Computational Conformal Mapping - Prem K. Kythe

FILIO • P R I M O
H V N C • LIBELLVM
D •D • D
Computational Conformal Mapping
Prem K. Kythe
Springer Science+Business Media, L L C

Prem K. Kythe
University of New Orleans
Department of Mathematics
New Orleans, L A 70148
Library of Congress Cataloging-in-Publication Data
Kythe, Prem K .
Computational Conformal Mapping/Prem K . Kythe.
p. cm.
I S B N 978-1-4612-7376-9 I S B N 978-1-4612-2002-2 (eBook)

D O I 10.1007/978-1-4612-2002-2
1. Conformal Mapping ~ Data Processing. I. Title
QA360.K97 1998 98-4748

516.3'6-dc21 CIP
Printed on acid-free paper

© 1998 Springer Science+Business Media New York
Originally published by Birkhäuser Boston in 1998
Softcover reprint of the hardcover 1st edition 1998
Copyright is not claimed for works of U.S. Government employees.
A l l rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical, photocopy-
ing, recording, or otherwise, without prior permission of the copyright owner.
Authorization to photocopy items for internal or personal use of specific clients is granted
by Birkhäuser Boston provided that the appropriate fee is paid directly to Copyright
Clearance Center (CCC), 222 Rosewood Drive, Danvers, M A 01923, U S A (Telephone:
(978) 750-8400), stating the ISBN, the title of the book, and the first and last page
numbers of each article copied. The copyright owner's consent does not include copying
for general distribution, promotion, new works, or resale. In these cases, specific written
permission must first be obtained from the publisher.
I S B N 978-1-4612-7376-9
9 8 7 6 5 4 3 2 1
Contents
Preface, ix
Chapter o. Introduction, 1
0.1. Historical Background, 1
0.2. Modern Developments, 8
Chapter 1. Basic Concepts, 15

1.1. Notation and Definitions, 15
1.2. Some Basic Theorems, 23
1.3. Harmonic Functions, 27
1.4. Conformal Mapping, 30
1.5. Problems, 38
Chapter 2. Conformal Mappings, 41

2.1. Polynomials, 41
2.2. Bilinear Transformations, 45
2.3. Schwarz-Christoffel Transformations, 51
2.4. Catalog of Conformal Maps, 64
2.5. Problems, 65
Chapter 3. Schwarz-Christoffel Integrals, 68

3.1. Parameter Problem, 68
3.2. Newton's Method, 72
3.3. Numerical Computations, 76
3.4. Kirchhoff Flow Problem, 83
3.5. Problems, 89
v
vi CONTENTS
Chapter 4. Polynomial Approximations, 92

4.1. Minimum Area Problem, 92
4.1.1. Bergman Kernel, 94
4.2. Numerical Methods for Problem I, 95
4.2.1. Ritz Method, 95
4.2.2. Bergman Kernel Method, 101
4.3. Minimum Boundary Problem, 103
4.4. Ritz Method for Problem II, 105
4.5. Orthogonal Polynomials, 109
4.5.1. Polynomials Orthogonal to the Boundary, 109
4.5.2. Polynomials Orthogonal to a Region, 116
4.6. Problems, 118
Chapter 5. Nearly Circular Regions, 120

5.1. Small Parameter Expansions, 120
5.2. Method of Infinite Systems, 125
5.3. Three Special Cases, 132
5.4. Exterior Regions, 140
5.5. Problems, 145
Chapter 6. Green's Functions, 147

6.1. Mean-Value Theorem, 147
6.2. Dirichlet Problem, 148
6.3. Numerical Computation, 152
6.4. Schwarz Formula, 158
6.5. Neumann Problem, 159
6.6. Series Representations, 162
6.7. Problems, 165
Chapter 7. Integral Equation Methods, 168

7.1. Neumann Kernel, 168
7.2. Interior Regions, 171
7.2.1. Lichtenstein's Integral Equation, 171
7.2.2. Gershgorin's Integral Equation, 172
7.2.3. Carrier's Integral Equation, 173
7.3.1. Banin's Integral Equation, 176
7.3.2. Warschawski-Stiefel's Integral Equation, 177
7.3.3. Interior and Exterior Maps, 177
7.4. Iterative Method, 178
7.5. Degenerate Kernel, 192
7.6. Szeg6 Kernel, 198
7.7. Problems, 204
CONTENTS vii
Chapter 8. Theodorsen's Integral Equation, 207

8.1. Classical Iterative Method, 207
8.2. Convergence, 209
8.3. Proofs, 211
8.4. Integral Representation, 222
8.5. Starlike Regions, 224
8.7. Trigonometric Interpolation, 228
8.8. Wegmann's Method, 229
8.9. Newton's Method, 234
8.10. Problems, 236
Chapter 9. Symm's Integral Equation, 237

9.1. Symm's Integral Equation, 237
9.1.1. Interior Regions, 237
9.1.2. Exterior Regions, 239
9.2. Orthonormal Polynomial Method, 240
9.3. Modified ONP Method, 247
9.4. Lagrange Interpolation, 251
9.5. Spline Approximations, 258
9.6. Problems, 265
Chapter 10. Airfoils, 269

10.1. Joukowski Function, 269
10.2. Generalized Joukowski Mappings, 274
10.2.1. Glauert's Modification, 275
10.2.2. Symmetric Joukowski Airfoil, 276
10.3. Nearly Circular Approximations, 277
10.4. James's Method, 279
10.4.1. Single-Element Airfoils, 280
10.4.2. von Karman-Trefftz transformations, 283
10.4.3. Two-Element Airfoils, 288
10.4.4. Multi-Element Airfoils, 291
10.5. Problems, 292
Chapter 11. Doubly Connected Regions, 295

11.1. Conformal Modulus, 295
11.1.1. Conformal Invariants, 296
11.1.2. Area Theorem, 298
11.2. Source Density, 305
11.3. Dipole Distribution, 308
11.4. Problems, 317
viii CONTENTS
Chapter 12. Singularities, 320

12.1. Arbenz's Integral Equation, 320
12.2. Boundary Corner, 325
12.3. Singularity Behavior, 328
12.4. Pole-Type Singularities, 330
12.6. Doubly Connected Regions, 345
12.7. Problems, 351
Chapter 13. Multiply Connected Regions, 358

13.1. Existence and Uniqueness, 359
13.2. Dirichlet Problem, 364
13.3. Mikhlin's Integral Equation, 367
13.4. Mayo's Method, 370
13.5. Fast Poisson Solver, 373
13.6. Problems, 377
Chapter 14. Grid Generation, 379

14.1. Computational Region, 379
14.2. Inlet Configurations, 384
14.3. Cascade Configurations, 390
14.4. Problems, 399
Appendix A. Cauchy P. V. Integrals, 401

A.1. Numerical Evaluation, 401
Appendix B. Green's Identities, 407
Appendix C. Riemann-Hilbert Problem, 410

C.l. Homogeneous Hilbert Problem, 410
C.2. Nonhomogeneous Hilbert Problem, 415
C.3. Riemann-Hilbert Problem, 417
Appendix D. Successive Approximations, 425

D.1. Tables, 425
Appendix E. Catalog of Conformal Mappings, 428
Bibliography, 431
Notation, 449
Index, 453
Preface
This book evolved out of a graduate course given at the University of New
Orleans in 1997. The class consisted of students from applied mathematics
and engineering. They had the background of at least a first course in complex
analysis with emphasis on conformal mapping and Schwarz-Christoffel trans-
formation, a first course in numerical analysis, and good to excellent working
knowledge of Mathematica* with additional knowledge of some programming
languages. Since the class had no background in Integral Equations, the chap-
ters involving integral equation formulations were not covered in detail, except
for Symm's integral equation which appealed to a subset of students who had
some training in boundary element methods. Mathematica was mostly used
for computations. In fact, it simplified numerical integration and other oper-
ations very significantly, which would have otherwise involved programming
in Fortran, C, or other language of choice, if classical numerical methods were
attempted.
Overview
Exact solutions of boundary value problems for simple regions, such as cir-
cles, squares or annuli, can be determined with relative ease even where the
boundary conditions are rather complicated. Green's functions for such simple
regions are known. However, for regions with complex structure the solution
of a boundary value problem often becomes more difficult, even for a simple
problem such as the Dirichlet problem. One approach to solving these difficult
problems is to conformally transform a given multiply connected region onto
*Mathematica is a registered trade mark of Wolfram Research, Inc.
ix
x PREFACE
simpler canonical regions. This will, however, result in change not only in
the region and the associated boundary conditions but also in the governing
differential equation. As compared to the simply connected regions, confor-
mal mapping of multiply connected regions suffers from severe limitations,
one of which is the fact that equal connectivity of regions is not a sufficient
condition to effect a reciprocally connected map of one region onto another.
There are though a few methods that carry out such mappings where most of
the computational details are done numerically.
The main purpose of this book is to provide a self-contained and systematic

introduction to the theory and computation of conformal mappings of simply
or multiply connected regions onto the unit disk or canonical regions. It
provides a comprehensive and systematic coverage of the basic theory and
related numerical analysis with applications to different areas in mathematical
physics and engineering. The material is presented at a level that can readily
be followed by graduate students and researchers. The prerequisites, besides
the theory of conformal mapping, include knowledge of basic methods of
numerical analysis, the theory of Fredholm and Stieltjes integral equations,
and a programming language. Depending on the background of students, this
book may cover two semesters of graduate course work, the first involving
classical developments (chapters 1 through 6, 10, and 14), and the second
involving the integral equation formulations and the study of singularities
(chapters 7 through 9 and 11 through 13).
Salient Features
Besides an exposition on the history of the subject, there are 14 chapters in

the book the first two provide a review of the theory of conformal mapping
and Schwartz-Christoffel transformations to map a polygon onto the upper
half-plane. The computation of improper Schwartz-Christoffel integrals and
the related parameter problem is one of the important classical topics which
is studied in detail in Chapter 3. An old problem, known as the Kirchhoff
flow problem that started in 1868, is presented with its very recent complete
solution. Chapter 4 deals with the methods of polynomial approximations,
first, with the minimum area problem with detailed accounts and algorithms
for the Ritz and Bergman kernel methods, and secondly, with the minimum
boundary problem with the related Ritz method and the Szego kernel and its
applications. Polynomials orthogonal to the boundary and those orthogonal
to the region are investigated. Nearly circular regions are studied in Chapter 5
through the method of infinite systems and successive approximations, where
PREFACE xi
special cases are treated separately, and computational algorithms with exam-
ples are presented. Techniques of numerical evaluation of Green's functions
for various kinds of regions are studied in Chapter 6, with special topics like
the Dirichlet and Neumann problems, Schwarz formula, and associated series
representation.
Various integral equation formulations of the conformal mapping problem

have been discussed in Chapters 7, 8, 9, 11, and 13. These equations are
mostly Fredholm equations of the first or second kind and provide boundary
correspondence functions and their computation by iterative methods. Chap-
ter 7 starts with the Neumann kernel and presents Lichtenstein's, Gershgorin's,
Carrier's, Banin's, and Warschawski-Siefel's integral equations. An algorithm
for the iterative method is provided with detailed explanations. The chapter
discusses the case of a degenerate kernel and ends with the Szego kernel and
the related method for solving the conformal mapping problem. Chapter 8
deals with the classical Theodorsen's integral equation, discusses convergence
of the iterative method, provides proofs for the convergence theorem due to
Warschawski, and investigates the cases of starlike and exterior regions. A
trigonometric interpolation method is outlined and the modern iterative and
Newton's methods are presented. Chapter 9 deals with Symm's integral equa-
tions for the interior and exterior regions, based on a single-layer potential.
The orthonormal polynomial method, Lagrange interpolation, and spline ap-
proximation method are discussed. Doubly connected regions are studied in
Chapter 11, and the related integral equation based on the dipole distribution
is presented with an algorithm for numerical computation. Chapter 13 deals
with multiply connected regions. Based on the dipole distribution, Mikhlin's
integral equation that determines the density function and related boundary
correspondence is solved by a fast Poisson solver. This method is very efficient
and solves problems for both simply and multiply connected regions.
A detailed account of airfoils is presented in Chapter 10. Various methods,

