stream function, that plots the rate of fluid
flow (the flux of fluid) through this surface as h varies from 0 to 4.
162 A MATLAB® Companion to Complex Variables

4.5 Mapping, Dirichlet, and Neumann Problems

and Line Sources
4.5.1  Dirichlet Problems
(See W chapter 8.5 and S chapter 9.)
Suppose a domain D in the x–y plane represents a region in which there is
an electric field or the conduction of heat. The region is the cross section of a
material extending to infinity. For the moment we assume that the electrical
voltage or the temperature is known on the boundaries of the domain. The
electrostatic potential or temperature, ϕ(x, y), inside D must satisfy Laplace’s
equation and agree with the given boundary conditions on the boundary
points of D. Such problems are often solved through the method of confor-
mal mapping. They are known as Dirichlet-type boundary value problems.
The domain D in the x–y plane is mapped, using a conformal transforma-
tion w = f(z) = u(x, y) + iv(x, y), into the u, v plane. The new domain is D1. The
boundaries are mapped into this plane as well. Thus, the point (x0, y0) on the
boundary of D is mapped into u(x0, y0) + iv(x0, y0) = u0 + iv0 on the boundary
of D1. Refer to Figure 4.22. The values that ϕ(x, y) assumes at each boundary
point of D are assigned to the image of that point in the w plane. Thus, the
values of ϕ(u, v) are known at the boundary points of D1.
Suppose the shape of the domain D1 is simpler or more familiar than D. We
find or recognize a solution to the problem, call it ϕ1(u, v), that is harmonic
in the w plane and satisfies the values of ϕ1(u, v) that were specified at the
boundary points. Let us take ϕ1(u, v) and replace u and v by their explicit
expressions in terms of x and y. Thus, ϕ1(u, v) = ϕ1(u(x, y), v(x, y)) = ϕ(x, y).
The function ϕ(x, y) is a harmonic function. A solution of Laplace’s equa-
tion remains a solution of this equation when transferred from one plane
to another by an analytic function (i.e., by a conformal transformation).

(x0 , y0) φ1(u0 , v0) = k

φ(x0 , y0) = k
(u0 , v0)

D D1

∂φ ∂φ
2 2
∂2φ1 ∂2φ1
+ =0 + =0
∂x2 ∂y2 ∂u2 ∂v2

C (boundary) C1 (boundary)
f(z) = w
z-plane w-plane
(a) (b)

FIGURE 4.22
A domain D (a) and its image D1 (b).
Harmonic Functions, Conformal Mapping, and Some Applications 163

This follows because an analytic function of an analytic function is analytic.

This function provides the solution to our problem—it is harmonic in the
domain D and satisfies the given boundary conditions.
In Neumann problems, which often occur in fluid mechanics where there
is a rigid boundary, or in heat transfer configurations involving insulation,
we again seek a function ϕ(x, y) that satisfies Laplace’s equation in a domain.
However, in this case the component of the gradient of ϕ(x, y) that is normal
to the boundary of the domain (written as ∂φ , where n stands for distance
∂n
along the normal) must take on certain prescribed values—often zero. The
method described above still applies—if we can solve this problem in the w
plane with the required boundary conditions on ∂φ satisfied at the image of
∂n
the original boundary, we can transform this solution back into the z plane
(with our analytic transformation).
Before demonstrating the power of the method, let us solve some simple
problems in the w plane without resorting to the preceding technique. These
results will be of use when we perform that method. Consider the complex
potential:
i
Φ(w) = − Log w (4.29)
π

which is analytic everywhere except on the negative real axis and the origin.
Separating into real and imaginary parts, we have

i arg ( w ) i
Φ (w) = − Log w = − Log w = φ ( u, v ) + iψ ( u, v ) (4.30)
π π π

We will confine our attention to the half space v ≥ 0 with the origin excluded.
Now
1 1 v
φ ( u, v ) = arg ( w ) = tan −1 (4.31)
π π u

where we use the principal value of the argument of w. Also,

1 1 1
ψ ( u, v ) = − Log w = Log 2 (4.32)
π 2π u + v2

1 v
Notice that the harmonic function φ ( u, v ) = arg ( w ) =
tan −1 is continu-
π u
ous in the upper half of the w plane and equal to zero on the positive u axis
and equals 1 on the negative real axis. It is undefined at the origin.
The equipotential surfaces are simply rays extending out from the origin,
the edges of the surfaces on which arg(w) is constant. A plot of this is pre-
sented in Figure 4.23.
164 A MATLAB® Companion to Complex Variables

0.6
6

0.4
0.8 2
0.
4

0.6
v

0.4
0.8
0.2
2
0.6

0.02 .4
0.8

0 1 1 0 0

–2
–6 –4 –2 0 2 4 6
u

FIGURE 4.23
 −i 
The contour lines for real Re  Log w  in the upper half plane.
 π 

Figure 4.23 is obtained (almost) from this code:

clf
x=linspace (−7,7,100);
y=linspace (0, 7,100);
[x,y]=meshgrid(x,y);
z=x+i*y;
phi=−i*log(z)/(pi);
phir=real(phi);
phim=imag(phi)
values=linspace(0 ,1,6);
figure (1)
[g,h]=contour (x,y,phir,values,'linewidth',2);
clabel(g,h);
grid
xlabel('u');ylabel('v');axis equal
x=linspace (−7,7,100);
y=linspace (0, 7,100);
[x,y]=meshgrid(x,y);
z=x+i*y;
phi=−i*log(z)/(pi);
phir=real(phi);
phim=imag(phi)
Harmonic Functions, Conformal Mapping, and Some Applications 165

values=linspace(0 ,−1,6);
figure (2)
[g,h]=contour (x,y,phir,values,'linewidth',2);
clabel(g,h);
grid
xlabel('u');ylabel('v');axis equal

If you run the above code, you see (with Release 2015a) that MATLAB fails
to correctly label the ray y = 0, x > 0 as being the contour on which our func-
tion is zero. We have added in this zero by editing Figure 4.23. I am told that
this weakness in MATLAB will likely be fixed in a future version.
The imaginary part of the complex potential is

1 1 1
− Log w = Log 2 (4.33)
π 2π u + v2

The streamlines are the lines or edges of the surfaces on which u2 + v2 is

constant. These are circles. Since we are focusing on the upper half of the w
plane, we are considering semicircular arcs. We can generate the streamlines
using the above code and get the curves as shown in Figure 4.24. Note that
we must use the imaginary part of the complex potential and also use differ-
ent numbers for the values assumed by the stream function on the contour
lines.

7
–0.6
–0.6
6 6
–0.
–0
5 .6
.6
–0

4
–0.4
v

–0.

3
6
–0
.4
–0
–0.6

.4

2 –0.2

1
–0.6
–0.4

–0.2

–0.4
–0.2

0
–6 –4 –2 0 2 4 6
u

FIGURE 4.24
 −i 
The contour lines for Im  Log w  in the upper half plane.
 π 
166 A MATLAB® Companion to Complex Variables

Comments:

1. Refer to Equations 4.10 through 4.12. In electrostatics we have that the

electric field is given by E = −∇ϕ(x, y). Because of the minus sign, the
arrows on the streamlines, if they are to show the direction of the elec-
tric field, should point from higher to lower values of electric potential.
Thus comparing the previous two figures, we see that arrows placed
on the streamlines, showing the direction of the electric field, should
be in the clockwise direction. This is also the case for the heat flux
density vector, while arrows showing the direction of fluid flow would
be oppositely directed because there is no minus sign. We have placed
a few arrows on the figure, using the plot editor, assuming this is a
physical configuration in electrostatics or heat flow.
2. In a region of space that has no electric sources (i.e., charge) or sources
of heat, the highest and lowest values of ϕ(x, y) (the electric potential
or the temperature) will occur on the boundaries. (See W section 4.6,
problems 13 and 14.) Thus, we might think of Figure 4.23 as a region
in the upper half space v ≥ 0. It is bounded below by a plane whose
right half is at ϕ = 0 and whose left half is at ϕ = 1. (These are tem-
peratures or voltages.) Note that in the upper half space we can say
that 0 ≤ ϕ ≤ 1, and the results confirm that. This kind of reasoning is
useful in choosing the values of ϕ(x, y) to be plotted in the contours
command in MATLAB. We can avoid seeking contours on which
ϕ(x, y) assumes values outside of the interval described by the poten-
tials or temperature on the boundary. This argument does not apply
when there are sources (e.g., charges) present.
If we were to use w − 1 in lieu of w in the previous complex potential,
Equation 4.29, we have

−i arg ( w − 1) i
Φ1 ( w ) = Log ( w − 1) = − Log w − 1 = φ1 ( u, v ) + iψ 1 ( u, v )
π π π

We would obtain equipotentials and streamlines identical to those in

Figures 4.23 and 4.24, except that the patterns would be displaced one unit to
the right. The equipotentials would be rays emanating from the point u = 1,
v = 0, while the streamlines would be semicircles centered at this same point.
Regarding Φ1(w) as a complex electrostatic potential, the voltage ϕ1(u, v) on
the boundary v = 0, u > 1 is now 0 while ϕ1(u, v) for v = 0, u < 1 is 1.
Similarly, for the complex potential
−i arg ( w + 1) i
Φ2 ( w ) =
Log ( w + 1) = − − Log w + 1 ,
π π π
the equipotentials for ϕ(u, v) are rays starting at u = −1, v = 0, while the stream-
lines are semicircles with centers at this point. The potential on the bound-
ary v = 0, u > −1 is now 0, while the potential for v = 0, u < −1 is 1.
Harmonic Functions, Conformal Mapping, and Some Applications 167

3 0.2

0.2
2
0.3
0.2
v

0.4

0.3
1

0.5

0.2
00. 4
.3
0.2

0.5
0 0 0 1 0 0

–1

–4 –3 –2 –1 0 1 2 3 4
u

FIGURE 4.25
 −i i 
The contour lines for Re  Log (w − 1) + Log (w + 1) in the upper half plane.
 π π 

Consider the subtraction Φ0(w) = Φ1(w) − Φ2(w) = ϕ0(u, v) + iψ0(u, v).

Thus,
Log ( w − 1) Log ( w + 1)
Φ0 ( w ) = − i +i (4.34)
π π
1
φ0 = arg ( w − 1) − arg ( w + 1)  (4.35)
π
Log ( w − 1) Log ( w + 1)
ψ 0 (w) = − + (4.36)
π π
To find the conditions satisfied by ϕ0(u, v) on the line v = 0, we find the differ-
ence between the conditions satisfied by the real parts of the complex potentials
Φ1(w) and Φ2(w). We find that ϕ0(u, v) = 0 for v = 0, u < −1, and ϕ0(u, v) = 0 for v = 0,
u > 1. Notice that ϕ0(u, v) = 1 for v = 0, −1 < u < 1. Let us use MATLAB to plot the
resulting equipotentials and streamlines in the space above the real axis.
Figure 4.25 shows the equipotential surfaces, in the space v > 0. If we imag-
ine that the earth is maintained at ground potential (zero volts) except for a
conducting strip of width 2, centered at the origin, and if this strip is kept
at 1  volt, the above figure indicates the electric potential above the earth.
A similar interpretation can be made involving temperature above the earth
if this is a problem in heat transfer. The code we used is as follows:
clf
clear
u=linspace (−4,4,100);
v=linspace (0, 4,100);
168 A MATLAB® Companion to Complex Variables

[u,v]=meshgrid(u,v);
w=u+i*v;
phi=−i*log(w−1)/(pi)+i*log(w+1)/(pi);
phir=real(phi);
phim=imag(phi);
[g,h]=contour (u,v,phir,[ .2 .3 .4 .5],'linewidth',2);
clabel(g,h);
xlabel('u');ylabel('v');axis equal
grid

Notice the line of code:

[g,h]=contour (u,v,phir,[.2 .3 .4 .5],'linewidth',2);

The elements in the vector [.2 .3 .4 .5] are the numerical values chosen that
apply on each equipotential surface. They were selected after some experi-
mentation to produce a useful plot. As mentioned above, we seek values for
the elements lying on or between 0 and 1.
The contours of potential 0 and 1 were plotted separately and given extra
width. The reader should be able to supply the additional code.
The streamlines created by the complex potential Φ0(w) are plotted with a
similar code and are shown in Figure 4.26.

4
0

3 -0.1 0.1
0

2
v

-0.

0.1
1

-0.2 0.2
1
0

0.3
-0.3
–0.2
–0.1

0.2
0.1
–0.2

–0.3

0.2
0.3

–1
–4 –3 –2 –1 0 1 2 3 4
u

FIGURE 4.26
 −i i 
The contour lines for Im  Log (w − 1) + Log (w + 1) in the upper half plane.
 π π 
Harmonic Functions, Conformal Mapping, and Some Applications 169

Assuming this is a problem in electrostatics, we have added arrows to

the streamlines in accordance with Figure 4.26. The electric field points in a
direction from higher to lower potential.

4.5.2  A Dirichlet Problem

Example 4.7
Using conformal mapping, find an expression for the electrostatic
potential in the space y ≥ 0, x ≥ 0 subject to the boundary conditions
shown in Figure 4.27. Use MATLAB to plot the equipotentials and the
streamlines.
We can think of the above as showing the cross section of a piece of
sheet metal that has been bent into a 90° angle. The sheet metal is kept
at an electrostatic potential of zero volts. However, a section of the metal
along both the x and y axes is maintained at 1 volt. The potential is not
defined at x = 1 on y = 0, or at y = 1 at x = 0.

Solution:
Consider the conformal mapping w = z2 applied to the configuration
shown in Figure 4.27. This would bend the boundary into the straight line
v = 0, −∞ < u < ∞. The image of a portion of the boundary, the L-shaped
conductor maintained at 1 volt in the z plane, is now the line segment
in the w plane, v = 0, −1 < u < 1. We assign to this the same potential,

2.5

1.5 0 volt
y

0.5 1 volt

1 volt 0 volt
0

0 0.5 1 1.5 2 2.5 3

x

FIGURE 4.27
An electrostatic Dirichlet problem.
170 A MATLAB® Companion to Complex Variables

the numerical value of 1, that its image had in the z plane. The remain-

der of the real axis in the w plane is assigned the value its image had
in the z plane: zero. The situation is shown in the w plane, as shown in
Figure 4.28.
The portion of the z plane in the first quadrant is now the region v ≥ 0.
The problem of finding the potential above this plane has been solved.
It is the one we just did and whose results are shown in Figures 4.25 and
4.26. We can transfer this result into the z plane by substituting z2 for w
in the set of Equations 4.34 through 4.36.
Thus,

Φ 0 (w) = − i
(
Log z 2 − 1 ) + i Log ( z 2
+1 ) (4.37)
π π

1
φ0 =
π
( ) (
arg z 2 − 1 − arg z 2 + 1  (4.38) )

ψ 0 (w) = −
(
Log z 2 − 1 ) + Log ( z 2
+1 ) (4.39)
π π

We can write MATLAB programs to plot the equipotentials and

streamlines. For example, the equipotentials are as shown in Figure 4.29.

2.5

1.5
v

0.5

0 volt 1 volt 0 volt

0

–3 –2 –1 0 1 2 3
u

FIGURE 4.28
A transformation of the previous figure.
Harmonic Functions, Conformal Mapping, and Some Applications 171

2.5

2 0.1
0.1 0.1

1.5
y

0.2
0.1

0.2
0.3
1 0.3

0.1
0.2

0.6

0.5
0.9 0.1
0. 0.3
0.6

0
0.9 0.1
0 0.5 1 1.5 2 2.5 3
x

FIGURE 4.29
The equipotentials for Figure 4.27.

Figure 4.29 is obtained from the following code:

clf
x=linspace (0,3,100);
y=linspace (0,3,100);
[x,y]=meshgrid(x,y);
z=x+i*y;
phi=-i*log(z.^2−1)/(pi)+i*log(z.^2+1)/(pi);
phir=real(phi);
phim=imag(phi);
values=[0 .1 .2 .3 .6 .9 1];
[g,h]=contour (x,y,phir,values,'linewidth',2);
clabel(g,h);
xlabel('x');ylabel('y');axis equal
grid

We notice that the equipotential at .9 volts nearly has the same shape as
the nearby L-shaped conductor at 1 volt. Some trial and error was used
before assigning the numbers in the line of code for values, in order to
get an interesting plot. The same goes for the code used for streamlines.
We can plot the streamlines for this configuration using nearly the
same code as above. We use phim in the above code (the imaginary part
of the complex potential) to get the contour lines, which are the stream-
lines. The result is as shown in Figure 4.30.
172 A MATLAB® Companion to Complex Variables

2.5
-0 0
.1

0
–0
1.5
y

.2

–0.1
1
.6

0 0.1
–0

–0.3
–0.2 0.1
0.5 0.2
–0.1
0.6

0.1
0.2
3

0
0.

0
0 0.5 1 1.5 2 2.5 3
x

FIGURE 4.30
The streamlines for Figures 4.27 and 4.29.

We have chosen values of ψ ranging between –1 and 1 for our contour val-
ues. We used the line of code: values=[−1−.6−.3−.2−.1 0.1.2.3.6 1];.
We have added arrows to the plot, placing them on the streamlines so
that they point in the direction of the electric field (i.e., going from higher
to lower potential). There is a streamline (along y = x) that goes from the
corner (at zero potential) to infinity.
A problem like the above can sometimes be done more easily if we use
the polar coordinates r, θ, where now z = reiθ. This could be the case, for
example, if the 90° angle in Figure 4.27 was changed to some other angle,
say 45° or 60°. To work in polar coordinates, we use the following code
which is written here for the case of a 90° bend. Note the values used
for theta:
clf
r=linspace (0,3,100);
theta=linspace (0,pi/2,100);
[r,theta]=meshgrid(r,theta);
z=r.*exp(i*theta);
phi=−1i*log(z.^2−1)/(pi)+1i*log(z.^2+1)/(pi);
phir=real(phi);
phim=imag(phi);
values=[ 0 .1 .2 .3 .6 .9 1];
x=real(z);y=imag(z);
[g,h]=contour (x,y,phir,values,'linewidth',2);
clabel(g,h);
xlabel('x');ylabel('y');axis equal
grid

This should produce the same output as we see in Figure 4.29.

Harmonic Functions, Conformal Mapping, and Some Applications 173

4.5.3  A Neumann Problem

Example 4.8
Fluid flows down into a closed channel as shown in Figure  4.31. The
fluid flow is Vo in the positive direction along the bottom wall at x = 0,
y = 0 +.
Find the velocity potential for this configuration, and plot the stream-
lines for the flow.

Solution:
The fluid cannot penetrate the walls of the channel. Hence, the veloc-
ity vector V can have no component along the normal to the walls
at any point. The velocity vector is given by V = grad ϕ, where ϕ is
the velocity potential for the fluid flow. Thus, the gradient of ϕ will
have no component normal to the walls of the closed channel. This
is ensured if the walls of the channel coincide with streamlines
because the velocity vector is tangent to the streamlines. This is
part of our boundary conditions, and because it involves the normal
component of V = grad ϕ, it is a Neumann problem. The remainder
c­onsists of our finding a velocity potential ϕ(x, y) that will satisfy the
prescribed ­condition on the velocity at the center of the bottom of
the channel.
If we can find a conformal mapping that will straighten the walls of the
channel into a line in the w plane (e.g., into the real axis), we have simpli-
fied the problem. A table of conformal mappings (see, e.g., S ­section 8.4)
is useful in discovering this transformation. We end up employing the
following:

π π π π π
w = u + iv = sin z = sin x cosh y + i cos x sinh y (4.40)
2 2 2 2 2

0.9
0.8
0.7
0.6
0.5
y

0.4
0.3
0.2
0.1 V0
0
–0.1
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 4.31
Flow into and out of a channel.
174 A MATLAB® Companion to Complex Variables

Notice that for x = ±1, y ≥ 0, which holds for points on the left- and right-
π
hand boundaries, cos x = 0. Points here are mapped onto the part of the
2
u axis where |u|≥1. The bottom of the channel, where y = 0, −1 ≤ x ≤ 1, is
π
mapped onto the line v = 0, −1 ≤ u ≤ 1. Notice that dw/dz = cos ( πz/2 )
2
vanishes at z = ±1, which means that the transformation is not conformal
at these two points. This enables the transformation to map the U-shaped
boundary into a straight line v = 0, −∞ ≤ u ≤ ∞.
Because the boundary is a streamline in the z plane, it follows that the
u axis in the w plane must also be a streamline. The simplest nonconstant
potential, satisfying such a condition, which is analytic in the upper half
of the w plane is

Φ(w) = ϕ(u, v) + iψ(u, v) = Aw = Ap (u, iv) (4.41)

where A is any real number. The stream function Av vanishes on the u

2
axis. For reasons that will soon be apparent, we choose A = V0. Using
π
this value, and combining Equations 4.40 and 4.41, we have for the com-
plex potential in the z plane,

2 π 2 π π π π 
Φ( z) = V0 sin z = V0  sin x cosh y + i cos x sinh y  (4.42)
π 2 π 2 2 2 2 

From Equation 4.24 and the above, we find that the fluid velocity in
the channel is

 π   π π π π 
Φ ′( z) = V0  cos z = V0  cos x cosh y + i sin x sinh y 
 2   2 2 2 2 

= Vx ( x, y ) + iVy ( x, y )

As required, at x = 0, y = 0+, we have that Vx = V0 and Vy = 0. (This

explains our choice of A.) A plot of the streamlines can be obtained from
MATLAB code. We have chosen to take Vx = V0 = 1 at the bottom of the
channel. The program is

clf
x=linspace (−1.1,1.1,100);
y=linspace (−.1, 3,100);
[x,y]=meshgrid(x,y);
z=x+i*y;
phi=2/pi*sin(pi/2*z);% this is the complex potential
phim=imag(phi);
values=[0 .5 1 3 5 10 15 20];
[g,h]=contour (x,y,phim,values,'linewidth',2);
clabel(g,h);
xlabel('x');ylabel('y');xlim([−2 2]);ylim=([0 3]);
axis equal
grid
Harmonic Functions, Conformal Mapping, and Some Applications 175

The choice of values of ψ on each streamline, given by the line of code

values=[0 .5 1 3 5 10 15 20], were determined by some experi-
mentation to produce the attractive and useful plot given in Figure 4.32.
Some arrows have been added to indicate the direction of fluid motion.

Comment:
The reader should note this subtlety. At the point x = 0, y = 0+, the
complex velocity vector is simply V0. The image of this point in the
w plane is w  =  0.  The complex fluid velocity vector in the w plane is
2
Φ ′ ( w ) = V0 . These are numerically different and illustrate a general
π
principle: Although the complex potential assumes identical numerical
values at a point in the z plane and at the image of that point in the w
plane, the same cannot be said in general for the magnitude (or direc-
tion) of the complex velocity vector at points that are images of each
dΦ
other. This is because in the w plane, we have a fluid velocity of ,
dw
while in the z plane it is

dΦ ( w ( z )) dΦ ( w ) dw
=
dz dw dz

and in general these are not the same.

3 20
1 0 0.5

1
15 0
53
1

15

20
2.5 15 10
5
0.5 0

10
2 5
3

3 5
1 0.5

1
0

1.5 3
y

0
0.5

1
1 1
0.5
0.5 0.5
0

0 0 0 0
0
0

–2 –1.5 –1 –0.5 0 0.5 1 1.5 2

x

FIGURE 4.32
Streamlines for flow into a closed channel.
176 A MATLAB® Companion to Complex Variables

Further Comment:
We were fortunate in doing this problem, which began in the z plane,
to find an explicit expression for w, as a function of z, that would map
our geometry in the z plane into a more tractable one in the w plane.
Sometimes we can find a function of z, call it z(w), that will change a sim-
ple geometry in the w plane into a more complicated but useful geometry
in the z plane. However, the inverse of this function w(z) may not be read-
ily available from known algebraic and transcendental functions. Such
an instance occurs in problems 5 and 6 of the exercises. However, we can
often still trace the equipotentials and streamlines in the given problem
(in the z plane) using a technique shown in problem 5 and, as shown
in problem 6, obtain the fluid velocity by using the solve command in
MATLAB to obtain numerical values of w for specified values of z.

4.5.4  Line Sources and Conformal Mapping

In the previous section, we discussed line sources of electric flux, heat energy,
or fluid flow.
If such a source passes though the point z = z0 in the complex plane, its
most general form for the resulting complex potential is
a
Φ ( z ) = A Log (4.43)
( z0 )
z −

Here the constants A and a are nonzero reals, and a is positive. We will call
A the strength of the line source. With

Φ(z) = ϕ(z) + iψ(z)

We have that

a
φ ( z ) = φ ( x , y ) = A log (4.44)
z − z0
ψ(z) = ψ(x,y) = −A arg(z − z0) (4.45)

For convenience, we have used principal values throughout the above.

This is not strictly necessary but is usually done. As noted in W (section 8.7),
the strength of the line source in electrostatics, heat conduction, or fluid flow
can be related to A by means of Equation 4.28. For example, a line charge car-
ρ
rying ρL coulombs per meter gives A = L , where ε0 is a known numerical
2 πε 0
constant. There are similar expressions giving A if the line source is shed-
ding heat or fluid.
Note from Equation 4.44 that the constant a establishes the distance from
the line charge, |z − z0|, at which the potential ϕ falls to zero. This number is
sometimes chosen arbitrarily and is related to the units in use.
Harmonic Functions, Conformal Mapping, and Some Applications 177

Suppose the line source, in the z plane, resides near a boundary on which
the value of ϕ is specified (a Dirichlet problem) or the boundary is required
to be a streamline (a Neumann problem). Because of the presence of the
boundary, we must assume that the complex potential created by the line
source is no longer identical to what the source would create were it iso-
lated. We can sometimes solve such a problem through a conformal mapping
into the w plane where w = f(z) creates a simpler or more familiar boundary.
The line source is transformed into a line source passing through w0 = f(z0).
The strength of the line source measured by A will remain the same as in the orig-
inal problem. This is because (see W section 8.7) the change in ψ measured
along a curve in the z plane and measured along the image of that curve in
the w plane will be identical. Example 4.9 presents the technique.

Example 4.9
Consider the line charge (or “line source”) that produces the complex
1
electric potential A Log , h > 0, when the line is in an infinite uni-
z − ih
form material; we take A as positive real which means that the charge
is positive. The source passes through x = 0, y = h. The source is placed
h units above an infinite “grounded” conducting plane as shown in
Figure  4.33. The x axis is the cross section of the plane that is at the
potential (or voltage) ϕ = 0. We use Φ(x, y) to describe the actual complex
potential that now exists above the plane. The real potential ϕ(x, y) must

1.8

1.6
y
1.4

1.2

1 y = h line source

0.8

0.6

0.4 φ = 0 on y = 0

0.2
x
0

–2 –1.5 –1 –0.5 0 0.5 1 1.5 2

FIGURE 4.33
A line source at y = h above a grounded plane.
178 A MATLAB® Companion to Complex Variables

vanish on the plane. Find the complex potential above the plane and use
MATLAB to sketch the equipotentials and stream lines.

Solution:
Suppose we do a conformal mapping that will have the line source at
z = ih pass through the origin in the w plane. We map the infinite line
y = 0 into the unit circle in the w plane. The symmetry thus obtained is
one in which we might find a solution to the problem of finding Φ(w) in
the w plane.
Referring to the previous section on bilinear transformations, we find
that one such transformation is

z − ih
w= (4.46)
z + ih

Note that there is no unique transformation that will accomplish this;

we can multiply the right side of the above by a complex number of the
form eiω (where ω is any real) and still have a valid transformation. You
should convince yourself that this mapping maps the upper half of the
z plane into the interior of the unit circle. Note that as required, the image
of z = ih is w = 0. With this transformation, we have Figure 4.34.

1.5

φ=0

0.5
1 Line source

0 *
v

–0.5

–1

–1.5
–1.5 –1 –0.5 0 0.5 1 1.5
u

FIGURE 4.34
A conformal transformation of the previous figure.
Harmonic Functions, Conformal Mapping, and Some Applications 179

A line source now appears at the center of a cylinder. The cylinder is

maintained at the same potential as its z-plane image: the x axis. Thus, on
the cylinder, ϕ(u, v) = ϕ(|w|= 1) = 0. Because the strength of the line source,
A, is preserved, we can say that the complex potential created by the line
a′ a′
is of the form Φ ( w ) = A Log = A Log − iA arg ( w ) = φ ( u, v ) + iψ ( u, v ).
w w
To make the potential ϕ(u, v) vanish at |w|= 1, we take a′ = 1. Thus,

1 1
Φ ( w ) = A Log = A Log − iA arg ( w ) = φ(u, v) + iψ ( u, v ) (4.47)
w w

Combining the preceding with Equation 4.46, we have for the complex
potential in the z plane:

z + ih z + ih z + ih
Φ ( z ) = A Log = A Log + iA arg (4.48)
z − ih z − ih z − ih

Let us use the above to plot the equipotentials and streamlines in the
upper half space of the z plane. Below the line y = 0, there is no electric
field. The infinite grounded conducting plane prevents any penetration
into y < 0. We take A = 1, h = 1 for our plots.
Thus, the following code generates the lines of constant ϕ. We did some
experimentation to discover that values of ϕ = 0, .1, .5, 1,  2,  5
­produce good plots. We have placed an o at the location of the line
source: (1,0). We present the code and results below:

clf
x=linspace (−3,3,100);
y=linspace (−.1,5,100);
[x,y]=meshgrid(x,y);
z=x+i*y;
phi=log((z+i)./(z−i));
phir=real(phi);
values=[0 .1 .5 1 2 5];
[g,h]=contour (x,y,phir,values,'linewidth',2);hold on
clabel(g,h);
xlabel('x');ylabel('y');axis equal
plot(0,1,'o')
grid

This results in the output shown in Figure 4.35.

Writing the code to plot the streamlines, the contours on which the
imaginary part of Φ is constant is slightly more complicated. This
z+i
is because Φ = log has a branch cut along the imaginary axis
z−i
between z = i and z = –i. To see this, notice that if x = 0, −1 < y <  1,
z+i
then is  negative real. Recall that MATLAB uses the principal
z−i
value of the logarithm. We need be concerned only with the portion of
180 A MATLAB® Companion to Complex Variables

4.5

4 0.5

3.5

2.5
0.5
y

0.5
2 1

1.5 2

1
1
0.5
0.1 0.5 0.1
0.5
0.1
0 0 0 0
–3 –2 –1 0 1 2 3
x

FIGURE 4.35
The equipotentials for line source above a ground plane.

the branch cut in the upper half plane, which is where we are seeking

a solution. Here is the required code:

clf;
clear;
x = linspace(−3, 3, 100);
y = linspace( 0, 5, 100);
[x, y] = meshgrid(x, y);
z = x + i*y;
phi = log((z+i)./(z−i));
phim = imag(phi);
del=.05
phim(abs(x) < del & y < 1) = nan; % added this line
values = [−2 −1, −.5,−.3 0,.3, .5, 1 2];
[g,h] = contour(x,y,phim,values,'linewidth',2);
hold on
clabel(g, h);
xlabel('x');
ylabel('y');
axis equal
plot(0, 1, 'o')
grid

The line of code phim(abs(x) < del & y < 1) = nan ensures that
MATLAB does not seek to find values of ϕ, in the contour command,
Harmonic Functions, Conformal Mapping, and Some Applications 181

within a strip surrounding the branch cut of width 2∗del. Here

2∗del  =  0.1 In the strip, MATLAB is told that ϕ is simply “not a
­number,” or nan. You should try running the code without this state-
ment, and you will see that the resulting streamlines, under the line
charge, are erroneous. This is because the contour function becomes
confused where the function changes discontinuously and seeks to
interpolate values between the numerical values on opposite sides of
the branch cut.
The line of code values = [−2 −1 −.5 −.3 0 .3 .5 1 2]; was
chosen after some experimentation so as to yield an attractive and sym-
metrical plot. The same is true of the value of del used here, .05. Making
this number smaller may result in contour seeking to make extrapola-
tions at the discontinuity, while making it too large can cause some of
the streamlines not to be plotted. Figure 4.36 is generated by the preced-
ing program.
Because this is a problem in electrostatics and we have assumed
that the charge at z = i is positive, we have added arrows to the
streamlines as shown. The direction of the streamline is the
­d irection of the electric field and emanates outward from the charge.
The infinite conducting plane at y = 0 shields the space y < 0 from
the electric field ­c reated by the line charge, and the field in this half
space is zero.

4.5
0

3.5
–0.3
0.3
3

2.5
y

2
–0.5 0.5

1.5
–0

0.5 0.3
.3
0

–1 1
1
–2

0.5

0
–3 –2 –1 0 1 2 3
x

FIGURE 4.36
The streamlines for a line source above a ground plane.
182 A MATLAB® Companion to Complex Variables

The complex potential can be written as

z+i 1 1
Φ( z) = A Log = A Log − A Log
z−i z−i z+i
It is customary to interpret this by saying that the potential cre-
ated above the grounded plane by the line source at z = i is the sum
of the potential created by the original line source, of strength A (in
the absence of the plane), plus the potential created by an image of that
source, located at z = –i, and having a strength of the opposite value,
namely, –A.
Similarly, if the line source were at z = ih, we would place its image at
z + ih
z = –ih and use the potential Φ ( z ) = A Log .
z − ih

Exercises
1. a. The space y > 0 is filled with a material that conducts heat. Let
ϕ(x, y) be the temperature in the space y ≥ 0. The boundary at y = 0
is maintained at prescribed temperatures as follows: −∞ < x < −1,
ϕ(x, 0) = 1; 1 < x < ∞, ϕ(x, 0) = −1; −1 < x < 1, ϕ(x, 0) = 0. Following an
argument like that used in solving the problem shown in Figures
4.25 and 4.26, find the complex temperature Φ(x,  y) = ϕ(x, y) +
iψ(x, y) in the space y ≥ 0.
i
Hint: Try an expression of the form Φ ( x , y ) = A
Log ( w − 1) +
π
i
B Log ( w + 1) + C , where A, B, and C are real constants. Notice
π
that a constant (like C) will satisfy Laplace’s equation.
b. Write a MATLAB program that will use the above result to plot
the isotherms on which ϕ assumes the values + (0, .1, .3, .5, .7, .9).
Try plotting in the space |x| ≤ 3, 0 ≤ y ≤ 3.
c. Write a MATLAB program that will use the same result to plot
the streamlines on which ψ assumes the values given above for ϕ,
except you will probably want to leave off positive values greater
than 0.3. to prevent bunching.
2. A piece of conducting sheet metal is bent into a 60° wedge as shown
in Figure 4.37 in the w plane.
On y = 0, 0 ≤ x < 1, the electric potential is maintained at ϕ = 1 volt,
on the locus y = 3 x the potential on the metal is also at 1 volt for a
distance of up to 1 unit from the origin, while on the remainder of
Harmonic Functions, Conformal Mapping, and Some Applications 183

2.5

φ=0
2

1.5
y

φ=1

0.5

φ=1 φ=0
60⁰
0
0 0.5 1 1.5 2 2.5 3
x

FIGURE 4.37
A piece of conducting sheet metal is bent into a 60° wedge.

the wedge the potential is maintained at zero. Find an expression for

the complex potential within the wedge and using MATLAB plot the
equipotentials and streamlines on separate plots.
Hint: Study Example 4.7 and use a similar argument, but be sure to
use polar coordinates in your code as described at the end of that
example. To make your plot more meaningful, include a heavy line
showing the line y = 3 x for 0 < x < 2. Some suggested values for the
equipotentials are ϕ = 0, .05, .1, .3, .5, .7, .9, 1 and for the streamlines
ψ = 0, ±.1, ±.3, ±.5.
3. Fluid flows downward into a corner as shown in Figure 4.38.
a. At x = 0+, y = 1, the flow is downward—it must be tangent to
the wall—and of magnitude one. As a complex vector this is the
velocity V = −i. The boundaries of flow on the positive x and
y axes must coincide with a streamline. Find the complex veloc-
ity potential Φ(z) in the corner, and write a MATLAB program
showing the streamlines of the flow. Show at least five stream-
lines and place arrows on them to show the direction of flow.
Which streamlines correspond to the boundary in the given
problem?
184 A MATLAB® Companion to Complex Variables

3.5

2.5

2
y

1.5

0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4
x

FIGURE 4.38
Fluid flow into a corner.