like James's method for single-element airfoils, von Karman-Trefftz transfor-
mation and Garrick's method of conjugated functions for two-element airfoils
are explained with related algorithms and examples. Chapter 12 presents an
important aspect of the conformal mapping problem, related to the location and
behavior of corner singularities on the boundary and pole-type singularities
of the mapping functions near the boundary in simply and doubly connected
regions. Their effect on the polynomial methods is analyzed and, by consid-
ering various cases, certain methods are presented to augment the bases in the
polynomial methods studied in Chapter 4 and 9. The last chapter provides
useful insight into the application of conformal mapping in adaptive grid gen-
eration. Effective programming aspects are discussed and cases examined to
exhibit the importance of this application.
xii PREFACE
Although it is not possible to include every written word on the subject, the
topics included in this book are carefully selected and meticulously presented,
thereby making it a book very useful to graduate students and researchers in
mathematics, physics, and engineering. The scope of the book can be judged
from the table of contents. The bibliography at the end of the book is extensive
and contains references, some not cited in the text, to computer programs and
other articles useful in theoretical and computational development. The book
ends with a notation and a subject index.
Intended Readers
The book is intended to contribute to an effective study program at the

graduate level and to serve as a reference book for scientists, engineers, and
mathematicians in industry. A partial list of readers of this book includes
persons interested in the study of acoustics, plane elasticity, electromagnetic
theory, fluid flows, inlet configurations, transonic flow problems, cascade of
blades in airfoil and wing designs, heat transfer, ion optics, solidification, solid
propellant rocket motors, plates, and vibrations. In most cases, the results are
developed throughout the book from the basics at a level consistent with the
mathematical background of the intended readers. In a few instances the
results are stated and the original references are cited. Each chapter ends with
exercises in the form of problems, some of which are more challenging and a
few are at the research level. Possible references are provided where solutions
to these problems may be found. Each chapter ends with a list of references
used in presenting the related subject matter. The intended readers fall into
one of several categories. First, they are students ready for a graduate course
in this subject. For them the book can be used as a textbook or a reference
book. The second category is that of graduate students engaged in research.
For them the book should become a constant companion, because it is filled
with a vast amount of information on methodology and an almost complete
bibliography on the subject. The third category consists of scientists and
researchers in various areas of applied mathematics, engineering and physics.
For them the book is a vital source of information in classical as well as
modern trends in research in the subject itself as well as in numerical methods
for Fredholm integral equations of the first and the second kind. Research
scientists, engineers, and mathematicians will appreciate this comprehensive
and up-to~ate account of all methods available on the subject. Some of the
techniques are very efficient, and, although not explicit in the title of this work,
PREFACE xiii
numerical treatment of Fredholm integral equations is dealt with in depth in

this book.
Computational Aspects
It has been my experience that Mathematica and a programming language, like

Fortran or C++, are sufficient to carry out any and all computational aspects
of the methods presented in the book. Many programs that may be needed
to complete computations are available in the public domain. Most of them
deal with numerical solution of Fredholm integral equations. Only a few need
be developed. The algorithms provided in the book are detailed enough to
generate computer programs with ease. Besides the algorithms, there are 74
cases studies presented in the book, supplemented by 96 suggested problems
with hints for finding solutions.
Acknowledgements
The help provided by some of my colleagues and students is gratefully ac-

knowledged. Thanks are also due to the editors at Birkhauser - Boston who
offered valuable suggestions for editorial improvements. I am very grate-
ful to my wife, Mrs. B. D. 'Kiran' Kulshrestha, for continuous support and
inspiration over the years, including the period for preparing this book.
New Orleans, Louisiana

March, 1998
"Die Ausfiihrung dieser Theorie, welche, wie bemerkt, einfache durch Grossen-
operationen bedingte Abhangigkeitsgesetze ins Licht zur setzen bestimmt ist, un-
terlassen wir indess jetzt, da wir die Betrachtung des Ausdrucken einer Function
gegenwartig ausschliessen.
Aus demselben Grunde befassen wir uns hier auch nicht damit, die Brauch-
barkeit unserer Satze als Grundlagen einer allgemeinen Theorie dieser Abhangigkeits-
gesetze darzuthun, wozu der Beweis erfordert wird, dass der hier zu Grunde gelegte
Begriff einer Function einer veranderlichen complexen Grosse mit dem einer durch
Grssenoperationen ausdriickbaren Abhangigkeit vollig zusammenfallt."
Bernhard Riemann
Inaugural Dissertation, Gottingen, 1851
xiv
Chapter 0
Introduction
Current research in computational conformal mapping has taken two major

directions. One direction involves the conformal mapping from a standard
region, like the unit disk or the upper half-plane, onto the problem region,
whereas in the other it is from the problem region onto a standard region.- In
the former case one solves a nonlinear integral equation involving the conju-
gate operator (e.g., Theodorsen's integral equation), by fast Fourier transform
(FFT), polynomial approximation, iteration, or Newton's method. In the lat-
ter case the integral equation, derived from the Dirichlet problem, is linear
or singular linear if it is derived from potential theory (e.g., Symm's inte-
gral equation). Depending on the nature of the problem region, these methods
sometimes use the Schwarz-Christoffel transformations. The historical devel-
opment of different methods for computational conformal mapping of simply
and multiply connected regions is sketched below.
0.1. Historical Background
The oldest transformation, known as the stereographic projection of the

sphere, was used by Claudius Ptolemy (ca. 150 A.D.) to represent the ce-
lestial sphere. In a totally different mapping of a sphere onto a plane, known
as Mercator's projection, the spherical earth is cut along a meridian circle
and conformally mapped onto a plane strip. Gerhardus Mercator which was
the Latinized name of the Flemish geographer Gerhard Kremer published the
first world map in 1569 using this technique, and ever since all sea maps are
P. K. Kythe, Computational Conformal Mapping

© Springer Science+Business Media New York 1998
2 CHAPTER O. INTRODUCTION
constructed by the method. These two projections, however, do not produce

similar maps of the same region of the earth, which shows that conformal
mapping does not imply similarity offigures. Johann Lambert (1728-1777) was
the first mathematician who made contributions to the mathematical projection
of maps. These and other similar considerations enabled Lagrange (1779) to
obtain all conformal representations of a portion of the earth's surface onto a
plane where all circles of longitudes and latitudes are represented by circular
arcs. Gauss (1822) was the first to state and completely solve the general
problem of determining all conformal mappings that transform a very small
neighborhood of a point on an arbitrary analytic surface onto a plane area.
However, Gauss' work did open the harder problem of finding the way whereby
a given finite portion of a surface can be mapped onto a portion of the plane. A
breakthrough came in 1851 when Riemann gave the fundamental result, known
as the Riemann mapping theorem, which has since been a turning point for all
subsequent developments in the theory of conformal mapping. In the proof of
this theorem he assumed that a variational problem, now known as the Dirichlet
problem, possesses a solution. It was fifty years later that Hilbert (190 I) proved
the existence of the solution of the Dirichlet problem. In the mean-time the
validity of Riemann's result was established rigorously by Schwarz (1890) by
using a number of theorems from the theory of logarithmic potential.
After the basic theoretical aspects of the theory of functions of a com-

plex variable and conformal mapping were established by Cauchy, Riemann,
Schwarz, Christoffel, Bieberbach, Caratheodory, Goursat, Koebe, and others
in the nineteenth and early decades of the twentieth century, the first numerical
research into developing a method for mapping a region bounded by finitely
many Jordan curves f i onto an n-sheeted Riemann surface where the curves
f i correspond to rectilinear slits was done by Burnside (1891). A minimizing
principle was established by Bieberbach (1914), namely, that among all suit-
ably normed conformal maps of a given simply connected region the one with
JL1I1'(z)11
the least area is the conformal map onto a circle. In particular, this principle
evolved as a result of minimizing the integral
2
dx dy, where f is
regular in D and normalized by f(O) = 0, f'(O) = 1, and the area theorem.
Bieberbach used the Ritz method to find an approximate solution in the form of
a polynomial for the above integral and used it to construct the conformal map
of a simply connected region onto a circle. An exposition of the Ritz method
can be found in Kantorovich and Krylov (1936). The estimates obtained by
this method were later improved by Hohndorf (1926) and Muller (1938) in the
problem of a conformal map of a nearly circular region onto a circle. Other im-
provements on the Bieberbach method were produced by Landau (1926), Julia
0.1. HISTORICAL BACKGROUND 3
(1926), and Kantorovich and Krylov (1936) who also developed a graphical
method of conformal mapping due to Melent'ev (1937).
The first integral equation method was developed by Lichtenstein (1917)

who solved the problem of conformally mapping a simply connected region
bounded by a Jordan contour onto a circle by reducing it to the solution of an
integral equation. Other attempts in this direction were made by Krylov and
Bogolyubov (1929) who reduced the Dirichlet problem to an integral equation
which is approximately solved by the Fredholm method with error estimates.
Nystrom (1930) gave a method for an approximate solution of integral equa-
tions, which is useful in conformal mapping. For conformal mappings of
nearly circular regions the first numerical work was done by Fock (1929) who
determined numerically the mapping function for a circular quadrilateral with
angles 7r /2, 7r /2, 7r /2, O. An important result, now known as Theodorsen's in-
tegral equation, was developed by Theodorsen (1931) and improved upon by
Theodorsen and Garrick (1933) for conformally mapping nearly circularregions
onto a circle. Another integral equation, known as Gershgorin's equation, was
developed by Gershgorin (1933) as a result of conformally mapping a simply
connected region bounded by a Jordan contour onto a circle, which is solved
by the Nystrom method. An exposition on Gershgorin's integral equation and
its application to the mapping of a simply connected region onto a circle, of
a doubly connected region onto an annulus, and of a multiply connected re-
gion, in general, onto slit planes is available in Kantorovich and Krylov (1936).
Various applications of conformal mapping are available in the monograph by
Kantorovich and Krylov (1936), where both harmonic and biharmonic func-
tions are studied. The Dirichlet and Neumann problems in ]R2 are solved by
conformal mapping for different boundary conditions, and the general Hilbert
problem is also investigated. Later Banin (1943) developed a method of ap-
proximately replacing Gershgorin's integral equation by a system of linear
differential equations. Krylov (1938) also reduced the problem of conformally
mapping an n-eonnected region bounded by n Jordan contours onto various
canonical regions to the problem of solving a system of simultaneous integral
equations.
Successive approximations in the integral equation method for simply and

multiply connected regions were used by Kantorovich (1933, 1937). Follow-
ing Theodorsen's integral equation method, Warschawski (1945) reduced the
problem of conformal mapping of a simply connected region bounded by a
Jordan contour onto the unit circle to that of solving a nonlinear integral equa-
tion which is then solved by the method of successive approximations where
the precise estimates for the convergence are also provided. Goluzin (1934)
investigated the region exterior to the circles G l , ... ,Gn and determined a
harmonic function u(x, y) for this region such that it takes preassigned values
on Gi , i = 1, ... ,n, and u( 00) < 00. The problem is then reduced to that of
solving a finite system of functional equations which are solved by successive
approximations. He applied this method to solve the Neumann problem for
the Laplace equation for such regions and determined the Green's function for
such regions and the conformal maps that carry such regions onto slit planes.
The Schwarz method is used to develop an integral equation which is solved
by successive approximations in Nevanlinna (1939) and Epstein (1948).
Julia (1927) determined a sequence of polynomials which converges to

a suitably normed mapping of a simply connected region. An application
of orthogonal polynomials to conformal mapping of simply connected re-
gions was first developed by Szego (1921). He defines a set of polynomials
Po(z), Pl(z), ... ,Pn(z), ... (subsequently known as the Szego polynomials)
for a closed Jordan contour r in the z-plane, satisfying the following two prop-
1
erties: (i) Pn(z) is a polynomial of degree n, i.e., Pn(z) = anz n +an_lZ n- l +
... + alZ + ao, such that an > 0, and (ii) ~ Pm(z)Pn(z) ds = om,n, where
l is the length of r, s its arc length, and om,n the Kronecker delta. Then the se-
=L
00
ries K(a, z) Pn(a) Pn(z) converges uniformly and absolutely in every

n=O
closed subregion of the interior of r. A formula is then obtained which defines
the mapping function of the interior of r onto the unit disk in terms of K (a, z).
Szego also gave a mapping function for the exterior of r by the polynomials
Pn(z). Further developments in the Szego polynomial method were produced
by Smirnov (1928) and later by Kantorovich and Krylov (1936) who investi-
gated the Szego and Bochner-Bergman type polynomials and applied them to
the minimizing problem. Some results about approximate mapping functions
for a simply connected region by means of certain polynomials were given by
Schaginyan (1944).
The problem of minimizing a functional was first solved by Hadamard

(1908) which later became known as the Hadamard variational method. Lowner
(1923) also developed an important variational method, whereas Julia (1926)
provided another characterization of the mapping function by a minimum prin-
ciple. The problem of reducing the mapping problem to that of minimizing of
a functional was solved by Douglas (1931) who used the Riemann mapping
theorem and the Osgood-Caratheodory theorem. Kufarev (1935-1937) inves-
tigated a minimal problem for a single-valued analytic function in an annulus
and discussed the mapping of the minimizing function. Later in 1947 he used
the Lowner method to study a polygonal problem region which consists of the
whole plane cut by a broken polygonal line with finitely many sides, one of
which extends to infinity. The problem of mapping a doubly connected region
onto an annulus was reduced to that of minimizing an area integral by Khajalia
(1940), who also showed that if the region is accessible from without, then
there exists a sequence of minimal rational functions that converges uniformly
to the desired mapping function. The problem of mapping a simply connected
region bounded by a Jordan contour is reduced by Shiffman (1939) to that of
minimizing a functional, almost similar to that of Douglas. This problem deals
with the Plateau problem, and the electrostatic characterization of the resulting
functional provides an effective method for determining the conformal maps.
The Hadamard formula and the variation of domain functions were used by
Schiffer (1946) to derive a new variational method.
In 1931 Grotsch solved the problem of conformally mapping a multiply

connected region onto some canonical regions based on the assumption that the
solution of a similar problem for a simply connected region is known. Itera-
tive methods were established by Goluzin (1939) who used to map a multiply
connected region conformally onto some canonical regions, thereby reducing
the problem to a sequence of conformal maps of simply connected regions.
Heinhold (1947) investigated the problem of conformally mapping the simply
connected region lying in the exterior of the unit circle onto the exterior of the
unit circle.
Green's function method for an arbitrary region was established by Leja

(1934, 1936) where the approximating functions were found closely related to
Lagrange polynomials. A set of polynomials was also obtained which were
used as mapping function for a region D, containing the point at infinity, onto
the exterior of the unit circle. It was determined that the map is univalent if D
is simply connected.
The Schwarz-Christoffel formula was used by Bergman in 1923-24 to

map a half-plane onto a particular polygon where a method for determining
the parameters in the formula from the lengths of the sides of the polygon is
given. In 1925 he investigated the conformal map of a special polygon onto
a rectangle and computed level curves and their orthogonal trajectories which
were presented in tabular and graphical forms. Bergman also gave the first
punch-card machine method in 1947 to solve the torsion problem where the
orthogonal polynomials were applied to solve the Laplace equation numerically.
The notion of the kernel functions was developed by Bergman in 1922 where
6 CHAPTER o. INTRODUCTION
the existence of a complete orthononnal system with respect to a region is

established and the kernel of the system is related to the conformal map of the
region. Based on Bergman's kernel function and complete orthonormal sets
of functions, Zarankiewicz (1934) found a method for effectively constructing
the conformal map of a doubly connected region onto an annulus; in 1934
he published details of this method. Schiffer (1946) found an expression for
Bergman's kernel function K(z, a) of a region in terms of Green's function
for the region and gave formulas for the variation of K(z, a). In 1948 he
extended the concept of kernel functions and orthonormal sets of functions to
a wider class of functions. Further study of the relationship between the kernel
function and conformal mappings of regions was done by Bergman and Schiffer
(1948-1951).
Hodgkinson and Poole (1924) used elliptic functions to map doubly con-
nected regions of certain types onto the whole plane with two slits on the real
axis. Using hyperelliptic integrals, a generalization of the solution by Hodgkin-
son and Poole was given by Vladimirsky (1941) for the problem of conformal
mapping of a doubly connected region bounded by rectangular segments or
circular arcs, who also extended the solution to n-tuply connected regions
(n > 2). Hodgkinson (1930) also established the relationship between the
theory of Lame differential equations and the Schwarz theory of conformal
mapping. An up-to-date survey of conformal mapping of multiply connected
regions onto canonical domains was published by Keldys (1939). Important
formulas for various confonnal maps of n-tuply connected regions in terms of
the kernel function of the region were established by Nehari in 1949. These
formulas served as tools for numerical computation of conformal maps since
the kernel functions are constructed more easily than the mapping function.
Gerabedian and Schiffer (1949) obtained many significant relations between
various domain functions of an n-tuply connected region and solved some
minimal problems.
The first systematic construction of conformal maps by the method of net-

works was done by Liebmann (1918). The Dirichlet problem in IR 2 and IR 3 was
first solved by this method by Phillips and Wiener (1923). The boundary prob-
lems in IR 2 with the Laplace equation were solved numerically by this method
by Luysternik (1947).
Relaxation methods were first applied to the problems of conformal map-

ping by Gandy and Southwell (1940) and Southwell (1946). Several examples
of technical interest were given, in which regions of arbitrary shape are mapped
onto the interior and exterior of circles and onto rectangles.
The small parameters method developed by Kantorovich (1933) was later

used by Rosenblatt and Turski (1936) for conformal mapping of a special type
of region. Rosenblatt (1943) constructed the conformal maps of regions onto
the unit disk by the Kantorovich method of small parameters and applied it to
the dynamic problems of airfoils.
Special conformal mappings were studied by Rothe (1908) who constructed

the mapping function for a circular quadrilateral with angles 1f /2, 1f /2, 1f /2, p1f
(p = l/VW). Hodgkinson's work (1924) on a similar problem has been
mentioned above. Wirtinger (1927) derived an explicit formula for computing
the conformal map of a triangle with circular arcs and arbitrary angles. The
mapping of an ellipse onto a circle and of a region bounded by two symmetrically
placed ellipses onto an annulus were investigated by Zmorovich (1935) who
also gave an approximate solution in the latter problem. Catalogs of various
types of explicit mappings are available in Koppenfels' papers (1937, 1939).
Other authors on special conformal mappings during this period were Muratov
(1937), Krylov (1937), Melent'ev (1937), and Goluzin (1937). Later Jeffreys
and Jeffreys (1946) and Wittich (1947) produced special conformal maps. A
very useful dictionary of conformal mappings, which is still a source work in
this area, was produced by Kober (1945-1948).
The theory of homogeneous boundary problems for analytic functions was

presented by Hilbert (1924) for a boundary that is a single Jordan contour.
Then he transformed this problem into a Fredholm integral equation, for which
Green's functions of the Neumann problem must be determined. Hilbert did
not give a complete solution of this integral equation. A complete solution
by means of Cauchy integrals for a particular case had already been given by
Plemelj (1908). The solution for the general case, based on the Plemelj method,
was later provided by Khvedelidze (1941). Picard (1927) had studied the
homogeneous Hilbert problem earlier. The nonhomogeneous Hilbert problem
which is solved to determine a sectionally analytic function (z) offinite degree
at infinity such that +(() = G( ()- (O+g( 0 on the boundary r, where G( 0
and g( 0 are defined on the boundary r and satisfy the HOlder condition there,
was first considered by Carleman (1922). Later Privalov (1934) investigated it
for the case when the boundary r is a rectilinear contour, G (() and g( 0 are
Lebesgue-integrable, G(() is bounded, and the limiting value of the solution
is taken along a nontangential path. He used Picard's method but did not find a
complete solution. A complete solution was first given by Gakhov (1937) for
the case when r is a single contour. Later Khvedilidze (1941) gave a complete
solution of the nonhomogeneous Hilbert problem. Carleman's and Gakhov's
methods are essentially alike. In 1851 Riemann considered a very general
boundary problem which is now known as the Riemann boundary problem.

This problem was also studied and solved by Hilbert in 1904, and therefore it is
also called the Riemann-Hilbert problem. It deals with determining an analytic
function in a given region with boundary values involving a relation between its
real and imaginary parts. Hilbert reduced this problem to a singular equation
and then applied it to the solution of two Dirichlet problems. Later Noether
(1921) used this solution to study the subject of singular integral equations.
One of the widely investigated practical aspects of conformal mapping

was the development of the airfoil theory which started with the pioneering
work by Joukowski (1890). He studied the flows around a variety of so-called
Joukowski airfoils. The developments in this area are very pertinent for com-
putational aerodynamic flows. Extensive research to develop computational
methods for solving the direct and inverse problems of airfoil theory has been
done by Theodorsen (1931), Glauert (1948), Andersen, Christiansen, M011er
and Tornehave (1962), Timman (1951), James (1971), Ives (1976), and Halsey
(1979,1982). In single and multi-elementairfoils, Theodorsen's integral equa-
tion has been solved with the von Karman-Trefftz transformation and the FFT
by Garrick's method of conjugate functions (Garrick, 1949) by Ives (1976).
James's method (James, 1971) is another approach in the conformal mapping
of single and multiple-element airfoils (Halsey 1979, 1982).
0.2. Modern Developments
During the last four decades several methods have evolved for numerical and
computational evaluation of mapping functions. The results of some of these
methods provide us with an explicit form of a function which approximately
evaluates the mapping function for a certain source region. This is possible
because of an important result which states that an analytic function defined
on a simply connected and bounded region D c <C can always be uniformly
approximated on every compact subset of D with any preassigned accuracy by
means of a polynomial. In most applications the mapping function is continuous
in D, and it can be uniformly approximated in D by a polynomial.
One major approach in developing methods for numerical conformal map-

ping is based on the following interpretation of the Riemann mapping the-
orem: there exists a conformal mapping f : D t--> U with f(zo) = 0
0.2. MODERN DEVELOPMENTS 9
and f'(zo) nonzero real, where Zo E D, and this function has a power se-
+L
00
ries expansion f(z) = Cl (z - zo) Cn (z - zot, with Cl nonzero real

n=2
and Zo ED, which converges uniformly in every closed disk with center
Zo and contained in D. However, a polynomial which is a good approxima-
tion of f in D is not the same as a truncated power series. If a polynomial
N
p(z) = c~ (z - zo) + L c~ (z - zo)n approximates f with accuracy € > 0,
n=2
then it is necessary that every term of p( z) must approximate the corresponding
term of the power series with accuracy € > 0 on the set D n B (zo, R), where
R = Iz - Zo I is the radius of convergence of the power series. All this means
is that a polynomial p which is a good approximation of the power series starts
in the same way as the power series, but the relative error in the coefficients
increases with increasing n.
Another direction for developing numerical methods for conformal mapping