Hint: Find a simple mapping that will map the corner into the
real axis in the w plane with the first quadrant of the z plane
mapped onto the upper half of the w plane. Now solve the prob-
lem of flow above the streamline Im(w) = 0.
b. Find an expression for the complex velocity vector along the line
y = x, x ≥ 0.
4. A Quonset hut in the shape of a semicylindrical conducting metal
cylinder sits on top of a perfectly conducting earth. The hut is main-
tained at a potential of 1 volt, and the earth is at ground potential (zero
volts). The half cylinder has a radius of unity (in some set of units).
There is a tiny gap between the semicylinder and the ground plane
so that they can be maintained at different potentials (Figure 4.39).
a. Find an expression for the complex electric potential, in the
region Im(z) ≥ 0, |z| ≥ 1, z ≠ ±1 for this problem.
1
Hint: Consider the mapping w = ( z + 1/ z ) applied to the given
2
configuration. How is the boundary consisting of the semicircular
roof of the hut and the ground (which is y = 0, |x| > 1) mapped by
this function? Note that on the roof we can use z = eiθ, 0 < θ < π. Also,
notice that w′(z) = 0 if z = ±1. Having applied this transformation,
Harmonic Functions, Conformal Mapping, and Some Applications 185

1.5

φ=1
0.5
y

0
φ=0 φ=0

–0.5

–1
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 4.39
A metal hut on a grounded plane.

now refer to the problem solved for Figure 4.25. Use the equations

derived in that problem to find Φ(x, y) in the present problem.
b. Write a MATLAB program that will plot the equipotentials for
the given problem. They should include ϕ = 1 and ϕ = 0.
Hint: Use polar coordinates to set up the grid on which you will
evaluate ϕ. Using polar coordinates r and θ, let r range from 1 to
around 3. Assume there is no electric field inside the hut, so there is
no reason to consider r less than unity. Let θ range from a number a
little less than zero to one slightly more than π. If you simply use 0
and π for the ends of the interval, then the command contour will
not have enough information to plot the contour on which ϕ = 0.
c. Using a similar program, plot the streamlines, considering both
positive and negative values of ψ.
5. A semi-infinite conducting plate is maintained at 1 volt as shown in
the z plane in Figure 4.40.
Its cross section occupies the line y = 1, −∞ < x ≤ 0. It is one unit
above a grounded conducting plate that is kept at zero volt potential.
This plane has a cross section occupying the whole x axis. Our goal
is to use MATLAB to plot the streamlines and equipotentials created
by the conductors.
186 A MATLAB® Companion to Complex Variables

1.5

φ=1
1

0.5
y

φ=0
0

–0.5

–1
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 4.40
An infinite plate and a semi-infinite one.

1
a. Consider the transformation z =
π
( w + 1 + Log ( w )), w = u + iv
applied to the line v = 0, 0 < u < ∞ in the w plane. What is the
image of this line in the z plane?
b. Under the same transformation, find the image of the line v = 0,
−∞ < u < 0 in the z plane. Be sure to prove that the most positive
value achieved by Re(z) on the image in the z plane is 0.
c. Assigning the values 0 and 1, respectively, to the potential ϕ(u, v)
on the line segments, in the w plane, in parts (a) and (b), show that
−i
the function in the upper half space Im(w) ≥ 0 of Φ ( w ) = Log w
π
will meet the required boundary condition.
d. We cannot solve the equation given in part (a), for w as a function
of z, in terms of elementary functions. Thus, we cannot neatly
obtain Φ(w(z)) for use in the z plane. However, there is a way to
use MATLAB to display both the equipotentials and streamlines
without using the contour function in MATLAB. Suppose that
Φ0 = ϕ0 + iψ0, where ϕ0 and ψ0 are specific real values assumed
by the potential and stream functions. Show that the value
of z where this value of complex potential occurs, is given by
1 i 1
z = −ψ 0 + iφ0 + e − πψ 0 cos ( πφ0 ) + e − πψ 0 sin ( πφ0 ) + .
π π π
Harmonic Functions, Conformal Mapping, and Some Applications 187

Or,
1 πiΦ0 1

z = iΦ 0 + e + .
π π
e. We know, because of the values prescribed on the given bound-
aries, that all values of the voltage must satisfy 0 ≤ ϕ ≤ 1. Suppose
we were to choose a value for ϕ0 satisfying this inequality, say
ϕ0 = 1/2. From our expression
−i 1 i
Φ ( w ) = Log w = arg ( w ) − Log w = φ + iψ
π π π
−1
we can see that Log w = ψ can assume any real value in the w
π
plane. Thus, fixing the value for ϕ0 in our expression for z in part
(d) and allowing ψ to vary through a set of real values, we can
trace out the equipotential on which ϕ0 has the prescribed value ϕ0.
Write a MATLAB program that will plot the equipotentials on
1 1 3
which φ = 0, , , , 1 . Some experimentation might be required
4 2 4
to choose the range of values for ψ. For a start, let ψ vary between
–1 and 10. Do not make the lower limit on ψ become very nega-
tive as the expression for z in part (d) shows that the magnitude
of z becomes exponentially unbounded as ψ  → −∞, which cre-
ates problems in plotting. Write your program in such a way that
each equipotential is labeled with the numerical value of ϕ on that
curve.
f. Write a MATLAB program that will generate the streamlines
for the given configuration. In electrostatics, these lines are tan-
gent to the electric field. To generate the lines, realize that on
each streamline ψ has a fixed real value (positive or negative),
while ϕ varies continuously between 0 and 1. Place arrows on the
streamlines so that they point in the direction of the electric field
(i.e., from higher to lower potential).
Suggestion: It should not be necessary to consider values of ψ
more negative than about –.5.
6. a. A Neumann problem—a continuation of problem 5 is as follows.
The transformation described in problem 5, part (a) can be used
to solve a problem arising in fluid mechanics. A fluid is guided
between two planes as shown in Figure 4.41.
The upper plane ends at x = 0, y = 1 allowing the fluid to flow
upward as well as from left to right. In the channel, on the far
left, we want the flow between the two plates to be uniform and
to the right. Thus, we say that the complex velocity between the
plates is V = 1 + i0 in the limit x → −∞, 0 < y < 1. The two planes,
the infinite one and the semi-infinite one, must coincide with
188 A MATLAB® Companion to Complex Variables

1.5

0.5
y

–0.5

–1
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 4.41
Fluid flow between two planes.

streamlines of the flow in the z space. If the velocity potential in

the image space is given by Φ(w), then the lines v = 0, u > 0 and
v = 0, u < 0 (the images of the original z space streamlines) must
also be streamlines.
Show that the complex potential Φ(w) = A Log w will satisfy
the required condition for the streamlines, in the w plane, if A
is real. What is the complex velocity vector V(w) for this velocity
potential? Answer in terms of A.
b. Show for the transformation given in part (a), in problem 5, that
if |w| → 0+, that Re(z) → −∞.
It is helpful here to switch to the polar representation w = ρeiα.
1
Explain why for the above limit we have z → Log ( w ) .
π
c. Show that as Re(z) → −∞, we have for the complex potential in
this limit, that Φ = Aπz. We wish to obtain uniform flow between
the plates, as x → −∞, which means that V = 1 + i0. Prove that this
implies that A = 1/π. Explain why the streamlines in this prob-
lem have the same shape as the equipotentials in the previous
one, and vice versa.
dz
d. With A = 1/π, show that = 1 + e πΦ and that the complex veloc-
dΦ
ity vector is given in general in the z plane by
 1  1
V ( z) =  πΦ( w( z )) 
=
 1+ e  1 + w ( z)
Harmonic Functions, Conformal Mapping, and Some Applications 189

e. Write a MATLAB program that asks you to first supply the

­coordinates of a point in the z plane and that will then give you
the complex velocity vector at that point. Use your program
to  find the fluid velocity at these coordinates (0,1/2), (−1,1/2),
(0,1/4).
1
Hint: From our transformation z = ( w + 1 + Log ( w )) , we can
π
use the solve function of MATLAB to obtain w for any given
value of z. Now use the result in part (c).
If z = –5 + i/2, we are between the guiding planes and far from
where the upper plane vanishes. The flow should be toward the
right and approximately equal to V = 1 + i0.
How well does your result and this asymptotic value agree?
Use long format.
7. a. For the following problem, it is convenient to take the results of
Example 4.9 and transform them into the w plane. Thus, we have
a line charge of strength A placed a distance h above a surface at
zero potential whose cross section is the real axis in the w plane.
The resulting complex potential in the w plane is

w + ih w + ih w + ih
Φ ( w ) = A Log = A Log + iA arg for Im(w) ≥ 0
w − ih w − ih w − ih

Suppose we have two infinite conducting planes separated by a

distance π as shown in Figure 4.42. Each plane is at zero electro-
static potential, ϕ = 0, and the strength of the electric line source,
midway between the planes, will be taken as A = 1. Show that
the  transformation w = ez applied to the configuration in the
z plane yields a more familiar (see Example 4.9) geometry in the
w plane. Where is the line source now located in the w plane?
State the complex potential in the w plane.
b. Now transform your result of part (a) back into the z plane and
obtain the complex potential for that problem. Write a MATLAB
program, using contour, that will plot the equipotentials for
that problem with A = 1. Choose your equipotentials so as to
obtain attractive plots. Note that ϕ cannot be negative because
the boundary potentials are zero. For the equipotentials, plot
values of ϕ between zero and 1. Verify that the contours for zero
coincide with the bounding planes.
c. Repeat the above but plot the streamlines. Place arrows on them
to indicate the direction of the electric field, recalling that stream-
lines leave from a positive charge.
Suggestion: Verify that the function for the complex potential has
a branch cut along the line y = π/2, x ≤ 0. The imaginary part
190 A MATLAB® Companion to Complex Variables

3.5 φ=0
π
3

2.5

1.5 Line charge

y

0.5
φ=0
0

–0.5

–1
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 4.42
A line charge between two planes at zero potential.

of this function will be discontinuous as we cross this line. We

must avoid values of z in a band containing this line—otherwise,
the command contour will produce erroneous curves. The line
of code

phim(abs(y−pi/2) < del & x < 0) = nan

can be helpful. Here phim is the imaginary part of the complex

potential, namely, ψ, while del is the width of a strip surround-
ing the branch cut. A good value here is .02. Values of ψ on the
streamlines should range between –3 and 3.
8. a. In this problem, we have a line of charge one unit away from
the edge of a semi-infinite conducting plane set at zero potential
(ϕ = 0) as shown in Figure 4.43.
Note that this half plane can be regarded as the limit as α →
0+ for the configuration shown in Figure 4.44, which employs a
wedge of angle 2α.
To begin, let us consider a problem in which there is a
grounded plane set at potential ϕ = 0 whose cross section coin-
cides with the entire imaginary axis in the w plane.
Harmonic Functions, Conformal Mapping, and Some Applications 191

1.5

0.5
φ=0
0 Line charge
y

–0.5

–1

–1.5

–2
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 4.43
A line charge and a half plane at zero potential.

1.5

0.5

α
0
y

–0.5

–1

–1.5

–2
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x

FIGURE 4.44
The half plane is this wedge as α → 0+.
192 A MATLAB® Companion to Complex Variables

A line charge of strength A > 0 is placed a distance h to the right

of this plane, and the charge passes through u = h, v = 0. Use the
results of Example 4.9 to argue that the resulting complex poten-
tial in the w plane for Re(w) ≥ 0 is

w+h w+h w+h

Φ ( w ) = A Log = A Log + iA arg
w−h w−h w−h

Hint: Do a simple transformation that will rotate the axes from
Example 4.9.
1
Log z
b. Consider the transformation w = z1/2 = e 2 applied to the
wedge shape (having angle 2α) shown in part (a). How is
the wedge mapped by this, and how is the domain outside
the wedge mapped by this transformation?
c. Combine the result above and that given in part (a) to show that
with A = 1 the complex potential for the given problem can be
z1/2 + 1
obtained by taking α → 0+ and results in Φ ( z ) = Log 1/2 .
z −1
Note for the next two parts that this function has a branch cut
along the negative real axis in the z plane because of the function
z1/2. It will also have a branch cut wherever the argument of the
z1/2 + 1
Log, namely, 1/2 , is negative real. This is where y = 0, −1 < x
z −1
< 1, which partly coincides with the first cut. In what follows, use
the NaN command for these cuts where needed.
d. Using MATLAB, plot the equipotentials for the line charge/­
conducting plane configuration.
e. In the same way, plot the streamlines, adding arrows so that they
point outward from the line charge.
9. A cylinder of unit radius, whose axis passes through the origin and
is perpendicular to the z plane, is maintained at a temperature of
T°. A similar cylinder of unit radius has an axis that passes through
x = 3, y = 0 and is maintained at 0°. Our goal is to find the temper-
ature anywhere outside the cylinders and to use MATLAB to plot
the isotherms and streamlines in the material. The configuration is
­presented in Figure 4.45.
a. A bilinear transformation with real coefficients must be found
that will map the circle (the cross section of the left-hand first
­cylinder) into a circle of unit radius, in the w plane, where the
points z = 1 and z = –1 are mapped into w = 1 and w = –1, respec-
tively. With the same transformation, the second cylinder is
mapped into a cylinder in such a way that z = 4 is mapped into
Harmonic Functions, Conformal Mapping, and Some Applications 193

2.5
2
1.5
1
T° 0°
0.5
0
y

–0.5
–1
–1.5
–2
–2.5
–2 –1 0 1 2 3 4 5
x

FIGURE 4.45
Two cylinders at different temperatures.

w = –R and z = 2 is mapped into w = R, where R is real and must

be determined. The two cylinders in the w plane are thus coaxial.
Using the invariance of the cross-ratio (see Section 4.2) and the
solve function of MATLAB, find the numerical value of R and
verify that it is positive. See Example 4.2, Section 4.2.1. Because
R results from the solution of a quadratic equation, MATLAB
should produce two numbers. It is not hard to show that their
product must be 1 and that they are both positive. Verify that
your value of R is positive and real. (You might also want to ver-
ify that had you mapped z = 4 into w = R and z = 2 into w = –R
that R would be negative and of no use.)
b. Using the more positive of the two values of R found above, and
the solve function of MATLAB, show that the bilinear trans-
formation that maps the two given cylinders (which are in the
z plane) into cylinders of radius 1 and R in the w plane is given
by –(4∗R∗z − 11∗R + R^2∗z + R^2)/(z − 11∗R + 4∗R∗z + 1).
Check your result by using this transformation to map each
of the given circles into the w plane and verify that the resulting
curves are the desired circles.
c. To solve our problem in the w plane, where the geometry is
now simpler, observe that the real part of the complex poten-
tial A Log(R/w) is A Log(R/|w|) (here A is real). This f­unction
is ­harmonic in the w plane and will vanish, as required where
194 A MATLAB® Companion to Complex Variables

|w|  =  R. Note  this circle is the cross section of the image of

the given cylinder at zero potential. Prove that if A = T/Log(R),
the requirements on the potential of the cylinder of unit radius,
in the w plane, are satisfied (i.e., it equals T). Now plot the con-
tour lines of the real part of the known complex potential, in the
z plane, by using the bilinear expansion that we have derived
above. These describe the temperature between the cylinders.
Take T = 1.
Realize that all the contours must show values satisfying 0 ≤ T
≤ 1, where T is the temperature. This follows from the maximum
and minimum principles in complex variable theory.
Be sure to plot the contours on which T = 1 and T = 0, and
verify that they coincide with the cross sections of the given cyl-
inders in the z plane.
d. Plot the contour lines of the stream function for the temperature
assumed above. These lines are tangent to the direction of heat
flow. On these contours, ψ can have both positive and negative
values, and zero. Try ψ = ±(0, .3, .5, .7, 1, 1.5). Add arrows to the
figure to show heat moving from the hotter to the colder cyl-
inder. Realize that you will be plotting a function of the form
 az + b 
ψ = Im  Log  , where the argument is a bilinear transfor-
 cz + d 
mation. This function will display discontinuities across the
branch cut of the function. They can be eliminated from your plot
if you include NaN statements for ψ in a narrow strip c­ ontaining
az + b
values where is zero or negative real.
cz + d
You do not want your plot to show streamlines inside the
circles |z| = 1 and |z − 3| = 1, as they have no physical mean-
ing and will only confuse the picture. These can be eliminated if
you include a statement like psi(abs(z)<1|abs(z-3)<1)=nan;.
Here psi is the imaginary part of the complex potential. To make
your plot more meaningful, plot circles indicating the cross sec-
tion of the cylinders.
5
Polynomials, Roots, the Principle of
the Argument, and Nyquist Stability

5.1 Introduction
The question of where an equation like f(z) = 0 is satisfied in the complex
plane, when f(z) is an analytic function, is a major subject in complex variable
theory. This problem often arises in engineering and scientific applications.
A value of z for which the equation is satisfied is said to be a “zero” of f(z).
Similarly, we must pay attention to the related problem of where g(z) = 1/f(z)
possesses pole singularities because of the vanishing of f(z). Some of these
questions are directly resolved in MATLAB®, while others require the more
sophisticated technique called the principle of the argument.
The Nyquist stability criterion, based directly on the principle of the argu-
ment, which we discuss, is a favorite technique of engineers; it is accessed
directly in a MATLAB toolbox.

5.2  The Fundamental Theorem of Algebra

One of the great pleasures of studying functions of a complex variable is to
see how easily one can prove the fundamental theorem of algebra (see W
section 4.6 and S problem 5.10). Here is a way to state the theorem: We have
a polynomial

P(z) = anzn + an–1zn–1 + … a0 (5.1)

in the variable z, where a0, a1, … are constants (real and/or complex); n is a
positive integer; and an ≠ 0. We say that the polynomial is of degree n, although
some authors prefer the term order instead of degree. The theorem states that

P(z) = 0 (5.2)

195
196 A MATLAB® Companion to Complex Variables

has exactly n solutions, or roots, in the complex z plane.* Equivalently, the

fundamental theorem can be stated that if z1, z2, … zn are the roots, then we
can factor the left side of Equation 5.1 so that

P(z) = an(z − z1) (z − z2) … (z − zn) (5.3)

In applying the fundamental theorem of algebra, we may have to count a

root more than once. For example, the equation z3 + (2 − i)z2 + z(1 − 2i) − i = 0
has roots at z = i and z = −1. If you try any other values of z, you will find that
the equation is not satisfied. We seem to have two roots, not the expected
n = 3 from our theorem. But if we factor the left side, we get (z + 1)2 (z − i) =
z3 + (2 − i)z2 + z(1 − 2i) − i = 0. The root at –1 is of multiplicity two—this fol-
lows from the exponent in (z + 1)2.
If we combine terms corresponding to repeated roots in Equation 5.3, any
term such as (z − p)N is said to contribute a root of order N or multiplicity N.
Thus, we assign N roots at p when applying the fundamental theorem.

5.2.1  MATLAB roots

To explore the fundamental theorem in MATLAB, it is easiest to use the func-
tion roots. The reader may wish to review the documentation for this function
using the MATLAB help. Briefly, to find the roots of a polynomial expression
like Equation 5.1, you must begin by creating a row vector whose elements are
the coefficients of the various powers of z listed in descending order of the associ-
ated powers. The first element would be the number an. If any power of z that
is less than n does not appear in the polynomial, then you must enter a zero
for that coefficient in the vector. If a polynomial has a root of multiplicity n, the
output of roots will show that root n times. To find the roots of the polynomial
z4 + z3 + 2z + 3, we proceed as follows in MATLAB:

a = [1 1 0 2 3]
>> roots(a)
ans =
0.7140 + 1.1370i
0.7140 − 1.1370i
−1.2140 + 0.4366i
−1.2140 − 0.4366i

Note the presence of the zero, the middle element, in the vector a. It arises
because there is no z2 term in the given polynomial. As the theorem pre-
dicted, we obtain four roots. Observe that because the coefficients in the

* In some texts, it is stated that the equation has one root. But the existence of the other roots
can be proven from this through a procedure of successive division of the polynomial on the
left in Equation 5.1 by (z – zj), where zj is any of the roots. See W section 4.6, problem 18.
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 197

polynomial are real, complex roots must appear in conjugate pairs as is the
case here.
We can use roots when the coefficients in the polynomial are complex. Let
us look at our first case from above: z3 + (2 − i)z2 + z(1 − 2i) − i = 0. Proceeding
as before, we have

a = [1 (2 − i) (1 – 2∗i) −i]
a = 1.0000  2.0000 − 1.0000i  1.0000 − 2.0000i  0 − 1.0000i
>> roots(a)
ans =
−0.0000 + 1.0000i
−1.0000 − 0.0000i
−1.0000 + 0.0000i

Note that −1.0000 − 0.0000i and −1.0000 + 0.0000i are numerically the same,
and so we really have only two numerically distinct roots. The root at –1 is of
multiplicity 2 and thus is printed twice. However, if you convert our above
answer to long format you will get

>> format long

>> ans
ans =
–0.000000000000001 + 1.000000000000000i
–1.000000016716123 – 0.000000014686955i
–0.999999983283878 + 0.000000014686955i

What we see here is round-off error, leading to the erroneous conclusion

that there are three different roots. Ideally the second and third elements
in this vector should be identical and equal to –1 + 0i. There are sometimes
advantages to using the MATLAB short format and, of course, disadvantages.
Comparing the right side of Equation 5.1 with the right side of Equation 5.3,
we have that

anzn + an–1zn–1 + … a0 = an(z – z1)(z – z2) … (z – zn) (5.4)

On the left side of the preceding, the only term not containing the vari-
able z is a0, while multiplying out the factors on the right shows that the
only  term not containing z is an(–z1)(–z2) … (–zn) = an(–1)nz1z2 … zn. Thus,
a0 = (–1)n anz1z2 … zn and so
a0
(−1)n = z1 z2 ... zn (5.5)
an
This gives us a simple way to check the roots obtained for a polynomial.
The product of the roots for the polynomial on the right side of Equation 5.1 is
the ratio of the last coefficient to the first, times plus or minus one, depending
on the degree of the polynomial. We can say, for example, that the product
198 A MATLAB® Companion to Complex Variables

of the roots of the polynomial 2z3 + z2 + i is (−1)3i/2 = −i/2, which the reader
should see is the case.
Suppose we were to multiply out all the factors on the right side in Equation
5.3. The root −z1 will multiply z in each of the other terms. There are n – 1
of these terms. Thus, −z1 will appear on the right side multiplied by zn–1.
Similarly, −z2 will be multiplied by the same factor, as will the other roots.
The upshot is that the product of the terms in the parentheses on the right
side of Equation 5.3 will contain the expression (−z1 −z2 … −zn)zn–1. The vari-
able zn–1 does not appear in any other way. The right side of Equation 5.3
when expanded contains an(−z1 −z2 … −zn)zn–1 as the only term containing
zn–1. The left side contains only an–1zn–1 with this same power of z. Equating
these expressions gives us another check on the sum of the roots of the poly-
− an−1
nomial: the sum of the roots must equal . Thus,
an

( z1 + z2 + ... + zn ) = − aan−1 (5.6)

n

The reader immediately should see that the sum of the roots of
2z3 + iz2 + 17 is –i/2, while the sum of the roots of z4 + z + i is zero because
there is no z3.
Aside from these above checks, it is interesting to verify that roots
works correctly by using it to give the roots of a large polynomial for
which the roots can be found by some other method. This is illustrated in
Example 5.1.

Example 5.1
1. Using roots, find all the roots of the polynomial 1 + z + z2 + … z9.
Use long format. Notice that there is a negative real root that is
available by inspection.
2. Using MATLAB, find the sum of the roots just obtained.
Confirm that the result is correct through the use of the coef-
ficients in the polynomial.
3. In a similar way, find the product of the roots obtained in (1),
and confirm that the sum is correct by using the coefficients in
the polynomial.
4. Check the values of your roots by using a simple formula equal
to this finite series and finding where this expression is zero.
How well do the results agree with part (1)?

Solution:
Here is the code for parts (1) through (3):

%parts 1-3
coeffs=ones(1,10);
('answer to part 1')
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 199

the_roots=roots(coeffs)
('answer to part 2')
sum(the_roots)
('this should equal ratio given in eq(5.6)')
('answer to part 3')
prod(the_roots)
('this should agree with eq(5.5)')

Here is our output:

ans = answer to part 1
the _ roots =
0.809016994374947 + 0.587785252292473i
0.809016994374947 − 0.587785252292473i
0.309016994374947 + 0.951056516295154i
0.309016994374947 − 0.951056516295154i
−0.999999999999999 + 0.000000000000000i
−0.809016994374948 + 0.587785252292473i
−0.809016994374948 − 0.587785252292473i
−0.309016994374948 + 0.951056516295154i
−0.309016994374948 − 0.951056516295154i
ans = answer to part 2
ans = −1.000000000000001
ans = this should equal ratio given in eq(5.6)
ans = answer to part 3
ans = −1.000000000000000
ans = this should agree with eq(5.5)

Comments:
It should be obvious that the given polynomial has a root at –1. Try pair-
ing the first and second terms, the third and fourth, and so on, and you
will see that they cancel at –1.
Notice that −0.999999999999999 in the vector giving the roots should
ideally have come out as –1, the value of the root.
The sum of the roots should be –a8/a9 = –1, while our result here is
–1.000000000000001, which is close.
9 a
The product of the roots should be ( −1) 0 = −1, which our program
a1
indeed produced.

Recall the finite series and its sum (see W section  5.2 and S
problem 2.41):

1 − z n+ 1
1 + z + z 2 + z 3 + ... z n = , n ≥ 0, z ≠ 1 (5.7)
1− z
The fact that we cannot use the formula on the right for z = 1
is of no consequence, because this value is clearly not a root of
the polynomial. Equation 5.7 can be verified by our multiplying
both sides by 1 – z.
200 A MATLAB® Companion to Complex Variables

The roots of the polynomial on the left must coincide with the solu-
tions of zn+1 = 1 (excluding the solution z = 1). We recall (see W section 1.4,
and S section 1.13) that roots of the preceding are given by z = ei2πk/(n+1)
k = 1, 2, … n. Notice that we do not have k = 0 here, otherwise we would
get the unwanted z = 1.
To treat the present problem, we take n = 9 in Equation 5.7 and in our
formula for z.
The numerical values of z are given by the following code and result.
We also sum these values.
k=[1:9];
format long
new_roots=exp(i*2*k*pi/10).
new_sum_roots=sum(new_roots)

The output is
for part 4
new _ roots =
0.809016994374947 + 0.587785252292473i
0.309016994374947 + 0.951056516295154i
−0.309016994374947 + 0.951056516295154i
−0.809016994374947 + 0.587785252292473i
−1.000000000000000 + 0.000000000000000i
−0.809016994374947 − 0.587785252292473i
−0.309016994374948 − 0.951056516295154i
0.309016994374947 − 0.951056516295154i
0.809016994374947 − 0.587785252292473i
new _ sum _ roots = −1.000000000000000

Notice the excellent—but not perfect—agreement between the values

obtained from roots and those obtained by our beginning with Equation 5.7.
The sum of the roots here is exactly –1, unlike in the previous calculation.
Incidentally, if you have the roots of a polynomial but not the polynomial
itself, MATLAB will allow you to create the polynomial by using the com-
mand poly. This is discussed in problem 6.

5.2.2  Use of Zeros

Suppose you wish to find the roots of a polynomial of a high degree n in
which many of the powers of z less than n do not appear (i.e., their coeffi-
cients in Equation 5.1 are zero). Rather than key all these zero values into the
row vector representing the coefficients of the polynomial, we can begin by
assigning all the coefficients the value zero and then correct the few excep-
tions. The use of zeros is given in this example where we solve (1 − i)z10 +
(1 + i)z3 + 2 = 0. We proceed as follows:
a=zeros(1,11);
%above creates a row vector of zeros;
%there are 11 elements in the row vector
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 201

%because of powers 0 thru 10

a(1,1)=(1-i);%coeff of z^10
a(1,11)=2; %coeff of z^0
a(1,8)=(1+i)%coeff of z^3
syms z
check=poly2sym(a,z)
% above checks that we have correct polynomial
rts=roots(a)%gives the roots

The above yields the required roots. Notice that if we have a polynomial in
which all or many of the coefficients are 1, we can do a similar trick: setting
the value of a by using the command ones instead of zeros. This was done
in Example 5.1.

5.2.3  Roots versus Factor

For some purposes, the MATLAB command factor is more useful than roots
in factoring a polynomial and revealing its roots. For example, to factor
z4 − 6z3 + 8z2 + 6z − 9 we write
>> syms z
>> factor(z^4−6∗z^3+8∗z^2+6∗z−9)

which yields
ans = (z − 1)∗(z + 1)∗(z − 3)^2

This clearly shows roots at 1, –1, and 3, the latter being a root of multiplic-
ity 2. The terms in the polynomial do not have to be entered into factor in
descending powers of z.
Had we sought the roots of the polynomial by using roots, and if we were
working in long format, we would have obtained this less satisfactory result:

>> roots([1 −6 8 6 −9])

ans =
3.000000086034111
2.999999913965896
−0.999999999999999
0.999999999999999
Because of rounding error, we do not clearly see that there are roots at
3, 1, and –1. In fact, roots seems to show that there are four different roots.
However, there are a number of reasons not to use factor. This function,
­factor, will factor polynomials into terms containing ( z − zk ) k, where zk is a
n

root and nk its multiplicity only when zk is rational (which of course means
it is real as well). The limitations of factor are illustrated in these examples:

>> syms z
>> factor(z^2−4)
202 A MATLAB® Companion to Complex Variables

ans = (z − 2, z + 2)
% the above is done successfully
>> factor(z^2−5)
ans = z^2 − 5
% the above factorization is not done as it would involve sqrt
(5), an irrational.
>> factor(z^2+4)
ans = z^2 + 4
he above factorization is not done because it would involve
%t
i2 which is not real and thus not rational
>> factor(z^3−z^2+4∗z−4)
ans = (z − 1, z^2 + 4)
% 
the above factorization is only partially successful. The
factors involving +i2 are not present.

5.2.4  Derivatives of Polynomials and the Function Roots

It is easy to show that if a polynomial has a zero of multiplicity 1 at zk, then
the derivative of this polynomial cannot have a zero there. A polynomial
P(z) with a zero of multiplicity one at zk can be written as P(z) = an (z − zk)
Q(z), where Q(z) is a polynomial of degree n – 1 such that Q(zk) ≠ 0. This
­follows  from Equation 5.3. The first derivative of this expression is P′(z) =
anQ(z) + an(z − zk)Q′(z), which does not vanish at zk.
Suppose the polynomial has a zero of multiplicity 2 at zk. Then we easily
show here that the first derivative of the polynomial will vanish at zk, while
the second will not. The polynomial assumes the form P(z) = an(z − zk)2Q(z),
where Q(z) is a polynomial of degree n – 2 that does not vanish at zk. We
have P′(z) = 2an(z − zk)Q(z) + an(z − zk)2Q′(z), which is zero when z = zk. And
P″(z) = 2anQ(z) + 4an(z − zk)Q′(z) + an(z − zk)2Q″(z), which does not vanish at
zk. In general, a polynomial with a zero of multiplicity m at some point will
vanish at that point as does its derivatives up to m – 1. The mth derivative
of the polynomial will not vanish at the point. We can investigate this with
MATLAB, roots, and diff as follows.
Before embarking on that, we study here two MATLAB functions sym-
2poly and poly2sym, which are useful when we are dealing with polynomi-
als. You may wish to read the documentation on them, although you may
recall that we used poly2sym in our discussion of row vectors whose ele-
ments are the coefficients of a symbolic variable. The numbers appear in
descending order and are thus suitable for use in the command roots. Here
is an example:

>> syms w z
>> w=−6∗i∗z^3+2∗z^2+1;
>> sym2poly(w)
ans = 0 − 6.0000i  2.0000  0  1.0000
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 203

We could easily now apply the command roots to the above answer and
obtain the roots of the given polynomial. Thus,

>> roots(ans)
ans =
0.0000 + 0.4587i
−0.4544 − 0.3960i
0.4544 − 0.3960i

The command poly2sym takes you from the row vector whose ele-
ments are the coefficients of a polynomial (in descending powers) to the
polynomial itself. The variable used in the polynomial (e.g., x or z) is
supplied by the user in calling poly2sym. Otherwise, x is employed by
default if you specify no variable. Here, we invoke poly2sum to recon-
struct the ­polynomial in z, used above, from the numerical values of the
coefficients.

>> poly2sym([ 0 – 6.0000i 2.0000 0 1.0000], z)

ans = z^3∗(–6∗i) + 2∗z^2 + 1

Thus, sym2poly and poly2sym undo each other.

To take the derivative of a polynomial, we can simply use the function
diff applied to the polynomial written with its variable (e.g., z), or apply the
function polyder to the row vector whose elements are the coefficients of the
polynomial, written in the usual descending order beginning with the coef-
ficient of the highest power.

Example 5.2
Consider the function

z4 − z3 + (−1 + i)z2 + (1 − 2i)z + i

For the above fourth-degree polynomial, use MATLAB to find not only
the roots and their multiplicity but also all the derivatives of this func-
tion, up to the fifth; the last should be zero.
Verify that where this function has a root of multiplicity n at a
point, the derivatives of the function of all orders up to n – 1 are zero
there  and  that  the function has a nonzero nth derivative at the same
point.

Solution:
syms w z
format short
a= [1 (-1) (-1+i) (1-2*i) i]
% note the use of parens in the above to avoid errors
204 A MATLAB® Companion to Complex Variables

Roots_of_poly=roots( a)
% above gives roots of given polynomial
k=1;
while k<6
('the order of the derivative')
k
a=polyder(a)
rts=roots(a)
A=poly2sym(a,z)
% the above is the polynomial in z
k=k+1;
end

This yields
.0000 + 0.0000i  −1.0000 + 0.0000i  −1.0000 +
a = 1
1.0000i  1.0000 − 2.0000i  0.0000 + 1.0000i
Roots_of_poly =
−1.3002 + 0.6248i
1.0000 − 0.0000i
1.0000 + 0.0000i
0.3002 − 0.6248i
ans = the order of the derivative
k = 1

a = 4
.0000 + 0.0000i  −3.0000 + 0.0000i  −2.0000 +
2.0000i  1.0000 − 2.0000i
rts =
−0.7699 + 0.3877i
1.0000 + 0.0000i
0.5199 − 0.3877i
A = z*(− 2 + 2*i) − 3*z^2 + 4*z^3 + 1 − 2*i
ans = the order of the derivative
k = 2

a = 12.0000 + 0.0000i  −6.0000 + 0.0000i  −2.0000 + 2.0000i

rts =
0.7562 − 0.1646i
−0.2562 + 0.1646i
A = 12*z^2 − 6*z − 2 + 2*i
ans = the order of the derivative
k = 3
a = 24  −6
rts = 0.2500
A = 24*z − 6
ans = the order of the derivative
k = 4
a = 24
rts = Empty matrix: 0−by−1
A = 24
ans = the order of the derivative
k = 5
a = 0
rts = Empty matrix: 0−by−1
A = 0
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 205

Comment:
We see that the given polynomial has roots of multiplicity 2 at z = 1,
a root of multiplicity 1 at z = –1.3002 + 0.6248i, and a root of multiplic-
ity 1 at .3002 − 0.6248i. After taking one derivative, we obtain a poly-
nomial with one root at 1, but now with roots at –0.7699 + 0.3877i and
0.5199 − 0.3877i.
Note that the new polynomial obtained by differentiating the
given  one does have two roots at a location where the first one did
not.  Now t­aking  the second derivative of the original polynomial,
we  obtain  a  polynomial with roots at 0.7562 − 0.1646i and −0.2562 +
0.1646i.
Taking the third derivative of the original polynomial, we obtain
a  polynomial that has just one root: at 0.2500. The fourth deriva-
tive of  our fourth-degree polynomial yields a constant, which has no
roots,  while  the fifth derivative (the first derivative of the constant)
is zero.