is based on computing a table of values for the mapping function f at several
points of D. The mapping function f always maps the boundary of the source
region D in a simple way onto the boundary of the unit disk U. Once the
mapping of the boundary is known, the mapping of D itself is determined by
the Cauchy formula with positive orientation of the boundary curve. Thus, it
is sufficient to determine the mapping of the boundary of D onto the boundary
of U, and once we obtain a parametric representation of the boundary of D,
we determine only one real-valued function of a real variable t. But in this
approach it so happens that this unknown real-valued function satisfies an
integral equation. Hence, solving the problem of conformal mapping reduces
to that of solving an integral equation.
The widely used current computational techniques are based on the integral
equation methods where an integral equation is developed to relate the bound-
aries of the problem region and the standard region like the unit disk. Once
the boundaries are discretized at n points, the integral equation reduces to an
algebraic system of equations. The majority of ongoing research in compu-
tational conformal mapping is divided basically into two groups: one where
the maps are constructed from a standard region such as the unit disk onto
the problem region, and the other where the maps are constructed the other
way around. In the first group the integral equation is nonlinear and involves
the conjugation operator, which can be solved by FFr on a discrete mesh in
O(n logn) operations. This method evolves with the numerical solution of
Theodorsen's integral equation, or a related equation, which is solved by us-
ing the fixed-point iteration method. The recent development of Newton's

method is faster for sufficiently smooth boundaries and the choice of a good
initial guess. The basic work in this group has been produced by Wegmann
(1978, 1986), Hubner (1986), and Gutknecht (1986). The first quadratically
convergent algorithm for Theodorsen's equation was presented by Wegmann
(1978) based on the following two ideas: an induction scheme along tangents
to the boundary of the problem region, and a computation scheme similar in
formulation to the Riemann-Hilbert problem for the unit disk at each iteration.
This method is efficient because the solution of the Riemann-Hilbert problem
can be represented by Cauchy integrals. A generalization of this algorithm
to the conformal mapping of an annulus onto a doubly connected region was
published by Wegmann in 1986. This algorithm is also based on the Riemann-
Hilbert problem and is so far the fastest known for this problem. HUbner (1979,
1986) studied Newton's method for the solution of Theodorsen's integral equa-
tion. His method is also based on the solution of the Riemann-Hilbert problem.
He established the quadratic convergence of Newton's method and obtained a
quadratically convergent conformal mapping. An extensive survey of almost all
known methods for numerical confonnal mapping of the unit disk onto a sim-
ply connected region is available in Gutknecht's work (1986). He has derived
integral and integro-differential equations involving the conjugation operator
for the boundary correspondence function. Then various iterative schemes for
solving these equations are presented in this work. The general theory is de-
scribed by specific methods, especially the successive conjugation methods of
Theodorsen (1931), Timman (1951), Freiberg (1951), the projection method
of Bergstrom (1958), Newton's method ofVertgeim (1958), Wegmann (1978),
and HUbner (1979).
In the other group where the maps are constructed from the problem region
onto the unit disk, the integral equations, mostly derived from the Dirichlet
problem, are generally linear, and require O(n 2 logn) operations. In cases
where the geometry is simpler, they may require a smaller number of operations.
The methods in this group are based on Symm's equation (1966) which is
a singular integral equation of the first kind derived by using a single-layer
potential as the basis of confonnal mapping. Symm (1966, 1969) investigated
an integral equation method, like the one he developed for the boundary integral
equation method, for computing the confonnal mapping of a simply connected
region onto the unit disk. Berrut (1976, 1985, 1986) solved Symm's equation
numerically by a Fourier method. This equation is a Fredholm integral equation
ofthe second kind for the derivative of the boundary correspondence function for
the conformal mapping of a Jordan region with a piecewise twice differentiable
boundary onto the unit disk. Kerzman and Trummer (1986) presented a new
method to compute the Riemann mapping function numerically. The solution

of the integral equation of the second kind is expressed in terms of the Szego
kernel and is based on an earlier work of Kerzman and Stein (1978) on the
Cauchy kernel, Szego kernel and Riemann mapping function, and of Kerzman
and Trummer (1986) on a method for the numerical solution of the conformal
mapping problem.
A variant of Symm's integral equation which is suitable for conformal

mapping of both simply and multiply connected regions was presented by
Reichel (1985). It uses a Fourier-Galerkin technique to produce an extremely
fast iterative solution of 0 (n 2 log n) which is due to the singularity of the kernel
so that the linear algebraic system becomes block diagonally dominant. Another
fast method for solving Symm's equation for multiply connected regions was
presented by Mayo (1986). This integral equation formulation is based on a
similar formulation in Mikhlin (1957) and has the advantage of reducing the
problems to integral equations of the second kind with unique solutions and
boundary kernels. Because the solutions are periodic, the trapezoidal rule can
be applied effectively. Once the integral equation is solved, a rapid method is
available to determine the mapping function in the interior of the region. This
method uses a fast Poisson solver for the Laplacian, thus avoiding the time-
consuming computation of integrals at singular points near the boundary.
Other significant modern research includes work by Hoidn (1982) which

deals with conformal mapping of simply connected regions where the sin-
gularity problem near and on the boundary is solved by reparametrization
of the boundary curve. Then this method is applied to Symm's equations
which are solved by using spline functions. The singularity problem has
been solved by Levin, Papamichael and Sideridis (1978), Hough and Pa-
pamichael (1981, 1983), Papamichael and Kokkinos (1981, 1982), Papamichael
and Warby (1984), and Papamichael, Warby and Hough (1983, 1986) by in-
tegral equation methods as well as expansion methods based on the Bergman
kernel or on the Ritz approximation. No research is available in this area for
multiply connected regions of higher connectivity. The Chebyshev approxi-
mation in conjunction with linear programming has been used by Hartman and
Opfer (1986) for conformal mapping of simply connected regions onto the unit
disk. A simple approximation formula has been derived by Zemach (1986) for
the boundary mapping function which gives a remarkably good fit for mappings
of regions with highly distorted boundaries. This method is based on reduc-
ing the nonlocal integral equation for the mapping function to a local equation
depending on the nature of the distorted regions. The Schwarz-Christoffel
formula has always been used in mappings of polygons and related regions,
but the treatment of the singularity problem in these cases has been only re-
cently investigated by Barnard and Pearce (1986), Elcrat and Trefethen (1986),
and Trefethen and Williams (1986). Boundary problems for analytic functions
and integral equations with transformations have been discussed by Lu (1994),
and a comprehensive theoretical account on conformal mapping and boundary
problems can be found in Wen (1992).
The advantages of integral equation formulation in conformal mapping can

be summarized as follows: All integral equations obtained in any conformal
mapping problem (except Arbenz's integral equation, see §12.1) are Fredholm
integral equations of the second kind with bounded kernel, except for Symm's
integral equation which is of the first kind with a kernel that has a logarithmic
singularity. The Fredholm integral equations of second kind are never iII-
conditioned, and there are reliable error estimates available for them. However,
the drawback with the equations of second kind is that the kernel has singularity
at points in the neighborhood of the boundary (but not on the boundary itself
unless it has a corner singularity) where computational difficulties often arise.
This situation, on the other hand, does not occur with the equations of the first
kind.
The best strategy for developing a computational method based on an in-

tegral equation formulation is to make sure that the solution is periodic and
unique. This permits an effective use of the trapezoid rule which is highly ac-
curate on smooth contours. Another feature to look for is that the mapping onto
canonical regions (unit disk, annulus, or slit disks) produces systems of linear
equations and avoids solving systems of nonlinear equations as in Fornberg's,
Guteknecht's, or Wegman's methods.
Some classical applications of conformal mappings to steady state prob-

lems of mathematical physics and especially for the solution of the Laplace
equation can be traced to the beginning of the twentieth century. A noteworthy
contribution to the theory of elasticity is by Muskhelishvili (1963). Modern
contributions can be found in areas of fluid flow, heat conduction, solidifica-
tion, electromagnetics, ion optics, acoustics, vibrations, wave guides, and grid
generation, to name a few; a detailed review and biography of the applications
through 1972 is available in Laura (1975). The problem of flow and heat trans-
fer in conduits of arbitrary shape in space vehicles was investigated by Sparrow
and Haji-Sheikh (1966). This study was extended to noncircular conduits with
uniform wall temperature by Casarella et al (1971). Unsteady heat conduc-
tion problems in bars of arbitrary cross section were investigated by Laura et
al.(1964, 1965, 1968). Ives (1976) analyzed the incompressible flow between
two concentric circles and computed the streamlines by using Garrick's method
of conjugate functions. The problem of solidification of steady-state and tran-
sient frozen layers in rectangular channels has been solved by Siegel, Goldstein
and Savino (1970). In transient solidification the shape of a frozen region is
determined by mapping it onto a potential plane and then computing the time-
dependent conformal map between the potential and the physical plane. The
thermoelastic problem of uniform heat flow distributed by an isolated hole of
ovaloid form was investigated by Florence and Goodier (1960) and extended
by Deresiewicz (1961) to holes which are mapped onto the unit circle and
approximated by polynomials.
Wilson (1963) and Richardson (1965) used conformal mapping to determine

the stresses in solid propellant rocket grains by solving an integral equation of
the Fredholm type and a system of coupled integral equations. The method of
computing the conformal mapping function and the related eigenvalue problem
for plane regions with irregular boundaries was published by Laura (1968).
The ion problem connected with the trajectory of a charged particle in a plane
electric field and a normal magnetic field was solved by Naidu and Westphal
(1966) by using the Schwarz-Christoffel transformation. Conformal mapping
techniques of simply and doubly connected regions have been used, among
others, by Kasin and Merkulov (1966), Laura (1967), and Laura et al. (1972).
Conformal mapping has been applied to acoustic waveguides of compli-

cated cross section where the Galerkin method is applied to obtain a functional
approximation for the solution of the boundary value problem. The grain of a
solid propellant rocket motor with a starlike internal propagation is in the form
of a circular cylinder bounded by a thin case. To solve any boundary value
or eigenvalue problem, the grain cross section is conformaUy mapped onto a
circle or an annulus. Studies on the shear vibrations of such rocket motors were
done by Baltrukonis et al. (1965) and Laura and Shahady (1966). Conformal
mapping techniques are used in a study on the Rayleigh-Taylor instability for
ideal fluid by Menikoff and Zemach (1980, 1983). Grid generation for cas-
cades of blades and inlet flows has been investigated by Inoue (1983, 1985).
The Kirchhoff flow problem past a polygonal obstacle was solved by Trefethen
(1986). Other contemporary applications can be found in the book on numerical
conformal mapping edited by Trefethen (1986).
Computational conformal mapping in the present decade is still progress-

ing steadily. Fast algorithms are being developed. Computational conformal
mapping is being used in engineering problems that require grid generation and
related domain simplification. Although the use ofconformal mapping in reduc-
ing or eliminating singularities in solutions of integral and integro-differential

equations is one important issue, the choice of an initial good guess in New-
ton's method is another. Much has developed since the evolution of computer
technology around 1955, but this subject has not yet reached maturity.
REFERENCES USED: Andersen et al. (1962), Barnard and Pearce (1986),

Berrut (1986), Caratheodory (1969), Elcrat and Trefethen (1986), Freiberg (1951),
Gaier (1964, 1983), Goluzin (1969), Gutknecht (1986), Hartman and Opfer (1986),
Hoidn (1986), Hiibner (1979, 1986), ryeS (1982), James (1971), Jawson (1963), Jaw-
son and Symm (1977), Kantorovich and Krylov (1958), Kerzman and Stein (1978),
Kerzman and Trummer (1986), Kober (1957), Laura (1975), Lawrentjew and Schabat
(1967), Lu (1994), Mayo (1986), Mikhlin (1957), Nehari (1949, 1952), Papamichael et
al. (1981, 1983, 1984, 1986), Riemann (1851), Seidel (1952), Symm (1966), Trefethen
(1980), Trefethen and Williams (1986), Wegmann (1979, 1986), Wen (1992), Zemach
(1986).
Chapter 1
Basic Concepts
Some basic concepts and results from complex analysis are presented. They
include harmonic functions, Cauchy's theorem, Cauchy kernel, Riemann map-
ping theorem, analytic continuation, and the Schwarz reflection principle.
Proofs for most of the results can be found in textbooks. References that were
mostly used are cited at the end of the chapter.
1.1. Notation and Definitions
A complex-valued function I is said to belong to the class Ck(D) if it is

continuous together with its k-th derivatives, in a domain D, 0 :::::: k < 00.
In this case we shall often write that I is a Ck(D)-function, or that I is
a Ck-function on D, and a CO-function is written as a C-function. The
function I in the class C k (D), for which all k-th derivatives admit continuous
continuations in the closure D, form the class of functions Ck(D). The class
COO(D) consists of functions f which are infinitely differentiable on D, i.e.,
continuous partial derivative of all orders exist. These classes are linear sets.
Thus, every linear combination >"1 + J.l9, where>" and J.l are arbitrary complex
numbers, also belongs to the respective class.
Let R n denote the Euclidean n-space, and R+ the set of nonnegative real
numbers. The complement of a set B with respect to a set A is denoted by
A\B (or compl (B) if the reference to set A is obvious), the product of the
sets A and B by A x B, and the closure of a set A by A.
15

16 CHAPTER 1. BASIC CONCEPTS
Let C denote the complex plane. If a E C and r > 0, then
B(a,r)={zEC: Iz-al<r},
B(a,r)={zEC: Iz-al~r}, (1.1.1)
aB(a,r) = {z E C : Iz - al = r},
denotes, respectively, an open disk, a closed disk, and a circle, each of radius
r and centered at a. An open unit disk B(O, 1) is sometimes denoted by U.
A connected open set A ~ C is called a region (or domain). The extended
complex plane is denoted by Coo' Then aooD is the boundary of a set D in
Coo, i.e.,
aD if D is bounded,
aooD = {
aD U {oo} if D is unbounded.
If D is a region in Coo, then the following statements are equivalent: (a) D is
simply connected, (b) Coo \D is connected, and (c) aooD is connected. Regions
that have more than one layer over the complex plane are called Riemann
surfaces.
Let z = x + i y be a complex number. Then z = x - i y, x = z ; Z, and

z-z
y= 2i' Also
(1.1.2)
A function f :D ~ C is analytic on D iff af = 0, which is equivalent to the

Cauchy-Riemann equations for the function w = fez) = u(x, y) +iv(x, y):
Ux = Vy , Uy = -V x , (1.1.3)
or, in polar form (z = re iO ),
1 1
Ur = -vo, Vr = --Uo· (1.1.4)
r r
Thus,
f'(z) = U x + iv x = vy - iuy. (1.1.5)
The Cauchy-Riemann equations are necessary conditions for f (z) to be analytic
on D. However, merely satisfying the Cauchy-Riemann equations alone is not
sufficient to ensure the differentiability of f (z) at a point in D.
1.1. NOTATION AND DEFINITIONS 17
1 - 1
The following results are obvious: 8 (log Izl) = 2z' 8 (log Izl) 2z'
8 f = [)1, and the Laplacian
2 _ 82 82 - -
V = 8x 2 + 8 y2 = 488 = 488. (1.1.6)
The Cauchy-Riemann equations (1.1.3) for the function f(z) = u(x, y) +

iv(x, y) satisfy the partial differential equations
u x vx +uy vy = 0,
2 2 (1.1. 7)
V u = 0, V v = O.
v = 1 8x
. the grad'lent vector M
Usmg · 8
+ J. 8y'
8 therst
fi equation
. .m (1.1.7) can be
written as the inner (scalar) product:
(Vu, Vv) = O. (1.1.8)
Then the Cauchy-Riemann equations yield IVul = IVvl = 1f'(z)l. Eq (1.1.8)

also signifies the orthogonality condition for the families of level curves defined
by u(x, y) = const and v(x, y) = const.
If w = u + iv, then
~{z}. ~{w} = ~ {Z2 -./zI2 w}.

2tz
(1.1.9)
Let E denote a closed bounded infinite set of points in the z-plane. For the
points Zl, Z2, ... , Zn E E the Vandermonde determinant is defined by
n
V(Zl,Z2,." ,zn)= II (Zi-Zj), n~2. (1.1.10)
i,j=l
i#j
Let us define the numbers

d n = v;/n(n-l), (1.1.11)
where dn+l :::; d n . Since d n does not exceed the diameter* of the set E for
any n, it follows that the sequence {d n } approaches a finite limit as n ~ 00.
*The diameter of the set E is defined as SUp{IZi - zjl : Zi,Zj E E}, i,j =
E
1,2, ... ,n (i i'j).
This limit is called the transfinite diameter of the set E and is denoted by
diam (E), i.e., diam (E) = lim dn . When the set E has finitely many points,
n--t(x,
we take diam (E) = O.
A simple closed curve f in C is a path "I : [a, b] f-? C such that 'Y( t) = "1(8)
iff t = 8 or It - 81 = b - a. In what follows, a simple closed curve shall be
called a Jordan contour. The Jordan curve theorem states that if f is a simple
contour, then C\f has two components, one called the interior of f, denoted
by Int (r), and the other called the exterior of f, denoted by Ext (f), each of
which has f as its boundary. Thus, if f is a Jordan contour, then Int (f) and
Ext (r) U {oo} are simply connected regions.
Fig. 1.1.1. Rectifiable curve.
Let a continuous curve f, defined by 'Y(t) = o:(t) + if3(t), be divided into

narcs (1k = Zk-'";' Zk, k = 1, ... , n, where Zk = 'Y(td for k = 0,1, ... ,n,
such that the end point of each are, except the last one, overlaps the initial point
of the next arc. If we join each each segment [Zk-l, Zk], k = 1, ... ,n, by
straight line segments (see Fig. 1.1.1 where n = 5), we obtain a polygonal line
L inscribed in f. The segments of L are the chords joining the end points of
the arcs (1k, and
z=
n
length of L = IZk - Zk-ll· (1.1.12)
k=l
The curve f is said to be rectifiable if
z=
n
sup IZk - Zk-ll = l < +00,
P k=l
where the least upper bound is taken over all partitions P = {a = to, t 1, . .. , t n
= b} of the interval [a, b], a ~ t ~ b. The nonnegative number l is called the
length of the curve f. The curve is said to be nonrecti!iable ifthe sums (1.1.10)
become arbitrarily large for suitably chosen partitions.
A piecewise smooth curve f, defined by 7 = [a, b] f-+ C, is called a

reparametrization of a curve f, defined by , = [a, b] f-+ C if there exists a
function a E C 1, a : [a, b] f-+ [a, b] with d(t) > 0, a(a) = a and a(b) = b,
such that ,(t) = 7(a(t)), t E [a, b]. Then i f = £f for any C-function f
defined on an open set containing the image of, (which is equal to the image
of7)·
Let f be a C-function on an open set Dee, and let r be a piecewise

C1-continuous curve in D. If If(z)1 ~ M for all points z E f, i.e., for all
z = ,(t) for t E [a, b], where M > 0 is a constant, then
Ii f dzl ~ M l(r), (1.1.13)
where
(1.1.14)
is the arc length of the path f. In general, by applying the triangle inequality,
(1.1.15)
If, : [a, b] f-+ C defines a piecewise smooth contour f and F is a function

defined and analytic on a region containing f, then
i F'(z) dz = F(r(b)) - F(r(a)). (1.1.16)
i
If,(a) = ,(b), then
F'(z) dz = O. (1.1.17)
This results is known as the fundamental theorem for line integrals (contour
integration) in the complex plane. Thus, if a function f is defined and analytic
on a region Dee and if f'(z) = 0 for all points zED, then f is a constant
on D. If j is a C-function on a region D, then the following three statements

are equivalent:
1
(i) If r I and r 2 are two paths in D from a point Zl E D to a point Z2 ED, then
1 j(z) dz = j(z) dz, i.e., the integrals are path-independent.

rl r2
(ii) If r is a Jordan contour lying in D, then l j(z) dz = 0, i.e., the integrals
on a closed contour are zero.
(iii) There exists a function F defined and analytic on D such that F ' (z) = j (z)
for all zED, i.e., there exists a global antiderivative of j on D.
A function If>(z), analytic inside a region with boundary r = u~=lrk, is

said to be sectionally analytic on r if If>(z) is continuous on each r k from
both left and right except at the end points where it satisfies the condition
1lf>(z)1 ::; IC I ' where c is the corresponding end point ofr k, and C and 0:
z-c'"
are real constant with 0: < 1.
HOLDER CONDITION. Let j(t) be defined on a Jordan curve r (open or

closed). If
(1.1.18)
for arbitrary points tl, t2 E r (tl :I t2), where A > 0 and 0: are real constants,
then j(t) is said to satisfy the Holder condition of order 0:, or simply the
condition H"', denoted by j(t) E H"'. The condition HI is known as the
Lipschitz condition. If j (t) E C (r) and I (t) E H"', then we say that I (t)
is HOlder-continuous on r. If j E C (r) and I E HI, then I (t) is said to be
Lipschitz-continuous.
FATOU'S LEMMA. Let In : If---' lR be nonnegative, extended real-valued,

measurable functions defined in an interval I and such that the sequence {In}
converges pointwise to the function I : If---' lR. If lim
n-+oo
r In <
Jr
00, then I is
integrable and
r j::;
JI
lim
n-+oo
r In.
Jr
(1.1.19)
METRIC SPACES. Let S denote the set of all real-valued sequences. Then
S is a vector space if addition and scalar multiplication of vectors Si, ti E S
for i = 1,2, ... are defined coordinatewise, i.e., {sil + {til = {Si + til, and
A{Si} = {A Si} for a scalar A. The Fnkhet metric (distance) d ({ Si}, {t i })

is defined for {sd, {ti} E S by
(1.1.20)
L ISijP < oo}. Since

00
The space £P, 1 ~ p< 00, is defined by £P = {{Si} :

i=l
the sum of any two elements in £P is also in £P, then fP is a vector subspace of
S. Define a norm II . lip on fP by
II{Si}llp = ( 2: ISi\P
00 ) lip
. (1.1.21)
i=l
Then it can be shown that fP is closed under vector addition and II . lip satisfies
the triangle inequality. Let {Si}, {til E fP be such that E ISilP = 1 and
E Iti\q = 1, where p and q are called conjugate exponents such that p > 1 and
1 1
-p + -q = 1. Then
Holder inequality:
(1.1.22)
Cauchy-Schwarz inequality:
00 00 00
Lisitil ~ 2: I il 2: It i l
S 2 2
. (1.1.23)
i=l i=l i=l
Minkowsky's inequality:
(flsi+tilpf/P ~ (fl siI P)l /P + (fltilpf/P, p?-1. (1.1.24)

i=l i=l i=l
Note that p = 1 gives the triangle inequality.
Let L 2 denotes the Hilbert space of all square-integrable analytic functions

j in a simply connected region D with boundary r. A function j(z) regular
in D is said to belong to the class L 2 (D) (we say, j E L 2 (D) ) if the integral
Jllf(z)1 2 dS z < +00, where dS z = dxdy denotes an area element in D.