Exercises
1. Using MATLAB roots find all roots of (1 + i)z4 + 2z3 + iz2 − 3 = 0, and
verify that the number of roots is that predicted by the fundamental
theorem of algebra.
2. a. Using MATLAB roots, find all the roots of the polynomial
(1 − i)z10 + (1 + i)z9 + z + 2. Did you find 10 of them, as predicted
by the fundamental theorem of algebra?
b. Find the products of the roots in part (a), and show that they have
the value predicted by the coefficients in the given polynomial.
Use the MATLAB command prod.
c. Find the sum of the roots found in part (a), and verify that they
have the value predicted by the coefficients in the polynomial.
Use the MATLAB command sum.
3. In using roots to find the roots of a polynomial, we can be fooled
into accepting a wrong answer if we are not aware of the rounding-
off properties of MATLAB.
a. Do the following in the MATLAB “short” format. Consider
this product w = (z − a)(z − b) whose roots are obviously a
and b. Let a  = 10 −6 and b = 106. Using MATLAB, create this
expression for w. Employ the command syms to make w
and z symbolic variables. Now use the command expand so
that w appears as a polynomial of degree 2 in the variable z.
Find the coefficients in the polynomial as a row vector using
206 A MATLAB® Companion to Complex Variables

the  command s­ym2poly, and use that result to obtain the

roots of the polynomial. Observe that one of the roots of w is
obtained incorrectly.
b. Repeat the above using the “long” format, and observe that the
correct roots are obtained.
c. Using the long format, repeat part (a) but take a = 10−8, b = 108;
observe that one of the roots obtained is incorrect. In general,
we must expect serious round-off errors in MATLAB when we
are dealing with numbers whose ratios are of the order of 1016 or
more, which is the case here.
d. MATLAB can do the preceding problem correctly if we ask it to
use “variable precision arithmetic.” In a MATLAB text or using
the Help feature of MATLAB, read up on the commands vpa and
digits. Now repeat part (c) and obtain the correct result.
Hint: 32 digits should be more than sufficient.
4. a. Consider the polynomial w = 1 + z + z2 + … z23 + z25. Note that
every nonnegative power of z is represented up to the 25th
except for the 24th power. Using roots and taking advantage of
the ones command in MATLAB, find all the roots of this expres-
sion. You should not have to key in all 26 coefficients into your
program. Verify that there are 25 roots. Explain why the sum of
the roots is zero. Sum your roots using MATLAB and see what
you obtain.
b. Suppose you do have the term z24 present in the above. Find the
roots now. How different are they from those found in part (a)?
Explain why there must be a root at z = −1. Do your results con-
firm this?
c. Consider the polynomial w = 1 + z + iz25. There are only three
terms. Using roots and taking advantage of the zeros command
in MATLAB, find all the roots of this expression. You should not
have to write in all 26 coefficients into your program. Verify that
there are 25 roots.
5. a. Consider the polynomial w(z) = 1 + 2z + 3z2 + … + nzn−1, where
n ≥ 2. This is obviously the derivative of the series W(z) = 1  +
z + z2 + … + zn. Earlier, we summed this series into a closed-
form expression, valid when z ≠ 1. Using this sum, show that
nz n+ 1 − z n (n + 1) + 1
w( z) = for z ≠ 1, n ≥ 2. If you wish, you can dif-
( z − 1)2
ferentiate using the function diff in MATLAB.
b. The above result is puzzling as it suggests that our polyno-
mial for w(z), which is of degree n – 1, has roots where the
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 207

numerator in the expression for w(z), above, is zero—that is,

where nzn+1 − zn(n + 1) + 1 = 0 is satisfied. This would imply—
from the Fundamental Theorem of Algebra—that w(z) has n +
1 roots. This equation does indeed have n + 1 roots, but two of
them are at z = 1, the very spot where our formula for w(z) is
not valid.
To prove the location of the roots, show that the function
A(z) = nzn+1 − zn(n + 1) + 1 vanishes at z = 1, that its first deriva-
tive vanishes at the same place but its second does not. Thus,
A(z) must be of the form A(z) = (z − 1)2Q(z), where Q(1) ≠ 0. Do
these steps in MATLAB using the function diff, and to make
your evaluation at z = 1, the function subs. To summarize, we
have shown that the roots of w(z) = 1 + 2z + 3z2 + … nzn−1 are the
solutions of the equation A(z) = nzn+1 − zn(n + 1) + 1 = 0 except for
the two roots in A(z) at z = 1.
c. Find the roots of the polynomial w(z) = 1 + 2z + 3z2 + … 10z9
by using the roots function in MATLAB. Now compare these
roots to those obtained from the equation nzn+1 − zn(n + 1) + 1 = 0,
solved with MATLAB roots, where you must choose the correct
value of n and exclude the two roots at z = 1.
6. Suppose you have the roots of a polynomial. MATLAB allows
you to find the polynomial. The easiest but not the only way is
to use the command poly, which you might want to read up on.
Essentially, you create a row vector, say v, whose elements are the
roots of the polynomial, in any order. You must enter a repeated
root as many times as it appears in the polynomial. The command
poly(v) will then yield the polynomial. The output is a row vec-
tor whose elements represent the coefficients of the polynomial
in descending powers (just as you enter the elements for roots).
Note that poly is designed so that the polynomial produced is
such that an = 1. If ans is the row vector obtained for poly(v), then
the command poly2sym(ans, z) will display the actual polyno-
mial having the prescribed roots, with z as the variable in the
polynomial, as in Equation 5.1, with an = 1. However, you will
first need the statement syms z to establish the symbolic nature
of z. Deleting  the z in poly2sym results in a polynomial in the
­variable x.
a. Using MATLAB, find the polynomial in the variable z that has
the two roots at 1 + i, one root at –i, three roots at the origin, and
one root at –4.
b. Apply the command sym2poly to the answer to part (a)
and  use  roots on that output to check your answer. Is there
agreement?
208 A MATLAB® Companion to Complex Variables

5.3  The Principle of the Argument

The principle of the argument is one of the most useful and elegant theo-
rems in complex variable theory (See W section 6.12 and S problems 5.16 and
5.17.). Suppose we have a function f(z) that is analytic on and inside a simple
closed contour C. We assume that f(z) ≠ 0 everywhere on C. Let us proceed
once around C in the positive direction (counterclockwise if C is just a circle),
beginning at any point on C. We call ΔC arg( f(z)) the increase in argument, or
angle (final value minus the initial value) of f(z) as we negotiate the contour.
Let N be the number of zeros of f(z) within C, where the zeros are counted
1
according to their multiplicities.* Then ∆ C arg f ( z) = N .
2π
If we permit f(z) to have poles inside C, but no other singularities (e.g.,
branch points, or essential singularities), but keep the other requirements,
then we may modify the above and have what is generally referred to as the
principle of the argument:

1
∆ C arg f ( z) = N − P (5.8)
2π

Here P is the total number of poles of f(z) inside C, where the poles are
counted according to their order. Thus, for example, if C were a circle |z| = 4,
( z + i)( z − 5)
and f ( z) = , then N = 1 because of the zero of order 1 at –i.
z( z − i)2 ( z − 12)
Also, P = 3 because of the pole of order 2 at z = i and the pole of order 1 at
z = 0. We ignore the zero at z = 5 and the pole at z = 12 because they are both
outside the contour.
In Figure 5.1 we have sketched a hypothetical contour C in the complex z
plane.
The arrow shows the direction (positive) in which we negotiate the contour.
In Figure  5.2, we show the locus assumed by w = f(z) as we go once
along the closed contour C in the z plane. In this case, the argument of f(z)
increased by 2π = ΔC arg  f(z). Notice that we encircled the origin of the w
plane exactly once in the positive sense. Thus in Figure 5.2, we show a situ-
ation in which N − P on the right in Equation 5.8 is one. This N − P is the
difference between the number of zeros and the number of poles of f(z)
within C. Observe that if f(z) had a zero on C, then the contour in Figure 5.2
would have passed through w = 0. And if f(z) had a pole on C, then the con-
tour in this same figure would have passed through infinity and could not
be drawn on the page.

* Recall that if an analytic function has a zero of order n at z 0, then it can be expressed as
f(z) = (z − z0)n ϕ(z) in a neighborhood of z0, where ϕ(z0) ≠ 0 and ϕ(z) is analytic at z0.
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 209

2.5

2
The contour C
1.5

0.5

z plane
0

–0.5

–1

–1.5

–2
0 1 2 3 4

FIGURE 5.1
A hypothetical simple closed contour.

1.5

0.5
arg f(z)
0 0
w plane
–0.5

–1

–1.5

–2
–1.5 –1 –0.5 0 0.5 1 1.5 2 2.5

FIGURE 5.2
The image of the preceding contour under the transformation w = f(z).
210 A MATLAB® Companion to Complex Variables

In Figure 5.3, we show another possibility.

Here, as we negotiate C in the z plane, the locus of w = f(z) takes us
twice around the origin in the w plane. Thus, 4π = ΔC arg f(z) and from
Equation 5.8 we have N − P = 2. In general, as the reader must have seen by
now, ΔC arg f(z)/(2π) is the number of encirclements in the positive sense that
w = f(z) makes of the origin in the w plane as z negotiates the contour C in the
z plane. Sometimes this number of encirclements will be negative as we will
see in a specific problem below where we look at f(z) = sin(πz)/z3. In general,
we may rewrite Equation 5.8 as

E = N − P (5.9)

where E is the total number of encirclements of the origin of the w plane

(where w = f(z)) as z encircles the origin in the positive sense on a contour C.
This number, of course, can be negative, positive, or zero.

Example 5.3
Part (a)
We know from the fundamental theorem of algebra that the polynomial
f(z) = z4 + iz3 + 4 + i has four roots. Using roots from MATLAB, we could
of course figure out where they are. Use the principle of the argument to
see how many roots lie inside the circle |z| = 3/2, and check your result
by using roots and establishing how far the roots are from the origin in
the z plane.

15

10

5
arg f(z)

0 0

–5

–10 w plane

–15
–10 –5 0 5 10 15 20

FIGURE 5.3
A contour going around the origin twice.
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 211

Solution:
Consider the following code:
theta=linspace(0,2*pi,1000);
%above establishes angles on a circle
z=3/2*exp(i*theta);
%above,are the points on a circle, radius 3/2,center origin
w=z.^4+i*z.^3+4+i;
plot(w)
coeffs=[1 i 0 0 (4+i)]
%above are coeffs of the polynomial we are given
the_roots=roots(coeffs)
abs_values=abs(the_roots)

Whose output is
coeffs = 1.0000  0 + 1.0000i  0  0  4.0000 + 1.0000i
the_roots =
1.0011 − 1.2337i
−0.8829 − 1.3673i
0.9136 + 0.8635i
−1.0318 + 0.7375i
abs_values =
1.5888
1.6276
1.2571
1.2683

These values are confirmed by the locus as presented in Figure 5.4.

10

0 0

–2

–4

–6

–8
–4 –2 0 2 4 6 8 10 12 14

FIGURE 5.4
3
Locus of f(z) = z4 + iz3 + 4 + i as it moves once around the circle z = .
2
212 A MATLAB® Companion to Complex Variables

We have placed arrows on the plot in the positive (counterclockwise)

direction and have placed a zero at the origin. We know the direction
because the given polynomial has no poles in the complex plane. The right
side in Equation 5.8 must be positive, and so the encirclement of the origin
must be in the positive direction.
The function roots produces, as expected, four distinct roots of z4 +
iz3 + 4 + i. Two of them have modulus less than 3/2 and are 1.2571 and
1.2683. Thus, the locus of w = f(z), as z moves once around |z| = 3/2,
must twice encircle the origin of the w plane, and it does, as shown in
Figure 5.4. Notice that there are two relatively small loops in the locus
that do not encircle the origin, and they can be ignored.

Part (b)
Using the principle of the argument, determine how many roots the
function in part (a) has in the domain |z − 1 −i| < 1.

Solution:
We can use the same code as was employed in part (a) but replace
z=3/2*exp(i*theta); with z=1+i+exp(i*theta);. This generates a
circle having center at z = 1 + i and radius 1. This is the boundary of the
given domain.
The locus generated in the w plane as we negotiate this circle while
evaluating f(z) is shown in Figure 5.5.

30

20

10

O
0

–10

–20

–30

–40
–50 –40 –30 –20 –10 0 10 20

FIGURE 5.5
Locus of f(z) for the contour |z–1–i| = 1.
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 213

Notice that we have encircled the origin of the w plane exactly once,
which indicates that f(z) = z4 + iz3 + 4 + i has exactly one root in the
domain |z − 1 − i| < 1. This is confirmed if we look at the list of roots of
the polynomial:

1.0011 − 1.2337i
−0.8829 − 1.3673i
0.9136 + 0.8635i
−1.0318 + 0.7375i
If you were to plot these in the complex plane, you would quickly see
that only 0.9136 + 0.8635i lies inside the given domain |z − 1 −i| < 1.

Let us now turn to a problem where the function studied has both zeros
sin πz
and poles. We look at f ( z) = . Notice that this function has a pole of
z3
order 2 at the origin. This is because the denominator has a zero of order
3 at z = 0, and the numerator has a zero of order 1 at the same point (see W
section 6.2). The order of the pole is the difference of these orders. In gen-
eral, sin πz has zeros of order 1 when z lies at any integer except zero, which
means that f(z) has zeros of order 1 at the same locations. Let us plot the locus
1
­ roceeds in the positive sense around the circle z − = 1. Notice
of f(z) as z p
2
that this circle will enclose the pole of order 2 at the origin and the zero of
order 1 at z = 1. No other zeros or poles of f(z) are enveloped. Thus, according
to Equations 5.8 and 5.9, if we plot w = f(z) as the circle is negotiated, then we
should go once around the origin in the w plane in the negative (clockwise)
sense. Here is the code and the output that confirms this (Figure 5.6):
theta=linspace(0,2*pi,1000);
%above establishes angle on a circle
z=.5+1*exp(i*theta);
%above, the points on a circle, radius 1, center .5
w=sin(pi*z)./z.^3;
plot(w);grid;hold on
w1=w(500)
w2=w(501)

The output is
w1 = 7.9987 + 0.1509i
w2 = 7.9987 − 0.1509i
Note that the contour approaches the origin but does not pass through
it. The arrow was assigned by our studying the results for w(500) and
w(501). The plot moves from having a positive imaginary part to a negative
one. We do, as expected, go once around the point w = 0 in the clockwise
direction.
At this point, the reader may be wondering if the principle of the argu-
ment is useful. So far we have used it to confirm that it establishes the zeros
214 A MATLAB® Companion to Complex Variables

10

0 0

–2

–4

–6

–8

–10
–12 –10 –8 –6 –4 –2 0 2 4 6 8

FIGURE 5.6
1
Image of z – = 1 under the given transformation.
2

and poles of functions whose zeros and poles were at known locations. But
consider the following situation. Suppose we want to find solutions of the
equation sin z = e−z. We might proceed with MATLAB as follows:
%solution of exp(-z)=sinz
syms z
solve(exp(-z)==sin(z))

The output is
ans = 0.5885327439818610774324520457029

The above result is entirely plausible. To realize this, we create the plots of
sin z and e−z, where z = x is real (Figure 5.7).
There is no point in making a plot for x < 0 as the exponential is greater
than one here while sinx ≤ 1. The intersection of the curves near the value
obtained through solve (approximately .588) should be evident. Notice too
that the curves will have an infinite number of intersections that grow
asymptotically closer to nπ, where n is positive integer, increasing without
bound. Thus, solve has found us just one of an infinite number of real solu-
tions. As for complex solutions, we still do not have a clue.
Our solution using solve employed the symbolic math toolbox in
MATLAB. This in turn is based on the software called MuPAD, which
went into use in September 2008. Before that, MATLAB used the MAPLE
software in its toolbox. If you use a version of MATLAB marketed before
the above date, you will obtain a different solution to this equation,
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 215

0.8

0.6
sinx
0.4
sin(x) and exp(–x)

0.2

–0.2

–0.4
exp(–x)
–0.6

–0.8

–1
0 2 4 6 8 10 12 14
x

FIGURE 5.7
Plots of sin x and e–x.

namely, −2.01+i∗2.70 (approximately).* It turns out that there are other com-

plex solutions; the conjugate of the one just given is also a solution as should
be obvious because e z = e z and sinz = sin z .
Thus, the principle of the argument allows us, in cases such as these, to
explore the complex plane for solutions of a given equation. In the present
instance, we would consider various closed curves and evaluate the change
in argument of e−z − sin z as we go around the curve. We can refine this pro-
cess to smaller curves and then seek a solution within a domain bounded by
the curve. In the final attempt to find a root, we again employ solve but sup-
ply it with a starting value that approximates the root being sought—a value
based on our findings through application of the principle of the argument.
Here is an example. Let us use the principle of the argument to see if there
is a solution to sin z − e−z = 0 near the point −5+5∗i. Any solution we find will
also have a conjugate that is a solution.
Here is some code that will help with our investigation.
z0=input('center of circle=')
r=input('radius of circle=')
theta=linspace(0,2*pi,1000);
z=z0+r*exp(i*theta);
w=sin(z)-exp(-z);
plot(w);grid ;hold on
plot(0,0,'*')

* See, for example, B. Hunt, R. Lipsman, and J. Rosenberg. A Guide to MATLAB for Beginners and
Experienced Users, 2nd ed. (Cambridge, 2007): 21. This book gave me the idea of discussing
this equation.
216 A MATLAB® Companion to Complex Variables

We supply the program with the center z0 and the radius r of a circle. The
code generates a plot of the locus of sin(z) − exp(−z) as the circle is negotiated.
Trying a circle of radius 1 with center at −5+5∗i, we find that the origin in the
w plane is enclosed. Shrinking the radius to one half, we find that the origin
is not enclosed. Trying a circle at −5.5+5.5∗i, with radius 1, in the z plane,
we again encircle the origin, and we also encircle the origin if the radius is
reduced to .5. You may wish to study this figure using the “zoom in” feature
of the MATLAB figure tools to make sure that the origin is indeed encircled.
To locate the root, we use the MATLAB command vpasolve that allows us to
make an initial guess for our solution. In this way, we know that we will get
a solution near where we are expecting one. The function vpasolve employs
the Newton–Raphson method and does occasionally fail—it might produce
a root but one that is not the closest one to our initial guess. An advantage
that vpasolve has over solve is that the latter does not permit an initial guess
as to the location of the solution, and it might produce a solution that is not
near to the one you are hunting for. If you want vpasolve to look for complex
solutions, begin by providing it with a complex value to begin hunting for
solutions. Here is the code for our present problem. The reader may wish to
read the MATLAB documentation for vpasolve.
syms z
vpasolve(exp(-z)==sin(z),z,-5+5*i)

Whose output is
ans =
−5
.1512194593898245955379661256296 +
5.8443612804879991094759668093818*i

This result is not far from the guess made available by the principle of the
argument.
For many or perhaps most purposes, it is more convenient to seek roots
inside a domain whose boundary is a rectangle with sides parallel to the x–y
axes. One reason is that we are seeking solutions in a Cartesian system, so it is
natural to use a Cartesian-type contour. The other, equally important, is that
the rectangular boundary has 90° corners, and we are interested in seeing
how these corners will be mapped. The following code can be used to enter a
rectangular contour bounding our domain. We supply the coordinates of the
center as well as the width and height of the rectangle. The values of z in the
row vector are such that we negotiate the contour in the positive direction. We
have chosen to try out our program by plotting the locus of w = sin z − sinh z
as z negotiates a rectangle, a square in this case, with center at the origin. The
rectangle has a width and height of 2. Here are the code and the results:

% use of rectangular contour, princ of the argument

cntr=input('real part center of rectangle')
cnti=input('imag part center of rectangle')
ht=input('height of rectangle')
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 217

wdth=input('width of rectangle')
x1
=linspace(-wdth/2,wdth/2,1000);​
x2=linspace (wdth/2,-wdth/2,1000);
y1=linspace(-ht/2,ht/2,1000); y2=linspace(ht/2,-ht/2,1000);
%t
he preceding places 1000 points on the 4 sides of
rectangle
zbot=x1-i*ht/2;%z on bottom side of rectangle;
ztop=x2+i*ht/2;%z on top of rectangle
zrt= wdth/2+i*y1;%z on right side of rectangle
zlft=-wdth/2+i*y2;%z on left side of rectangle.
[zbot zrt ztop zlft]+cntr+i*cnti;%this centers the rectangle
z=
%where desired
%enter the function w(z) below
w=sin(z)-sinh(z);
plot(w,'linewidth',2);grid;hold on
plot(0,0,'*')

The output is Figure 5.8 below.

The arrow has been added to the plot. Figure 5.8 indicates that the function
w(z) = sin z − sinh z has three zeros inside the given square in the z plane;
we encircled the origin three times. If we were to expand w(z) in a Maclaurin

0.8

0.6

0.4

0.2

–0.2

–0.4

–0.6

–0.8
–0.8 –0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8

FIGURE 5.8
Locus of sin z – sinh z as z negotiates a square of side length z with center at z = 0.
218 A MATLAB® Companion to Complex Variables

series, we would find that the leading term (try this) is −2z3/3! Thus, there
is a zero of order 3 at the origin, and this accounts for the three zeros of w(z)
which were found inside the given contour. Of course, this function might
have had other zeros in the contour, but our procedure shows that there are
no others. Notice here the four cusps in the curve where the locus suddenly
changes direction by 90°. These arise from the four corners of the square that
we are mapping using this analytic function. The angles of the square are
preserved because this is a conformal transformation.
Suppose we apply the above program to the function w(z) = (sin z − sinh z)2,
using the same square in the z plane. We would expect the locus of f(z) to encir-
cle the origin of the w plane six times. This is because the leading term in the
Maclaurin expansion of the new w(z) would be (−2z3/3!)2 = 4z6/(3!)2.
However, we obtain the plot as shown in Figure 5.9.
There is something puzzling about this result: We expected to find six
encirclements of the origin, but the figure shows only three. Another puz-
zling feature, which is a clue to explaining our result, is that the plot exhibits
only two corners of 90° and not the expected four.
Notice that w = f(z) = (sin z − sinh z)2 is an even function of z (it is the square
of an odd function). For every point on the square contour (in the z plane)
we employed, there is another point on the contour that is its negative. Both
these points are mapped into the same point in the w plane by the given
function. Thus, if we proceed just halfway around the given square, we gen-
erate the entire curve shown in Figure  5.9. Proceeding the rest of the way

0.8

0.6

0.4

0.2

–0.2

–0.4

–0.6

–0.8

–1
–0.4 –0.3 –0.2 –0.1 0 0.1 0.2 0.3 0.4

FIGURE 5.9
The locus of (sin z – sinh z)2 as z negotiates the square being discussed.
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 219

around causes us to revisit these same points in the w plane. This is why we

see only two right angles in Figure 5.9. We trace each of them out twice. We
go along the above curve twice and encircle the origin six times as we let z
move once around our given rectangle in the z plane.
Thus, we must be cautious in interpreting a contour. The matter can be
resolved if we generate a table of numerical values of the values of z that
we employ and list next to them the corresponding values of w. Of course,
in the problem just presented, we would not need to list all 4,000 values
used for z as well as the corresponding values of w. We could have used, for
­example, just 16 values in our list by adding on the following additional lines
of code:

x1=linspace(-wdth/2,wdth/2,4);x2=linspace(wdth/2,-wdth/2,4);
y1=linspace(-ht/2,ht/2,4); y2=linspace(ht/2,-ht/2,4);
zbot=x1-i*ht/2;%z on bottom side of rectangle;
ztop=x2+i*ht/2;%z on top of rectangle
zrt= wdth/2+i*y1;%z on right side of rectangle
zlft=-wdth/2+i*y2;%z on left side of rectangle.
z=[zbot zrt ztop zlft]+cntr+i*cnti;

w=(sin(z)-sinh(z)).^2;
('z and w')
disp([z.' w.'])

Adding this code shows in the program’s output that every value of w that
we generate is encountered twice. Thus, we negotiate the curve shown in the
preceding figure twice.
Suppose we had to investigate the locus of (sin z − sinh z)2 as z proceeds
around a rectangle whose center is displaced from the origin. Then it would
no longer be true that for every point on this rectangle, there is a point on
the rectangle that is its negative. Thus, the locus traced in the w plane would
not, unlike the preceding case, be covered twice. This matter is investigated
in problem 3.

5.3.1  Rouché’s Theorem

This theorem is a nice result, easily proven (see W section 6.12 and S prob-
lem 5.18), and is useful in determining the location of zeros of analytic func-
tions. The theorem refers to two functions f(z) and g(z), both of which are
analytic in and on a simple closed curve. We assume that on C we have
everywhere |f(z)| > |g(z)|. The theorem states that with these conditions
ΔC(arg(f(z) + g(z)) = ΔC arg(f(z)) from which it of course follows that the num-
ber of zeros (or roots) of f(z) + g(z) inside C is equal to the number of zeros
of f(z) inside C. Again we count zeros according to their multiplicities. Note
that the theorem does not state that the locations of the zeros of f(z) + g(z)
and f(z) are identical.
220 A MATLAB® Companion to Complex Variables

The theorem can provide a concise proof of the Fundamental Theorem of

Algebra. Here is a simple example solved with Rouché’s theorem.

Example 5.4
Show that the function w(z) = z4 + z3 + i has four zeros inside |z| = 1.5 and
no zeros inside |z| = .8.

Solution:
Let us first take f(z) = z 4 and g(z) = z3 + i. We take the contour C as
the circle |z| = 1.5. On this contour, |f(z)| = (1.5)4 = 5.0625. Also,
|g(z)|  =  |z3  +  i| ≤ |z3| + |i| = |(1.5)3| + |i| = 3.375 + 1 = 4.375, from
which we see that indeed |f(z)| > |g(z)| on the contour. It follows that
the number of zeros of z 4 inside the contour must be the same as the
number of zeros of z 4 + z3 + i. It is obvious that z4 has four zeros, all at
the origin. From this, it follows that z4 + z3 + i has four zeros inside the
contour |z| = 1.5. Of course, the locations of the zeros are different in
these two cases.
The preceding should give an indication of how to prove the
Fundamental Theorem of Algebra. We must argue that the polynomial
anzn + an−1zn−1 + … a0 has n roots in the complex plane. We take f(z) = anzn
and g(z) = an−1zn−1 + … a0. We then choose a circle, |z| = r centered at the
origin, whose radius is sufficiently large that |anzn| > |an−1zn−1 + … a0|.
One can always find such a circle—the reader should convince herself of
this. The number of roots f(z) inside the circle is obviously n. Invoking
Rouché’s theorem, we proved the theorem.
To return to the problem, we now take g(z) = z4 + z3 and f(z) = i. On the
contour |z| = .8 we have |g(z)| ≤ |z|3 + |z|4 = .83 + .84 = 0.9216. On the
same contour, we have |f(z)| = |i| = 1. Thus, it is clear that on the contour
we have |g(z)| < |f(z)|. This tells us that f(z) + g(z) = z4 + z3 + i has the
same number of zeros inside |z| = .8 as does the number i. Thus, there
are no zeros. We can confirm all of the above by applying the function
roots to the polynomial z4 + z3 + i. We have the following:

>> roots([1 1 0 0 i])

ans =
−1.2865 + 0.3197i
−0.6382 − 0.8183i
0.1834 + 0.8614i
0.7413 − 0.3628i
>> abs(ans)
ans =
1.3256
1.0378
0.8807
0.8253

The absolute values of all four roots exceed .8 and are less than 1.5, as
predicted.
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 221

Exercises
z3 + 1
1. a. Consider w( z) = . Identify the zeros and poles of this func-
z
tion, and give the orders of each.
b. Using MATLAB plot the locus of w(z) as z negotiates the circle
|z| = .9 once in the counterclockwise direction. Explain how
your result confirms the principle of the argument by referring
to part (a). Put an arrow on your locus to show the direction in
which it is negotiated.
c. Repeat part (b) but use as your locus |z| = 1.2.
2. a. Consider the function w(z) = z3, which obviously has a zero of
order 3 at the origin. Using MATLAB plot the locus assumed by
w(z) as z goes once around the circle |z| = 1. Notice that this
locus apparently encircles the origin in the w plane just once,
but the principle of the argument would suggest three encircle-
ments. Explain this apparent contradiction.
b. Repeat part (a) but use the circle |z − 1| = 3/2. Observe that your
locus does indeed encircle the origin three times as expected.
Explain why this differs from the result in part (a).
c. Repeat part (c) but use as your locus in the z plane the square
centered at the origin with sides parallel to the x and y axes. Each
side has length one. Did you get three encirclements?
3. Plot the locus of f(z) = (sin z − sinh z)2 as z goes, in the positive sense,
along the boundary of a domain bounded by a square. The length of
each side is 2. The sides are parallel to the x–y axes. The center of the
square is at x = .1, y = .1. How does your result confirm the conclusion
obtained from a Maclaurin expansion that f(z) has a zero of order 6
at z = 0, and how does it show that there are no other zeros for this
function in this domain? How many right angles are in your locus?
Explain why?
4. a. Using the principle of the argument and a MATLAB plot, show
that the equation sin(ez) = ez has no solution in the z plane inside
the domain |x| < 1/2, |y| < 1/2.
b. Repeat the above problem and show that there is no solution
inside the domain |x| < 1, |y| < 1. You will have to use the zoom
in feature of MATLAB to see that your locus does not encircle
the origin.
c. Use vpasolve to try to find a solution near the upper-right-hand
corner of the square domain (i.e., in the vicinity of x = 1, y = 1). Is
your solution outside the square, as it should be?
222 A MATLAB® Companion to Complex Variables

Log(1 + z)
5. a. Where are the zeros and poles of w = f ( z) = ? Give the
z2
location and the order. Note that this is the principal branch of
the logarithm.
b. Using MATLAB, plot the locus of this function as z goes around
the circle |z|= .9 in the positive sense. Be sure to place an arrow
on the plot showing the direction in which it is negotiated.
Explain how the number of encirclements of the origin confirms
the principle of the argument.
Log(1 + z)
c. The function w = f ( z) = has a removable singularity at
z
z = 0. How should this function be defined at this point if  the
function is to be analytic in a neighborhood of the origin?
d. Repeat part (b) for the function given in (c). Discuss the number
of encirclements of the origin.
e. The function w = f(z) = Log z has a zero of order 1 at z = 1. Yet a
plot generated by evaluating f(z) as z proceeds once around the
circle |z|= 2 will show no encirclements of the origin in the w
plane. Try this. Why is there no encirclement?
6. a. Using the principle of the argument, show that the equation
sin(i  cos(z)) = ez has just one root inside the square domain
−1 < x < 1, −1 < y < 1.
b. Does this root lie in the left half of this domain, the right half,
or on the imaginary axis? Answer using the principle of the
argument.
c. Using solve or vpasolve, find the location of this root. You may
have to experiment to find the root we are seeking. Did you find
other roots?
7. a. Show that on the circle |z|= 1.5 |z4| > |ez|.
b. From the above, state how many solutions the equation z4 + ez
has inside this circle. How many roots does z4 − ez have inside the
same circle?
c. Plot the locus of z4 + ez as z goes around the given circle.
Also plot z4 − ez. How do your results confirm your findings of
part b?
d. A plot of z4 as z goes around the circle once seems to show a
single encirclement of the origin. Does this contradict what you
have found in part (b)?
e. Using solve, find four solutions of z4 + ez = 0 inside the given cir-
cle. Note that solve yields solutions in terms of the lambert func-
tion. To get numerical (decimal) expressions from these results,
apply the MATLAB command eval to the results containing the
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 223

lambert function. Work in long format to make comparisons

with part (f) easier.
f. Using vpasolve, find a solution of the above equation. Begin with
the starting (trial) value of z = 1 + i and see whether you obtain
one of the values obtained in part (e). Note that you have not
proved in this problem that there are only four solutions to the
given equation. You have shown only that there are exactly four
solutions inside the given circle.
8. The purpose of this problem is to show that 5 sin z + ez and 5 sin z
have an identical number of zeros inside a certain square domain
centered at the origin.
a. Recall that (see W section 3.2) sin z = sinh 2 y + sin 2 x . Consider
the square-shaped domain, centered at the origin described by
|x| < π/2, |y| < π/2.
Show that on the square boundary of this domain, which we
call C, that |ez| < 5|sin z|. Thus, from Rouché’s theorem, we can
conclude that on C we have ΔC arg(5 sin z + ez) = ΔC arg(5 sin z).
Verify this by plotting the locus of 5 sin z + ez and 5 sin z as z
goes in the positive direction around the given C. Verify that the
increase in argument is identical for both plots. How many roots
does 5 sin z + ez = 0 have inside the given square?
b. Study this equation and conclude with the aid of sketches of
5 sin z and −ez that it has an infinite number of real roots in the
complex plane.
c. Solve the equation using the solve function. Is the solution inside
the square of part (a)?
9. a. Show mathematically that on the circle |z − 2| = 1, we have |e1/z|
< 3|z|.
b. Explain why this establishes that in the domain bounded by this
circle we have that the equation e1/z = 3z has no solution.
c. Plot the locus of e1/z − 3z as z progresses around this circle in the
positive sense, and verify that the origin is never encircled, as
confirmed by Rouché’s theorem.

5.4  Nyquist Plots, the Location of Roots, and the Pole-Zero Map
A useful special case of the principle of the argument deals with the count-
ing of the number of zeros and poles of an analytic function in a half plane
(e.g., the right half of the z plane, the left half of the same plane, or the top or
224 A MATLAB® Companion to Complex Variables

bottom halves). Let us focus on the right half of the z plane, as this is of most
interest in engineering applications.
Suppose we have a polynomial of the form

w(z) = anzn + an−1zn−1 + … a0 (5.10)

Refer now to Figure 5.10. Imagine that we have a semicircle whose diam-

eter is the imaginary axis satisfying −R ≤ y ≤ R. We move downward along
the imaginary axis from z = iR to z = −iR and then close the contour by
moving upward (counterclockwise) along the arc |z| and return to the
starting point. While on this trip, we compute the values assumed by w(z)
and plot them in the w plane so as to obtain the locus of w. We assume
that the value of R is sufficiently large that no roots of w(z) in the right
half plane lie outside this circle. Now assuming that w(z) has no roots on
the imaginary axis, we can say, from the principle of the argument, that
the number of encirclements of the origin in the w plane is equal to the
number of roots of w(z), where roots are counted according to their mul-
tiplicities. If we were to allow w(z) to have roots on the imaginary axis,

15

y=R
10

5
R

x=R
0

–5

–10
y = –R z plane

–15
–5 0 5 10 15

FIGURE 5.10
Semicircle with diameter on the imaginary axis and center at the origin.
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 225

then when we pass through these roots, the locus would go through the
origin in the w plane. The above, with some modification shown later, is
the essence of the Nyquist method, a technique for finding the number of
roots of a function in the right half plane (RHP). We could, of course, use
a similar technique to find the number of roots in the left half plane or the
upper or lower half planes.
It is an interesting little exercise to see how large we must take R in order
for there to be no roots of w(z) in the RHP outside the semicircle. Suppose
we write

w(z) = P(z) + Q(z) (5.11)

where P(z) = anzn and Q(z) = an−1zn−1 + an−2zn−2 + … a0. Note that where w(z) = 0,
then

P(z) = −Q(z) (5.12)

Notice that |P(z)| = |an||zn| and |Q(z)| = |an−1zn−1 + an−2zn−2 + … a0| ≤

|an−1||zn−1| + |an−2||zn−2| + … |a0|, where we use a generalization of the tri-
angle inequality.
Now we assume that |z| ≥ 1. Thus, |Q(z)| ≤ |an−1||zn−1| + |an−2||zn−1| + …
|a0||zn−1| = |zn−1|(|an−1| + |an−2| + … |a0|). Let A be the largest of the n num-
bers |an−1|, |an−2|, … |a0|. Using this in the preceding inequality, we have
|Q(z)| ≤ |zn−1|nA. Suppose we require that |P(z)| > |Q(z)|. This is ensured if
nA
|an||zn| > |zn−1|nA. Or z > . Thus, if we take
an

nA
R> (5.13)
an

|z| ≥ R, and |z| ≥ 1, then |P(z)| > |Q(z)|, and Equation (5.12) is not satisfied,
which means that w(z) = anzn + an−1zn−1 + … a0 ≠ 0 for |z| ≥ R.
If the right side of Equation 5.13 is less than or equal to one, it is not hard to
show that w(z) has no roots outside |z| = 1.
Let us verify the above by establishing the number of roots of w(z) = 9z5 +
z + 2 in the right half of the z plane. Of course, the MATLAB function roots
4

will do this immediately, but this little exercise will lead us into the kind of
thinking involved in the Nyquist plot. From Equation 5.13, we see that the
radius of the circle used must exceed (5 × 2)/9 = 1.111 …. We choose R = 1.2.
There are advantages to making R not much bigger than we  have to. If
R greatly exceeds its minimum value, then from Equation 5.10, we  have
(approximately) on the semicircular portion of the ­contour, which is of
radius R, that w(z) ≈ anR neinθ, where θ = arg(z). Thus, the mapping of this
226 A MATLAB® Companion to Complex Variables

semicircle into the w plane will result in an arc of large radius anRn, which
might cause portions of the rest of the locus in the w plane to appear very
close to w = 0. The shape of this locus might not be discernible.
The following code generates the locus describing our polynomial w(z) as
z negotiates the closed semicircle of Figure 5.10 in the positive (counterclock-
wise) sense.

R=1.2;
y=linspace(R,-R,1000);
theta=linspace(-pi/2,pi/2,1000);
z1=i*y;%gives points on the diameter
z2=R*exp(i*theta);%gives points on the arc
w1=9*z1.^5+z1.^4+2;% the function at points on the diameter
w2=9*z2.^5+z2.^4+2;%the function at points on the arc
u1=real(w1);v1=imag(w1);u2=real(w2);v2=imag(w2);
plot(u1,v1);hold;
plot(u2,v2); grid;axis equal
plot(0,0,'o');%puts a o at the origin
roots([9 1 0 0 0 2])
%the above tells us where the roots are

The output is shown in Figure 5.11 and the roots are:

ans =
−0.7639 + 0.0000i
−0.2513 + 0.7027i
−0.2513 − 0.7027i
0.5777 + 0.4344i
0.5777 − 0.4344i

We have added arrows to the plot to show the direction in which it

is negotiated as we move along the contour of Figure  5.10. It is easy to
establish their direction. Because the function w(z) has only zeros, and no
poles, the locus can encircle the origin in the w plane, only in the positive
sense.
It is clear that in this case the origin is enclosed twice, indicating two roots
in the RHP. This result is confirmed if we use roots to find the zeros of the
polynomial. The results are above.
We see that the last two roots are in the RHP, and there are no others
there. There are no roots on the imaginary axis. Had there been, the locus in
Figure 5.11 would have passed through the origin.
The Nyquist technique, which has been much used by control system engi-
neers, is a graphical method designed to answer whether a certain class of
functions has zeros in the right half of the complex plane. The method was
invented in 1932 by Harry Nyquist of Bell Telephone Laboratories. A sys-
tem described by a function having one or more zeros in the RHP is said to
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 227

20

15

10

5

10

15

20

–20 10 0 10 20 30

FIGURE 5.11
Locus of w(z) = 9z5 + z4 + 2 as the semicircle in the z plane is negotiated in the z plane.

be unstable. The technique has to a considerable extent been superseded by

other functions in the MATLAB toolboxes. Employing the function nyquist,
which is in the Control Systems Toolbox of MATLAB, gives us a figure called
the Nyquist diagram.
In the case of nyquist, the Toolbox does not employ the variable z but
instead the complex variable s = σ + iω, where σ and ω are real. The tech-
nique is applied to an expression like

F(s) = 1 + T(s) (5.14)

where T(s) is a rational function of s, with real coefficients, and is often mul-
tiplied by an exponential function of s. In an elementary form, it is express-
ible as

an sn + an−1sn−1 + ... + a0 −τs

T ( s) = e (5.15)
bm sm + bm−1sm−1 + ... + b0

In MATLAB’s nyquist, T(s) is called the “transfer function” of an electri-

cal or mechanical system. In some texts (e.g., W pages 494–498), the term
transfer function is applied to an expression containing Equation 5.14 in its
228 A MATLAB® Companion to Complex Variables

denominator and a polynomial in s in its numerator. We use the language

of MATLAB here.
To use nyquist, the quantity τ in the exponential in Equation 5.15 must be
real and not negative. The function nyquist will accept more complicated
expressions than those shown in Equation (5.15), provided they are combi-
nations of polynomials in s (with real coefficients) and exponentials of the
form e−sτ, τ ≥ 0. Some examples appear in the problems below. The function
T(s) above is assumed to have no poles in the right half of the s plane or on
the imaginary axis.
To determine the number of zeros of F(s) = 1 + T(s) in the right half of the
s plane, we move around a semicircular boundary like that in Figure 5.10,
but one created in the complex s plane. Then, computing the values assumed
by F(s), we plot them in a complex plane. If the locus of 1 + T(s) encircles the
origin in the complex F plane, then we know that there are zeros in the RHP.
The number of encirclements is equal to the number of zeros that 1 + T(s) has
in the right half of the s plane. Rather than plot 1 + T(s), we can simply plot
T(s), which is the previous plot but moved one unit to the left. Rather than
see if the origin is enclosed, we see if the point −1+i0 is enclosed. This is the
procedure followed by the function nyquist in MATLAB when it generates
the Nyquist plot. The point −1+i0 is placed directly on the plot so that we can
see if it is enclosed.
Another feature of the function nyquist is that the plot mentioned above
is generated by our moving in the clockwise direction, rather than in the
counterclockwise direction we employed in Figure 5.10. We must reverse the
arrow on that plot. Thus, in the s plane, we begin at σ = 0, ω = −R, proceed
upward to σ = 0, ω = R, and then move downward (clockwise) along the arc
of radius R. The function nyquist places an arrow on the locus of T(s)e−sτ, and
we now look for the number of times that this locus encloses −1+i0 in the
clockwise sense.
In physically realizable situations, as we let R → ∞ in Figure 5.10, so that
our contour encloses the entire right-hand plane, all the values assumed
by T(s) on the curved portion of the contour become in this limit, a con-
stant. The function nyquist automatically chooses R sufficiently large
so that we can assume that T(s) is essentially constant, on the semicir-
cular portion of the contour in the s plane, for purposes of drawing the
plot. Thus, in studying the plot generated by nyquist, you will not see a
curve ­corresponding to the image of the curved portion of the contour in
Figure 5.10.
At this point you should read the documentation in MATLAB help for
nyquist. We use only a small part of the information provided there. Here is
an example of how to use Nyquist. In every case, you must enter the transfer
function that is created from the MATLAB function tf; you might wish to
look this up in the help feature.
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 229

Example 5.5
Obtain a Nyquist plot to determine the number of roots in the RHP of
−6(2 s 2 + s + 1) − sτ
F( s) = 1 + e
s3 + 5s2 + 9s + 5

where τ = .01.