The (surface)-norm Ilflb,D of fez) is defined by
J
IlflltD = llf(z)1 2 dSz ·
If two functions f,g E L 2 (D), then their inner product is defined by
(1.1.25)
Let f be the rectifiable Jordan boundary of the region D, of length l, and let f p
denote the image ofthe circle Iwl = p under the mapping z = g( w) = f- 1 (w),
o ,
27r
= p J<t>=Jf w = pei<t>, (1.1.26)
r
where ds z denotes a line element on f, is a monotone increasing function. We

say that f E L 2 (f) if this integral remains bounded as p --+ R, i.e.,
where Ilfll~ r is the (line)-norm of f. For any two functions f, 9 E L 2 (f), we

define their 'inner product as
(1.1.27)
If the disk B (zo , r) CD, then
(1.1.28)
Let a region have a piecewise continuous boundary f, and let fez), g(z) be
regular in D, and f'(z) and g'(z) continuous in D. Then
(1.1.29)
1.2. SOME BASIC THEOREMS 23
This is known as Green '8 formula. It is useful in converting a surface integral

into a line integral. This formula also holds for multiply connected regions.
The integral along r is taken in the positive sense, i.e., the region D remains to
the left as one traverses the contour r. The following inequality is also useful:
If f E L 2 (D), then f E L 2 (r), and
(1.1.30)
We shall denote by Loo the Hilbert space of 21r-periodic and bounded

functions f with the norm
Ilflloo = w,~ If(x)l, (1.1.31 )
and by W the Sobolev space of21r-periodic and absolutely continuous functions

f, with l' E L 2 [0,21r]. Then Ilfll = max (1Iflloo, 111'112), and (W, II· II) is a
Banach space. Thus, if f, g E (W, II . 11), then (i) Ilfll ~ 0; (ii) II fll = 0 implies
that f = 0; (iii) Iia fll = laillfil for a scalar a; and Ilf + gil::; Ilfll + Ilgll
which is known as the triangle inequality.
1.2. Some Basic Theorems
Some basic theorems from the theory of functions of a complex variable are
presented without proofs which can be found in standard textbooks on the
subject.
THEOREM 1.2.1 (CAUCHY'S THEOREM). Let f be analytic on a

region D, and let r be a closed contour which is homotopic to a point in
D. Then l f = O.
Note that a set is said to be simply connected if every closed contour reD
is homotopic (as a closed curve) to a point in D, i.e., to some constant curve.
Some local versions of Cauchy's theorem are as follows:
THEOREM 1.2.2 (CAUCHY-GOURSAT THEOREM FOR A DISK). Let

f: B I-t C be analytic on a disk B = B(zo, r) c D. Then
(i) there exists a function F : B I-t C which is analytic on B and is such
l
that F ' (z) = f (z) for all z E B (i. e., f has an antiderivative on B) i and
(ii) Iff is a closed contour in B, then f = O.
This theorem also holds if f is continuous on B and analytic on D\ {Zl}

for some fixed Zl E D.
THEOREM 1.2.3 (CAUCHY-GOURSAT THEOREM FOR A RECTAN-

GLE). Let R denote a rectangle with sides parallel to the coordinate
l
axes, and let f be a function defined and analytic on an open set D
containing R. Then f = O.
Even if the function f is analytic on D except at some fixed point Zl E D

which does not lie on the contour R, and if lim (z - zl)f(Z) = 0, then also
l
Z-+Zl
f = o. The two theorems 1.2.2 and 1.2.3 hold (a) if f is bounded on a

deleted neighborhood of Zl, or (b) if f is continuous on D, or (c) if lim fez)
Z-+Zl
exists.
The index of a curve f with respect to a point Zo E C is the integer n

that expresses how many times f winds around Zoo This index is denoted by
I(f, zo) and is called the winding number off with respect to Zoo Thus,
1
I(f,zo) = -2.
t7T
1
r
-dz
-.
Z - Zo
(1.2.1)
In fact,
±n if Zo E Int(f),
I (f, Zo ) = {
o if Zo E Ext (f).
THEOREM 1.2.4 (CAUCHY'S INTEGRAL FORMULA). Let f be ana-

lytic on a region D, and let f be a simple closed contour in D that is
homotopic to a point in D. Let Zo E D be a point not on f. Then
f(zo) . I(r, zo) = -1.

2Z7T
1 f(C)-".-
r" - Zo
dC· (1.2.2)
1.2. SOME BASIC THEOREMS 25
The integrand in (1.2.2) is known as the Cauchy kernel defined by
H(z, zo) = J..- f(z) . (1.2.3)

2m z - Zo
The formula (1.2.2) is a special case of integrals of the Cauchy type. If we set
F(z) =~
2m
r g(() de,
Jr ( - z (1.2.4)
where r : [a, b] f-+ C is a simple contour and 9 a C-function defined on the

image r([a, b]), then F is analytic on C\r([a, b]) and is infinitely differentiable,
such that its k-th derivative is given by
F(k)(Z) = ~
2i7r Jrr ((-g((jk
z +1 de, k = 1,2, .... (1.2.5)
Then Cauchy's integral formula for the derivatives is
f
_ ~ r
f(()
(z) I(r, zo) - 2i7r J (( _ z)k+l de,
(k).
k = 1,2, .... (1.2.6)
r
Let f be analytic on a region D and let r = oB(zo, R) be a circle lying in D.
If If I :s; M for all z E r, then for k = 0,1, ...
k'
If(k)(zo)1 :s; R~ M. (1.2.7)
This result is known as Cauchy's inequality. A corollary is the Liouville

theorem which states that the only bounded entire functions are constants. A
l
partial converse of Cauchy's theorem is known as Morera's theorem which
states that if f is continuous on a region D and if f dz = 0 for every closed
contour r in D, then f is analytic on D, and f = F' for some analytic function

FonD.
Two very useful corollaries of Cauchy's integral formula (1.2.2) are

(i) The maximum modulus theorem which states that if f is a nonconstant
analytic function on a region D with a simple boundary r, then If I cannot have
a local maximum anywhere in Int (r).
(ii) The mean value property of an analytic function f defined on the circle
oB(zo, r) states that
f(zo) = -1
27r
1
0
2
71"
f (zo + re i6 ) dB. (1.2.8)
An application of the maximum modulus theorem is
SCHWARZ LEMMA. Let j be analytic on the open unit disk U, and suppose
that Ij(z) ::; 1 for all z E U and j(O) = O. Then Ij(z)1 ::; Izi for all z E U
and Ij'(O)1 ::; 1. If Ij(zo)1 = Izol for some Zo E U, Zo =J 0, then j(z) = ez
for all z E U, where eis some constant such that lei = 1.
BOUNDARY VALUES FOR CAUCHY'S INTEGRAL. If the boundary r

of a simply connected region D is rectifiable and if j(z) is regular in D and
continuous on D, then Cauchy's integralformula (1.2.2) holds for every Zo ED.
For the boundary values we have the following result: Let the boundary r of
D be smooth, and let j(z) be regular in D and continuous on D. Then for a
point Zo E r which is not a corner point
j(zo) = ~
~1r
r j(() de.
Jr ( - Zo (1.2.9)
In the case when Zo E r is a corner point with inner angle em, 0 < a < 2, the
boundary value j(zo) is given by
j(zo) = ~
W1r
r j(()
Jr (- Zo
de, (1.2.10)
where the integral is taken as a Cauchy p.v. If D is the region exterior to the
Jordan curve r which is traversed in the positive sense and if j (z) is regular at
z = 00, then the boundary value at a point Zo E r which is not a corner point
is given by
j(zo) = -~
~1r
r /(()
Jr" - Zo
+ 2 j(oo) de (1.2.11)
and at a corner point Zo E r with the inner angle a, 0 < a < 2, by

J(zo) = '(2
~
1
-a1r
) 1 J(()
-r-d(+2J(00).
r,,-Zo
(1.2.12)
The integrals in (1.2.9)-( 1.2.12) are taken as Cauchy p.v's.
PLEMELJ FORMULAS: If r is a contour, then
P+((o) = ~ j((o) + ~ r J(()

Jr ( - (0
de,
1
2 2~1r
(1.2.13)
1 1
P-((o)=--J((o)+-. -j(()
-de
2 2m r (- (0
1.3. HARMONIC FUNCTIONS 27
where P+ ((0) and P- ((0) are the limiting values from the right and left of r,
respectively, f (() satisfies the Holder condition on r, and (0 does not coincide
with those end points where f ((0) I=- o. If (0 coincides with an end point where
f((o) = 0, then F+((o) = F-((o) = f((o). A proof for these formulas can
be found in Muskhelishvili (1946,1992).
Another set of useful formulas is as follows:
sin 7rZ = 7rZ g (1 - ~:) , (1.2.14)
7r cot 7rZ = L --
001 =
Z- n
n=-oo
1+
-
Z
L (--
00
n=l
1 + -1 -) ,
Z- n Z+ n (1.2.15)
where the series converges in the Cauchy sense.
1.3. Harmonic functions
The functions whose Laplacian is zero are known as harmonic functions. Thus,
a real-valued function u(x, y) E C 2 (D) is said to be harmonic in a region D
if V'2 u = O. Some properties of harmonic functions in 1R2 are as follows:
(i) The function

1 1
r = -Vr7(x=-=x:::=:;o =+=;(=y=-=Y:::=:;o):=;;;:2
):=;;;:2
(1.3.1 )
is harmonic in a region that does not contain the point (xo, Yo).
(ii) If u(x, y) is a harmonic function in a simply connected region D, then there

exists a conjugate harmonic function v(x, y) in D such that u(x, y) + i v(x, y)
is an analytic function of Z = x + iy = (x, y) in D. In view of the Cauchy-
Riemann equations (1.1.3),
l
X ,Y
v(x, y) - v(xo, Yo) = (-u y dx + U x dy) , (1.3.2)
Xo,Yo
where (xo, Yo) = Zo is a given point in D. This property is also true if Dis
multiply connected. However, in that case the conjugate function v(x, y) can
be multiple-valued, as we see by considering u(x, y) = log r = log x 2 + y2 vi

defined on a region D containing the origin which has been indented by a small
circle centered at the origin. Then, in view of (1.3.2),
vex, y) - v(xo, Yo) = tan- 1 JL ± 2mr + const, n = 1,2, ... ,

x
which is multiple-valued.
(iii) Since derivatives of all orders of an analytic function exist and are them-
selves analytic, any harmonic function will have continuous partial derivatives
of all orders, i.e., a harmonic function belongs to the class Coo(D), and a partial
derivative of any order is again harmonic.
(iv) A harmonic function must satisfy the mean-value theorem, where the mean
value at a point is evaluated for the circumference or the area of the circle around
that point. If u is harmonic on a region containing the closed disk B (zo, r),
where Zo = Xo + iyo, then
u(xo, Yo)
1
= 271" r 2tr
io u(zo +re
iO
) dB. (1.3.3)
(v) In view of the maximum modulus theorem (§ 1.2), the maximum (and also
the minimum) of a harmonic function u in a region D occurs only on the
boundary of D. This result is known as
THEOREM 1.3.1 (MAXIMUM PRINCIPLE). A nonconstant function

which is harmonic inside a bounded region D with boundary r and con-
tinuous in the closed region [) = D u r attains its maximum and mini-
mum values only on the boundary of the region.
Thus, u has a maximum (or minimum) at Zo E D, i.e., if u(z) ::; u(zo)

(or u(z) 2 u(zo) for z in a neighborhood B(zo, c) of zo, then u = const in
B(zo, c).
(vi) The value of a harmonic function u at an interior point in terms of the

boundary values u and ~~ is given by Green's third identity (B.8).
(vii) If u and U are continuous in [) and harmonic in D such that u ::; U on r,

then u ::; U also at all points inside D. In fact, the function U - u is continuous
1.3. HARMONIC FUNCTIONS 29
and harmonic in D. Hence U - u :2: 0 on r. Then, in view of the maximum

principle (Theorem 1.3.1), we require that U - u :2: 0 at all points inside D.
(viii) If u and U are continuous in D and harmonic in D for which lui :::; U
on r, then lui :::; U also at all points inside D. In fact, the three harmonic
functions -U, u, and U satisfy the relation -U :::; u :::; U on r. Then, by (vii),
-U :::; u :::; U at all points inside D, or lui:::; U inside D.
HARNACK THEOREM: Suppose that {un(z)} is a monotone increasing

sequence of harmonic functions on a region D, which is convergent at a point
Zo ED. Then {Un (z)} is uniformly convergent on closed sets in D.
Let Zo i- 00 be any point inside D, and let K denote the closed disk
B(zo, R) such that KeD. Then, if r = Iz - zol :::; R, the Harnack inequality
R-r R+r
un(zo) - R :::; un(z) :::; un(zo) - R (1.3.4)
+r -r
holds for any annulus 0 < r < R with center at Zo, provided that un(z) are
harmonic and nonnegative on the disk B(zo, R).
IDENTITY THEOREM: If 1(z) is analytic on a region D, Zo E D, and

{Zk} '1 is a sequence of distinct points in D such that Zk -> zo, and 1(Zk) = 0
for k = 1,2, ... , then 1(z) == 0 on D. This theorem is useful in establishing
certain identities, including the concept of analytic continuation.
CAPACITY: Let D* denote the complement of the region D that includes

the point Z = 00. Then the region D* can be covered by regions D*(n) with
boundaries r*(n), n = 1,2, .... The Green's function 9n(z, 00) of the regions
D*(n) is a harmonic function in D*(n) except at the point Z = 00 which
assumes the value zero on r*(n) and in a neighborhood of 00 behaves such that
lim [9n (z, 00) - log Izl] = In exists (and is finite). The quantity In is called
z-->oo
Robin's constant for the regions D*(n). Hence, in a neighborhood of Z = 00
9n(z,00) = log Izi + un(z) + In, (1.3.5)

where un(z) is a harmonic function in D*(n), including the point z = 00, and
un(z) -> 0 as Z -> 00. Since D*(n) c D*(n+1) , the function 9n+1(z, 00) -
9n(z, 00) is harmonic in D*(n), including the point 00. In view of the maximum
principle (Theorem 1.3.1), the last statement is true everywhere in the region
D*(n), i.e., for all z E D*(n)
(1.3.6)
In view of the Harnack theorem, the sequence of functions {un + 'Yn}, where
Un (00) = 0, either converges to 00 in D* or converges to a harmonic function
u(z) +'Y such that u( 00) = O. In the latter case, the quantity 'Y is called Robin's
constant for the region D*, the quantity C = e--Y is cal1ed the capacity of the
region D* (denoted by cap (D*)), and the function 9(z, 00) is called Green's
function for the region D*, which assumes nonnegative values everywhere in
D* , but these values need not be zero. For example, at isolated boundary points
of D* the value of 9 (z, 00) is positive. In the case when Un (z) + 'Yn ----t 00 for
z E D*, we have 'Yn ----t 00 because Un (00) = O. In this case the capacity of
the region D is taken as zero. The fol1owing results are useful:
THEOREM 1.3.2. For the region D*, containing the point at infinity,
to have a Green's function 9 (z, 00), it is necessary and sufficient that
the capacity of its boundary r be positive.
THEOREM 1.3.3. The capacity of an arbitrary closed and bounded

region D is equal to its transfinite diameter, i.e., cap (D) = diam(D).
1.4. Conformal Mapping
If we place a sphere such that the complex plane is tangent to it and the origin
coincides with the south pole, then we can transfer al1 points of C to the sphere
by projection from the north pole. This is called a stereographic projection
which is a one-tQ-Qne map of C onto the sphere, such that its image is the
whole sphere except the north pole which then corresponds to the point at
infinity z = 00.
A mapping f of a region D onto a region G is called analytic iff it is

differentiable. The mapping f is cal1ed conformal if it is bijective and analytic.
The conformal mapping theorem states that if a mapping f : D f-+ G is
analytic and f' (zo) -I 0 for each zED, then f is conformal. Thus, f rotates
e
tangent vectors to curves through Zo by a definite angle and magnifies (or
contracts) them by a factor r. The mapping f is conformal ifit is analytic with
a nonzero derivative. Two important properties are the fol1owing:
(i) If f : D f-+ G is conformal and bijective (i.e., one-tQ-Qne and onto), then
f- 1 : G f-+ D is also conformal and bijective.
1.4. CONFORMAL MAPPING 31
(ii) If I : D 1-+ G and 9 : G 1-+ E are conformal, then the composition

log : D 1-+ E is conformal and bijective.
Property (i) is useful in solving boundary value problems (e.g., the Dirichlet
problem) for a region D. The method involves finding a map I : D 1-+ G such
that G is a simpler region on which the problem can be first solved, and then the
result for the original problem is provided by 1-1. Since the Dirichlet problem
involves harmonic functions (see §1.3), the following result on the composition
of a harmonic function with a conformal map is useful: If u is harmonic on a
region G and if I : D 1-+ G is conformal, then u 0 I is harmonic on D. In
fact, let zED and w = I(z) E G. Then there is an analytic function 9 on the
open disk B(w, p) c G such that u = ~{g}. Thus, u 0 9 = ~{g 0 n, which
is harmonic since 9 0 I is analytic. A mapping in which both the magnitude
and the sense of the angles between the curves and their images are preserved
is said to be a conformal mapping of the first kind, but if the sense of the angles
is reversed, then it is called a conformal mapping of the second kind.
g
-----+
Fig. 1.4.1.
THEOREM 1.4.1 (RIEMANN MAPPING THEOREM). Let Dee be a

simply connected region. Then there exists a bijective conlormal map
I : D 1-+ U, where U is the open unit disk. Moreover, the map I is
unique provided that I(zo) = 0 and f'(zo) > 0 for Zo E D.
This theorem implies that if D and G, both contained in C, are any two
simply connected regions, then there exists a bijective conformal map g: D 1-+
G. If I : D 1-+ U and h : G 1-+ U, then 9 = h- 1 0 I is bijective conformal
(Fig. 1.4.1). Thus, the two regions D and G are said to be conformal if there
exists a bijective conformal map between them.
A bijective conformal map is also called a univalent map. A function

w = f (z) defining a univalent mapping is called a univalent function. Its
inverse image is also a univalent function defined on the image region. In
the study of univalent mappings the first question asked is whether a given
region can be mapped univalently onto another region. In the case of simply
connected regions it is necessary that the two regions have the same connectivity.
Once this condition is met, we can inquire about the possibility of univalent
conformal mapping of various regions onto a given simply connected region. To
determine how a simply connected region is mapped onto a simply connected
region, we must know their mappings onto a standard region, such as the unit
disk. This enables us to obtain the required mapping first by mapping the
given region onto the unit disk and then mapping the unit disk onto the other
region. Now the question arises whether an arbitrary simply connected region
can be mapped onto the unit disk. It turns out that this is always possible except
for two cases, namely, when the region is the entire plane and when it has a
single boundary point. Note that the requirement that the boundary orientation
be preserved even in the mapping of regions with non-Jordan boundaries is
important. Consider, for example, the map w = sin z which is analytic on
the strip -1f / 4 < x ~ 31f / 4 and maps bijectively the two boundaries of this
strip onto the two boundaries of the hyperbola u 2 - v 2 = 1/2, which make the
boundary of a simply connected region (curved strip). But the boundary of this
curved strip is not traversed in the positive sense. Other simple examples of
functions that are not univalent in the lower or the upper half-plane are w = z2,
and w = sinh z, although they are analytic in C and map bijectively the real axis
~{z} = 0 onto the real axis ~{w} = O. Thus, thefunction w = f(z), analytic
°
on the half-plane ~{z} > and continuous on the closed region ~{z} 2': 0,
grows as Iz I --+ 00 at most as fast as C z2, Le., the ratio 1(:) is bounded for
° ~ arg{ z} ~ 1f and a sufficiently large 14
z
The Riemann mapping theorem also implies that there exists a unique func-
°
tion w = F(z) that is regular in D, that is normalized at a finite point Zo E D
by the conditions F(zo) = and F'(zo) = 1, and that maps the region D
univalently onto the disk Iwl < 1. In fact, the function F(z) = f(~}) is
°
such a function, where f(z), with f(zo) = and f'(zo) > 0, is the function
mentioned in the Riemann mapping theorem, and the radius of the disk onto
which the function w = F(z) maps the region D is R = f,tzo)' If there exists
another function w = FI (z), with FI (zo) = 0 and F{ (zo) = 1, that maps
D onto a disk Iwl < R I , then, by the Riemann mapping theorem, we could
F1(z) 1,. f(z)

have ~ = f(z), and hence, Rl = f (zo), I.e., F1(z) = f'(zo) = F(z),
which proves the uniqueness of the mapping function w = F( z). The quantity
R = f' (~o) is called the conformal radi us ofthe region D at the point Zo ED.
Since D* = Ext (r) is simply connected, Green's function Q(z, 00) for D*
coincides with log If(z)l, where the function w = f(z) maps D* univalently
onto Iwl > 1 such that f( 00) = 00. Then, Robin's constant, for the region
D* is equal to log 1f'(00)1, and cap (D*) = If'(~)1 = R, where R is the
conformal radius of the region D* (with respect to 00), i.e., the number R is
such that the region D* is mapped univalently onto Iwl > R by a normalized
function w = F( z) with F( 00) = 00 and F' (00) = 1. Thus, in view of
Theorems 1.3.2 and 1.3.3, we have the following theorem.
THEOREM 1.4.2. The capacity, and hence the transfinite diameter

of a bounded simply connected region D, is equal to the conformal radius
of the region D* which is the complement of the region D in Cex> and
contains the point at infinity.
CHAIN PROPERTY: Let Do, ... ,Dn be regions inC, andletfk : D k - 1 >->
Dk denote conformal mappings for k = 1, ... ,n. Then the mapping g =
fn 0 . . . 0 fI, defined by
g(z) = fn (fn-l (... 12 (fI(z))·· .)) (1.4.1)
is a conformal mapping of Do onto Dn . The mapping g is said to be composed

of a chain of mappings fI, ... , f n and is represented by the scheme
it D 1 ~
Do ~ h D2 ~
h ...
In-1
~
Dn-l ~
In Dn' (1.4.2)
Thus, the set of regions on C can be divided into mapping classes such that
two regions can be mapped conformally onto each other iff they belong to the
same mapping class. Fig. 1.4.2 represents an example of this chain property.
The practical applications of conformal mappings are related to the problem

of constructing a function which maps a given region onto a given region. Often
we find an explicit expression for the mapping function and determine it by
applying the chain property.
y y
:-"'_.JIL_ _• X --
Z''''-' -iz
---+----~x
/z........- Z2_ 1
y y
--t----+x -
z........- .JZ
---==3----+ x
1-
z I+Z
Y
y
----f--+--+c--'I> X
- - - t - -.... x
Fig. 1.4.2. Chain property of conformal mappings.
ANALYTIC CONTINUATION: If A and B are two regions that overlap, i.e.,