Solution:
We use the following code:

s = tf('s');%establishes that s is to be used to create a

%transfer function.
tau=.01;
transfer_func=-6*(2*s^2+s +1)/(s^3+5*s^2+9*s+5)*exp(-tau*s)
nyquist(transfer_func)
%the above will create the Nyquist plot

Notice the especially important line of code s = tf('s'). This estab-

lishes that a transfer function is to be established whose variable is s.
Having stated this, you do not have a choice of variables here; you must
use s.
The output of this program is Figure 5.12 and
−12 s ^ 2 − 6 s − 6)
transfer_func = exp(−0.01 * s) *
s ^ 3 + 5 s ^ 2 + 9s + 5
We have added the notation “minus one” to Figure 5.12. The Nyquist
plot always shows you the location of –1 via a plus sign. Since minus one
is encircled, clockwise, three times by the locus of

−6(2 s 2 + s + 1) − sτ
F( s) = e , τ = .01
s3 + 5s2 + 9s + 5
we can conclude that

2 s2 + s + 1
F( s) = 1 − 6 e − sτ , τ = .01
s + 5s2 + 9s + 5
3

has three zeros in the right half of the s plane. Engineers would say
that this transfer function describes an unstable system. This expres-
sion for F(s) as may be verified has no poles in the RHP. We apply the
MATLAB function roots to the denominator of the preceding fraction to
establish this. In this example, the resulting Nyquist plot does not pass
through –1, confirming that there are no zeros of F(s) on the imaginary
axis in the s plane.

If the preceding problem is changed to one where

2 s2 + s + 1
F( s) = 1 + 6 e − sτ
s + 5s2 + 9s + 5
3
230 A MATLAB® Companion to Complex Variables

Nyquist diagram
1.5

0.5
–1
Imaginary axis

–0.5

–1

–1.5
–3 –2.5 –2 –1.5 –1 –0.5 0 0.5
Real axis

FIGURE 5.12
Nyquist plot for Example 5.5.

with the same value of τ as before, it is found that the resulting Nyquist plot
does not encircle the origin. There are no zeros of F(s) in the right half of the
s plane. The reader should verify this.

5.4.1  The Use of pzmap

Another approach to finding the roots of an expression like the one given
in Equations 5.14 and 5.15 is to use the MATLAB function pzmap, which is
in the Signal Processing Toolbox. We use this in much the same way as we
used nyquist, except we regard the transfer function as the function F(s) in
Equation 5.14. We can ask pzmap to plot the zeros and poles of this function.
Again s is used as the variable. The plot obtained uses an x to show where
the poles of a function are and a 0 to show the locations of zeros. Here is
Example 5.5 solved with pzmap.
The following is the code and the resulting figure:

clear; clf
s = tf('s');
tau=.01;
transfer_func=1-6*(2*s^2+s +1)/(s^3+5*s^2+9*s+5)*exp(-tau*s)
pzmap(transfer_func)
Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 231

Pole–zero map
1.5

1
Imaginary axis (seconds–1)

0.5

–0.5

–1

–1.5
–3 –2 –1 0 1 2 3 4 5 6 7
Real axis (seconds–1)

Note that, as predicted from the example, the function

2 s2 + s + 1
F( s) = 1 − 6 e − sτ
s3 + 5s2 + 9s + 5

with the given τ has three zeros in the RHP. All three poles are in the left
half plane.

5.4.2  A Caveat on the Use of pzmap

The function whose poles and zeros are being investigated in pzmap has
certain restrictions that arise from communications theory.
Rational functions and exponentials can be used; these are functions that
are analytic except for poles. You cannot use functions that contain a branch
cut, like Log s or sqrt(s), and you cannot employ functions with an infinite
sin s
number of zeros like 2 , or poles, like 1/sin s. When a function is not suit-
s +1
able for the pzmap, you may get an error message and/or simply a wrong
answer.
At this juncture, the attentive reader might wonder what point there is in
using nyquist since pzmap shows you the actual location of poles and zeros,
232 A MATLAB® Companion to Complex Variables

while the former function merely establishes the number of poles in the RHP
of a narrower class of functions.
To answer this question, we must see how Nyquist diagrams are used in
an electrical laboratory. Recall that the variable s is σ + iω. We can some-
times make a series of electrical measurements on an electrical system by
measuring T(s) (see Equations 5.14 and 5.15) while varying ω from 0 to such
a large positive value that T(s) becomes essentially constant. This constant is
lim T ( s) in the RHP. While making these measurements, we keep σ = 0. The
s →∞
results are typically complex numbers. Because the coefficients in the poly-
nomials in Equation 5.15 are real numbers, it follows that T ( s) = T ( s). With
s = iω, this means T (iω ) = T (− iω ). Thus, if we measure the complex values of T
for positive ω and take their conjugates, it follows that we have the values for
negative ω. For example, if T measured at ω = 1000 is 3+i4, then T measured
at ω = −1000 would be 3−i4.
Shown below is a hypothetical set of measurements for ω ≥ 0. We have dis-
played the data points, with a small o, and connected them with a smooth
curve. We have also plotted with a broken curve the conjugate of the curve for
positive ω. Thus, we have a hypothetical Nyquist plot. In this case, we see that
our plot encloses the point –1+i0 two times as we proceed along the ω axis from
−∞ to ∞. There are two zeros of 1 + T(s) in the right half of the complex plane.
Notice that Nyquist plots made from experimental data do not use the
function nyquist but just the plot function.

2.5

1.5

1 w= 0
w = 100,000
0.5
–1
w= 5
0
w = 5,000
–0.5

–1

–1.5

–2
w = .5
–2.5
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5

Hypothetical experimental Nyquist plot

Polynomials, Roots, the Principle of the Argument, and Nyquist Stability 233

Exercises
1. a. Explain why all roots of the polynomial 3z4 + z3 + 1 must lie inside
any circle in the complex plane, centered at the origin, whose
radius R exceeds 4/3.
b. How many of the roots lie in the RHP? Answer this not by finding
the roots but by moving along a contour, like that in Figure 5.11,
and plotting the results. Make a suitable choice of radius for the
semicircle. How does your contour show that the polynomial has
no roots on the imaginary axis?
c. Check your answer by using the MATLAB function roots.
2. a. Use the command Nyquist to determine if

( s2 + 1)e − s/10
F( s) = 1 + 20
s5 + 10 s 4 + 40 s3 + 80 s2 + 80 s + 33
has any zeros in the right half of the s plane or on the imaginary
axis.
b. Make a pole–zero map for the above function, and see whether it
confirms your answer.
3. a. Consider the function
β( s2 + 1)e − s/10
F( s) = 1 + 20
s5 + 10 s 4 + 40 s3 + 80 s2 + 80 s + 33
where β is a real number that can be positive or negative. Study
the Nyquist plot obtained in the preceding problem (which is
this problem except that in problem 2 we have β = 1), and use that
result to estimate (approximately) the smallest value for |β| that
will result in the function having at least one zero in the right half
of the s plane. Prove your result by generating a Nyquist plot
with this value of β, which you should have found is <0.
b. Confirm the result of part (a) by means of a pole–zero map of the
given function where you employ the value of β found above.
c. For the value of β that you selected, use the function fzero to find
the numerical value of s in the right half of the s plane that causes
the function F(s) to have a zero in that plane. The pzmap can help
you to choose where to search in the plane.
4. a. Suppose in Equation 5.14 that we have
−β
T ( s) =
( s + 1)(1 + e −1 e − s )
where β is a positive real number.
234 A MATLAB® Companion to Complex Variables

How can you be certain that this expression has no poles in

the right half of the s plane?
b. Taking β = 1, obtain a Nyquist plot for F(s).
c. By studying the above plot, figure out approximately how large
β can be made so that F(s) will not have any zeros in the right
half of the s plane. Confirm this with a Nyquist plot using such a
value of β.
6
Transforms: Laplace, Fourier, Z, and Hilbert

6.1 Introduction
Anyone working in engineering or the physical sciences will know some
of the transformations named in this chapter’s title. Laplace and Fourier
transforms are much used in electric circuit theory, electromagnetic theory
(including optics), and signal processing. The Z transformation has in the
past two generations proved a major tool in the field of digital signal pro-
cessing, while the Hilbert transform is useful in subjects as diverse as the
modulation of radio signals as well as the analysis of brain waves. Because
of the similarity of Laplace and Fourier transforms, we treat them first and
begin with the easier of the two, Laplace.

6.2  The Laplace Transform

The Laplace transform of f(t), a function of the real variable t, is generally
written as
∞

Lf (t) =
∫ f (t)e
0
− st
dt (6.1)

The function produced by this integral is written as F(s).

The preceding equation will be defined as follows, where some special
attention is paid to the limits
L

Lf (t) = F( s) = lim L → ∞ lim ε → 0 −

∫ f (t)e
ε
− st
dt (6.2)

The order of taking the two limits on the right should be immaterial if the
definition is to be valid. Note the lower limit is achieved by our letting ε shrink
toward zero through negative values. For functions that are continuous at t = 0,

235
236 A MATLAB® Companion to Complex Variables

this subtlety can be ignored. Otherwise it should be observed. Note that some
textbooks use a lower limit in which ε shrinks toward zero through positive
values, and this can cause different results when we are transforming func-
tions that are discontinuous at the origin. The convention we use here is con-
sistent with MATLAB®. In general, we should think of s as a complex variable:

s = σ + iω  σ and ω real (6.3)

Thus, the Laplace transformation takes a function (which might be com-

plex, e.g., eit) of a real variable and creates a function of a complex variable.
In most cases, there are restrictions on s if the integral is to exist. For exam-
ple, the reader should verify that if f(t) = 1 and if s is purely imaginary or
has a negative real part, then the integral fails to exist. But, if f(t) = 1/t, then
the integral will not exist for any value of s owing to the singularity of this
function at t = 0.
Typically, when F(s) does exist, it is an analytic function in some domain
in the complex s plane. At this point, the reader should peruse the MATLAB
documentation on the function laplace, which creates Laplace transforms.
He or she might also wish to review a textbook showing how Laplace trans-
forms are useful in solving linear differential and sometimes integral equa-
tions. Note that because the defining integral does not require our knowing
or using values assumed by f(t) for t < 0, many authors and practitioners sim-
ply take f(t) = 0 for t < 0, and we will follow that here even though it means
that f(t) may now be discontinuous at t = 0. In using MATLAB to find Laplace
transforms, it is usually best to supply a function of the variable t, as is used
in the above definition. This is the “default variable.” If you wish to use some
other variable like x, you should read the documentation for laplace to see
how this is accomplished.
Here is a simple example of using MATLAB to find
∞

Le − bt
=
∫e
0
− st − bt
e dt

>> syms b t
>> laplace (exp(−b∗t))
ans = 1/(b + s)

Notice the answer gives no indication of what restrictions must be placed

on s for this result to be valid. Clearly, s ≠ −b. Let us perform this simple
integration to see what limitations might be required. We assume that b is
complex so that b = α + iβ. Thus,

L
e − ( s+b ) L − e − ( s+b ) ε e − (α+σ )L e − i(β+ω )L − 1

Le − at
= Llim lim
∫
→∞ ε→ 0 −
ε
e − bt e − st dt = lim lim
L→∞ ε→0− ( −b − s )
= lim
L→∞ − (b + s)
Transforms: Laplace, Fourier, Z, and Hilbert 237

Studying the expression on the above right, we see that the limit exists
only if α + σ > 0 or Re(s) > −Re(b). Thus, the result obtained by MATLAB
is valid only to the right of a vertical line in the complex s plane. Here, as is
typically the case, the Laplace transform is an analytic function of s. Thus,
one must be cautious in using the laplace operation in MATLAB.
The function 1/(b + s) is the Laplace transform of e−bt only in the domain of
the s plane where Re(s) > −Re(b). However, we see that this function, 1/(b + s),
is defined and analytic in any domain in the s plane from which the point
s = −b is excluded. The function 1/(b + s) evaluated anywhere in such a domain
where Re(s) > −Re(b) is not satisfied is said to be the analytic continuation of the
Laplace transform of e−bt. Usually we are not aware of this subtlety and say
simply that 1/(b + s) is the Laplace transform of e−bt and use it, except at s = −b,
in the s plane. Such analytic continuations will occur for other Laplace trans-
forms and are especially useful when we are inverting Laplace transforms
(recovering f(t)) by means of residues as described below. See W section 5.7
and S section 10.1 for more information on analytic continuation.
There is a sufficient condition that can guarantee that F(s) = Lf(t) is analytic
in a half plane to the right of a vertical line in the complex s plane. Briefly
stated, assuming that the integral in Equation 6.2 exists for some values of s
and that there exist real constants k, p, and T such that

|f(t)| < kept for t ≥ T (6.4)

then it can be shown* that the function F(s) is analytic in the half space
Re(s) > p.
Let us use MATLAB to find the Laplace transform of cos(bt), where
b = α + iβ is complex. We proceed as follows:

>> syms b t
>> laplace(cos(b∗t))
ans = s/(b^2 + s^2)

Where in the s plane is this expression for F(s) valid?

Notice that

cos ( bt ) =
1 i(α+ iβ )t
2
(
e ) 1
2
(
+ e − i(α+ iβ )t = e −βt e iαt + eβt e − iαt )

1 −βt iαt 1 βt − iαt 1 −βt iαt 1 βt − iαt 1 −βt 1 βt

cos(bt) ≤ e e + e e = e e + e e = e + e ≤ eβt
2 2 2 2 2 2

Thus, referring now to Equation 6.4, we see that by taking T = 0, p > |β| =

|Im(b)|, the inequality is satisfied, and we are guaranteed that the Laplace

* See, for example, R.V. Churchill. Operational Mathematics, 3rd ed. (McGraw-Hill, New York,
1971).
238 A MATLAB® Companion to Complex Variables

transform obtained by MATLAB is analytic in the half space Re(s) > |Im(b)|.

A simple additional calculation can show you that in this half space the
denominator in s/(b^2 + s^2) is nonvanishing, which of course means that
F(s) is analytic in this domain, as predicted.
In general, such calculations are not necessary each time you employ
MATLAB to find a Laplace transform. To find the half space in which the
result holds, you should look for the singularities of F(s) and assume that
the result is valid in the infinite half space that is to the right of all these
singularities.
In using MATLAB to obtain Laplace transforms, you should be aware that
the result might be in terms of a perhaps unfamiliar transcendental function,
as in the following example:

>> syms n s t
>> laplace( t^(1/n))
ans = piecewise([−1 < real(1/n), gamma(1/n + 1)/s^(1/n + 1)])

The gamma in the preceding refers to the Γ function that you may wish
to review (see W section 6.11 and S section 10.8). Recall that it is defined by
∞

Γ(w ) =
∫e
0
−t w−1
t dt

where w is a complex variable whose real part is positive.

Γ(1/n + 1)
Thus, we learn that the Laplace transform of the function t1/n is .
s(1/n+1)
There is an additional constraint, −1 < real(1/n), which ensures that the argu-
ment of the gamma function in the preceding has a positive real part; this is
a requirement of the definition of the gamma function via an integral.
MATLAB in Release 2015a is still unable to obtain the Laplace transform of
some relatively simple functions as shown by the following:

>> laplace(cosh(sqrt(t)))
ans = laplace(cosh(t^(1/2)), t, s)

We are trying to get the Laplace transform of cosh(t1/2). MATLAB merely

returns our question to us. In fact, the transform is to be found in any good
set of tables of Laplace transforms, but it does involve the error function,
which is defined further in this section.
Sometimes you must be careful in employing MATLAB to obtain a
Laplace transform of a function. The statement laplace takes as the default
variable of integration the variable t. If this is not your intention, you must
so instruct MATLAB. Suppose you wish to obtain the Laplace transform of
the function f(x) = cos(x + 1/t), where t is a parameter and x is the variable
of integration.
Transforms: Laplace, Fourier, Z, and Hilbert 239

Consider

>> syms s x t
>> laplace (cos(x+1/t))
ans = laplace(cos(x + 1/t), t, s)

Notice that MATLAB returns the question to us. It is unable to do the inte-
gration on the variable t. We wanted the integration done on the variable x.
We proceed as follows:

>> syms s x t
>> laplace (cos(x+1/t),x,s)
ans = −(sin(1/t) − s∗cos(1/t))/(s^2 + 1)

MATLAB obtains the relatively simple answer. The characters x,s in the
laplace command instruct MATLAB to do the integration in Equation 6.1 on
the variable x, not t, and to produce a function of s.

6.2.1  Branch Points and Branch Cuts in the Laplace Transform

Sometimes in using MATLAB to find a Laplace transform you will receive a
function that must be defined by means of a branch cut. This would be true
in the above case where we transformed t1/n. Here is an example where we
et
get the transform of and must also deal with branch cuts:
t
>> syms s t
>> laplace(t^(−1/2)∗exp(t))
ans = pi^(1/2)/(s − 1)^(1/2)

Our result is
π
F( s) =
( s − 1)1/2

How do we define this function in the complex s plane? Notice that the
given function of t, namely, e t/ t , can be made to satisfy Equation 6.4 if we
take p = 1 and k > 1 and T = 1. Thus, we are guaranteed that this Laplace
transform is analytic in the half plane Re(s) > 1. We may use any branch cut
of (s − 1)1/2 in the complex s plane that originates at s = 1, extends to infinity,
and does not pass into the domain Re(s) > 1. We might have guessed this,
because we know that if the Laplace transform exists, it must be analytic in
a domain to the right of a vertical line in the complex s plane. We can choose
any vertical line to the right of s = 1 and achieve this.
To establish a branch of a multivalued function, we require not only a
branch cut (or cuts) but also the numerical value of the function at one point.
240 A MATLAB® Companion to Complex Variables

Suppose s = 2 or any real value greater than 1. Using s = 2 in Equation 6.1,

with our given f (t) = e t/ t must result in a positive real value, and this com-
π
pletes the specification of the branch of F( s) = , because F(2) = π .
( s − 1)1/2
It is entirely consistent with the preceding, and convenient, for us to
take the principal branch of (s − 1)1/2 in our result for F(s). It is usually safe
to  take the principal branch of F(s), but to be absolutely sure, one should
verify the choice as was done here.

6.2.2  Heaviside and Dirac Delta Functions

Both the Heaviside and Dirac delta functions are often used in ­conjunction
with Laplace transforms, especially in electric circuit theory and mechanics.
Following the conventions of MATLAB, we use the notation heaviside(t)
to mean a function that is equal to 1 for t > 0 and equal to zero for t < 0.
In MATLAB, the function is defined as .5 for t = 0, but this is by no means
universal, and many authors simply leave it undefined there. The reader
should verify by pencil-and-paper integration that the Laplace transform of
this function is given by L(heaviside(t)) = 1/s for Re(s) > 0. The Heaviside
function is also often known as the unit step function, and we use the common
notation u(t) to mean the Heaviside function as one soon tires of writing out
the word Heaviside. Thus,

u(t) = heaviside(t)

The plot shown in Figure 6.1 should make clear where the language “unit
step” comes from. The function u(t − τ) or heaviside (t − τ) (here τ is real) would
just be a step like the one shown in Figure 6.1 but with the discontinuous rise
displaced τ units to the right if τ > 0. Here we obtain the Laplace transform
of the Heaviside function with MATLAB:

>> syms s t
>> laplace (heaviside(t))
ans = 1/s

The Heaviside function is defined only as a function of a real variable.

Thus, to take the Laplace transform of the function f(t) = heaviside(t − t1), we
proceed as follows:

>> syms s t t1
>> assume(t1>0)
>> laplace (heaviside(t−t1))
ans = exp(−s∗t1)/s
Transforms: Laplace, Fourier, Z, and Hilbert 241

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
t

FIGURE 6.1
Heaviside function for −2 < t < 2.

Notice that the line of code assume(t1>0) establishes that t1 is positive and
that it is real. Without such an assume statement, MATLAB cannot evaluate
the Laplace transform if t1 is a symbol whose numerical value has not been
specified. If you used assume(t1<0), the answer would be 1/s, and t1 would
not appear. You should explain to yourself why this is true. In the preced-
ing work, you may use any real number instead of t1. But MATLAB will not
evaluate the Laplace transform of heaviside(t−i) because of the presence of
the non-real i. Verify this.
Studying Figure 6.1, we would expect that the derivative of the function
heaviside(t) might be thought of as infinite at t = 0 and zero for all t ≠ 0. A rig-
orous treatment, using the usual definition of the derivative as the limit of a
difference quotient, would show that this function has no derivative at t = 0.*
However, we nevertheless use the following:

d(heaviside(t))
dirac(t) =
dt

* A branch of advanced calculus called generalized functions permits the repeated differentia-
tion of the Heaviside function.
242 A MATLAB® Companion to Complex Variables

The function on the left, the Dirac delta function, is often written as δ(t) and
is nonzero only at t = 0 and is to be thought of loosely as “positive infinity”
at t = 0. Thus, we have also

δ(t) = dirac(t)

du(t) d
δ(t) = = heaviside(t)
dt dt

The integral of this expression must yield the Heaviside function. Thus,
t t

∫
L
δ(t ′) dt ′ = u(t) or
∫ dirac(t′) dt′ = heaviside(t)  
L
L < 0 and t > L

We can verify the preceding by means of MATLAB as follows:

>> syms b t
>> diff(heaviside(t−b))
ans = dirac(b − t)
>> int(dirac(t))
ans = heaviside(t)

Note that dirac(b − t)=dirac(t−b) as the dirac function is an even function of its
argument. Observe also that the operator diff treats t as the variable for which
differentiation is to take place. MATLAB assumes that you do not wish to dif-
ferentiate with respect to b and takes t as the default variable for differentiation.
A useful result is called the sampling property of the dirac delta function:
∞

∫ dirac(t − τ) f (t)dt = f (τ) (6.5)
−∞

The integral returns the value (or sample) of the function f(t) at t = τ. This
is sometimes known also as the sifting property as it sifts through possible
values of the function and supplies one. We assume that f(t) is continuous at
t = τ. Because the dirac function is zero except where its argument is zero,
the preceding equation is still valid if −∞ is replaced by the symbol b, and ∞
is replaced by the symbol c, provided b < τ < c.
Because of the sampling property, we have that
L

L dirac(t) = lim L → ∞ lim ε → 0 −

∫ dirac(t)e − st
dt = 1
ε

which shows the significance of having ε approach zero through negative

values.
Transforms: Laplace, Fourier, Z, and Hilbert 243

Similarly, if τ is a nonnegative real number, we have

L

L dirac(t − τ) = lim L → ∞ lim ε → 0 −

∫ dirac(t − τ)e − st
dt = e − sτ
ε

You should see that if τ is negative real, then the Laplace transform is zero.
Here is the same result from MATLAB where we use t1 instead of τ.

>> syms s t t1
>> laplace (dirac (t−t1))
ans = piecewise([t1 < 0, 0], [0 <= t1, exp(−s∗t1)])

Notice the additional information in this “piecewise” answer: that if t1 is

negative, the Laplace transform is zero. Observe also that the form of the
answer shows that MATLAB assumes that t1 is real.
The Heaviside function is useful in describing pulse-type functions.
Suppose a function f(t) is zero for t < t1, and also for t > t2, where t1 < t2. Also
suppose that f(t) = 1 for t1 < t < t2. The reader should verify that this is con-
veniently written as f(t) = heaviside(t − t1) − heaviside(t − t2). Here MATLAB
finds the Laplace transform of f(t), where t1 = 1, t2 = 4. The function is shown
in Figure 6.2.

1.8

1.6

1.4

1.2

0.8

0.6

0.4

0.2

0
0 1 2 3 4 5 6
t

FIGURE 6.2
Plot of f(t) = heaviside(t − 1) − heaviside(t − 4).
244 A MATLAB® Companion to Complex Variables

And here is its transform:

>> syms s t t1 t2
>> t1=1;t2=4;
>> laplace(heaviside(t−t1) − heaviside(t−t2))
ans = exp(−s)/s − exp(−4∗s)/s

A function that is similar to the Heaviside function is the sign function.

It is defined only for real arguments. Thus, sign(x) = 1, x > 0 and sign(x) = −1,
x < 0, while sign(0) = 0. Thus, sign(x) assumes the values 1 or –1 according to
whether its argument is positive or negative.
The reader who is at all familiar with Laplace transforms knows that much
of their utility resides in how easily one can find the derivatives of the func-
tions f′(t), f′′(t), f′′′(t), and so on, once the Laplace transform of f(t) is known.
With Lf(t) = F(s), we have

Lf′(t) = sF(s) − f′(0−) (6.6)

Lf′′(t) = s2 F(s) − sf′(0−) − f′′(0−) (6.7)

The first of these formulas is obtained by our integrating Equation 6.1 by

parts, using f′(t) in place of f(t) in that equation. A comparable procedure
yields the second of these. We can also obtain a formula for the Laplace
transform of the integral of a function in terms of the Laplace transform of
that function:
t
F( s)
L
∫ f (t′) dt′ =
0
s
(6.8)

This is also obtainable if we do an integration by parts of Equation 6.1 but

t

use
∫ f (t′) dt′ in place of f(t).
0
cos at
If we know the Laplace transform of the function f (t) = − , then it
a
should be easy to find the Laplace transform of the function which is its first
derivative: f′(t) = sin(at).
 − cos at 
From Equation 6.6, we have L sin( at) = sL   − (−1/a). From MATLAB,
 a 
we obtain

>> syms s a t
>> s∗laplace (−cos(a∗t)/a) − (− 1/a)
ans = 1/a − s^2/(a∗(a^2 + s^2))
>> simplify (ans)
Transforms: Laplace, Fourier, Z, and Hilbert 245

ans = a/(a^2 + s^2)

>> laplace( sin(a∗t))
ans = a/(a^2 + s^2)

Note the agreement between the two ways of getting the Laplace trans-
form of sin(at).
The derivative of the Heaviside function is the Dirac function. Thus,
the Laplace transform of the Dirac function should be obtained if we use
MATLAB to obtain the Laplace transform of the Heaviside function, multi-
ply this result by s, and from this subtract the value of the Heaviside function
at 0− (this value is zero). We verify this as follows:

>> syms s t
>> s∗laplace (heaviside(t))
ans = 1

which is indeed the Laplace transform of dirac(t).

The derivative of the sign function is twice the Dirac function. This is
because the discontinuity of heaviside(t) at t = 0 is half the discontinuity of
sign(t) at t = 0. The size of the first jump is unity, and the size of the second is 2.
A sketch of the two functions illustrates this.
Notice that from MATLAB we have

>> syms t
>> diff(heaviside(t))
ans = dirac(t)
>> diff(sign(t))
ans = 2∗dirac(t)

The Dirac function has, if we permit ourselves the use of generalized func-
tions, a first derivative. Generalized functions include the Dirac function
and/or any of its derivatives. In the language of MATLAB, we have that
d
dirac(t) = dirac(t , 1), where the notation dirac(t,1) means simply the first
dt
derivative of the Dirac function. MATLAB can handle higher-order deriva-
tives of the Dirac function and uses dirac(t,n) to mean the nth derivative of
dirac(t). If the number n is left unspecified, it is taken as zero—no derivative
is taken. Just as the function dirac(t − τ), where τ is real, samples the value of a
function that is continuous at τ when we do an integration (see Equation 6.5),
the function dirac(t−τ, 1) will sample the negative of the first derivative of a
function with a continuous first derivative. Thus,
∞

∫ dirac(t − τ, 1) f (t)dt = − f ′(τ) (6.9)
−∞
246 A MATLAB® Companion to Complex Variables

The preceding equation is still valid if −∞ is replaced by the symbol b and

∞ is replaced by the symbol c, provided b < τ < c. The above equation can be
derived by our performing this integration
∞
d
∫ f (t) dt dirac(t − τ)dt
−∞

by parts and employing Equation 6.5. A generalization of the above is

∞

∫ dirac(t − τ, n) f (t)dt = ( −1)

n
f ( n) (τ) (6.10)
−∞

Thus, the nth derivative of the Dirac function, δ(n)(t − τ), samples the nth
derivative of the function f(t) and multiplies it by either 1 or –1.

6.2.3  The Inverse Laplace Transform

If f(t) has Laplace transform F(s), then we say that F(s) has inverse Laplace trans-
form f(t).
Using complex variable theory, we can obtain the inverse Laplace trans-
form for a wide class of functions F(s) (see W section 7.1).
We assume that the function F(s) is analytic in a right half portion of the
complex s plane described by Res ≥ a. We also assume that there exist positive
constants m, R0, and k such that

|F(s)| ≤ m/|s|k (6.11)

for all |s| > R0 in this half plane.* Under these conditions, the function f(t), for
t > 0, can be obtained by a contour integral along an infinite vertical line in
the complex s plane as follows:

a + i∞
1
−1
f (t) = L F( s) =
2 πi ∫ F(s)e
a − i∞
st
ds (6.12)

The integration is done along the line Re(s) = a or along any other contour
into which this line can legitimately be deformed without affecting the value
of the integral. The preceding is known as the Bromwich integral and is per-
formed along a straight infinite vertical line like that shown in Figure 6.3.
The line is called the Bromwich contour. If we perform the integra-
tion for t  <  0,  it can be shown that the function of t obtained is zero.

1
* The reader might wish to verify that the function F( s) = will satisfy the preceding if we
s+1
were to take m = 1, k = 1, a = 0, and R0 as any positive number.
Transforms: Laplace, Fourier, Z, and Hilbert 247

Original
contour

Singularities
of F(s)

a σ

A valid
alternative
contour

FIGURE 6.3
A Bromwich contour and its valid deformation.

If the function f(t) has a step-type discontinuity at t = t0 > 0, and if we obtain

F(s) and then seek to get back f(t) by means of Equation 6.12, there is a sub-
1
tlety. We obtain a function whose value at t0 is ( f (t0 +) + f (t0 −)), which is the
2
average of the values on either side of the step.
Suppose the function F(s) satisfies the requirements for using Equation 6.12,
and suppose also that its singularities are all poles and that they are finite in
number. They of course lie to the left of the line in Figure 6.3. Assume also
that there exist positive constants m, R0, and k such that F(s) ≤ m/|s|k when
|s|  > R0 is in the half plane lying to the left of the vertical line described
above. Then the integral can be evaluated using the method of residues (see
W section 7.1) with this result:

f(t) = the sum of the residues of F(s)est  at all poles of F(s) (6.13)

If there are branch cut singularities of F(s) lying to the left of the vertical
line, the method cannot be used.
1
Let us see how this applies to the function F( s) = 2 . Notice that this
s ( s + 1)
function is analytic in a half plane Re(s) ≥ a, where a > 0. Also, in this half
1 1 1
plane |s| < |s + 1|, which means that in this same space ≤ 2 = 3.
s2 ( s + 1) s s s
Thus, we may employ the Bromwich integral to find the inverse transform.
248 A MATLAB® Companion to Complex Variables

We may evaluate this integral using the residues described in Equation 6.13.

Note that if |s| > 1, we have that |s + 1| > |s| −1 > 0. Thus, |s + 1| > |s| − 1 =
|s/2| + |s/2| − 1 > |s/2| > 0 provided |s| > 2. Furthermore, in this same
domain,

1 1 1 2
= 2 ≤ 2 = 3 if s > 2 .
s2 ( s + 1) s s+1 s s /2 s

The conditions for using Equation 6.13 are satisfied, and we assert that

1 e st e st
L−1 = Res @ s = 0 + Res @ s = −1
s 2 ( s + 1) s 2 ( s + 1) s 2 ( s + 1)

The first pole is of order 2 and has residue

 d s2 e st 
lim  = t − 1.
s → 0  ds s 2 ( s + 1) 


The second pole is of order 1 and has residue

 ( s + 1) e st 
lim  2 = e − t.
s →−1  s ( s + 1)  

Summing residues, we have finally that f(t) = t − 1 + e−t.

All of the preceding can very easily be performed with MATLAB by means
of the function ilaplace in the symbolic math toolbox. The reader should read
the documentation for this function. Here we perform the above problem:

>> syms s t
>> ilaplace (1/(s^2∗(s+1)))
ans = t + exp(−t) – 1

which agrees with what we obtained using the calculus of residues.

The use of ilaplace did not seem particularly helpful when we did the
above problem, as an answer using residues also works. But there are many
functions for which Equation 6.12 or Equation 6.13, or both cannot be applied.
This is especially true when f(t) is composed of one or more Dirac functions
or their derivatives. The function f(t) = δ(t − 1) has Laplace transform F(s) = e−s.
Yet, we cannot apply the Bromwich integral, Equation 6.12, to F(s) and get
back f(t) because F(s)  =  e−s cannot be made to satisfy Equation 6.11, as the
reader should verify.
On the other hand, the function F(s) = e−s/s can be made to satisfy this same
equation, but we cannot apply Equation 6.13 to find its inverse transform,
because there do not exist constants such that F(s) ≤ m/|s|k, |s| > R0, is satis-
fied to the left of the vertical line used in the Bromwich integral.
Transforms: Laplace, Fourier, Z, and Hilbert 249

1
The function F( s) = , where b is a constant and the function is
( s − b)1/2
defined by the principal branch, cannot be found from Equation 6.13 because
of the presence of branch point singularities to the left of the Bromwich con-
tour. Yet in all three cases just cited, MATLAB will yield the inverse Laplace
transform. Here is one example:

>> syms s b t
>> ilaplace (1/(s−b)^(1/2))
ans = exp(b∗t)/(pi^(1/2)∗t^(1/2))

1
The result is that the inverse Laplace transform of F( s) = is
( s − b)1/2
e bt
f (t) = . Notice that MATLAB recognizes that s is the variable and b the
πt 1/2
constant in the expression that it is asked to invert. It always takes s as the
default variable in these situations unless instructed otherwise. The branch
of (s − b)1/2 is assumed to be the principal one, and in general ilaplace uses the
principal branch of F(s). The square root of t in the preceding answer should
be interpreted as the principal value.
The preceding inverse transform can be found by analytic means (see W
section 7.1). Some idea of the method is outlined in problem 6 where b = –1.
In the present release of MATLAB (R2015a) that the author is using, there
are some quirks that perhaps will be ironed out in future versions. For exam-
ple, consider this operation:

>> syms s t
>> ilaplace (1)

whose output is
Undefined function ‘ilaplace’ for input arguments of type ‘double’.

MATLAB refuses to provide the desired inverse Laplace transform because

the ilaplace function is anticipating a function of s. The number one is not
such a function. The matter can be dealt with as follows:

>> ilaplace(sym(‘1’))
ans = dirac(t)

The number 1 is here treated as a symbol. But an easier ploy is this:

>> syms s t
>> ilaplace (s/s)
ans = dirac(t)

where the number 1 is disguised here as a function of s. This subtlety is also

considered in problem 3.
250 A MATLAB® Companion to Complex Variables

Sometimes when we use the functions laplace or ilaplace, we may receive

from MATLAB a result that might not be familiar. Consider the following, where
2
we want the Laplace transform of f (t) = e − at , where a is a positive real number.

>> syms s t a
>> assume(a>0)
>> laplace (exp(−a∗t^2))
ans = (pi^(1/2)∗exp(s^2/(4∗a))∗erfc(s/(2∗a^(1/2))))/(2∗a^(1/2))

Our result is

π s2 /(4 a ) s 
e erfc 
2 a  2 a 

The function erfc is known as the complimentary error function. It is defined as

∞
2
erfc( x) =
π ∫e − t 2 /2
dt
x

and appears often both in Laplace transformations and their inverses.

A related function is the error function which is
x
2
erf( x) =
π ∫e
0
− t 2 /2
dt.

The sum of these functions is one.

Incidentally, had we left off the line of code, assume(a > 0), we would still
have obtained the same result from MATLAB. However, the result is not
valid if a < 0, a fact not stated by the output from MATLAB.
To remind the reader of the power of the Laplace transform in solving dif-
ferential equations or integro-differential equations, we look at the following
problem:

Example 6.1
Solve the equation.
t
d2 f

dt 2
+
∫ f (τ) dτ = e sin t (6.14)
0
t

Assume f′(0) = 0 and f(0) = 0.

Solution:
Recalling that the Laplace transform is a linear operator, we take the
Laplace transform of the left-hand side of the above equation by add-
ing together the Laplace transform of each of the two terms on the left.
Transforms: Laplace, Fourier, Z, and Hilbert 251

We have with the aid of Equation 6.7 and 6.8 that the Laplace t­ ransform
of the left side of Equation 6.14 is s2 F(s) + F(s)/s. The Laplace transform
of the right side is best obtained from MATLAB, as follows:
>> syms s t
>> laplace (exp(t)∗sin(t))
ans = 1/((s − 1)^2 + 1)
Thus, equating the Laplace transform of the right- and left-hand sides
of Equation 6.14, we have
1
s 2 F( s) + F( s)/ s =
( s − 1)2 + 1

which yields
1 1
F( s) =
( s − 1)2 + 1 s 2 + 1/ s

We find f(t) by applying ilaplace as follows:

>> syms s t
>> ilaplace (1/((s−1)^2+1)∗1/(s^2+1/s))
ans = (2∗exp(t/2)∗cos((3^(1/2)∗t)/2))/3 − exp(−t)/15 − (3∗exp(t)∗(cos(t)
− sin(t)/3))/5
This is more comfortably displayed as

2 t/2  3  e −t 3 t  1 
f (t) = e cos  t − − e cos t − sin t 
3  2  15 5  3 

Of course, there are other more direct ways to solve equations like the one
just treated by using MATLAB. Differentiating both sides of Equation 6.14
results in a differential equation that can be solved with the MATLAB func-
tion dsolve, but this topic takes us far from complex variable theory.

Exercises
1. By evaluating the Bromwich integral with the method of residues,
find the inverse Laplace transform of the following functions and
compare your results with those obtained by MATLAB.
1
a. F( s) = 2
( s − 1)
1
b. F( s) =
( s − 1)2 + 1
1 1
c. F( s) = 2 +
s +s+1 s+1
252 A MATLAB® Companion to Complex Variables

2. a. Suppose a function f(t) has Laplace transform F(s). Show that the
Laplace transform of the function f(t − τ)u(t − τ) is e−sτ F(s), where
τ is a nonnegative real.
Hint: Write the Laplace transform of f(t − τ)u(t − τ) as an integral,
and make a simple change of variable. Our result is known as
a translation theorem. The inverse transform of e−sτ F(s) is thus the
inverse transform of F(s) shifted τ units to the right on the t axis
and multiplied by u(t − τ). The latter ensures that the resulting
function is zero for t − τ. Thus, L −1e−sτ F(s) = u(t − τ)f(t − τ), where
f(t) = L −1F(s).
e st
b. Find the residue of the function at its pole and use that
( s − 1)2
result together with Equation 6.13 as well as the result derived in
e −2 s
part (a) to find the inverse Laplace transform of .
( s − 1)2
c. Check the result by using MATLAB to find the inverse transform
e −2 s
of F( s) = .
( s − 1)2
3. This problem explores some peculiarities of MATLAB in its han-
dling of inverse Laplace transforms.
a. Using ilaplace, find the inverse Laplace transform of the func-
s+1
tion .
s
b. Verify that you get the same result by applying ilaplace to the
1
function  1 +  .
 s
c. Since the inverse Laplace transformation is a linear operation,
1
you should get the same result by using ilaplace(1) + ilaplace  .
 s
Try this. You will see that there is a difficulty. This arises because
the operation ilaplace expects a symbolic function in its
argument.
d. Show that you can resolve this difficulty by first stating
one=sym(‘1’) in your code. This establishes that one is now a
symbol equal to unity. Now try taking the sum ilaplace(one) +
1
ilaplace  and show that you get an inverse Laplace transform
 s
s+1
of .
s
 s 1
Also try taking the sum ilaplace  + ilaplace  and verify
 s  s
that this works, too.
Transforms: Laplace, Fourier, Z, and Hilbert 253

1
4. a. Consider the Laplace transform F( s) = . The principal
( s + 1)1/2
2

branch of the function is used. Where in the s plane is F(s) ana-

lytic (i.e., where is the branch cut?)? Explain why F(s) satisfies
Equation 6.11 for suitable m and k.
Use MATLAB to find the inverse Laplace transform. Notice
that the answer is given in terms of the Bessel function of the
first kind, order zero: J0(t). You can learn more about this function
by typing help besselj in MATLAB.
1
b. Repeat the above using F( s) = 2 . The principal branch of
( s − 1)1/2
the function is used. Notice that the answer is again given in
terms of the Bessel function of the first kind, order zero: J0(t), but
the argument of the function is purely imaginary. Some books
use the function I0 but with a real argument to mean the same
thing.
5. Let f(t) and g(t) be two functions having Laplace transforms
F(s) and G(s). The convolution of f(t) and g(t), written f(t) ⊗ g(t), is
t

defined as f (t) ⊗ g(t) =

∫ f (τ)g(t − τ)dτ. It is a simple matter to show
0
that f(t) ⊗ g(t) = g(t) ⊗ f(t). This means that the order of the convolu-
tion is immaterial. In standard texts on complex variable theory
or Laplace transforms, it is shown that L[f(t) ⊗ g(t)] = F(s)G(s) (i.e.,
the Laplace transform of the convolution of two functions is the
product of the Laplace transform of each). In this problem, we use
MATLAB to verify this property in a specific case. A more general
definition of the convolution of two functions is given in the fol-
lowing section of this book.
a. Use the function int in MATLAB to find the function of t that is
the convolution of the functions f(t) = sin(t + 1) and g(t) = cos 2t.
Compute both f(t) ⊗ g(t) and g(t) ⊗ f(t), and verify that you get the
same answer. You may wish to read the documentation for int in
the help window.
b. Using MATLAB, find the Laplace transforms of f(t) and g(t).
c. Find the Laplace transform of the function found in part (a).
Verify that it is indeed the product of the Laplace transforms of
f(t) and g(t).
d. Show that the general relationship L[f(t) ⊗ g(t)] = F(s)G(s)
can  be  used to derive Equation 6.8 if we make a suitable
choice of g(t) and know the Laplace transform of the unit step
function.
254 A MATLAB® Companion to Complex Variables

6. a. Using MATLAB, find the inverse Laplace transform of the function

1
F( s) =

( s + 1)1/2

where the principal branch of the function is used.

We check this result by using contour integration. Follow the
steps below.
b. Explain why the inverse Laplace transform can be obtained by
your doing the integration in Equation 6.12, when a = 0.
c. Explain why if we integrate

1 e st

2 πi ∫ ( s + 1)1/2
ds

along the closed contour, in the s plane, shown with arrows

in Figure 6.4, we get zero. Note that we make a detour around
s = −1 by using a nearly circular contour of radius ε. The hori-
zontal portions of the contour are just above and below the
branch cut for (s + 1)1/2. This cut lies along the line Im(s) = 0,
−∞ < Re(s) ≤ −1.

3 R

0 –1 0

–1

–2 Radius ε

–3

–4

–5
–6 –4 –2 0 2

FIGURE 6.4
Contour for Part (c).
Transforms: Laplace, Fourier, Z, and Hilbert 255

d. In the next two steps we prove that as R → ∞ the portion of the

integral along the arc in the second quadrant above goes to zero.
Hint: Make the polar substitution s = Reiθ, where π/2 ≤ θ ≤ π.
Recall the inequality

q q

∫ u(θ) dθ ≤ ∫ u(θ) dθ
p p
if q > p

We may also substitute a function g(θ) for |u(θ)| on the right

provided that g(θ) ≥ |u(θ)| everywhere along the interval of inte-
gration p ≤ θ ≤ q. Show that the magnitude of the integral along
the arc satisfies the inequality

π
1 e st 1 R
∫ ∫e Rt cos θ
1/2 ds ≤ dθ
2 πi ( s − 1) 2 π R−1
π/2

Hint: Make the change of variable s = Reiθ for the integral on this
iθ
arc, where π/2 ≤ θ ≤ π. Find e t Re .
e. Observe that, by making a sketch, for π/2 ≤ θ ≤ π, we have the
inequality 1 − 2θ/π ≥ cosθ. Thus, show that

π
1 e st 1 R
∫ ∫e Rt (1− 2 θ/π )
1/2 ds ≤ dθ
2 πi ( s − 1) 2 π R−1
π/2

along the arc in the second quadrant.

Now perform the right-hand integration to show that

1 e st 1

2 πi ∫ ( s − 1) 1/2 ds ≤
4t R − 1
 e − Rt − e −2 Rt 

Argue that for t > 0, we must have that the integral on the left
goes to zero as R tends to infinity.
If we perform this integration along the arc in the third quad-
rant, we will again get zero, in this same limit, because the inte-
grand assumes the conjugate of the values it assumed in the
second quadrant at conjugate points on the two arcs.
256 A MATLAB® Companion to Complex Variables

f. It is a similar but easier exercise to show that the integral

1 e st

2 πi ∫ ( s + 1)1/2
ds

around the small circle of radius ε, on the contour, goes to zero as

ε → 0+. Take s = εeiθ. Carry out the steps using the “ML inequality.”
g. Having passed to the two limits ε → 0+, R → ∞, show that the
inverse Laplace transform of 1/(s + 1)1/2 consists of the sum of the
integrals:
−∞ −1
1 e xt 1 e xt
2 πi ∫ x+1 i
1/2 dx +
2 πi ∫ 1/2
x + 1 ( − i)
dx
−1 −∞

which arise, respectively, from integrations along the top and bot-
tom of the branch cut in the above figure in the limits. Combine
these integrals and make the change of variable x′ = −x to show
that
∞
1 1 e − x ′t
L−1
( s + 1)1/2
=
π ∫ x′ − 1
dx ′
1

You can evaluate this integral by using the well-known result

∞
1

∫e
0
− au2
du =
2
π/a a > 0.

Let u = x′ − 1. Thus, finally verify that the inverse Laplace trans-

1 e−t
form of 1/2
is .
( s + 1) πt
7. The present version of MATLAB (R2015a) will not give the Laplace
t
2
transform of the error function defined as erf(t) =
π ∫e
0
−τ 2
d τ.

Investigate this for your version. Now show how to get the Laplace
transform of this function by using MATLAB to obtain the Laplace
2 − t2
transform of the function e and then employing Equation 6.8.
π
It is helpful in this problem to treat π as a symbol using the statement
syms pi. In this way, the program will not reduce the square root of
π to a number. Notice that your answers will be in terms of the com-
plimentary error function: erfc(t) = 1−erf(t).
Transforms: Laplace, Fourier, Z, and Hilbert 257

6.3  The Fourier Transform

The Fourier transform (or transformation) is a tool that is much used by
engineers and scientists in such disciplines as signal analysis, system theory,
electromagnetics, optics, and mechanics. Like the Laplace transform, it can
be used to solve both linear differential and integral equations. Suppose we
are given a function f(t) of a real variable t. Then its Fourier transformation,
Ff ( t ), is obtained by converting f(t) to a function f(w) defined by the integral

Ff (t) = F(w) =
∫ f (t)e
−∞
− iwt
dt (6.15)

Here w is to be regarded as a real variable, although we often employ the

analytic continuation of the function defined above and continue to call it
F(ω). A sufficient condition for the existence of the above integral is that the
∞

integral
∫ f (t) dt exist and that discontinuities of the function f(t), if any, be
−∞
limited to a finite number of jump discontinuities in any finite interval on the
t axis. In reality, the engineer or scientist very often deals with functions that
do not satisfy this necessary condition but whose transforms do in fact exist.
A simple function sin t is such an example.
The function f(t) can be recovered from F(ω), and we write F −1 F(w) = f (t).
The procedure requires an integral:
∞
1
−1
f (t) = F F(w) =
2π ∫ F(w)e
−∞
iwt
dw (6.16)

The integration is done along the real axis in the complex w plane, and we
employ the analytic continuation of F(ω) as mentioned above. Like the inver-
sion formula for the Laplace transform, this formula comes with a caveat: If
f(t) has a jump discontinuity at a point, say t0, then the integral on the right
1
yields f (t0 ) =  f (t0 +) + f (t0 −) , the average of the values assumed by f(t) on
2
each side of the jump.
The reader should note that there are a variety of ways in which to con-
sistently define the Fourier transform and its inverse. We have used the one
1
employed by MATLAB. In other definitions, is swapped from the right
2π
side of Equation 6.16 to the right side of Equation 6.15. This is the procedure in
W (pages 405–411). Sometimes the minus sign is swapped from the exponent
in the integral Equation 6.15 to the exponent in the integral in Equation 6.16.
Note that in the Wikipedia tables of Fourier transforms, three different
258 A MATLAB® Companion to Complex Variables

options are offered for defining the transform, see http://en.wikipedia.org/

wiki/Fourier_transform#Distributions.
The integrals in Equations 6.15 and 6.16 are both to be interpreted as
Cauchy principal values—that is,
∞ R

∫ ...(dt or dw) = lim ∫ ...(dt or dw).
−∞
R →∞
−R

The limits move out from the origin symmetrically. If f(t) has a singularity at
the real value t0, then we must perform our integration as follows:

 t0 −ε R 
lim lim 
R →∞ ε→ 0 + 
 −R
∫ − iwt
f (t)e dt +
t0 +ε
∫
f (t)e − iwt dt 



The limits move toward t0 symmetrically, and this is also known as a

Cauchy principal value. In all practical cases, the result does not depend on
the order in which we take the limits on R and ε. The method is readily
extended to cases where there are more than one such singularity for real t.
If the limit as ε shrinks to zero does not exist, then the function f(t) does not
have an integrable singularity and the integration cannot be done. However,
as we see, both MATLAB and the community that uses Fourier transforms
take liberties with this fact. If F(w) in Equation 6.16 has a singularity as some
real value, say, ω0, it is handled in an analogous way.
Unlike the Laplace transformation, the Fourier transformation does not
directly take into account conditions, present at t = 0− in an electrical or
mechanical system. However, the Laplace transformation will not treat the
behavior of a system that has been receiving excitation for t < 0−.
Much of the utility of the Fourier transform, like that of the Laplace trans-
form, resides in our ability to easily find the Fourier transform of derivatives
of any order of a function f(t) in terms of F(w). We can also find the Fourier
transform of the integral of f(t) in terms of F(w). These results can be found in
any standard text on Fourier transforms.

df d2 f dn f
= iwF(w), F 2 = ( iw ) F(w) ... F n = ( iw ) F(w) (6.17)
2 n
Ff (t) = F(w), F
dt dt dt

t
F(w)
F
∫ f (τ)dτ =
−∞
iw
+ F(w)δ(w) (6.18)

The first group, Equation 6.17, can be derived formally if we permit our-
selves to differentiate both sides of Equation 6.16 with respect to the variable
t and bring the derivative under the integral sign.
Transforms: Laplace, Fourier, Z, and Hilbert 259

Equation 6.18 can be derived from the properties of the convolution inte-
gral described below. It is especially helpful if we are seeking the Fourier
transform of a function such that we can readily transform its derivative.
A similar procedure was used for Laplace transforms.
These formulas in Equations 6.17 and 6.18 are useful for converting dif-
ferential/integral equations in the variable t to algebraic equations in the
variable w. These equations involve the unknown F(w), and the technique is
similar to that employed for Laplace transforms.
It is instructive for us to perform a Fourier transformation using the calcu-
lus of residues and verify these results with MATLAB. Here is an example.

Example 6.2
1
Using complex variable theory, find F 2 and check your result by
using MATLAB. t +1

Solution:
From Equation 6.15, we must evaluate
∞
1 − iwt
F (w ) =
∫t 2
+1
e dt
−∞

Let us think of t here as a complex variable and integrate around

the semicircular contour shown in Figure  6.5 in the complex t plane.

2
t plane
1

0 R

–1
R
–2

–3
Pole at –i
–4

–5

–6

–7
–6 –4 –2 0 2 4 6

FIGURE 6.5
Contour for Example 1.
260 A MATLAB® Companion to Complex Variables

We assume at first that w is a nonnegative real quantity. Note that our

contour is in the lower half of the t plane so that we can argue using
Jordan’s lemma (see W section  6.6 and S problem 7.7) that the integral
e − iwt
along the arc of radius R vanishes as R → ∞. The function 2 has a
t +1
simple pole in the lower half of the complex plane, at –i, with residue
e − iwt e−w
, which, evaluated at the pole, yields . Thus, passing to the limit
2t −2 i
in R, we are left with
∞
1 − iwt 1 −w
F (w ) =
∫t
−∞
2
+1
e dt = −2 πi
−2 i
e = πe − w (6.19)

Note the presence of the minus sign in front of 2πi, which arises
because the pole is enclosed in the negative (clockwise) direction. The
preceding result holds when w is a nonnegative real. We can write
∞ ∞ ∞
1 − iwt 1 1
F (w ) =
∫ t2 + 1
e dt =
∫ t2 + 1
cos wt dt − i
∫t 2
+1
sin wt dt
−∞ −∞ −∞

where we have employed Euler’s identity for the exponential. The sec-
ond of these integrals is zero as it represents the integral of an odd
function between symmetrical (even) limits. The first integral on the
∞
1
right,
∫
−∞
t2 + 1
cos wt dt , where w is real, is unaffected by a change in the

sign of w because cos wt is an even function of w. Our result for F(w),

Equation 6.19, which was derived from residue calculus, required our
assuming that w ≥ 0. We now realize that we can adapt our answer to
the case w ≤ 0 by simply placing absolute values around w wherever it
may occur. This renders our result, from residues, applicable for w ≤ 0.
Thus, to summarize,
∞ ∞
1 1 − iwt 1
F 2
t +1
= F (w ) =
∫ t2 + 1
e dt =
∫t 2
+1
cos wt dt = πe − w
−∞ −∞

where w is real.
Let us verify the preceding result with the use of MATLAB. To take
a Fourier transform of a function of t, we use the command fourier.
This yields, unless MATLAB is instructed otherwise, a function of the
­variable w. To recover the function of t from the function of w we use the
command ifourier. MATLAB will return, unless told otherwise, a func-
tion of the variable x. Because we are often solving problems involving
time, or t, we may wish to have a function of t. This is accomplished as
shown below.

>> syms w t
>> fourier (1/(t^2+1))
ans = pi∗exp(−w)∗heaviside(w) + pi∗heaviside(−w)∗exp(w)
Transforms: Laplace, Fourier, Z, and Hilbert 261

>> ifourier (ans, t)

ans = −(pi/(t∗i − 1) − pi/(t∗i + 1))/(2∗pi)
>> simplify(ans)
ans = 1/(t^2 + 1)

The answer, pi∗exp(−w)∗heaviside(w) + pi∗heaviside(−w)∗exp(w), for

the Fourier transform, requires some study. If w > 0, then heaviside(−w)
is zero, while heaviside(w) = 1, and we obtain πe−w. Similarly, if w < 0,
then heaviside(w) is zero, while heaviside(−w) = 1. Notice that now
exp(w) = exp(−|w|), and we obtain πe−|w| for w < 0. If w = 0, then MATLAB
uses heaviside (0) = 1/2, and the answer pi∗exp(−w)∗heaviside(w) +
pi∗heaviside(−w)∗exp(w) is equal to π, which will agree with πe−|w| at
w  =  0. Thus, MATLAB’s answer for the Fourier transform agrees with
our result of πe−|w|.
Consider the line of code ifourier (ans, t). We have included the vari-
able t so as to obtain a function of t after the inversion. Had we not
included t, then MATLAB would have given us a function of x. This is
the default variable if none is specified.

We could have performed the inverse transformation using elementary

calculus as follows:
∞ ∞
1 1
F πe−1 −w
=
2π ∫ πe
−∞
− w iwt
e dw =
2π ∫ πe
−∞
−w
(cos wt + i sin wt)dw

∞
1
=
2π ∫ πe −w
(cos wt)dw
−∞

The integral involving sin wt vanishes because it contains on odd function.

Now, we evaluate the final integral by noticing that e−|w|cos wt is an even
function of w. We also employ Euler’s identity:
∞ ∞ ∞
1
−1
F πe −w
=
2 ∫e
−∞
−w
(cos wt)dw =
∫e
0
−w
∫
(cos wt)dw = Re e − w (e iwt )dw
0

∫
= Re e − w(1− it ) dw
0

The last integral on the right is elementary, and we have

∞
−1 −w e − w(1− it )  1 1 + it 1
F πe = Re  = Re = Re 2
=
−1 + it  0 1 − it 1+ t 1 + t2

which is the desired result.
262 A MATLAB® Companion to Complex Variables

Here is a caveat in using the MATLAB command fourier. If you supply

MATLAB with a function that contains the variable x, it will perform the
Fourier integration by integrating on the variable x, even if, say for example,
the variable t is present. Suppose we wish to take the Fourier transform of
1
the function f (t) = 2 by using the definition in Equation 6.2. We might
t +x
carelessly proceed as follows:

>> syms t x
>> fourier(1/(t^2+x))
ans = −pi∗exp(t^2∗w∗i)∗i − pi∗exp(t^2∗w∗i)∗sign(imag(t^2))∗i + pi∗heaviside(−w)
∗exp(t^2∗w∗i)∗2∗i

This is incorrect, as MATLAB treated t as a parameter and integrated on x.

The correct method is this:

>> syms t x
>> fourier(1/(t^2+x),t,w)
(pi∗exp(−w∗x^(1/2))∗heaviside(w))/x^(1/2) + (pi∗heaviside(−w)∗exp(w∗x^(1/2)))/
x^(1/2)

In fourier(1/(t^2+x),t,w) the t,w instructs MATLAB to integrate on t and

produce a Fourier transform that is a function of w.
Practitioners of applied mathematics frequently use Fourier transforms in
situations where a mathematician, accustomed to rigor, might be appalled
because he or she would quickly realize that one or both of the integrals in
Equations 6.15 and 6.16 do not exist. Here is a simple example. Suppose f(t) =
δ(t), the Dirac delta function. Then using the sampling property of the delta
function, we have from Equation 6.15,
∞

Fδ(t) =
∫ δ(t)e
−∞
− iwt
dt = 1 = F(ω ) (6.20)

Now suppose we wish to recover our function of t, namely, δ(t), from F(ω).
Employing Equation 6.16, we have

∞ ∞ ∞
1 1 i
f (t) = F −1 (1) =
2π ∫ e iwt dw =
2π ∫ cos wtdw +
2π ∫ sin wtdw
−∞ −∞ −∞

Evaluating these as Cauchy principal values, we have

 1 R i
R
 2 sin Rt 
lim 
f (t) = lim 
R →∞  2 π
 −R
∫cos wtdw +
2π
−R


∫
sin wtdw  =
2 π R →∞  t 

Transforms: Laplace, Fourier, Z, and Hilbert 263

R
i
We have used the fact that the second integral above,
2π
−R
∫
sin wtdw, which

involves an odd function sin wt, vanishes because of the symmetric limits.

2 sin Rt 
Notice that the limit on the above far right, lim   , does not exist.
2 π R→∞  t 
Thus, we might argue that the inverse Fourier transform of 1 does not exist.
However, MATLAB has other ideas as the following shows:

>> syms t
>> ifourier(1,t)
ans = dirac(t)

The delta function has been recovered. To deal with this contradiction, we
might simply postulate that because the Fourier transform of δ(t) is 1, then
by definition the inverse Fourier transform of 1 must be δ(t). In other words,
we dispense with the integral definition of the inverse transform. Sometimes
we will be forced to do this—a widespread practice in engineering but one
generally abhorrent to mathematicians.*
Here is a further example: The reader should verify with Equation 6.15 that
the Fourier transform of δ(t − τ) is e−iwτ, where τ is a real number. The reader
should also verify that the inverse transformation of e−iwτ cannot be obtained
through the use of Equation 6.16 as the integral does not exist. We have, how-
ever, from MATLAB:

>> syms t tau

>> assume(tau,’real’)
>> fourier(dirac(t−tau))
ans = exp(−tau∗w∗i)
>> ifourier(ans,t)
ans = dirac(t − tau)

The assume statement above tells MATLAB that tau is real. This is a pre-
caution. In the present version that we are using (R2015a), MATLAB assumes
as a default that the argument of the Dirac function is real.
The reader might find it helpful to memorize not only that Fδ(t − t0 ) =
−1 e iω 0t
e − iωt0 (which includes that Fδ(t) = 1), but also F δ(w − w0 ) = which means
1 2π
that F −1δ(w) = .
2π
Sometimes we can be surprised, in the subject of Fourier transforms,
when comparing results obtained from calculus with the results produced

* A presentation of a rigorous treatment of such practices can be found in the book by Hans
Bremermann, Distributions, Complex Variables, and Fourier Transforms. (Reading, MA: Addison
Wesley, 1965).
264 A MATLAB® Companion to Complex Variables

by MATLAB. For example, let us look at the transform of f(t) = 1/(t − 1)2. The
definition of Fourier transforms requires that we do this integration:
∞ − iwt
e
∫ (t − 1)2
dt
−∞

Let us look at the real part:

∞
cos wt
∫ (t − 1) dt 2

−∞

and consider an integration along the subinterval 1 − τ < 1 < 1 + τ. We consider

1+τ
cos wt
∫ (t − 1) dt 2

1−τ

and assume that cos wt > 0 in the interval of integration. If cos wt > m > 0,
we have
1+τ 1+τ
cos wt m
∫ (t − 1)2
dt >
∫ (t − 1) dt
2

1−τ 1−τ

If we attempt the second integral, we find the result is infinite. Thus, the
integral on the left of the inequality cannot exist and, therefore, neither does
the Fourier transform we are seeking. Note that it does not exist as a Cauchy
principal value either. However, consider this from MATLAB:

>> syms t
>> fourier(1/(t−1)^2)
ans = −pi∗w∗exp(−w∗i) + 2∗pi∗w∗heaviside(−w)∗exp(−w∗i)
>> simplify (ans)
ans = −pi∗w∗exp(−w∗i)∗(2∗heaviside(w) − 1)

1
According to MATLAB, the function has the Fourier transform
(t − 1)2

 1 
F
( − ) 2  = −πwe
− iw
( 2u( w ) − 1)
 t 1 

How can we explain this? Notice that the Fourier transform of the function
1 e − iwt
g(t) = does exist as a Cauchy principal value. The expression has
(t − 1) (t − 1)
an integrable singularity at t = 1.
Transforms: Laplace, Fourier, Z, and Hilbert 265

We can find the Fourier transform

∞ − iwt
e

∫
−∞
(t − 1)
dt

by means of residue calculus or MATLAB.

Now recall (see Equation 6.17) that if we know the Fourier transform of a
function, then the transform of its derivative can be found simply by multi-
plying this known transform by iw. The derivative of 1/(t − 1) is −1/(t − 1)2.
Thus, −iw times the Fourier transform of 1/(t − 1) is the Fourier transform
of 1/(t − 1)2. We use residues and verify that this is the procedure consistent
with MATLAB.
Let us assume that w > 0. Noting the presence of the simple pole on the real
axis at t = 1 in the complex t plane and invoking Jordan’s lemma for the lower
half of the t plane, we have

1  e − iwt
F   = −πi Res @ t = 1 = −πie − iw
 t − 1 t−1

Figure  6.6 illustrates the contour of integration. We ultimately pass to a

limit where R → ∞ and ε → 0+.

2
t plane
1
0 1
0 R

–1
Radius ε

–2 R

–3

–4

–5

–6

–7
–6 –4 –2 0 2 4 6

FIGURE 6.6
An indented contour for this problem.
266 A MATLAB® Companion to Complex Variables

Similarly, if w < 0, we use the above contour reflected about the

real axis, and Jordan’s lemma for the upper half plane, and obtain
1  e − iwt
F  = πi Res @ t = 1 = πie − iw. To summarize
 t − 1  t−1

1 
F  = −πie − iw (2 u(w) − 1)
 t − 1 

We have used the Heaviside function to bring about the required change
in sign. Multiplying the preceding by −iw, we obtain −πwe−iw(2u(w)−1), which
1
is indeed the result that will be produced by MATLAB for F . The pre-
(t − 1)2
ceding is also verified with MATLAB as follows:

>> syms w t
>> a1=fourier(1/(t−1))
a1 = pi∗exp(−w∗i)∗(2∗heaviside(−w) − 1)∗i
% the above is the Fourier transform of the given function.
>> a2=−i∗w∗a1
a2 = pi∗w∗exp(−w∗i)∗(2∗heaviside(−w) − 1)
% the above is –iw times times the transform of 1/(t−1)
>> a3=fourier(1/(t−1)^2)
a3 = −pi∗w∗exp(−w∗i) + 2∗pi∗w∗heaviside(−w)∗exp(−w∗i)
% the above is the transform of 1/(t−1)^2
>> a2−a3
ans = pi∗w∗exp(−w∗i) + pi∗w∗exp(−w∗i)∗(2∗heaviside(−w) − 1)
− 2∗pi∗w∗heaviside(−w)∗exp(−w∗i)
>> simplify(ans)
ans = 0
% the preceding shows that the Fourier transform of 1/(t−1)^2 is identical to –iw
%times the transform of 1/(t−1)

6.3.1  Changing Variables: Fourier to Laplace and Vice Versa

A technique that requires shifting between Laplace and Fourier transforms
by means of a change of variables must be applied with some caution (see
W page 380). For example, we can use the MATLAB function laplace to find
the Laplace transform of the Heaviside function. The Laplace transform of
heaviside(t) is given by

∞
1

L heaviside(t) =
∫e
0
− st
dt =
s
Transforms: Laplace, Fourier, Z, and Hilbert 267

while from MATLAB we have that

∞
i
F heaviside(t) =
∫e − iwt
dt = πδ(w) −
w
0

Here is the code and output:

>> syms t
>> laplace(heaviside(t))
ans = 1/s
>> fourier(heaviside(t))
ans = pi∗dirac(w) − i/w

If we replace s in the first result with iw, we do not get the second result—
the Fourier transform. Simply put, the function obtained from the first inte-
gral is not defined at s = 0, so the variable change w = −is, when applied to
this function, is not valid at w = 0, which is exactly where the two transforms
disagree. In general, if a result obtained from a Laplace transform is used
to obtain a Fourier transform by the substitution s = iw, then we should, if
possible, perform the inverse transform on the Fourier result, as a check,
using the command ifourier. The same applies when going from Fourier to
Laplace transforms.

6.3.2  Convolution of Functions

If you did problem 5 in the previous section, you would have been exposed
to a somewhat restricted definition of the convolution of two functions.
The more general definition is this. The convolution of functions f(t) and g(t),
written f(t) ⊗ g(t), is given by

f (t) ⊗ g(t) =
∫ f (t − τ)g(τ) dτ (6.21)
−∞

This definition is consistent with the one used in the previous section.
This is because if we agree to employ functions f(t) and g(t) that are 0 for
t < 0, as we did in dealing with Laplace transforms, then the lower and
upper integration limits in Equation 6.21 can be replaced by 0 and t, respec-
tively. It can be shown that the Fourier transform of the convolution of two
functions is the product of the Fourier transform of each. Conversely (and less
frequently used), it can be shown that if H(w) is a Fourier transform obtained
by the convolution of the functions F(w) and G(w), then the inverse Fourier
transform of H(w) is f(t)g(t), where f(t) and g(t) are the inverse transforms
268 A MATLAB® Companion to Complex Variables

of F(w) and G(w), respectively. Let us try out Equation 6.21 with MATLAB.
1 1
We take f (t) = 2 and g(t) = 2 . We use MATLAB to form the convolu-
t +1 t +9
tion of these functions, and take the Fourier transform of this result, using
MATLAB. Then we use MATLAB to find the Fourier transform of f(t) and
g(t) and show that the product of these results is the Fourier transform of
the convolution of f(t) with g(t).
Here is the MATLAB code:

syms t tau
assume(t,'real')
assume(tau,'real')
%f(t)=1/(t^2+1)and g(t)=1/(t^2+9)our funcs of t.
ftau=1/((tau)^2+1);
gtau=1/((t-tau)^2+9);
h=ftau*gtau;
the_conv=int(h,tau, -inf,inf)
% the above is the convolution of two functions
syms w
assume(w,'real')
Fourier_of_conv=fourier(the_conv,t,w)
%t
he above is the Fourier transform the convolution of the two
%functions
F=fourier(ftau,tau,w)
G=fourier(1/(t^2+9),t,w)
prod_transforms=simplify(F*G)
%t
he above is the product of the Fourier transforms of each
%function
check_me=prod_transforms-Fourier_of_conv
% 
the above should work out to zero, it checks to see that the
%product
%of the transforms is the transform of their convolution.

whose output is

the_conv = (4∗pi)/(3∗(t^2 + 16))

Fourier_of_conv = (pi^2∗exp(−4∗abs(w)))/3
F = pi∗exp(−abs(w))
G = (pi∗exp(−3∗abs(w)))/3
prod_transforms = (pi^2∗exp(−4∗abs(w)))/3
check_me = 0

The line check_me should produce zero, and it does.

Thus, our conjecture that the Fourier transform of a convolution is equal to
the product of the Fourier transform of each function is valid in this problem.
Transforms: Laplace, Fourier, Z, and Hilbert 269

Exercises
1. a.  Using MATLAB, find the Fourier transform of the function
t
f (t) = 2 , where a is assumed to be a positive real. Be sure
t + a2
to use the statement assume (a > 0) in your code. Use the sim-
plify statement to get your final result. Now check your work
by using the statement ifourier on the transform, and verify that
you recover the given f(t). Explain why the Fourier transform of
your f(t) must be a pure imaginary odd function.
b. Find the Fourier transform requested above by using residue
calculus. Consider w positive and also negative.
sin( at)
2. a. Repeat problem 1, parts (a) and (b), but take as f (t) = and
t
a is again real. Explain why the Fourier transform of f(t) must be
an even function of w.
b. Taking a = 1, use MATLAB to plot F(w) for –5 < w < 5.
3. a. Using the definition of the inverse Fourier transform (Equation 6.16),
find the inverse transform of δ(w − w0), where w0 is a real ≥ 0. Give
your result as a function of t, not x (the default). Repeat this for
the function δ(w + w0). Check these results using MATLAB and
the function ifourier.
b. Using the above results for the inverse Fourier transform, state
the Fourier transform of the function sin(w0t).
Hint: Write this as the sum of two exponentials. Now check this
result by applying the operation fourier to this function.
c. Find the Fourier transform of cos(w0t) from the transform for
sin(w0t), which you found in part (a), by using the second entry
in Equation 6.17. Check your result by using MATLAB to find the
Fourier transform of cos(w0t).
4. Assuming that the convolution property for Fourier transforms,
following Equation 6.21, is true, use it to prove Equation 6.18.
Hint: Let f(t  −  τ)  = heaviside(t − τ). Simplify the integral in
Equation 6.21 for the Heaviside function in the integrand and find
the Fourier transform of heaviside(t − τ).
5. a. The present release of MATLAB, R2015 a, will not produce the
Fourier transform of the function arctan(t). Verify that this is
so in the version that you are using. If indeed your MATLAB
will compute the transform of this function, take note of the
result as it will confirm the calculations to be made in what
follows.
270 A MATLAB® Companion to Complex Variables

b. There is a way in which MATLAB can easily compute the Fourier

d arctan(t) 1
transform of this function. Note that = so that
dt 1 + t2
t
dτ
arctan(t) =
∫
−∞
1 + τ2
− π/2 . Confirm this expression for arctan(t)

by doing a symbolic integration in MATLAB.

c. Use this result to get the Fourier transform of arctan(t) by employ-
ing Equation 6.18 and using MATLAB to first get the Fourier
1
transform of 2 . You will also need the Fourier transform of
t +1
π
which you should be able to figure out, or get from MATLAB.
2
6. a. This problem explores how the assume command in MATLAB
can simplify the results obtained from MATLAB. Consider the
1
problem of getting the Fourier transform of f (t) = 2 . Let
t + 2t + a
us assume that a is real. Explain why the question of whether
a > 1 or a < 1 affects the nature of the answer.
b. Using MATLAB, compute the Fourier transform of the function
1
f (t) = 2 , where a is real and a < 1. Use the assume com-
t + 2t + a
mand on a. Use simplify to simplify your answer.
c. Using MATLAB, compute the Fourier transform of
1
f (t) = 2 , where no restrictions are placed on b. It can be
t + 2t + b
any complex ­number. Was it advantageous to use assume in
part (b)?
7. Use Fourier transforms to solve the following integro-differential
equation:
∞
d2 f
∫ f (τ)e − t −τ dτ =
dt 2
+ f (t) + cos(t)
−∞

Proceed as follows:
a. Take the Fourier transform of both sides, recognizing that the left
side is a convolution, and obtain an algebraic equation in F(w).
b. Solve this equation for F(w).
c. Using MATLAB, obtain the inverse Fourier transform of F(w),
thus solving the problem. Simplify your answer and show that it
is real, as it must be, as the equation that it satisfies involves real
functions of a real variable.
Transforms: Laplace, Fourier, Z, and Hilbert 271

6.4  The Z Transform

The Z transform is useful in solving linear difference equations; such equa-
tions arise in the field of digital signal processing. Not surprisingly, the his-
tory of the Z transform is relatively recent and has its origins in the twentieth
century.
A difference equation uses differences (and sums) of the values assumed
by a function for specified finite uniform intervals of its independent real
variable. This is in contrast to differential equations that employ derivatives
of functions that are more or less continuous functions of an independent
variable. Let n ≥ 0 be an integer, and let f(n) be a function that is defined for all
the nonnegative values of n. If we should need f(n) for a negative value of n,
which sometimes happens, we use f(n) = 0, n < 0. This is analogous to f(t) = 0,
t < 0 in our work in Laplace transforms.
Here are some examples of difference equations:

f(n + 1) − f(n) = 2 (6.22)

f(n + 1) − 2f(n) = 0 (6.23)

f(n + 2) − f(n + 1) = f(n) (6.24)

f(n + 1) = (n + 1)f(n) (6.25)

Equation 6.22 says, with slight rearrangement, that the value of f(n + 1) is 2
added to f(n). Thus, if we take as a “boundary condition” that f(0) = 1, then it
is apparent that f(1) = 3, f(2) = 5 … or f(n) = 2n + 1.
Equation 6.23 says that each term in the sequence f(0), f(1), … is twice the
one before it. Taking the zeroth (n = 0) term as 1, you quickly see that the
nth term is f(n) = 2n. Equation 6.24 says that each term is the sum of the two
preceding ones—that is, f(n + 2) = f(n + 1) + f(n); this is the basis of the famous
Fibonacci sequence that often appears in biology and population growth.
By definition, Fibonacci numbers assume that f(0) = 0, f(1) = 1, which would
mean that f(2) = 1, f(3) = 2, f(4) = 3, f(5) = 5, ….
In Equation 6.25, we have that the n + 1 element, f(n + 1) is n + 1 times the
preceding element. Taking f(0) = 1, you quickly discover that f(n) = n!.
Except for Equation 6.24, the problem of finding the nth term was quite
simple. The method of Z transforms provides us with a systematic way
of solving equations such as these, especially, as in Equation 6.24, where
the answer is not readily apparent. Notice that in all these equations f(n),
f(n + 1), … appear only to the first power and that these functions are not
multiplied together as, for example, f(n + 1) f(n − 1). This is why we say that
we are dealing with linear difference equations.
272 A MATLAB® Companion to Complex Variables

In the sequence f(0), f(1), f(2), …, the first element is of course f(0), and the
second is f(1), and so on. But we also refer to the zeroth element, the first
element, and so on, in correspondence with the number n. The usage should
be clear from the context.
Sometimes one is concerned not with f(n) but with f(nT), where T is a posi-
tive real number that represents a time interval from which samples of a
time-varying function f(t) are collected.
Expressions involving f(nT) instead of f(n) frequently appear in
the field of digital signal processing. Such an expression might be
1 1
f (( n + 2 ) T ) = f (( n + 1) T ) + f ( nT ), which would indicate that each new
2 2
value of the sample is the average of the preceding two samples.
Figure 6.7 illustrates a signal from which values f(T), f(2T), f(3T), f(4T), have
been sampled.
MATLAB is set up so as to assume that T = 1, and we follow that convention.
A Z transformation creates an analytic function of the complex variable z
by first forming a Laurent series in the variable z whose coefficients are found
from f(0), f(1), … f(n), …. The series contains no values of z raised to positive
powers. The closed-form expression for the sum of this series, an analytic
function, is said to be the Z transform of the sequence. (See W section 5.8.)
Here is the form of the series:

f (0) f (1) f (2)

+ + 2 + ... (6.26)
z0 z z

0.35

0.3
f(T) f(2T )

0.25
f(3T )
0.2
f(t)

0.15 f(4T )

0.1

0.05

T 2T 3T 4T
0
t

FIGURE 6.7
Sampling a signal or function at intervals of length T.
Transforms: Laplace, Fourier, Z, and Hilbert 273

As we know from complex variable theory, when such a Laurent series

converges, it will do so in a ring-shaped domain, centered at the origin, of
the form, a < |z| ≤ ∞. It is possible that a can be as small as zero. We also
know from complex variable theory that if the series converges, it does so
to an analytic function. If we determine a closed-form expression for the
sum of the series in Equation 6.26 and if we call this analytic function F(z),
we say that

Z[f(n)] = F(z) (6.27)

that is, “the Z transform of f(n) is F(z).” Equivalently,

f (0) f (1) f (2)

F ( z) = + + 2 + ... (6.28)
z0 z z
We also say that “the inverse Z transform of F(z) is f(n).” Mathematically,

Z−1[F(z)] = f(n) (6.29)

Even if we are unable to determine a closed form for the series in

Equation 6.26, we can refer to the sum of this series—an analytic function in
a domain—as the Z transform of f(n) and, likewise, that the inverse Z trans-
form of the sum of this series is f(n).
A property of the Z transform (and its inverse) that is nearly self-evident
and that can be useful is this: the Z transformation is a linear operation,
and so the Z transformation of the sum of two or more functions is the
sum of the Z transform of each. The inverse Z transformation has the same
property.
Here is another useful result: Suppose in Equation 6.26 we make the
change of variable w = 1/z. We then have

F(1/w) = f(0) + f(1)w + f(2)w2 + …

which is valid in a disc-shaped domain centered at the origin. This is a

Maclaurin series. If we let w → 0 in the preceding, we obtain on the right f(0).
Thus,

f (0) = lim F(1/w),

w→ 0

or equivalently,

f (0) = lim F( z) (6.30)

z →∞

If F(z) has no limit as z → ∞, for example, it becomes unbounded, then it

cannot be a Z transformation of f(n).
274 A MATLAB® Companion to Complex Variables

Sometimes a Z transformation is obvious by inspection. An easy exam-

ple is if f(n) = 1 for all n ≥ 0. Then, our Laurent series in Equation 6.26 is
1 1 1
1 + + 2 + 3 + .... Recalling the Maclaurin series,
z z z
1
= 1 + w + w2 + w3 + … w <1
1− w

we replace w with 1/z and find that

1 z 1 1 1
= = 1+ + 2 + 3 + … z >1
1 z−1 z z z
1−
z

z
Thus, Z(1) = is the Z transformation of our given function. The inverse
z−1
z z 
Z transform of is simply 1. In other words, Z −1  = 1.
z−1  z − 1 
Suppose f(n) = 1/n!, n = 0, 1, 2, …. Thus, our Laurent series in Equation 6.19 is

1 1 1 1 1
+ + + +…+ +…
z 0 z 2! z 2 3! z 3 n! zn

The alert reader will recognize this as the Laurent series for e1/z. We say
1 1
that Z   = e 1/z. Of course, the inverse Z transform of e1/z is .
 
n ! n !
It should quickly become apparent that we can choose an expression for
f(n) such that we cannot find its Z transform by inspection. Here is where
MATLAB comes in handy. The reader should now read the documentation
for the MATLAB function ztrans.
Suppose we want to obtain the Z transform of the function f(n) = sin (n).
We proceed as follows:

>> syms n
>> ztrans(sin(n))
ans = (z∗sin(1))/(z^2 − 2∗cos(1)∗z + 1)
>> pretty(ans)

z sin(1)
z − cos(1)z 2 + 1
2

z sin 1
Thus, the Z transform of sin n is 2 , which is the sum of the
Laurent series z − 2 z cos 1 + 1
Transforms: Laplace, Fourier, Z, and Hilbert 275

sin 1 sin 2 sin 3

+ 2 + 3 + ...
z z z

Suppose you want to use MATLAB to find the Z transform of f(n) = 1.

If  you try ztrans(1), you get an error message because the function ztrans
is expecting a symbolic function, not a number, in its argument. However,
ztrans(sym(1)) does work because 1 is now interpreted as a symbolic func-
tion. Try this on your computer, and verify that your result agrees with what
we got by a mathematical calculation.
You can use the Heaviside function to take the Z transform of a constant
in MATLAB. Suppose you want the Z transform of f(n) = 1. We proceed as
follows:

>> syms n
>> ztrans(heaviside (n))
ans = 1/(z − 1) + ½

The preceding is not the Z transform of unity. From earlier in this sec-
z
tion, we know the correct answer is . The problem is that when n = 0,
z−1
the  Heaviside function yields one half. However, we can do the following
trick:

>> ztrans(heaviside (n+1))

ans = z∗(1/(z − 1) + 1/2) − z/2
>> simplify(ans)
ans = z/(z − 1)

This does yield the desired result because heaviside (n+1) is unity for
n = 0, 1, 2, ….
The question of the inverse Z transform (see Equation 6.29) is poten-
tially vexing. The reader should at this point read the help documentation
for the MATLAB function iztrans, which yields the inverse Z transform.
If we  have a function that is analytic in the domain |z| > a > 0, then this
­f unction has a unique Laurent expansion in integer powers of 1/z—that
is, the coefficients f(0), f(1), … in Equation 6.26 are determined. The Z
transform is unique. However, there can often be more than one closed-
form expression that will generate these coefficients. For example, sup-
1 1
pose these numbers are 1, , ,.... These could have been derived from
1 2 3
f (n) = or from
n+1

sin 2 (( 2 n + 1) π/2 )
f (n) =
n+1
276 A MATLAB® Companion to Complex Variables

or from a limitless number of similar expressions. In taking the inverse Z

transform of an analytic expression or its Laurent expansion (which is also
analytic), the best we can say is that the ordered set of numbers that are the
coefficients in the Laurent expansion are uniquely determined.
In general, MATLAB deals with this issue by giving us the formula f(n)
that is obtained from the method of residues in complex variable theory, as
1
follows. Consider
C z m∫
dz, where m is any integer, positive, negative, or zero,
and C is any simple closed contour whose interior contains z = 0. If m = 1, the
value of this integral is 2πi, otherwise its value is zero. (See W section 4.3 and
S section 5.1.) Thus,

∫ C
z − mdz = 2 πi, m = 1,
∫ C
z − mdz = 0, m ≠ 1 (6.31)

Now we have, taking a Z transform,

f (0) f (1) f (2) f (n)

Zf (n) = F( z) = + 1 + 2 +…+ n
z0 z z z

We assume that this series converges for |z| > a > 0.

Suppose we multiply both sides of the above by zn−1, where n ≥ 0 is an inte-
ger. We have

z n− 1 f (0) z n− 1 f (1) z n− 1 f (2) z n− 1 f (n)

z n − 1 F ( z) = + + + … +
z0 z1 z2 zn

and now suppose we integrate both sides around a simple closed curve C
that encloses the origin and lies entirely in a domain |z| > a.

∫ z ∫ z ∫ z ∫ z
n− 1 n− 1 n− 2 −1
F( z) dz = f (0) dz + f (1) dz + … + f (n) dz
C C C C

∫ z
−2
+ f (n + 1) dz + …

C

From Equation 6.31, we see that every integral on the right has the value
zero except the one containing z−1. Thus, from Equation 6.31,

∫ z
n− 1
F( z)dz = 2 πif (n)
C
Transforms: Laplace, Fourier, Z, and Hilbert 277

and we have a contour integration that will provide the inverse Z

transformation:

1
∫ z
n− 1
f (n) = F( z)dz   n = 0, 1, 2, … (6.32)
2 πi
C

The preceding integration is most often done with residues.

Here is a very simple example that we first perform with iztrans and later
verify with residues.

Example 6.3
z
What is the inverse Z transform of F( z) = ?
1+ z

Solution:
>> syms z
>> iztrans(z/(1+z))
ans = (–1)^n

Thus, the answer is f(n) = (−1)n.

Using residues we see from Equation 6.32 that we must evaluate

1 1
f (n) =
∫ z
n
dz
2 πi 1+ z
C

The contour C must enclose all poles of the integrand. There is but one
pole—it is simple—and it lies at z = −1. Using the calculus of residues,
we have that

2 πi zn
f (n) = Res @ z = −1
2 πi 1+ z

The preceding is zn evaluated at z = −1, which yields (−1)n.

Here is a more sophisticated example.
We learned from MATLAB that the Z transform of f(n) = sin n is

z sin 1
F ( z) =
z 2 − 2 z cos 1 + 1

Let us use residues to take the inverse Z transform of F(z) and try to
obtain our f(n). We must find

1 z sin 1
∫ z n− 1
dz n = 0, 1, 2, … (6.33)
2 πi z 2 − 2 z cos 1 + 1
C
278 A MATLAB® Companion to Complex Variables

We choose as a contour a circle of radius r centered at z = 0. The radius

is chosen large enough so as to include all singularities of the integrand.
Otherwise, we would not be justified in using a Laurent expansion to
represent F(z) outside this circle. Notice that all the singularities of

z n − 1z sin 1
n = 0, 1, 2, …

2
z − 2 z cos 1 + 1

are poles and lie where z2 − 2z cos 1 + 1 = 0. Thus

2 cos 1 ± 4 cos 2 1 − 4
z= = e±i
2

Hence, our inverse Z transform is the sum of the residues of

z sin 1
z n− 1 dz @ z = e ± i
z 2 − 2 z cos 1 + 1

Notice the canceling of the 2πi factors when we apply residue calculus
to the expression in Equation 6.33. The expression

z sin 1
z n− 1

2
z − 2 z cos 1 + 1

has only simple poles at its two singularities. The residues are the
numerator divided by the first derivative of the denominator, evaluated
at the poles (see W section 6.3 and S section 7.2).
The residues are
z sin 1
z n− 1
2 z − 2 cos 1
evaluated at ei and e−i. Because these numbers are conjugates of each
other, it should be seen that the residues themselves are also (try plug-
ging these numbers into this expression for the residue). The sum of the
residues is twice the real part of the residue at ei. You may use the other
pole if you prefer. You should convince yourself that the residue at ei is

e in

2 sin 1
(
e − i − cos 1 )
Twice the real part of this expression is sin n, which you should confirm.
This is the desired answer.
The preceding example was not entirely simple, and it would have been
much easier to obtain the desired inverse Z transform with MATLAB. Here
is the procedure:

>> syms a b z
>> a=sym(‘sin(1)’);
>> b=sym(‘cos(1)’);
>> iztrans(a∗z/(z^2–2∗z∗b+1))
ans = sin(n)
Transforms: Laplace, Fourier, Z, and Hilbert 279

There is an important subtlety here: We used symbols to represent sin 1

and cos 1. Had we simply placed sin 1 and cos 1 in the argument of iztrans,
then MATLAB would have sought to replace these irrational numbers by
the quotient of rationals. The final answer would then contain an elaborate
collection of numbers, and the result sin(n) would in no sense be apparent.
As an exercise, try running this piece of code where you use sin1 and cos1
in the argument and they are regarded as numbers, not symbols. Notice the
confusing answer, which I will not supply here as it takes up a lot of space.

>> syms z
>> iztrans(z∗sin(1)/(z^2 – 2∗cos(1)∗z + 1))

In obtaining Z transforms from the function iztrans, we might obtain an

unfamiliar expression.
Here is an example:

>> syms z
>> iztrans(1/(z^2–1))
ans = (–1)^n/2 – kroneckerDelta(n, 0) + ½

The function kroneckerDelta(a, b) contains two real arguments. If they

agree (a = b), the function assumes the value 1. If they disagree (a ≠ b), the
function has value zero. Some versions of MATLAB prior to R2015a may not
contain this function.
1
Let us study the answer just obtained. Note that (–1)^n/2 = (−1)n …
2
1 1
Thus, f (n) = (−1)n − kroneckerDelta(n, 0) + . With n = 0, this is equal to
2 2
1 1
− 1 + = 0.
2 2
When n = 1 it is again zero. Some study will show you that when n > 0 is
even, we have that f(n) = 1; otherwise, f(n) = 0. In Laurent series form, we have

1 1 1
F ( z) = + + + ...
z 2 z 4 z6

This result is confirmed with residue calculus in problem 4.

Sometimes MATLAB will not allow us to perform an inverse Z trans-
formation simply because the inverse Z transform does not exist. In other
words, the given function F(z) cannot be the sum of a series like that in
Equation 6.26. Consider the following:

>> syms z
>> iztrans(exp(1/z))
ans = 1/factorial(n)
>> iztrans(exp(z))
ans = iztrans(exp(z), z, n)
280 A MATLAB® Companion to Complex Variables

Notice that MATLAB gave us the inverse Z transform of e1/z without any
trouble but refused to give us the inverse transform of ez. This is because
(refer to Equation 6.30) lim e z does not exist. In fact, this function has an
z →∞
essential singularity at infinity.
How can we use Z transforms to solve difference equations? Note the
following.
Suppose the function f(n) has the Z transform shown on the right of
Equation 6.28. What is the Z transform of f(n − 1)? Recall that f(n − 1), by defi-
nition, will be zero if n − 1 < 0, or equivalently if n < 1. Thus,

f (0) f (1) f (2)

Zf (n − 1) = + 2 + 3 + ...
z z z

and in similar fashion, we have

f (0) f (1) f (2)

Zf (n − 2) = + 3 + 4 + ...
z2 z z

In general,

f (0) f (1) f (2)

Zf (n − k ) = + k + 1 + k + 2 + ...
zk z z

And if we refer back to Equation 6.28, we have

F ( z)
Zf (n − k ) = (6.34)
zk

where k ≥ 0 is an integer. This is the first of our shifting or translation

­formulas. Now suppose we want Zf(n + 1). We have from our definition of
Zf(n) in Equation 6.28 that

f (1) f (2) f (3)

Zf (n + 1) = + 1 + 2 + ...
z0 z z

Now refer to Equation 6.28; combining it with the above, we have

f (1) f (2) f (3)

Zf (n + 1) = + 1 + 2 + ... = zF( z) − zf (0) (6.35)
z0 z z

Now in a similar fashion, we find

f (2) f (3) f (4)

Zf (n + 2) = + 1 + 2 + ... = z 2 F( z) − z 2 f (0) − zf (1) (6.36)
z0 z z
Transforms: Laplace, Fourier, Z, and Hilbert 281

And in general we have the second translation formula:

f ( k ) f ( k + 1) f ( k + 2)
Zf (n + k ) = + + +…
z0 z1 z2 (6.37)
k −1 k−2
= z F( z) − z f (0) − z
k k
f (1) − z f (2) … − zf ( k − 1)

where k ≥ 0 is an integer.

With the preceding formulas (Equations 6.34 through 6.37), we can solve
some simple difference equations.

Example 6.4
Suppose each term in a sequence is simply the average of the two pre-
ceding terms, and that the initial term in the sequence is 1 and the sub-
sequent term is 2. Calling these the zeroth and first terms, what is the
general (nth) term?
We must solve this difference equation:

1 1
f (n + 2) = f (n + 1) + f (n).
2 2

We have f(0) = 1, f(1) = 2. The Z transform of the left side is according

to Equation 6.36: Zf(n + 2) = z2 F(z) − z2f(0) − zf(1) = z2 F(z) − z2 − 2z. The Z
transform of the right side is, from the linearity property, the sum of the
1
Z transforms of each term. The Z transform of f (n + 1) is according to
Equation 6.36, 2

1 z z
( zF( z) − zf (0)) = F( z) − ,
2 2 2

1 1
while obviously the Z transform of f (n) is F( z). We are left with this
transformed equation: 2 2

z z 1
z 2 F ( z) − z 2 − 2 z = F ( z) − + F ( z)
2 2 2
whose solution is

2 z 2 + 3z
F ( z) = .
2z2 − z − 1
We need the inverse Z transform of this. The easy way is with MATLAB.
We have

>> syms z
>> iztrans((2∗z^2+3∗z)/(2∗z^2–z–1))
ans = 5/3 – (2∗(–1/2)^n)/3
282 A MATLAB® Companion to Complex Variables

>> pretty(ans)
n
5 2  −1 
−  
3 3 2 

n
5 2  −1 
Thus, we see that our answer f(n) is −   . As n → ∞, this has a
3 3 2 
5
limit of . We should check our answer by trying the values n = 0 and 1
3
and seeing that we do indeed obtain the initial values of 1 and 2, respec-
tively. It is interesting to plot this function against n to see how rapidly
the sequence converges to its limit.
Here is the code we used for Figure 6.8 on the next page:

n=[0:10];
y=5/3–2/3∗(–1/2).^n;
plot(n,y,’k∗:’);grid

We might check our answer with residue calculus. According to

Equation 6.32,

f (n) = ∑ residues z n− 1 2 z 2 + 3z
2z2 − z − 1
@ all poles.

This is the same as finding the sum of the residues of

z + 3/2
zn @ its poles.
z 2 − z/2 − 1/2

The zeros of the denominator are readily found to be at 1 and –1/2. Thus,
the poles are simple. The residue at 1 is 5/3, while the residue at –1/2 is
2
− (−1/2)n. The sum of these residues is exactly what MATLAB gave us
3
above.

The preceding problem led to a sequence that has a limit as n → ∞. Here is

a more interesting sequence, whose limit is unbounded, and which leads to
the famous Fibonacci sequence of numbers.* Each element is the sum of the
previous two. Thus,

f(n + 2) = f(n + 1) + f(n)

This is the same equation as the one in the previous problem, but there is
no one half on the right side. The convention is to take f(0) = 0, f(1) = 1, which
would of course mean that f(2) = 1, f(3) = 2, f(4) = 3, f(5) = 5. We want the

* The Wikipedia entry on this subject is short and good. See http://en.wikipedia.org/wiki/
Fibonacci_number
Transforms: Laplace, Fourier, Z, and Hilbert 283

1.9

1.8

1.7

1.6
f (n)

1.5

1.4

1.3

1.2

1.1

1
0 1 2 3 4 5 6 7 8 9 10
Index n

FIGURE 6.8
n
5 2  −1 
Plot of −   .
3 3 2 

general, nth term. We take Z transforms of both sides of the above equation
and apply the conditions on f(0) and f(1). The problem is very similar to the
previous one. We have, after transforming,

z2 F(z) − z = zF(z) + F(z)

z
The solution is F(z) = . Let us try to get the inverse transform of this
2
z −z−1
function with MATLAB.

>> syms z
>> iztrans(z/(z^2–z–1))
ans = (2∗(–1)^n∗cos(n∗(pi/2 + asinh(1/2)∗i)))/i^n – (2∗(–1)^(1 – n)∗(–1)^n∗5^(1/2)∗(1/2
– 5^(1/2)/2)^(n – 1))/5 + (2∗(–1)^(1 – n)∗(–1)^n∗5^(1/2)∗(5^(1/2)/2 + 1/2)^(n – 1))/5
>> simplify(ans,100)
ans = 2∗(–1)^(n/2)∗cos(n∗(pi/2 + asinh(1/2)∗i)) + 1/2^n∗(5^(1/2)/5 – 1)∗(5^(1/2) +
1)^n – 1/2^n∗(5^(1/2)/5 + 1)∗(1 – 5^(1/2))^n

The above is certainly confusing. We rewrite it more clearly as

 nπ  1  5  1  5 
− 1 ( 5 + 1) − n  + 1 ( 1 − 5 )
n n
f (n) = 2 i n cos  + ni arcsinh(1/ 2) + n 
 2  2  5  2  5 

284 A MATLAB® Companion to Complex Variables

The first term on the left is still daunting. One could simplify it by first
assuming n is an odd integer, applying the formula cos(a + b) = cos a cos b −
sin a sin b, and also by using the logarithmic representation of the arcsinh.
Later one assumes that n is even and repeats the process. A more fruitful
approach is to use MATLAB to again compute the inverse Z transform, but
one first simplifies the problem by using the method of partial fractions
(see W section 5.5, or any standard calculus book).
Notice that z2 − z − 1 = 0 has solutions

1 5 1 5
z1 = + and z2 = − ,
2 2 2 2

which means that we have the representation

z 1 ( 5 + 1)/2 1 ( 5 − 1)/2
F ( z) = = +
z2 − z − 1 5 z − 1 (1 + 5 ) 5 z − 1 (1 − 5 )
2 2

The reader may wish to review the method of partial fractions to see how
this result was obtained. We then find the inverse Z transform of F(z) by tak-
ing the inverse transform of each of these fractions separately and adding
the results. To make our answer useful, we should enter the square root of 5
as a symbol in MATLAB, otherwise it will be converted to a decimal expres-
sion. Here is our code for finding f(n):

syms a z b c f_n
a=sym(‘sqrt(5)’);
b=iztrans((a+1)/2∗1/(z–(1+a)/2))/a;
c=iztrans((a–1)/2∗1/(z–(1–a)/2))/a;
f_n=b+c;
simplify(f_n)

And here is our output:

–(5^(1/2)∗((1/2 – 5^(1/2)/2)^n – (5^(1/2)/2 + 1/2)^n))/5

We can simplify the above by multiplying through by the minus sign on

the left and putting 51/2/5 = 1/ 5 . We also extract the factor 2n in the denomi-
nator. Our final answer is rather simple:

1 ( 1 + 5 )n − ( 1 − 5 )n  (6.38)
f (n) =
2 n
5  
Transforms: Laplace, Fourier, Z, and Hilbert 285

We expect the Fibonacci numbers to be integers for n = 0, 1, 2, …, and they

are, despite the presence of the 5. Using the binomial theorem on each of
the binomial expressions in the brackets, you should be able to prove that
there are ultimately no terms in the brackets (after you do the indicated sub-
traction) that do not contain an integer times 5. This square root is canceled
by the one outside the brackets in the denominator.
Here is a little program that prompts you to enter an integer. The output
will be all the Fibonacci numbers up to and including that integer.

nmax=input('nmax =')
N=[0:nmax];
a=sqrt(5);
for j=1:length(N)
n=N(j);
nth_fibo(j)=1/a*((1+a)^n-(1-a)^n)/2^n;
end
'number= fibonacci number='
disp([N' round(nth_fibo)'])

Notice the use of the function round. This ensures that we do not see any
round-off errors arising from the use of the square root of 5 raised to various
powers.
Here are the first 10 Fibonacci numbers from this program:

ans =
number = fibonacci number =
0 0
1 1
2 1
3 2
4 3
5 5
6 8
7 13
8 21
9 34
10 55

Although as n → ∞ the nth Fibonacci number tends to infinity, the ratio of

the n + 1 number to the nth approaches a finite limit as n tends to infinity.
From Equation 6.38, we have that

f (n + 1) 1 + 5  1 − p n+ 1 
=
f (n) 2  1 − p n 

286 A MATLAB® Companion to Complex Variables

where

1− 5
p=
1+ 5

Notice that |p| < 1 and observe that as n → ∞ the expression in the brackets
has a limit of 1. Thus, the ratio of a Fibonacci number to the one preceding
( )
it has in the limit n → ∞ the value 1 + 5 /2. This irrational expression is
the golden ratio often written as φ and is approximately 1.618033988749895…
There is a vast amount of literature on this number, and the reader may wish
to begin with the Wikipedia entry on this subject to get started. A rectangle
having these proportions has historically been said to be the most pleasing
possible and to serve as the basis for classical Greek architecture.* The ratio
of the 10th Fibonacci number to the 9th, which we obtain from the above list,
is obtainable from MATLAB as follows:

>> vpa(55/34)
ans = 1.6176470588235294117647058823529

which looks almost like an irrational number (it cannot be, of course, and
note the repeating decimal starting with 1764…) and which is remarkably
close to the value φ. Note that we used vpa (i.e., variable precision arithme-
tic) so as to obtain as many decimal places as we could for our comparison.
One can see that the use of Fibonacci numbers is a way to generate num-
bers whose numerical ratio (the (n+1)th divided by the nth), which unless
you can find the repeating decimal, appears to be irrational. Of course, as
n tends to infinity, the limit of this ratio will be irrational. The code that we
used to generate the Fibonacci numbers, which is based on Equation 6.38,
may not be the best way to find the Fibonacci numbers, as demonstrated in
problem 12.
Here is another useful formula. Consider the Z transform:

f (1) 2 f (2) 3 f (3) nf (n)

Znf (n) = + 2
+ 3 +…+ n
z z z z

Recall that

f (0) f (1) f (2) f (n)

F ( z) = + 1 + 2 +…+ n +…
z0 z z z

* H. E. Huntley, The Divine Proportion: A Study in Mathematical Beauty. (Mineola, NY: Dover
Books on Mathematics, 1970).
Transforms: Laplace, Fourier, Z, and Hilbert 287

is the Z transform of f(n). We differentiate both sides of this equation within

the domain in which the given series is valid. We have

dF − f (1) 2 f (2) 3 f (2) nf (n)

= − − 4 … − n+1 − …
dz z2 z3 z z

Notice that the preceding right side can be turned into the Z transform of
nf(n) (see above) if we multiply it by –z. Thus,

dF
Z(nf (n)) = − z (6.39)
dz

where F(z) = Zf(n).

Since the Z transform of nf(n) is –z times the derivative of F(z), it must fol-
low that the Z transform of the expression n2f(n), or n × nf(n), is –z times the
derivative of the Z transform of nf(n) from which it follows:

dF d2 F
Z(n2 f (n)) = z + z 2 2 (6.40)
dz dz

Here is a simple application of Equation 6.39. Suppose we want the Z trans-

form of the function n. We regard n as the product of n and the ­number 1,
z
and we saw at the start of this section that the Z transform of 1 is . Thus,
we have z−1

d  z  z
Z ( n × 1) = − z  =
dz  z − 1  ( z − 1)2

We can check this result with MATLAB by either asking for ztrans(n) or
 z 
iztrans  2 
.
 ( z − 1)  z
We can also take the inverse Z transform of with residues as
( z − 1)2
described in Equation 6.32.
We have that the inverse Z transform of

 z 
 ( z − 1)2 
 

is

1 z
∫ z
n− 1
dz
2 πi ( z − 1)2
288 A MATLAB® Companion to Complex Variables

where we might use a circle of radius exceeding 1 and centered at the ori-
gin. Notice that the integrand has a pole of order 2 at z = 1. We use a stan-
dard formula for computing the residue at this pole (see W section 6.3 or
S ­problem 7.3). Recall that if a function f(z) has a pole of order 2 at a point
z0, then

d
Res[ f ( z), z0 ] = lim (( z − z0 )2 f ( z))
z → z0 dz
d n
Employing this, we have that our inverse Z transform is z evaluated at
z = 1. The result is simply n, as required. dz
It is useful to have the Z transform of n2 for a problem that is to follow. We
use Equation 6.40, taking, as in the example just done, f(n) = 1 (whose Z trans-
z z2 + z
form is ). We obtain as the required transform. To summarize,
z−1 ( z − 1)3

z
Z(1) = (6.41)
z−1

z
Z(n) = (6.42)
( z − 1)2

z2 + z
Z(n2 ) = (6.43)
( z − 1)3

Example 6.5
Problem
Find the sum of the squares of the first N integers (i.e., find a closed-form
expression for S = 12 + 22 + 32 + … N2).

Solution:
The reader is probably already aware of the simpler result

1
1 + 2 + 3 + ... + N = N ( N + 1).
2

This result was known to the ancient Greeks (e.g., Pythagoras). In the
early 18th century, a young Swiss boy, Leonhard Euler, who was to
become a great mathematician, figured it out by writing down the
sum of the terms on a horizontal line and then placing the same sum
­u nderneath in reverse order. He then added the two lines, term by
term. The present problem is harder.
We begin with the difference equation

f(n + 1) = (n + 1)2 + f(n),

Transforms: Laplace, Fourier, Z, and Hilbert 289

where f(0) = 0 and n = 0, 1, 2, …. We now take the Z transform of the right

side and left side. Note that from Equations 6.35 and 6.43 that

z 2 ( z + 1)
Z(n + 1)2 = .
( z − 1)3

We use Equation 6.35 on the left as well. We obtain, after transforma-

tion and some simplification, this formula:

z 2 ( z + 1)
F ( z) = .
( z − 1)4

In problem 8, we evaluate the inverse Z transform with residues. Here,

we use MATLAB as follows:

syms z
r=iztrans(z^2*(z+1)/(z-1)^4)

The output is

r = 4∗n + 5∗nchoosek(n – 1, 2) + 2∗nchoosek(n – 1, 3) – 3

a!
The expression nchoose(a,b) means , which is obviously a
b ! ( a − b)!
binomial coefficient.
We thus rewrite our answer as

5 ( n − 1)! ( n − 1)! − 3 = 4n + 5 (n − 1)(n − 2) + 2 (n − 1)(n − 2)(n − 3) − 3

4n + +2
( n − 3)!2! ( n − 4)!3! 2 6

After placing the last expression over the common denominator 6 and
doing a little more simplification, we have the result that

n(n + 1)(2 n + 1)
f (n) = 1 + 2 2 + 32 + ... n2 =
6

which you should check by putting n = 0, 1, 2.

We can skip the preceding by using the following in MATLAB. We add
one more step to the code shown above:
expand(r)

This yields

ans = n^3/3 + n^2/2 + n/6

n(2 n2 + 3n + 1) n(n + 1)(2 n + 1)

This is which agrees with .
6 6
290 A MATLAB® Companion to Complex Variables

6.4.1  The Z Transform of a Product

Suppose we have functions f(n) and g(n), both defined for the nonnegative
integers. Consider now h(n) = f(n)g(n) defined for these same integers. We
have

f (0) g(0) f (1) g(1) f (2) g(2) f (n) g(n)

Zh(n) = + + +…+ + … (6.44)
z0 z1 z2 zn

If we know the Z transforms of f(n) and g(n), is there some way we can
obtain the Z transform of h(n) without doing the calculation in Equation 6.44?
The answer is yes, as we now show.
Now f(n) has Z transform F(z), and g(n) has Z transform G(z). Here,

∞
F ( z) = ∑ c /z
m= 0
m
m
cm = f (m) (6.45)

G( z) = ∑ d /z
n= 0
n
n
dn = g(n) (6.46)

We assume that both F(z) and G(z) are analytic in the domain |z| > a > 0.
Next we introduce a new variable w and work with F(w) and G(z/w). We
place these two variables in the above series and have

F (w ) = ∑c w
m= 0
m
−m
(6.47)

and

∞ ∞

G( z/w) = ∑ n= 0
dn ( z/w )− n = ∑ dnw nz− n (6.48)
n= 0

Multiplying these functions and their corresponding series together, we

have

∞ ∞

F(w)G( z/w) = ∑∑c d w

m= 0 n= 0
m n
n− m − n
z (6.49)

We now use a circle of radius b centered at the origin in the w plane.

We  assume that b > a. Thus, F(w) is analytic on and outside this circle.
We require that |w| ≥ b > a. Now let us agree to choose |z/w| ≥ b or |z| ≥
b|w|. With these constraints, both the series in Equations 6.47 and 6.48 are
Transforms: Laplace, Fourier, Z, and Hilbert 291

uniformly ­convergent in the region |w| ≥ b. We now multiply both sides of

Equations 6.47  and 6.48 by w−1 and integrate as follows around |w| = b.
We  can  exchange  the order of integration and summation because of the
uniform convergence of both series, whose terms are continuous, and thus
obtain

∞ ∞
cmdnw n−m
∑ ∑ ∫
1 F(w)G( z/w) 1

2 πi ∫
w =b
w
dw =
2 πi m= 0 n= 0
z− n
w =b
w
dw (6.50)

From the theory of residues, we know that the integral, which is the in
1
variable w, will vanish except if n = m. Recalling that
w
dw = 2 πi when the
∫
integral is taken around any simple closed contour enclosing the origin, we
have that

∞ ∞

∑ ∑ f (n)g(n)z
1 F(w)G( z/w)
∫ cndn z − n = −n
dw = (6.51)
2 πi w n= 0 n= 0
w =b

Looking at Equation 6.44, we see that the right side of the above equation is
the Z transform of the product f(n)g(n). Thus, to summarize,

1 F(w)G( z/w)
Z  f (n) g(n)  =
2 πi ∫
w =b
w
dw (6.52)

where F(z) and G(z) are the Z transforms of f(n) and g(n), and the integration
is taken around a circle centered at the origin such that F(w) and G(w) are
analytic on and outside of it. With b as the radius of the circle, we require
that |z| > b|w|. But since |w| = b on the path of integration, this becomes
simply |z| > b2.

Example 6.6
Using MATLAB find the Z transforms of

1
f (n) = e in and g(n) = .
n+1

With these results as well as Equation 6.52 and the calculus of resi-
e in
dues, find the Z transform of h(n) = . State any restrictions that
( n + 1)
must apply to the variable z in your answer. Check your result by using
MATLAB to get Zh(n).
292 A MATLAB® Companion to Complex Variables

Solution:
From MATLAB,

>> syms n
>> ztrans(exp(i∗n))
ans = z/(z – exp(i))
>> ztrans(1/(n+1))
ans = –z∗log(1 – 1/z)

z z
Thus, F( z) = and G( z) = − z log(1 − 1/z) = z log .
z − ei z−1
e in
According to Equation 6.52, the Z transform of h(n) = is the
integral n+1

1 F(w)G( z/w) 1 1 z  z/w 

2 πi ∫
w =b
w
dw =
2 πi ∫
w =b
w − ei w
log   dw
 z/w − 1 
(6.53)
1 1 z  z 
=
2 πi ∫
w =b
w − ei w
log 
 z − w 
dw

w
Notice that F(w) = is analytic on and outside any circular con-
w − ei
tour in the w plane given by |w| = b > 1. This is because the only singu-
larity of this function is at w = ei, which lies on the unit circle. Now

w
G(w) = w log .
w−1
We use the principal branch of this function. This is defined by means
of a branch cut on the real axis of the w plane that is given by 0 ≤ w ≤ 1. On
this segment, the argument of the log will be negative real, 0, or infinity.
Notice that G(w) is also analytic for |w| ≥ b. As discussed in our deriva-
tion, we take |z| > b2 in Equation 6.53. Notice that

z 1
log = log
z−w 1 − w/z

w
and that because < 1 on the path of integration |w| = b, the path does
z z
not intersect the branch cut for log .
z−w
Now the reader should evaluate the integral in Equation 6.53, noting
the apparent pole at w = 0. In fact, this is not a pole but a removable singu-
larity (see W section 6.2 and S section 3.11), which has no residue. There
is a residue to be found at w = ei. Our final answer is
e in z z 1
Z = i log = ze − i log
n+1 e z−e i
1 − e i/z

which is valid when |z| > 1.

Transforms: Laplace, Fourier, Z, and Hilbert 293

We check this with MATLAB as follows:

>> syms n
>> ztrans(exp(i∗n)/(n+1))
ans = –z∗exp(–i)∗log(1 – exp(i)/z)

1
The above is identical to ze − i log .
1 − e i/z

6.4.2  Inverse Z Transform of a Product and Convolutions

Suppose f(n) has Z transform F(z) and g(n) has Z transform G(z). Thus,

c1 c 2
F ( z) = c 0 + + +… c0 = f (0), c1 = f (1), … cn = f (n) (6.54)
z z2

d1 d2
G( z) = d0 + + +… d0 = g(0), d1 = g(1), … dn = g(n) (6.55)
z z2

Let us multiply the two Laurent series together. We get the Laurent series
for the product of the Z transforms:

c0 d1 + c1 d0 c0 d2 + c1 d1 + c2 d0
F( z)G( z) = c0 d0 + + + … (6.56)
z z2

Note that the coefficient of 1/zn in the above is

k=n

∑c d
k=0
k n− k = an n = 0, 1, 2, … (6.57)

and so

a1 a1
F( z)G( z) = a0 + + +…
z z2

We say that the sequence of numbers a 0, a1, a2, … results from the con-
volution of the sequences c 0, c1, … and d 0, d1, …. Alternatively, we say that
the function f(n) is convolved with the function g(n). We write this as
f(n) ∗ g(n). The convolution is commutative—it is easy to see that f(n) ∗ g(n) =
g(n) ∗ f(n). This definition of the convolution, for functions defined only for
discrete values of their independent variable (n = 0, 1, 2, 3, …), is the ana-
log of the convolution for continuous functions that we discussed for the
Laplace transform.
What we have discovered is that the Z transform of the convolution of two
sequences is the product of the Z transforms of each sequence. But perhaps
294 A MATLAB® Companion to Complex Variables

just as important, we see that the inverse Z transforms of the product of two
functions is the convolution of the sequences giving rise to each function.
Here is an example.

Example 6.7
We saw in our work above that
z z
Z(1) = and Z(e in ) = .
z−1 z − ei

Use convolution to find

z2
Z −1 .
( z − 1)( z − e i )
Check your result with residue calculus.

Solution:
Note that our result can be obtained through a convolution of f(n) = 1 and
g(n) = ein. Thus, our resulting sequence is a0 = 1 × ei0, a1 = 1 × ei1 + 1 × ei0,
a2 = 1 × ei2 + 1 × ei1 + 1 × ei0 …
It should become apparent that

( ) + (e ) ( )
0 i 1 n
an = 1 + e i + e 2 i + … e in = e i + … + ei

Recalling the sum of the finite geometric series:

1 − r n+ 1
1 + r + r2 + … rn = , r≠1
1− r
we see that
1 − e i( n + 1)
1 + e i + e 2 i + … e ni = .
1 − ei
This is

z2 1 − e i( n + 1)
Z −1 = .
( z − 1)( z − e )
i
1 − ei

We can check this result with Equation 6.32. We have

z2 1 z2
Z −1 =
( z − 1)( z − e ) 2 πi
i ∫ z n− 1
( z − 1)( z − e i )
dz, n = 0, 1, 2, …
C

We will take the contour C as a circle centered at the origin and having
radius greater than 1. Notice that there are simple poles of the integrand
at z = 1 and z = ei. Our result is

z n+ 1 z n+ 1
Res @ 1 + Res @ ei
( z − 1)( z − e )
i
( z − 1)( z − e i )
Transforms: Laplace, Fourier, Z, and Hilbert 295

This becomes

1 e i( n + 1)
−
1 − e (1 − e i )
i

which agrees with our answer obtained above. We can also check our
answer by using the command iztrans in MATLAB, but there are some
subtleties involved, as shown in exercise problem 13.

Exercises
1. a. Using MATLAB find the Z transform of f(n) = eian, where a is an
arbitrary constant that in MATLAB you must treat as a symbol.
Check this result by summing the series

∞ n
 e ia 
∑  z 
n= 0

which is easily done as it is a geometric series.

b. Use our result from part (a) to find the Z transform of g(n) = e−ian.
c. Use the two preceding results to find the Z transform of
h(n) = sin (an).
Check this result by using MATLAB.
2. a. Certain functions do not have a Z transform. Show that f(n) = n!
does not have such a transform. To do this apply the ratio test (see
S theorem 6.11 or any standard calculus book) to the series whose
n!
nth term is n and see where this series converges in the complex
plane. z
b. Verify your finding by trying to find the Z transform of f(n) with
MATLAB. Recall that in MATLAB we use factorial (n) to p
­ roduce n!.
What does MATLAB return to you?
c. Can you think of another well-defined function of n, not
involving factorials, that does not have a Z transform? Try
to get this function’s Z transform with MATLAB to see what
happens.

3. a. Determine the Z transform of the function f (n) =

( −1)n . Do this
n!
by writing down the defining Laurent series. The sum of this
series should be familiar.
296 A MATLAB® Companion to Complex Variables

b. Check your answer by using MATLAB.

c. Provide another check by applying Equation 6.32 to the Z trans-
form that you obtained above, and verify that you get back
your f(n).
4. We used MATLAB in this section to find the inverse Z transform of
1
the function F( z) = 2 . Check that answer by finding the inverse
z −1
Z transform using the calculus of residues.
5. a. In a certain sequence, each successive number is obtained as
a weighted average consisting of two thirds of the previous
number and one third of the number before that one. Suppose
the initial number, call it f(0), is zero and the next ­number
f(1) is  one. Find a general formula for the nth number by
writing down a difference equation that you will solve with
Z transforms.
b. Obtain a graph like that shown in Figure  6.8 that shows how
quickly this weighted average procedure approaches its limit.
What is the limit? Use the formula you derived for the nth term
in the sequence.
6. a. Use MATLAB to find the Z transform of the function defined
as f(n) = 0, n ≤ 1, f(n) = 1, n ≥ 2. It is helpful to use the Heaviside
function here.
b. We showed in this section that the Z transform of the function
z
that is equal to unity for all n ≥ 0 is . Combine this result with
z−1
the  translation formula, Equation 6.34, to check your answer to
part (a).
c. Check your answer by recovering f(n) by applying Equation 6.32
to the F(z) you found in parts (a) and (b).
7. The formula 0 + 1 + 2 + 3 + … + n = n(n + 1) is well known and can be
derived by elementary school students if given some help. We derive it
here with Z transforms. If you put n = 0, 1, 2, 3, … on the left in the pre-
ceding, you generate a sequence of numbers. Note that the elements
in the sequence satisfy the difference equation f(n + 1) − f(n) = n + 1,
where f(0) = 0.
a. Take the Z transform of both sides of this equation and solve
for F(z).
b. Find the inverse of this transform by using the calculus of resi-
dues (i.e., Equation 6.32).
c. Check your answer by finding the inverse of the Z transform
with MATLAB.
Transforms: Laplace, Fourier, Z, and Hilbert 297

8. In solving the problem of obtaining the formula for the sum of the
squares of the first n integers, in one of our worked examples, we
had to find the inverse Z transform of

z 2 ( z + 1)
F ( z) = .
( z − 1)4
We did this using MATLAB. Confirm your result by using resi-
dues in the complex plane.
9. a. Use MATLAB to obtain the Z transform of f(n) = ein.
b. Obtain the Z transform of g(n) = n2ein by using your result from
part (a) as well as Equation 6.40.
c. Check your result by using MATLAB to find the Z transform of g(n).
10. a. We found the Z transform of the function f(n) = n in this sec-
tion. Use one of the translation formulas we derived to find the Z
transform of g(n) = n + 1.
b. Now use the transforms of n and n + 1 to find the Z transform of
the function h(n) = n(n + 1) by means of Equation 6.52. Evaluate
the integral with residues.
c. Check your answer to part (b) by asking MATLAB to compute
the Z transform of n(n + 1).
11. a. Use MATLAB to find the Z transform of the function given by

1
f (n) = , n ≥ 1 and f (n) = 0, n ≤ 0.
n
The Heaviside function is useful here. Notice that an expres-
sion such as heaviside(x − 1/2)/x will, when evaluated at x = 0 by
MATLAB, yield the result NaN (not a number) because of the ratio
0/0. MATLAB skips this point when evaluating a Z transform.
b. Check your answer to part (a) by recovering f(n) by applying
Equation 6.32 to the answer to part (a). Notice that the answer to
part (a) requires the use of a branch cut. What are the requirements
on this cut? The integral that you will get by applying Equation
6.32 cannot be evaluated with residue calculus because the inte-
rior of the contour you use contains branch cut singularities of the
integrand. However, you should make the change of variables w =
1/z and perform your integration in the w plane. Notice the branch
cut is now outside the contour of integration, and you are now
integrating in the clockwise (negative) direction. The series

1 2 1 3 1 4
log(1 − w) = − w − w − w − w −… w <1
2 3 4
is useful here.
298 A MATLAB® Companion to Complex Variables

12. We have seen that the Fibonacci numbers can be obtained as the
coefficients in the Laurent series expansion
∞

2
z
z −z−1
= ∑ zc n
n
n= 0

1
Suppose we make a change of variables w = in both sides of the
z
preceding equation. The right side becomes a Maclaurin series in the
variable w, which converges to the rational function in w that now
appears on the left. Now look up the MATLAB documentation for
the function taylor. Recall that a Maclaurin series is a Taylor series
expansion about the origin.
a. Use taylor to find the first 11 Fibonacci numbers (i.e., the first 10
nonzero ones). Note that MATLAB will give you them in reverse
order.
b. Where does this infinite Taylor series converge in the complex w
plane?
13. a. Try to use MATLAB to obtain the inverse Z transform of the
function

z2 1
F ( z) =
z − 1 z − ei

Notice that you will get a confusing and messy answer

because MATLAB will convert ei to a numerical value.
b. Repeat part (a) but use the symbolic expression exp(sym(i)) in
place of ei in MATLAB. Be sure to apply the operation simplify
to your answer. Which is the more satisfying expression, the one
from part (a) or part (b)?
14. Find a closed-form expression for the sum of the series:

n
e 0 cos 0 e 0 cos 1 + e 1 cos 0 e 0 cos 2 + e 1 cos 1 + e 2 cos 0
z0
+
z
+
z2
+ ... + ∑
k=0
e k cos(n − k )
zn
+ ...

by taking the Z transform of two simple functions of n that you

can find with MATLAB. State the conditions on z for this series to
converge.
15. a. Use Equation 6.40 to find the Z transform of the function g(n) = n3
by starting with the transform of n, which we have already
derived. You may use MATLAB to take the needed derivatives.
Use the function diff.
Transforms: Laplace, Fourier, Z, and Hilbert 299

Check your answer by finding the transform with MATLAB.

Without using MATLAB, find the Z transform of h(n) = (n + 1)3.
b. Suppose f(0) = 0, f(1) = 1, and that f(n + 1) = (n + 1)3 + f(n). Starting
with n = 0, the terms generated are 03, 13, 13 + 23, 13 + 23 + 33, …,
13  +  23  + 33 + …  n3, …. Take the Z transform of both sides of
the difference equation, and use the result derived in part (a) to
show that

z( z 2 + 4 z + 1)
F ( z) =
( z − 1)5

c. Use either MATLAB (the expand function is useful here) or resi-

dues as in Equation 6.32 to show that

n2
f (n) = (n + 1)2
4
How can you be sure this will always produce an integer?
If you elect to use Equation 6.32, you have to compute a fourth
derivative. It is best to do this with MATLAB to avoid mistakes.
Check your f(n) by putting in n = 0, 1, 2, 3, ….

6.5  The Hilbert Transform

Unlike the Laplace and Fourier transforms, which emerged in the eighteenth
and nineteenth centuries, respectively, the Hilbert transform is relatively
recent and came into use in the twentieth century. It ultimately proved itself
useful in the latter twentieth century in analysis of electrical systems and
in the study of modulated electrical signals such as radio waves. Students
of complex variable theory will see it as a means of relating the real and
imaginary parts of many analytic functions, while practitioners of electri-
cal engineering will see its value in relating the real and imaginary parts of
the system function of a broad class of electrical systems. In contrast to the
Laplace and Fourier transforms, the present version of MATLAB (R2015a)
does not yield the Hilbert transform of a continuous function. It will perform
a Hilbert transformation of a function that is defined for uniformly spaced
discrete time intervals, but this takes us too far from the world of functions
of a complex variable and will not be discussed here.
Some earlier versions of MATLAB would perform the Hilbert transform
of a function that is defined over the real axis by a mathematical expression.
This capability was contained in the Signal Processing Toolbox. Because the
present version of MATLAB does not contain this feature, we write our own
300 A MATLAB® Companion to Complex Variables

software to do the Hilbert transform of a broad class of functions of a real

variable. This is a useful exercise in MATLAB programming.

6.5.1 Definition
Suppose g(t) is a function of the real variable t. In most applications this is a
real function, but this is not a requirement. It will, however, simplify our dis-
cussion if we make this assumption. We call the function’s Hilbert transform
gˆ ( x) and define it by the following integral, assuming that x is real and that
the integral exists:
∞
1 g(t)
gˆ ( x) =
π ∫ x − tdt (6.58)
−∞
We also write

gˆ ( x) = Hg(t) (6.59)

Here, H indicates that we are performing the Hilbert transform (transfor-

mation). The integral in Equation 6.58 is to be thought of as a Cauchy prin-
cipal value both with respect to how its infinite limits are treated and with
how we deal with the singularity at t = x.
The inverse of the above equation is

g(t) = H−1 gˆ ( x)

where H−1 indicates that we are to take the inverse Hilbert transform. (See
W section 6.10.) Let us use the calculus of residues to find Hsin t. We must do
the integration
∞
1 sin t

π ∫ x − tdt
−∞

This is a familiar integral from residue calculus which is performed using

the method of indented contours and is evaluated as
∞
1 e it
Im
π ∫
−∞
x−t
dt

We use the contour shown in Figure 6.9, which is in the complex τ plane, not
the z plane. The real axis in the τ plane is the t axis. The integral

1 e iτ

π ∫ x−τ
dτ
Transforms: Laplace, Fourier, Z, and Hilbert 301

6
CR τ plane
5

2 R Radius ε
CE
1

0 –R R
t axis x–ε x x+ε
–1

–2
–6 –4 –2 0 2 4 6

FIGURE 6.9
An indented contour to compute our Hilbert transform.

taken around this contour is zero because the integrand is analytic on and
in the contour.
We pass to the limits R → ∞ and ε → 0+. Notice, from Jordan’s lemma, that

1 e iτ
lim R → ∞
π ∫ x−τ
dτ = 0

over the arc CR, while over the arc Cε, we have

1 e iτ πi e iτ
lim ε → 0 +
π ∫ x−τ
dτ = − Res
π x−τ
@τ = x

After evaluating this, we have

1 e iτ

π ∫ x−τ
dτ = ie ix

over the indentation. Thus, with a little rearrangement, we have

∞
1 e it
Im
π ∫
−∞
x−t
dt = − Im ie ix = − cos x
302 A MATLAB® Companion to Complex Variables

or
Hsint = −cos x
for our Hilbert transform.
Once a function, say g(t), has been Hilbert transformed into gˆ ( x), how do
we recover g(t)? In other words, how do we find the inverse transform? The
answer is that for a wide class of functions, we can take the Hilbert transform
of gˆ ( x) and put a minus sign in front of the result. There are two proofs in W
(section 6.10). The simplest is based on Fourier transforms and requires that
we accept the validity of the Fourier transform inversion formula. A  more
complicated proof requires complex variable theory and the use of indented
contours and residues. The Fourier transform proof has the advantage that it is
less restrictive as to what functions can be transformed (e.g., delta functions).
To summarize the results, if
∞
1 g(t)
gˆ ( x) = Hg(t) =
π ∫ x − tdt
−∞

then
∞
−1 gˆ ( x)
g(t) =
π ∫ t − xdx = H
−∞
−1
gˆ ( x) (6.60)

The middle expression in the above equation is the negative of the Hilbert
transform of g(t).
Of course, the preceding assumes that both integrals exist as Cauchy prin-
cipal values.
How can we use MATLAB to evaluate a Hilbert transform and its inverse?
We do not use the MATLAB command hilbert for the problems we have
been doing, as it is intended to be used with functions defined for discrete
uniformly spaced values of some independent variable.
Study Equation 6.58 for a moment. If
∞
1 g(t)
gˆ ( x) =
π ∫ x − tdt
−∞

1
then the function gˆ ( x) is seen to be the convolution of the functions g(t) and
.
πt
See Equation 6.21 as a reminder, and recall that the Fourier transform of a
convolution of two functions is the product of the Fourier transform of each.
Let us now work with gˆ (t) instead of gˆ ( x). Thus,

1
Fgˆ (t) = Fg(t)F
πt
Transforms: Laplace, Fourier, Z, and Hilbert 303

The Fourier transform of gˆ (t) is the product of the Fourier transforms of

1
g(t) and .
πt
Suppose we take the inverse Fourier transform of both sides. The left
 1
side is simply gˆ (t), and we have gˆ (t) = inverse Fourier transform  Fg(t)F .
 πt
In the language of MATLAB, we have

1
gˆ (t) = ifourier fourierg(t) × fourier  (6.61)
 πt 

1
The Fourier transform of is easily calculated with either residues or
MATLAB, and we have πt

>> syms t
>> Fourier(1/(pi∗t))
ans = heaviside(–w)∗2∗i – i

This function is equal to –i for w > 0 and i for w < 0. It is zero for w = 0. The
following MATLAB script M file will determine the Hilbert transform of a
function that we supply. We have elected to give this M file the name hiltrial.
After creating the file and saving it, you can type this word in the command
window. You will be prompted to produce the function of t that you want to
be transformed.
If you elect to use this M file, you may save it with any other name that
MATLAB does not use for a function or file. Do not use the name hilbert—it
has been taken.
Here is the M file:

syms t
g=input('g is equal to ')
G=fourier (g);
syms pi
h=1/(pi*t);
H=fourier(h);
C=ifourier (G*H);
assume(x,'real')
The_Hilbert=simplify(C)

Notice that we treat pi as a symbol so that it will not be reduced to a

numerical value in our calculation. We did choose to have MATLAB find
1
the Fourier transform of , although we might have used the transform we
πt
found above; in a later example we do that.
304 A MATLAB® Companion to Complex Variables

We use the M file, as follows, in the command window. We demonstrate its

use in obtaining the Hilbert transform of sin t:
>> hiltrial
g is equal to sin(t)
g = sin(t)
The_Hilbert = –cos(x)
We were given the prompt g is equal to, and we then supplied sin(t). Our
result The_Hilbert confirms our calculation by residues: namely, –cos(x).
Another way to evaluate Hilbert transforms in MATLAB is by creating
a function to do the job. We employ this function in the same way that we
employ any MATLAB function, such as the sin, log, or int. We have chosen
to call our function hlbtrn, and we write it in the following code:
function [output ] = hlbtrn( func )
% this gives you the hilbert transform of the
%function that is the argument of hlbtrn.
%the input is a function of t, the output is a function of x.
syms t w h x
H=
@(f)fourier(f(t),t,w);%this creates a nested in line
%function% of the function f(t)
t)=func; % this establishes that f(t) is the argument you
f(
%fed in %when calling hlbtrn
h=H(f);% this evaluates the Fourier transform
% of the input f(t)or func
%(2*heaviside(-w) - 1)*i is the Fourier transform of 1/(pi*t)
output=simplify(ifourier(h*(2*heaviside(-w) - 1)*i));
end
% the end is needed because you have a nested function within
%the one we created

The function will evaluate the Hilbert transform of sin t if we proceed as

follows in the command window:
>> syms t
>> hlbtrn(sin(t))
ans = –cos(x)
The reader may wonder why we did not use the MATLAB function int
to perform the integration in Equation 6.58. This should directly yield the
Hilbert transform without our taking a route through Fourier transforms.
Try getting the Hilbert transform on your computer of the function sin t by
using the following:
>> syms x t pi
>> assume(x,’real’)
>> H=1/pi∗int(sin(t)/(x–t),t,–inf,inf,’PrincipalValue’,true,’IgnoreAnalyticCons
traints’,true)
Transforms: Laplace, Fourier, Z, and Hilbert 305

You will be surprised at how complicated and clumsy the answer looks. It
does not resemble the correct answer of −cos t. MATLAB is much more sophis-
ticated in handling Fourier and inverse Fourier transforms than in perform-
ing integrations using int. This may change in later versions.
Suppose we wish to use either hiltrial or hlbtrn to find the Hilbert trans-
form of the number 1. Here we try both. Notice that both hiltrial and hlbtrn
expect to be fed a symbolic function, not a number. Thus, we create a symbol
to stand for the number 1, and then summon either the M file or the function:

>> a=sym(‘1’)
a=1
>> hiltrial
g is equal to a
g=1
The_Hilbert = 0

Now we try out the function that we created called hlbtrn

>> hlbtrn(a)
ans = 0.

In both cases, we obtain zero. In fact, if you try any constant, you will find
that its Hilbert transform is equal to zero. You may confirm this by using
real variable calculus to evaluate Equation 6.58, taking the integrand as a
real constant. This is alarming as it tells us that the Hilbert transform is not
unique and therefore neither is its inverse transform. Thus, you could argue
that the inverse Hilbert transform of zero is any number you wish.
How can we avoid this situation? From W (see pages 420–421) we learn
the following: Consider the function f (t) = g(t) + igˆ (t). Here the imaginary
part is the Hilbert transform of the real part. Now consider the function
f (τ) = g(τ) + igˆ (τ), where τ is a complex variable whose real part is t.
Now assume that

1 f (τ)
lim R → ∞
π ∫ x − τdτ = 0,
where the integration is performed over a semicircle of radius R in the com-
plex τ plane, and x has any real value. The center of the circle is at the origin
of the τ plane, and the semicircle lies in either the upper or lower half plane.
The function f(τ) must be analytic on and inside at least one of these semi-
circles. If this behavior is satisfied, the transform is unique. A function g(t)
equal to a constant will not satisfy this condition as f(τ) will not satisfy the
requirement on the arc of the semicircle when R → ∞.
Because MATLAB has more sophistication in handling Fourier and
inverse Fourier transforms than it does in ordinary symbolic integration,
306 A MATLAB® Companion to Complex Variables

a third method—probably the best—is presented here for finding Hilbert

transforms. Recall the definition of the Fourier transform:
∞

Ff (t) = F(w) =
∫ f (t)e − iwt
dt
−∞

Suppose we take
g(t)
f (t) = (6.62)
x−t
If we evaluate the Fourier transform of the above and then set w = 0, we
have arrived at the Hilbert transform of g(t). In other words,
∞
g(t)
gˆ ( x) = lim w → 0
∫ π(x − t) e
−∞
− iwt
dt (6.63)

Typically, the Fourier transform of a real function of t is a complex function.

But the Hilbert transform resulting from the above calculation must be real.
We have created a MATLAB M file called hilfor that will prompt the user to
supply a function g(t) and return its Hilbert transform. Here is the code that you
1
might wish to experiment with by trying various functions, for example, 2 :
t +1
% the script file hilfor
syms t x w
g=input('g(t) is equal to ')
syms pi
h=(1/pi)/(x-t);
G=@(w)fourier (g*h,t,w);
assume(w,'real')
The_hilbert=simplify(real(G(0))

Exercises
1. Do the following problems by using either one or the script files or
functions described above, like hiltrial, or hilfor or hlbtrn, or a pro-
gram of your own devising that does Hilbert transforms. Find the
Hilbert transforms of these functions, g(t). Check your answers by
doing contour integration directly.
a. cos 2t
b. sin(t + 1)
1 − cos 2t
c.
t
Transforms: Laplace, Fourier, Z, and Hilbert 307

2. We asserted that if you take the Hilbert transform of a function g(t),

then take the Hilbert transform of the resulting function, and multi-
ply the result by –1, you should recover the original g(t). Verify this
in the case of the function g(t) = sin t by using one of the programs
described above in Problem 1.
sin t
3. a. Find the Hilbert transform of the function g(t) = by using
residues. t

b. Find the Hilbert transform of this same function by using

­hiltrial, hlbtrn, and hilfor given above. Which gives the most
useful result? Does it agree with the result of part (a)?
c. Using hilfor, find the Hilbert transform of the result of part (a)—
that is, transform gˆ (t)—and verify that you get the negative of
the function g(t) supplied in part (a) except that the independent
variable is now x.
4. The Hilbert transform can be used to solve certain integral equa-
tions. Consider the equation

∞
1 f (t) sin x
π ∫ x − tdt + x
= 2 cos x
−∞

Notice that the integral on the far left is the Hilbert transform of
the function f(t). Thus, if we take the inverse Hilbert transform of
the expression, we recover f(t). Take the inverse Hilbert transform
of both sides of the above equation (using the linearity property
of the Hilbert transformation and its inverse) and find f(t). Check
your answer by using either residue calculus to find the Hilbert
transform of your answer or by using the program hilfor devel-
oped in this section.
This page intentionally left blank
7
Coda: Fractals and the Mandelbrot Set

A FRACTAL GIRAFFE from The New Yorker magazine

In this book, I have sought to create a bridge between complex variable the-
ory and the MATLAB programming language. It seems fitting to end with
a branch of mathematics, which, although dating in its origins to the early
1900s, really did not come into flower until the widespread use of high-speed
digital computers in the 1980s and cannot really be appreciated without a
programming language such as MATLAB®. This is the subject known as
fractals, which is a branch of a larger field in mathematics called chaos theory.
The public is exposed to chaos theory in newspaper articles that speak of
“the butterfly effect.” The term was coined in 1969 by Edward Lorenz, an
MIT Professor of Earth Sciences, who is famous for his research in weather
prediction. He is usually credited with the observation that if a butterfly
flaps its wings, it might trigger a storm appearing many hundreds of miles
away. To a mathematician, this is saying that the solution of a differential
equation, or some iterative process, is acutely sensitive to the initial conditions
imposed on the equations. Lorenz was at the time working on solutions of

309
310 A MATLAB® Companion to Complex Variables

the Navier–Stokes equations, which are nonlinear differential equations that

are at the heart of the theory of the behavior of fluids and therefore of the
weather.
We will not be studying these equations, or in fact any differential equa-
tions. The reader might unknowingly already be familiar with chaos theory,
which is about the acute sensitivity of the outcome of a real or computer
experiment to initial conditions. Go to the top of a hill on a windless day
and release, one at a time, a series of seemingly identical newly bought rub-
ber balls from a particular spot. You will find, of course, that the balls take
different paths, and that the paths differ increasingly as the balls make their
ways down the hill. The differences are due to small errors arising because
you did not place each ball in exactly the same spot, or because the balls
have subtle differences in their manufacture, or because you inadvertently
imparted a slightly different velocity to each ball as you set it adrift.
Our glimpse of chaos theory will come from some computer experiments
in the complex plane. We perform the experiments using MATLAB. We will
be using MATLAB programs to generate what are called fractals. The term
was coined in 1975 by the Polish-born mathematician, Benoit Mandelbrot,
who spent most of his professional life in the United States. It has the same
Latin root as the words fractured and fraction and suggests something that is
not whole (i.e., broken).
What is a fractal? There is no neat universally accepted definition of the
term. We describe here some of the attributes of those fractals that can, in
theory, be drawn on a piece of paper, provided you have a pen with an infi-
nitely fine nib. With a real pen or a computer’s printer, you can construct a
fair approximation to many fractals. There are fractals that you might build
out of clay and hold in the palm of your hand, but these fractals possessing
mass will not concern us.
Look at a map of England or the state of Maine and follow the coastline
depicted there. With the aid of a ruler, or perhaps a piece of thread and some
pins, you can, if you know the scale of the map, estimate the length of the coast.
But suppose you decided to walk along the actual coast, following the high
water mark of the tides. You would come up with a greater measure for the
coastline’s length. An ant following the high water mark would walk an even
greater distance than you, while an insect smaller than the ant, and more sen-
sitive to the fluctuations of the boundary, would log an even greater trip. This
procedure of using a finer and finer measure could proceed indefinitely (we
stop short of the molecular scale) so that we might think that the coastline is
of infinite length. Moreover, we might conclude that the curve describing the
coastline could never be said to possess at any point a definable tangent, since
at some scale of scrutiny the curve would be seen as rough, not smooth. A frac-
tal is typically composed of an infinite set of points. If they are committed to a
piece of paper, plotted with our fine-pointed pen, they will create a figure that
like the coastline has no definable tangent. The language of fractals is used by
some mathematicians to describe coastlines of countries and states.
Coda: Fractals and the Mandelbrot Set 311

You and I are accustomed to speaking of one-, two-, and three-dimensional

objects. A fractal, by contrast, often, but not always, has a dimensionality that
is not a whole number—possibly fractional or irrational. We do not have the
space here to explain what a fractional dimension is, so the reader might
want to refer to other books for additional information.*
We explore a number of fractal sets in the exercises. In what follows we
study the fractal set that has received the most publicity and which is the
most interesting one known to this writer. This is called the Mandelbrot set.
It is obtained from an iterative process. Mandelbrot has taken credit for his
eponymous set, which he claimed to discover in 1979, but there is some fas-
cinating controversy about the set’s origins which the reader might wish
to pursue at this web address http://www.scientificamerican.com/article/
mandelbrot-set-1990-horgan/
An iterative process can take the form

zn+1 = f(zn) (7.1)

where n is a nonnegative integer and f(z) is a suitably defined function. We

begin with an initial value, typically z0, and then obtain z1 from the above
equation, and from this z2, and so on. By way of illustration, we use the func-
tion used in creating the Mandelbrot set:

zn+ 1 = zn2 + c (7.2)

This does look very simple, but you are in for a surprise. Before showing
how this formula defines the Mandelbrot set, we need to review the concept
of a sequence of complex numbers that remains bounded.
Suppose the elements of the sequence are the set of numbers z0, z1, z2, ... zn,
zn+1, .... This sequence is bounded if and only if the following is true: given any
positive real number M there exists a real number N such

|zn| < M for all n > N (7.3)

Thus, the sequence i, i2, i3, ... in, ... is bounded (the magnitude of each term
is one), while the sequence 1+i, (1+i)2, (1+i)3, ... (1+i)n, ..., whose magnitude for
( )
n
the nth term is 2 , is not bounded. You should be able to show that the
sequence consisting of principal values 1i, 2i, 3i, ... ni, ... is bounded, but if we
replace i with 1 + i, we do get a sequence that is not bounded.
Here is how we find the elements of the Mandelbrot set, the most famous
fractal. We choose a numerical value for c (any number, real or complex)
and choose for our “seed” the number z0 = 0. We plug this into Equation 7.2
and compute z1, which is of course equal to c. Knowing z1, we return to this

* Fractals, Chaos, Power Laws: Minutes from an Infinite Pradise by Manfred R. Schroeder, New York:
WH Freeman, 1990.
312 A MATLAB® Companion to Complex Variables

equation and see that z2 = c2 + c, and in the same way, we have z3 = (c2 + c)2 + c =
c4 + 2c3 + c2 + c. The expressions become increasingly complicated. The
Mandelbrot set is defined as follows:

It is the set of values of c for which the sequence of values obtained from
Equation 7.2 is bounded when we use as our starting value z0 = 0.

Notice that we said nothing about the convergence of the sequence.

The sequence might very well diverge, but c could still be in the Mandelbrot
set. The sequence whose nth term is isin(nπ/4) does not converge (you might
want to look up the definition of convergence), but if you found such a
sequence in the course of using Equation 7.2, you would say that the value of
c in use does belong to the Mandelbrot set.
It should be obvious that certain values of c do or do not belong to the
Mandelbrot set. For example, c = 1 results in the sequence 0, 1, 2, 5, 26, …,
which you should verify by means of Equation 7.2. The terms are getting
bigger and bigger. Thus, c = 1 is not in the set.
What about c = −1? You should verify that the sequence looks like 0, –1, 0,
–1, 0, …, which although a divergent sequence, is bounded. Thus, c = –1 does
lie in the set.
And what about c = i? This is quite interesting. The sequence is easily
found to be 0, i, −1+i, −i, −1+i, −i, −1+i, .... Once we have passed the first two
terms, every term is either −i or −1+i. Again, we have a divergent sequence
but not an unbounded one. The reader should see that if we took c = –i, our
sequence would consist of the conjugate of every term found for the case
c = i. Thus, c = ±i are both in the Mandelbrot set. You should notice that if
a value of c is found to be in the Mandelbrot set, then so is the conjugate of
that value, since the sequences z0, z1, z2, ... zn, zn+1, ... obtained in each case
are conjugates of each other, and their being bounded (or not) is identical
in each case. And of course, if a value of c is not in the set, then neither is
its conjugate. You should also observe the truth of the following: in ­trying
to determine whether a certain value of c belongs to the Mandelbrot set,
you encounter two values zp and zq that are identical (zp = zq, p ≠ q), then the
sequence will not be unbounded but will cycle between these two values.
Thus, c is in the Mandelbrot set.
How can we tell if a number c belongs to the Mandelbrot set without a lot
of tedious numerical calculation? At this point you should read the impor-
tant results contained in the appendixes.
From the first we see that a necessary condition (but not sufficient condition)
is that |c| ≤ 2. Thus, the Mandelbrot set is contained entirely on and inside a circle
of radius 2 centered at the origin in the complex plane. From the second appen-
dix, we find that if we are testing to see whether c is in the Mandelbrot set
(first assuring ourselves that |c| ≤ 2) and discover a value of |zn| > 2, then we
can conclude that the value of c is not in the set. Indeed, we can conclude that
the sequence |zn|, |zn+1|, |zn+2|, ... is monotonically increasing without bound.
Coda: Fractals and the Mandelbrot Set 313

Imagine that we begin, by means of a MATLAB program, to compute the

values z1, z2, z3, ... . Suppose after computing 10 or 20 such numbers we find
none whose magnitude exceeds the number 2. Can we conclude that we have
found a value of c that lies in the Mandelbrot set, or might we be worried that
the next number we compute might be one whose magnitude exceeds 2, thus
indicating that c is not in the set.
This is a real concern that mathematicians have thought about. In the 1980s,
when pure and applied mathematicians began using computers to explore
fractals and especially the Mandelbrot set, a number of popular articles and
books assured the reader that he or she could stop the computation employ-
ing Equation 7.2 if after reaching the 200th term a value of |zn| exceeding 2
had not been reached. They asserted that it was highly unlikely that further
iteration would result in a term whose magnitude exceeds 2.* In 2015, as I
write this, computers are many orders of magnitude faster, and we will typi-
cally go to a much higher level of iteration (e.g., 10 million iterations). We will
see why this is sometimes necessary.
It is easy to write a MATLAB program that prompts us to supply a value
of c to be tested and that will go through 107, namely 10 million, iterations of
Equation 7.2 and then tell us if c lies in the Mandelbrot set.
Here is the code, which we call mandelcheck:

q=1;
while q>0
c=input('the complex value of c to be checked')
tic
nmax=10e6;
%this is the max number of iterations, but
%the iterations will stop if |zn|>2
n=1;
z=0;
while(abs(z)<=2)
z=z^2+c;
n=n+1;
if n>nmax
break
end
end
if n<=nmax
disp('the value of zn that causes early termination if n<nmax')
z
end
disp('number of iterations used')
n−1
if n−1==nmax

* See A. J. Crilly, R. A. Earnshaw, and H. Jones. Fractals and Chaos (New York: Springer, 1991): 36.
314 A MATLAB® Companion to Complex Variables

disp('the number c is in the set')

else
disp('the number c is not in the set')
end
toc
q=input('q is neg to stop')
end

Let us try the program out by seeing if the point .2+.5∗i is in the Mandelbrot
set. Here is what occurs on our computer screen when we run the above
program:

>> mandelcheck
the complex value of c to be checked .2+.5∗i
c = 0.2000 + 0.5000i
number of iterations used
ans = 10000000
the number c is in the set
Elapsed time is 8.165984 seconds.

The calculation took around 8 seconds. If the point chosen lies outside the
set, the calculation takes less time because the iteration procedure termi-
nates before the 10 millionth value of zn is found. We see that in determining
whether c = .2 +.7∗i lies in the set—it does not—we require a lot less time.
Here are our results:

>> mandelcheck
the complex value of c to be checked .2+.7∗i
c = 0.2000 + 0.7000i
the value of zn that causes early termination if n<nmax
z = −1.0957 + 1.7956i
number of iterations used
ans = 6
the number c is not in the set
Elapsed time is 0.003120 seconds.

We see that by the time we reach z6, whose magnitude exceeds one, we can
conclude that .2+.7∗i is not in the Mandelbrot set.
Even though we are allowing for 10 million possible iterations, the pro-
gram is not perfect. It can be shown that if c is positive real, then it must
satisfy c ≤ 1/4. This is derived in problem 3. Thus, we might wish to test our
program by choosing c just to the right of one quarter. Here is an instance
where we try both c = .25 + 1000∗eps and also .25+100∗eps. In the second
case, we got an incorrect answer. The number should have been found to not
Coda: Fractals and the Mandelbrot Set 315

lie in the set. Recall that eps is the smallest positive real that MATLAB can
handle. For my version of MATLAB, this was found as follows:

>> eps
ans = 2.2204e−16

If you type eps you will find the value in use on your computer. Our results
for these two values of c are as follows:

>> mandelcheck
the complex value of c to be checked .25+1000∗eps
c = 0.2500
the value of zn that causes early termination if n<nmax
z = 2.3790
number of iterations used
ans = 6666992
the number c is not in the set
Elapsed time is 0.136802 seconds.
q = negative to stop, positive to keep going 2
q=2
the complex value of c to be checked .25+100∗eps
c = 0.2500
number of iterations used
ans = 10000000
the number c is in the set
Elapsed time is 0.202285 seconds.

As noted, the second answer is erroneous. But, if you change the number
of iterations to something larger (e.g., 100 million), the computer determines
that the number is not in the set.
At this point, we might recall Lorenz’s statement about a butterfly flapping
its wings. Here is an analogous finding from the Mandelbrot set. Suppose
c = i. We saw above that c = i does lie in the Mandelbrot set. If you try out
our program Mandelcheck with any of these four values i±10−6 or i±i10−6,
you will find that none of them lies in the Mandelbrot set. In other words,
if we use the iterative formula Equation 7.2 with any of these four values of
c, we develop sequences the magnitudes of whose elements grow without
bound. Yet these points seem to lie “close” to z = i. These small deviations
from i might make us think of the flapping of the butterfly’s wings causing
a storm.
The reader might be thinking, “What’s so special about that? After all, the
series 1 + z + z2 + z3 + ... will converge if z = .99999999 but will diverge to infin-
ity if z = 1.” The difference is that we can state quite simply where this series
will converge, namely, it converges if and only if |z| < 1. If asked to find all
316 A MATLAB® Companion to Complex Variables

points in a small neighborhood of c = i that lie in the Mandelbrot set, we are

facing a very difficult problem. In the exercises at the end of this section, we
see how to find some such points by numerical, not analytic, means.
At this juncture, we have found a way to determine whether a point is in
the Mandelbrot set but we have not created a graphical picture of the set,
which would be our first example of a fractal. What is a fractal? There is no
uniformly agreed-upon on definition. The reader might wish to refer to the
book by Falconer in the footnotes for an overview of the definition and might
also want to look in a complex variables text to review the meaning of such
words as open set, closed set, boundary point, interior point, connected set, and
bounded set.* The following is a good start at a working definition of a fractal.

1. A fractal exhibits some degree of self-similarity at all scales of magnifi-

cation. This means that if you examine a portion of a fractal through a
magnifying glass, it will look much, or exactly, like, some other portion.
A familiar example might be the hourly change of the Dow Jones stock
average as reported in the newspaper. Without the axes scales that
were provided, you might think that this shows the same quantity over
a year. If you look at the photograph of a cloud and there is nothing else
in the picture, you have no way of knowing how big the cloud is. The
ears of the giraffe at the start of this Coda look like the giraffe. And
if you could study the ears on the ears you might see another giraffe,
although I doubt if the cartoonist bothered with such a small scale.
2. A fractal exhibits complexity at every degree of magnification. This
reminds one of the coastline of Maine, which looks complicated on
a map and which will look complicated if you study the line left by
the high water mark while you are walking. Look at the photograph
of the edge of a cloud and then look at the photograph through a
magnifying class. It still looks complicated.
3. The classical methods of calculus, used to determine the slope of a
curve (or its tangent), do not apply. We cannot say in general what
the local tangent is to a curve generated from fractals.
4. Mathematical fractals (as opposed to fractals in nature, like clouds)
are usually obtained from a fairly simple recursive formula. We have
just seen the one that Mandelbrot used, Equation 7.2.

It is easy and fun to use MATLAB to generate a curve exhibiting self-­

similarity. Here is an example of such a program and its output:
x=[0:100];%x assumes the integer values from 0 to 100
y=(1-(-1).^x)/2;%at even x values this is zero,at odd x is one
r=
(.9).^((x-1)/2);%at x=1,3,5,...gives values of
%1,.9,.9^2,,.9^3....

* Kenneth Falconer, Fractals: A Very Short Introduction (New York: Oxford University Press, 2013).
Coda: Fractals and the Mandelbrot Set 317

y=y.*r; %multiplies y at the peaks by 1,.9,.9^2,.9^3 etc

xx
=(1-sqrt(.9).^(x+1))/(1-sqrt(.9));% this is the partial sum
%of the series
%1
+a+a^2+a^2...,a=sqrt(.9) each term in the sum is sqrt(.9)
%times the one preceding it,
plot(xx,y,'linewidth',2);grid

This program produces the output as shown in Figure 7.1.

Notice that the height of each peak is 9/10, the height of the peak just to
the left of it. The base of each triangular peak is 9/10, the width of the peak
just to the left of it. Magnifying any portion of the curve the right amount
can make it look like a portion of the curve to the left of it. This is the essence
of self-similarity. But this is not a fractal—it lacks the necessary complexity;
except where the curve abruptly changes slope, it has a well-defined tangent.
To illustrate what a fractal can look like, we now write and use the
code that will illustrate the Mandelbrot set. We call this program simply
mandel_one.

q=1;
while q>0
xo=input('x coord center of box=')
yo=input('y coord center of box=')
dx=input('half width of box, x direction=')
dy=input('half width of box, y direction=')
nx=input('number of x divisions')
ny=input('number of y divisions')
tic

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 2 4 6 8 10 12 14 16 18 20

FIGURE 7.1
Self similarity.
318 A MATLAB® Companion to Complex Variables

cr=linspace(xo-dx,xo+dx,nx);
ci=linspace(yo-dy,yo+dy,ny);
[Cr,Ci]=meshgrid(cr,ci);
c=Cr+i*Ci;%this creates a grid of complex numbers for c
nmax=200;% we use 200 iterations of the recursion relation
j=1;
z=
zeros(size(c));%this starts off z, for each value of c,
%at the value zero
ile j<=nmax;%iterates the expression below nmax times;​
wh
%200 here;
z=z.*z+c;
j=j+1;
end
ck
=abs(z)<=2;%puts a symbol of 1 in the matrix where |z|<=2;
%otherwise puts %0
=1.0*ck;%converts symbolic elements to numerical in above
dk
%matrix.
p=pcolor(cr,ci,dk);
se
t(p,'EdgeColor','none');colormap(gray);grid;%use gray for a
%black and white %picture
set(gca,'layer','top'); %required if you hope to see the grid
axis([xo-dx xo+dx yo-dy yo+dy])
q=input('negative q to stop')
end

In this program, we are asked for the center and dimensions of a rectan-
gular box. The program then creates a plot on your screen where points in
the box that lie in the Mandelbrot set are given the color white, while the
other points are given the color black. Recall that we introduced the function
pcolor at the end of Chapter 2. If you wish to reverse this convention, use the
line of code p=pcolor(cr,ci,-dk). Notice the minus sign in front of the symbol
dk. However, there is then the risk that a spot of dark dirt on your screen
might be erroneously taken for an element of the set.
We know that the Mandelbrot set must lie on and within a circle of radius
2 centered at the origin of the complex c plane. We generate our set inside a
square box centered at the origin. The sides are parallel to the real or imagi-
nary axes, and each side is of length 4. Thus, we are guaranteed that the set
will be inside the box and no part will be left out (Figure 7.2). We will use
slightly over 1 million date points, as follows:

>> mandel_one
x coord center of box=0
xo = 0
y coord center of box=0
yo = 0
half width of box, x direction=2
dx = 2
Coda: Fractals and the Mandelbrot Set 319

1.5

0.5

–0.5

–1

–1.5
–2 –1.5 –1 –0.5 0 0.5 1

FIGURE 7.2
The Mandelbrot set.

half width of box, y direction=2

dy = 2
number of x divisions1001
nx = 1001
number of y divisions1001
ny = 1001
negative q to stop (−1)
q = −1

We chose odd numbers (namely 1001) for nx and ny, the number of x and y
subdivisions of our box in the x and y directions, to ensure that there would
be elements lying on the horizontal and vertical lines passing through the
center of the box. This is especially useful if the center of the box lies on the
x axis.
By choosing ny as odd, we ensure that we are choosing elements that lie on
the x axis, which is an axis of symmetry for the Mandelbrot set.
In 1981–1982, two mathematicians, A. Douady (a Frenchman) and
J.H.  Hubbard (an American) proved that the Mandelbrot set is simply
connected and closed. The set has no holes and contains all its boundary
points. Choosing any two points in the set, we should be able to con-
nect them by a chain of curved and straight segments all of whose points
lie in the  Mandelbrot set. We know that z = i is in the Mandelbrot set.
320 A MATLAB® Companion to Complex Variables

And so is its conjugate z = −i. Yet neither of these points appears as a white

dot in the above plot, and there is no path going from these points to the
points depicted in white in the figure. There is in fact a slender thread
going from the points z = ±i to the main body of the set, but the number
of points that we might discover in any section of the thread are so few
that if they could be reduced to pixels they cannot be seen on the screen.
There are several ways out of this dilemma. We present one here.
If we apply the iterative formula Equation 7.2 to test whether a value of c lies
in the Mandelbrot set and if that value of c does not belong to the set, then the
application of Equation 7.2 will eventually result in a value of |zn+1| > 2. It has
been found that the further this value of c lies from a boundary point of the
Mandelbrot set, the sooner the condition |zn+1| > 2 is encountered. Similarly,
the closer this value lies to a boundary point, the later this condition is
encountered. Thus, a plot showing not only the Mandelbrot set but showing
points that narrowly escape being in the set can help us see the set, especially
close to its boundaries. This procedure is written in a code employing what
is called “an escape time algorithm.” Such an algorithm assigns a color to not
only points in the set but to points that nearly made it into the set. Because
for a narrow filament connecting the main (cardioid shaped) body of the
Mandelbrot set to the point z = i, there are numerous nearby points that do
not quite fall into the Mandelbrot set but do surround the filament, a plot of
them suggests the path of the filament. Such a technique can help us study
the complicated boundary of the Mandelbrot set that contains innumerable
such filaments.
Here is an example of a program that will plot not only the Mandelbrot set
but nearby points that did not quite qualify for being in the set. We call this
program Mandelescape:

q=1;
while q>0
xo=input('x coord center of box=')
dx=input('half width of box, x direction=')
yo=input('y coord center of box=')
dy=input('half width of box, y direction=')
nx=input('number of x divisions')
ny=input('number of y divisions, preferably use odd=')
tic
cr=linspace(xo-dx,xo+dx,nx);
ci=linspace(yo-dy,yo+dy,ny);
[Cr,Ci]=meshgrid(cr,ci);
c=Cr+i*Ci;
figure(1)
z=zeros(size(c));
for j=1:20
z=z.*z+c;
end
Coda: Fractals and the Mandelbrot Set 321

D=abs(z);
d=(D<=2);%this gives a matrix having ones for those values of c
%that are apparently in the Mandelbrot set after 20 iterations
grid;axis equal;
w=z;
for j=21:1000
w=w.*w+c;
end
DD=abs(w);
=(DD<=2);%this gives a matrix having ones for those
dd
%values of c
%t
hat are taken to be in the Mandelbrot set after 1000
%iterations
p=
pcolor(cr,ci,d/2+dd/2);%the matrix d/2+dd/2 will have
%ones at those
%values of c that are in the matrix after 1000 iterations
% 
it will have the value 1/2 at those values of c that are
%in the
% 
matrix after 20 iterations but not after 1000; these
%values are not
%i
n the Mandelbrot set;values of c that are eliminated
%after just 20
%iterations are represented by zeros
set(p,'EdgeColor','none')
colormap(hot);hold on
grid on;axis equal
xlim([xo-dx xo+dx])
ylim([yo-dy yo+dy])
set(gca,'layer','top');% this allows you to see the grid
toc
q=input('choose a negative q to stop')
end

From our previous plot, we have some idea of the rough outlines of the set.
Thus, we will run the program so that xo=−.5, yo=0, dx=1.5, dy=1.5. We have
then a box centered at [−.5,0]. The sides of the box are each of length 3. We
choose nx=ny=501 so that the grid on which we determine the Mandelbrot
set has slightly over a quarter million data points. We used a “hot” colormap
whose middle tones are orange. The reader might wish to study the topic
“colormap” in MATLAB help. We have chosen to place points in the set if
after 1,000 iterations of Equation 7.2 we have not achieved |zn| > 2. These are
portrayed with a white dot. Points that appeared to be in the set after 20 itera-
tions but are found not to be in the set (they escaped from it) after we were
somewhere between 21 and 1,000 iterations are portrayed by this program
with an orange dot. All the points that were found to be outside the set in the
first 20 iterations appear in black. The reader might wish to experiment with
these settings.
322 A MATLAB® Companion to Complex Variables

The following plot is a gray-scale version of the color plot generated by

the code given above. The plot gives us some idea of the narrow filaments
surrounding the central cardioid set—they are embedded within the set
of points that escaped. It can be shown that all the points on the interval
y = 0, –2 ≤ x ≤ 1/4 lie in the Mandelbrot set, and this is clearly indicated.
Notice, too, that we can begin to see the shape of the filaments that extend
outward from the cardioid to the points z = ±i, which we already knew to
be in the set.
Our result is presented in Figure 7.3.
Compare this figure with Figure 7.2. Notice that we have some points
here that are neither black nor white but are shown in gray. These points
did not make the cut—they are not in the Mandelbrot set—but after only
20  ­iterations they appeared to be. However, after more iterations, we
realized they were not. Their presence helps us to locate the positions of
points in the Mandelbrot set which exist on such a narrow thread that
they cannot be shown on the screen of your computer. The gray points
surround these. If you run the code provided above, you will not obtain
a gray-scale plot but a more dramatic color plot where there is orange
instead of gray.

1.5

0.5

–0.5

–1

–1.5
–2 –1.5 –1 –0.5 0 0.5 1

FIGURE 7.3
The Mandelbrot set enhanced with escape points.
Coda: Fractals and the Mandelbrot Set 323

7.1  Julia Sets

Suppose we just assign any number for the value of c in Equation 7.2. We
begin the iteration progress in this equation not with z0 = 0 (the value used
to determine the Mandelbrot set), but with some arbitrary value, real or com-
plex. Starting the iteration process we will find, as in our previous work, that
we obtain a sequence of values for |z1|, |z2| ... whose terms either become
unbounded or do not become unbounded. If the sequence remains bounded
(i.e., does not diverge to infinity), we take note of the value of z0. Now sup-
pose we go throughout the complex z plane, continuing to use the same
value of c, and take each point in turn as a possible value for z0.
We note which values of z0 do not result in a sequence whose elements
become unbounded. If we plot these numbers as points in the complex plane,
we obtain a certain set pertaining to c. It has been shown that there are two
possibilities for our set:

(a) The set is connected, bounded, and closed

or
(b) The set is composed of points, none of which can be connected to any
other point by a contour composed of elements lying entirely in the set.

The set described in (b) may be thought of as mathematical “dust particles”

comprising the elements of the set.
The set of points in (b) is composed entirely of its own boundary points.
Every neighborhood of every element contains points not belonging to the
set. And a neighborhood of every element can be found such that there are no
other points (of the set) in the neighborhood except the given e­ lement. The
set described in (b) is called a Julia set and is named after a French mathema-
tician Gaston Julia who discovered it in the early decades of the twentieth
century.
The set of points in (a) has a subset—its boundary points. These boundary
points are also called a Julia set. The entire set in (a) is called a “filled-in Julia
set.” The term Julia set (without filled- in) refers to the set of boundary points.
This Julia language can be applied to iterative processes other than the one
used in Equation 7.2. Julia sets are often but not always fractals, as we later
see (e.g., problem 6).
We do not provide the software for generating Julia or filled-in Julia sets.
The MATLAB code is sufficiently similar to that given for generating the
Mandelbrot set that readers should be able to write their own code. Note that
in Appendix C we prove that no value of z0 whose magnitude exceeds the
larger of |c| and 2 can belong to any kind of Julia set for c. We also assert
that if we try any value of z0 (irrespective of its magnitude) and if using the
iterative formula (Equation 7.2), we encounter a value of |zn| that exceeds the
larger of |c| and 2, then z0 is not in a Julia set, filled-in or not.
324 A MATLAB® Companion to Complex Variables

Shown below is the filled-in Julia set for the case c = .25 + .45i. Note that
|c| < 2. The points in the plane are to be interpreted as values of z0. We have
used 1,000 iterations in the formula (Equation 7.2) and have decided that if
the magnitude of the value of zn has not exceeded 2 after this many iterations,
then the starting value z0 is in a Julia set.
The filled-in Julia set is in white. The border consisting of the boundary
points is complicated. This is the Julia set for c = .25 + .45i. The filled-in Julia
set—the interior plus boundary—looks like the map of a country. Think of
France, for example. The number of data points used in creating this image
is (1001)2—slightly over a million.
Here is another example of a Julia set (see Figure 7.4). This one uses c = –1 +
.287i. We obtain the Figure 7.5. It looks like a constellation of stars in the sky.
Note that although there appear to be portions of the set containing points
that are not isolated (i.e., interior points whose every neighborhood contain
elements other than the point in question), all members of this set are indeed
isolated; the computer is not capable of showing the true dust-like nature of
this Julia set. How do we know that all the elements are isolated? We answer
this in a moment.
The image was created using a grid for possible values of z0 equal to (501)2
data points. There is an art to choosing the fineness of the grid. If you make
the grid too fine, the number of pixels assigned to each value of z0 will be so
few that a white spot will not appear on the computer screen or be reproduc-
ible from your printer.

Julia set for c = .25 + .45i

2

1.5

0.5

0.5

–1

1.5

–2
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2

FIGURE 7.4
An example of a Julia set. This one uses c = .25 + .45i.
Coda: Fractals and the Mandelbrot Set 325

0.5

0.4

0.3

0.2

0.1

–0.1

–0.2

–0.3

–0.4

–0.5
–1.5 –1 –0.5 0 0.5 1 1.5

FIGURE 7.5
Julia set for c = –1 + .287i.

The use of “escape time algorithms” which enhanced our pictures of the
Mandelbrot set can also be advantageously applied also in the code used to
display Julia sets.
When does a value of c yield a connected set, and when does it yield
“dust”—a set consisting entirely of boundary points? The answer to this
is sometimes called The Fundamental Theorem of the Mandelbrot Set. The
Julia set is connected (and this applies to the filled-in set as well) if c lies
in the Mandelbrot set. The Julia set is not connected (nor is any subset
of it connected) if c lies outside the Mandelbrot set. The proof was made
independently, circa 1919, by the two French mathematicians Gaston Julia
and Pierre Fatou. The theorem was proved long before the Mandelbrot set
received its name—the naming did not take place until the early 1980s and
originated with the mathematicians John Hubbard and Andrew Douady,
who wished to honor Mandelbrot for his contribution to our understand-
ing the set.
The point c = –1 + .287i does not lie in the Mandelbrot set. You can verify
this with our program called Mandelcheck. Thus, Figure 7.5 shows a Julia
set—every point is boundary point and every point is an isolated member
of the set. The figure preceding that shows a closed connected set. This is
because .25 + .45i lies in the Mandelbrot set, as can be verified with our pro-
gram Mandelcheck.
Here is a hint for writing the software to generate Julia sets and filled-in
Julia sets. We use Equation 7.2, zn+1 = zn2 + c; our Julia plot is for some chosen
326 A MATLAB® Companion to Complex Variables

value of c. We then explore the complex plane; each point of interest is

assigned the value z0, and Equation 7.2 is used to iterate this formula. If c is
known to be in the Mandelbrot set, then you may stop the procedure as soon
as you encounter a value of zn+1 whose magnitude exceeds 2. The z0 in ques-
tion is thus not in a Julia set. If you do 1,000 iterations and this magnitude
has not been achieved, it is highly likely that the z0 chosen does lie in a Julia
set. We make a mark in the complex plane (we chose white in the above plots)
having a distinctive color at the point we called z0. We chose black for points
found not to be in a Julia set.
If you know that c is not in the Mandelbrot set, or if you are not sure,
then you perform the operation described above. As soon as you encounter a
value of zn+1 whose magnitude exceeds the larger of 2 and |c|, you can argue
that the starting value z0 is not included in a Julia set. If you reach 1,000 itera-
tions and this situation has not occurred, then it is highly likely that this z0
does lie in a Julia set. It should be noted that the number 1,000 quoted here
and in the previous paragraph is very conservative. Many authors say that it
is safe to stop after 100 iterations.

Appendices to Coda
Appendix A
We prove here that if |c| > 2, then c is not in the Mandelbrot set.
We make considerable use in this discussion of a version of the triangle
inequality (see W section 1.3 or S section 1.5).
If u = v + w with |v| > |w|, then we have

|u| > |v| − |w| > 0 (7.4)

Let us assume that |c| > 2.

Now from our iterative formula,

zn+ 1 = zn2 + c (7.5)

and with our taking z0 = 0 we have that z1 = c; and so

|z1| = |c| > 2 (7.6)

Consider now z2 = z1 + c = z1 ( z1 + c/z1 ). Thus, |z2| = |z1||(z1 + c/z1)|.

2

c
Now z1 > from Equation 7.6. Thus, from our triangle inequality, |z2| >
z1
|z1|(|z1| − |c|/|z1|).
Coda: Fractals and the Mandelbrot Set 327

With Equation 7.6, this reduces to

|z2|>|c|(|c| − 1) (7.7)

And, since |c| > 2, we have from Equation 7.7,

|z2| > 2  and |z2| > |c| (7.8)

Also, from Equation 7.7 and with |c| > 2, we obtain

c 1
< < 1 (7.9)
z2 ( c − 1)

Continuing, z3 = z2 + c = z2 ( z2 + c/z2 ), so that |z3| = |z2||z2 + c/z2|.

2

Recognizing from Equation 7.9 that |c|/|z2| < 1 and |z2| > 2 we have with
our triangle inequality |z3| = |z2||z2 + c/z2| > |z2|(|z2| − |c|/|z2|), so that

|z3|>|z2|(|z2| − 1) (7.10)

Since |z2| > |c|(|c| − 1) (see Equation 7.7 and |z2| > |c|), we can rewrite

Equation 7.10 as

|z3| > |c|(|c| − 1)2

One should see that a pattern is emerging. We now make the following
proof by the method of induction. Our goal is to prove if |c| > 2 and if for
some integer n > 0 that

|zn|>|c|(|c| − 1)n−1 (7.11)

These conditions will ensure that

|zn+1|>|c|(|c| − 1)n (7.12)

In other words, we will have proved that Equation 7.12 remains valid for
all values of n larger than the one where we already knew it to be valid.
Notice by the way that we have already proved that Equation 7.11 is true
for n = 2,3.
From our iterative formula, taking magnitudes:

zn+ 1 = zn2 + c = zn ( zn + c/zn ) (7.13)

328 A MATLAB® Companion to Complex Variables

Going back to Equation 7.11, which we assume to be true, for some n, and
recalling that |c| > 2, it should be apparent that |zn| > |c| > 2 and also from
Equation 7.11 that

1
c / zn < < 1 < zn
( c − 1)n−1
Using our triangle inequality, we have from Equation 7.13,

|zn+1| > |zn|(|zn|−|c|/|zn|) > |zn|(|c|–1)

Replacing |zn| on the far right with our assumption in Equation 7.11, we have
finally our desired result |zn+1| > |c|(|c|− 1)n. We can repeat this process indefi-
nitely, each time using the latest result, so that |zn+k+1| > |c|(|c|− 1)n+k, where
k ≥ 0 is any integer. Notice that as n → ∞ in Equation 7.12, we find that |zn+1|
grows without bound, in other words, the sequence diverges to infinity, a subject
we touched on in Chapter 3. Thus, if |c| > 2, then c is not in the Mandelbrot set.

Appendix B
In this appendix, we prove the following. If we assume that |c| ≤ 2 (which
means that c might lie in the Mandelbrot set), and if we perform our ­iterative
procedure (Equation 7.2) and discover a value of |zn| > 2, then |zn+1| >
|zn|, |zn+2| > |zn+1|, and so on, so that the magnitude of each term in the
sequence exceeds that of any term preceding it. This, of course, does not
prove that the sequence becomes unbounded. But we will in fact prove that
the sequence |z0|, |z1|, |z2|, ... does grow without bound and thus c is not in
the Mandelbrot set.
From Equation 7.5, zn+ 1 = zn2 + c, which gives us the familiar

c  c 
zn+ 1 = zn2 + c = zn zn + > zn  zn − > zn ( zn − 1) (7.14)
zn  zn 

Since |zn+1| > |zn|(|zn| − 1), we have that |zn+1| > |zn|. And since |zn| > 2,
we can argue from the preceding that |zn+1| > 2. Notice that for this reason it
is now permissible to change the index from n to n + 1 in Equation 7.14 to get

|zn+2| > |zn+1|(|zn+1| − 1) (7.15)

We now use Equation 7.14 and the fact that |zn+1| > |zn| to rewrite
Equation 7.14 as

|zn+2| > |zn|(|zn| − 1)2 (7.16)
Coda: Fractals and the Mandelbrot Set 329

Continuing along in this way, we have the generalization that

|zn+k| > |zn|(|zn| − 1)k (7.17)

provided |zn| > 2 and k ≥ 0. Since |zn| > 2, the limit of |zn+k| is infinity as k → ∞.
Thus, as soon as we find a |zn| > 2, we know that the value c we are using
cannot belong in the Mandelbrot set.
The equation |zn+1| > |zn|(|zn|− 1) can be applied whenever |zn| > 2. Thus,
it can be applied to the sequence z1, z2, …, zn, …, provided you choose a term zn
whose magnitude exceeds 2. Hence, for this term or any term to the right of
it, you can say that the succeeding term is larger in magnitude than the one
before it. As soon as you find a term whose magnitude exceeds 2, you will
know that from then on the terms will get bigger and bigger in magnitude
and grow without bound.

Appendix C
Here we prove that if a point z0 satisfies the condition that |z0| is greater than
the larger of 2 and |c|, then this point is not in a Julia set.
We have that z1 = z02 + c , so that

c  c 
z1 = z02 + c = z0 z0 + > z0  z0 − (7.18)
z0  z0 

c
Recall that < 1 by assumption. Also |z0| > 2 by assumption. Thus, it
z0
must be true that |z1| >|z0| and that |z1| is greater than the larger of 2 and
|c|, since z0 is. Continuing in this way we could argue that

 c 
z2 > z1  z1 − (7.19)
 z1 

and that |z2| is greater than the larger of 2 and |c|. Combining our two
inequalities, we have

 c  c 
z2 > z0 z1  z0 −   z1 −
 z0   z1 

and continuing on, we have

 c  c   c 
zn > z0 z1 ... zn− 1  z0 −   z1 −  ...  zn− 1 − 
 z 0  z1   z n− 1 
330 A MATLAB® Companion to Complex Variables

Suppose we pass to the limit as n → ∞. The product |z0||z1| ... |zn−1| ­consists
of terms each of which exceeds 2 and thus it grows without bound. Each of
the other terms in the product

 c  c   c 
 z0 − z0   z1 − z1  ...  zn − zn 

exceeds one. Thus, our limit goes to infinity—the product is unbounded
and the iterative process leads to an unbounded result. As in the above two
appendixes, we obtain a sequence diverging to infinity.
Notice that had we begun with a value of z0 whose magnitude was not
greater than the larger of 2 and |c| but, through a process of iteration,
encountered a value of zn whose magnitude did exceed these two quanti-
ties, we could readily modify the preceding proof to show that the resulting
sequence did become unbounded, and we could then conclude that z0 is not
in a Julia set.

Exercises
1. Consider the sequence |z0|,|z1|,|z2|, ... For the expressions zn given
below, explain whether this sequence is bounded. If the expression is
a function of z, state the domain or set of points where the sequence
becomes unbounded.
a. zn = ncos(nπ/2)
b. zn = enz
c. zn = nz
2. a. We established that z = i is in the Mandelbrot set. Since the
Mandelbrot set is connected, every neighborhood of this point
must contain at least one other point in the set. Consider z = x + iy
­satisfying −.1 ≤ x ≤ .1, .9 ≤ y ≤ 1.1. Using our program Mandelcheck,
try to find a point in this region other than z = i that lies in the
Mandelbrot set. This could be quite tedious. It is doubtful that you
will succeed by ­proceeding by trial and error.
b. A systematic way to solve the preceding problem is with the follow-
ing code. In order to understand it, read the MATLAB documen-
tation for the command find. Try running the code for the region
described below using nx=ny=1001 subdivisions for the  square
region. Find at least one point in the Mandelbrot set (the program
should yield many). Test that this point lies in the Mandelbrot set by
using the code Mandelcheck given ­earlier in this chapter.
Coda: Fractals and the Mandelbrot Set 331

q=1
format long
while q>0
xo=input('x coord center of box=')
yo=input('y coord center of box=')
dx=input('half width of box, x direction=')
dy=input('half width of box, y direction=')
%below use odd numbers for nx and ny to include center of box
%in your calculations
nx=input('number of x divisions')
ny=nx; %this is number of y divisions
% the two lines below give the real and imaginary parts of c.
tic
cr=linspace(xo-dx,xo+dx,nx);
ci=linspace(yo-dy,yo+dy,ny);
[Cr,Ci]=meshgrid(cr,ci);
c=Cr+i*Ci;%this creates a grid of complex numbers for c
nmax=1000;% we use 1000 iterations of the recursion relation
j=1;
zeros(size(c));%this starts off z, for each value of c,
z=
%at the value zero
while j<=nmax;%iterates the expression below nmax times;
z=z.*z+c;
j=j+1;
end
=abs(z)<=2;%puts a symbol of 1 in the matrix where |z|<=2 ;
ck
%otherwise puts zero
dk
=1.0*ck;%converts symbolic elements to numerical in above
%matrix.
ows,cols,vals] = find(dk);%this finds the nonzero elements
[r
%in dk
for k=1:length(rows)
locations=c(rows(k),cols(k))
figure(1)
plot(locations,'k.'); hold on
%the above is optional and will plot the points
%you just found as dots
%t
he "locations" are points in the complex plane lying in
the % mandelbrot set
end
toc
q=input('negative q to stop')
end
332 A MATLAB® Companion to Complex Variables

3. We can prove that if c is real and c > 1/4, then c is not in the
Mandelbrot set.
a. Why is zn+ 1 − zn = zn2 − zn + c ?
Note that all values of zn obtained through iteration are real if c
is real.
b. Using the result of (a), argue that zn+1 − zn is always greater than
or equal to some positive constant that is independent of n.
Hint: Look at the equation in the x–y plane y = x2 − x + c, c >
1/4 and argue that this is always positive. One way to do this
is to find where the first derivative vanishes and show that the
second derivative is positive there. Thus, the function displays a
minimum. It is easy to show that the function is positive at the
minimum. What is its value at the minimum? It is convenient to
1
take c = + ε, where ε > 0.
4
c. Since zn+1 − zn ≥ ε, how does this show that as m → ∞ zm grows
positive without bound? This completes the proof. Notice that
we already know that if c is real and c < −2, then c is not in the
Mandelbrot set because |c| > 2.
4. Our code Mandelescape shows points that are presumed to be in the
Mandelbrot set when we have made 1,000 iterations with our working
iterative formula (Equation 7.2). It also shows points that are presumed
to be in the set after 20 iterations but are found not to be in the set when
we have made between 21 and 1,000 iterations. It also shows points
found to not be in the set when we have made 1 to 20 iterations.
Improve upon the code as follows. It should show points that are
presumed to be in the Mandelbrot set when we have made 1,000
iterations with our working iterative formula (Equation  7.2). Also
show points with a distinctive color that are presumed to be in the
set after 5 iterations but are found not to be in the set when we have
made between 6 and 15 iterations. Similarly, show points found to
not be in the set when we have made 16 to 1,000 iterations. And show
points found not to be in the set when we have made 1 to 5 iterations.
5. Show that no matter what value is chosen for c, the correspond-
ing Julia set or filled Julia set must contain the two points given by
1 1 1 1
z0 = + ( 1 − 4c ) c ≠ . If c = , this reduces to the fact that the
1/2

2 2 4 4
1
filled Julia set contains z0 = .
2
Hint: Study Equation 7.2 and look for fixed points (i.e., points which
the iterative procedure transforms back into themselves).
6. a. Write code similar to our code called Mandelbrot, above, to
­generate a Julia set. The code will prompt you for the value of c
Coda: Fractals and the Mandelbrot Set 333

to be considered and will iterate (Equation 7.2) throughout a grid

placed in a rectangular box in the complex z plane. We take z0
as the points lying at the intersections of the grid. Check your
program by generating the Julia sets for .25+.45i and −1+.287i and
compare them with the results given in this section. Notice that
1
in the previous problem (number 5) we saw that if c = , then the
4
1
Julia set must contain z = . Verify that this is so by using your
2
program to generate the filled Julia set for this value of c.
b. Not every Julia set obtained from Equation 7.2 is a fractal. If c = 0,
find mathematically the corresponding filled Julia set and the
Julia set as well. Why is the latter not a fractal set? Consider
issues of self-similarity and tangents. Now verify the Julia sets
(filled and ordinary) by running the program found in part (a)
for c = 0.
7. a. Show that no matter what value is chosen for c, the correspond-
ing Julia set or filled Julia set must contain the values of z0 that
are solutions of the equation: z04 + 2 cz02 − z0 + c 2 + c = 0.
Hint: Suppose we have a value of z0 in Equation 7.2 such that
z0  =  z2. Why does this guarantee that z0 belongs to a Julia set?
How does the preceding lead to the quartic equation?
b. There is a closed-form solution available for quartic equations;
it is quite complicated. One is generally better served by the
MATLAB function roots (which you might wish to review) into
which one enters the numerical values of the coefficients in the
preceding equations. Using MATLAB, find the values of z0 in the
above quartic corresponding to c = i/2.
c. Run the code you wrote in problem 6, to generate Julia sets,
choosing c = i/2 to see if the four values found above appear to
be in the set. Put them on your plot. Note that some appear to be
boundary points.
8. There are other iterative schemes besides Equation 7.2 that generate
fractal shapes. Here are some to experiment with. In the absence
of other information, you may assume that a point being studied
does not belong to the set being sought if after 200 iterations we have
|fn| > 10. Obtain fractal plots for the following, finding the regions of
the complex c plane where the sequence does not become unbounded,
provided we begin with z0 = 0.
zn+ 1 = zn3 + c
a.
b.
zn+1 = cos(zn + c)
c.
zn+1 = cos(czn)
zn+ 1 = e iczn. Plot in the complex plane for 0 ≤ Re(c) ≤ 6, 0 ≤ Im(c) ≤ 6.
d.
334 A MATLAB® Companion to Complex Variables

9. Write a program to create a curve that has self-similarity but

is different from the one using the sequence of diminishing
positive triangles that we used earlier in this section. Try, for
­
example, a damped triangular wave having peaks and troughs:
1, −.9, .92, −.93, .94, −.95, ….
This page intentionally left blank
 Mathematics / Real, Complex & Functional Analysis

TEXTBOOKS IN MATHEMATICS TEXTBOOKS IN MATHEMATICS

A MATLAB® Companion to Complex Variables

A MATLAB®
ABET, the accrediting board for engineering programs, makes it clear that
engineering graduates must be skilled in the art of programming in a lan-
guage such as MATLAB®. Supplying students with a bridge between the

COMPANION
functions of complex variable theory and MATLAB, this supplemental text
enables instructors to easily add a MATLAB component to their complex
variables courses.

to COMPLEX
A MATLAB® Companion to Complex Variables provides readers with a
clear understanding of the utility of MATLAB in complex variable calculus.
An ideal adjunct to standard texts on the functions of complex variables,
the book allows professors to quickly find and assign MATLAB program-
ming problems that will strengthen students’ knowledge of the language
and concepts of complex variable theory.
VARIABLES
The book shows students how MATLAB can be a powerful learning aid in
such staples of complex variable theory as conformal mapping, infinite se-
ries, contour integration, and Laplace and Fourier transforms. In addition to
MATLAB programming problems, the text includes many examples in each
chapter along with MATLAB code.

Fractals, the most recent interesting topic involving complex variables, de-
mands to be treated with a language such as MATLAB. This book concludes
with a Coda, which is devoted entirely to this visually intriguing subject.

MATLAB is not without constraints, limitations, irritations, and quirks, and

there are subtleties involved in performing the calculus of complex variable
theory with this language. Without knowledge of these subtleties, engineers
or scientists attempting to use MATLAB for solutions of practical problems
in complex variable theory suffer the risk of making major mistakes. This
book serves as an early warning system about these pitfalls.

A highly detailed solutions manual is also available upon qualified course

adoption.
Wunsch

A. David Wunsch
K27481

w w w. c rc p r e s s . c o m

K27481_cover.indd 1 3/16/16 1:54 PM