An B =1= 0, and if f is analytic on A, 9 is analytic on B, and f(z) == g(z) on
AnB, then 9 is called the analytic continuation of f from A into B. This analytic
continuation is unique. In fact, suppose that g1 and g2 are analytic continuations
of f from A into B. Theng1 (z) == g2(Z) on AnB, and thus, by identity theorem
(§1.3),91 = g2 throughout B. Thus, in practice we shall determine a single
1
function F(z) that is analytic on Au B and is given by F(z) = f(z) on A,
00
and F(z) = g(z) on B. For example, let f(z) = e- zt dt. Notice that
f (z) is analytic on ~ { z} > 0 because after evaluating the improper integral
we find that f(z) = liz is analytic on ~{z} > o. So we take g(z) = liz.
Since g(z) is analytic on C\{O}, g(z) becomes the analytic continuation of
f (z) from the right half-plane into the whole plane indented at the origin.
Similarly, the Laplace transform of cos at is analytic for ~{z} > 0; but the
function 2 Z 2 is its analytic continuation from the right half-plane into the
z +a
whole plane indented at the points ±ia.
x
a b
V'
f'
Fig. 1.4.3. Schwarz reflection principle.
SCHWARZ REFLECTION PRINCIPLE: If fez) is analytic on a region

containing a segment of the real axis and is real-valued on this segment, then
f (z) = f (z). This leads to the concept of analytic continuation across the real
axis when the function is known only on one side. Let f denote a segment
a < x < b, and let r be a Jordan arc joining a and b and lying in ~{z} > O.
Then C = fur is a contour that encloses a region D lying entirely in ~{z} > O.
If we reflect this region into the real axis, we get a region D' that lies in the
lower half-plane and is bounded by f and r' which is the reflection of r into
the real axis. Then the region f U D' is symmetric about the real axis. If fez)
is continuous on D U f, analytic on D, and real-valued on f, then the function
F(z) defined by
F(z) = {f(Z), if zED U f,
fez), if zED'
is analytic on D U f U D' (see Fig. 1.4.3). If f is analytic on one side of a curve
r and cannot be continued analytically across r, then r is called a natural
boundary for f. For example, the unit circle Izi = 1 is a natural boundary for
L
00
k
the function f (z) = Z2 •
n=O
Let S+ (S-) denote the upper (lower) half-,-plane ~{z} > 0 (~{ z} < 0),
respectively, or vice-versa, with L as their common boundary (i.e., the real axis,
Fig. 1.4.4). Let f(z) be a function defined at z E S+, and let it be connected
with the function 1* (z) defined in S- by the relation
j*(z) = f(z), (1.4.3)

i.e., f(z) and 1*(z) take conjugate complex values at points symmetric with
respect to the real axis, since the points z and z are reflections of each other
in L. We can rewrite the relation (1.4.3) as j*(z) = /(z), i.e., if f(z) =
u(x, y) + iv(x, y), then /(z) = u(x, -y) - iv(x, -y). If f(z) is regular (or
meromorphic) in S+, then 1* (z) = f (z) is regular (or meromorphic) in S- ,
and the relation (1.4.3) is symmetric as regards f and 1*, i.e.,
f(z) = f*(z), U*(z))* = f(z).
If f(z) is a rational function
f(z) - an z n + an-l zn-l + + ao

(1.4.4)
- bm zm + bm - 1 zm-l + + bo '
then 1* (z) = /( z) is obtained by simply replacing the coefficients by their
conjugate complex values. Let t be a real number, and assume that f(z) takes
a definite limit value f+ (t) as z --+ t from S+. Then 1*- (t) exists and
(1.4.5)
becausez --+ tfromS- asz --+ tfromS+,andhence,j*(z) = f(z) --+ f+(t).
Fig. 1.4.4.
We shall assume that f(z) is regular on S+, except possibly at infinity, and
continuous on L from the left. Define a sectionally regular function F(z) by
F(z) = {f(Z) for z E S+, (1.4.6)

j*(z) for z E S-.
Then, in view of (1.4.5)
(1.4.7)
These relations are useful when transforming the boundary conditions in any
boundary problem containing r(t) and f+(t), or f- (t) and f- (t) into those
involving F+(t) and F-(t).
In view of the Schwarz reflection principle, another property is that of

extending f(z). If ~{f*(t)} = 0 in any interval I of the real axis, then
the function 1* (z) is the analytic continuation of f (z) through the interval I
because 1*- (t) = r(t) on this interval.
A conformal equivalence between two regions D and G in the complex

plane is a one-to-one analytic function f with f(D) = G. Thus, I'(z) =1= 0 for
all z in D. Conversely, if f : D f--' C is analytic such that l' never vanishes,
then f is not necessarily a conformal equivalence. As an example, consider
f (z) = eZ • However, if l' (z) =1= 0 on D, then f is locally one-to-one and
conformal. The Riemann mapping theorem (Theorem 1.4.1) establishes the
conformal equivalence of two regions. If f is a conformal equivalence between
the open sets D and G, then
Area (G) = JL11'1 2

dz. (1.4.8)
Note that if f is an analytic function, then the Jacobian of f, regarded as a

mapping from JR2 into JR 2, is 11'1 2 . Hence, if G is a simply connected region,
9 : D f--' G is a Riemann map, and g(z) = L
anz n in D, then from (1.4.8)
n
(1.4.9)
In fact, since g'(z) = L nanzn- 1, then, for r < 1,

n
= L nmano,mrn+m-2 ei(n-m)6,
n,m
which converges uniformly in B. Since 1

2
" ei (n-m)9 dB = 0 for n i- m,
11 19'1 2
= ~ n 1an 1 21r
2 2
1 1
r
2n
-
1
dr
1
= 21r" n 2 lan l2 -
L..J 2n
n
This relation is useful in solving the minimum area problem in conformal

mapping.
1.5. Problems
x 4 _ y4 x4 + y4
PROBLEM 1. 5.1. Show that the function f (z)
x +y
=
x +y
3 3 +i 3 3'
f(O) = 0, z E C, is not differentiable at the origin, although it satisfies the

Cauchy-Riemann equations there. [Hint: ux(O,O) = 1 = vx(O, 0) = vy(O, 0),
uy(O,O) = -1, but by limit definition, I' (0) = (1 + i) /2 along the line y = x.]
PROBLEM 1.5.2. Prove that if a function w = f(z) maps conformally

a region D onto another region G, and if Zo is an isolated boundary point of
D, then Zo is a removable singularity or a simple pole of f(z). That is why
we assume the regions to be without isolated boundary points. (Goluzin, 1969,
p.205; Wen, 1992, p.95.)
PROBLEM 1.5.3. Show that the chain of mappings II : z f-7 z - 1,

iz : z 2z, h :
f-7 z f-7 Z + 3i/2 maps the circle Iz - 11 = 1 onto the circle
Iw - 3i/21 = 2.
PROBLEM 1.5.4. Let w = f(z) be an analytic function, regular in a
simply connected region D. Let 0: be an interior point of D and a simple zero
of f(z) such that f(o:) = 0 and 1'(0:) i- O. Moreover, let Zo E D be a first
rough approximation of the zero n, i.e., Zo is close to 0:. Show that
1.5. PROBLEMS 39
= exp { - f(zo) [ drl(W)]

dw
}
'
w=f(zo)
where z = f-l(w) denotes the inverse function of fez) and the exponen-
tial function operates symbolically on the differential symbol. (Blaskett and
Scherdtfeger, 1945-46, p.266.)
PROBLEM 1.5.5. Prove the Schwarz lemma: If fez) is analytic on the

disk Izl < R such that f(O) = 0, and If(z)1 :'S M, where M > 0 is a constant,
then If (z) I :'S ~ IzI, 11' (0) I :'S ~ on this disk, where the equality holds only
M
for f (z) = e
iQ
Ii z, Q: real. (Wen, 1992, p.26-27.)
PROBLEM 1.5.6. Let S denote the family of univalent functions analytic

00
on the unit disk Izl < 1, which are of the form fez) = z + Lan zn. Show
n=2
that the coefficients an satisfy the inequality Ian I :'S n for n = 2,3, ... , and
this bound, known as de Branges estimate (see de Branges, 1985), is sharp for
the Koebe function fez) = ( Z )2' (Wen, 1992, p.54.)
l-z
z
PROBLEM 1.5.7. Ifx = auand
y = a f (v), where z = x + i y and
w = u + i v, then the meridians
and parallels of the earth will be
mapped onto the coordinate lines
in the (x, y)-plane. The equator y
will be mapped onto the x-axis if

f(O) = O. To make a rhumb line on
the sphere, which cuts all the meri-
dians at the same angle Q: map onto
x
a straight line in the z-plane cutting
the lines x =const at the same angle Q:, we must have
cot Q: = dvd
cosv u
= dd
y
x
= l' ~) dV,
u
which yields f (v) =
Jo
r sec v dv =
log tan (~ + ~). The resulting mapping is called Mercator's projection,
given by = z a[u+ log tan (~ + ~) ]. Show that this mapping is conformal

in the sense that angles are preserved and lengths are multiplied by a factor
depending on the latitude. Mercator's projection can also be written as z =
a [u + gd- 1v], where gd denotes the Gudermannian of u. (Franklin, 1944,

p.140-141.)
PROBLEM 1.5.8. Show that if a region whose boundary is a piecewise

analytic Jordan contour is mapped conformally onto the unit disk, the mapping
is continuous on the boundary. (Nevanlinna and Paatero, 1969, pp.339-340.)
REFERENCES USED: Ahlfors (1966), Betz (1964), Blaskett and Scherdtfeger

(1945-46), Boas (1987), Caratheodory (1969), Carrier, Krook and Pearson (1966),
Franklin (1944), Gaier (1964), Goluzin (1969), Jeffrey (1992), Kober (1945-48),
Lawrentjew and Schabat (1965), Marsden and Hoffman (1987), Muskhelishvili (1948,
1992), Nehari (1952), Wen (1992).
Chapter 2
Conformal Mappings
The central problem in the theory of conformal mapping is determining a

function f which maps a given region Dee conformally onto another
region G c Co The function f does not always exists, and it is not always
uniquely determined. The Riemann mapping theorem (§1.4) guarantees the
existence and uniqueness of a conformal map of D onto the unit disk U under
certain specific conditions. Besides some elementary mappings we shall study
linear, bilinear, and Schwarz-Christoffel transformations.
2.1. Polynomials
The general complex polynomial Pn(z) of degree n is an entire function for

all z E C and has derivatives of all orders such that p~k) (z) = 0 for k > n.
If Pn(z) has a zero of multiplicity p at zo, then Pn(z) = (z - zo)P g(z),
where g(z) is a polynomial of degree n - p, g(zo) 1= O. The special cases
PI (z) = az+b, where a and bare complex constants, represent a magnification
(or dilatation) by lal and a rotation by arga, followed by a translation by b.
A rational function, as the quotient of two complex polynomials Pn (z) and
Qm(z) of degree nand m, respectively, n :::; m -1, with no common factors,
is analytic for all z that is not a zero of Qm (z). If the polynomial Qm (z) has a
zero Zo of multiplicity p, then the partial fraction development of the rational
function corresponding to this zero has the form
(2.1.1)
41

42 CHAPTER 2. CONFORMAL MAPPINGS
Under the mapping w = Pn(z), n > 1, there are at most n points wo in the
extended w-plane with fewer than n distinct inverse images. In fact, the point
w = 00 has just one inverse image. If Wo =1= 00 has fewer than n distinct inverse
images, then the equation
(2.1.2)
where Wo = Pn(zo), must have multiple roots which satisfy the equation
P'(z) = O. But since the polynomial P'(z) is of degree at most (n - 1) and
has at most n - 1 distinct zeros z~, 1 ~ v ~ n - 1. Hence Eq (2.1.2) can have
a multiple root only at the numbers Pn(zD, Pn(z~), ... , Pn (00), at most n of
which are distinct.
Let Zo be a root of multiplicity k > 1 ofEq (2.1.2). Then under the mapping
w = Pn (z), every angle with its vertex at Zo is enlarged k-times, whereas every
angle with vertex at z = 00 is enlarged n-times.
Now we shall study the mapping
w=(z-a)n, n>1. (2.1.3)
This function maps the extended z-plane into the extended w-plane such that
every point w has n distinct inverse images, except at the two points w = aand
w = 00, for which the n inverse images coalesce into a single point z = a and
z = 00, respectively. For w =1= a and w =1= 00, the n inverse images of w are
obtained by solving (2.1.3) for z, which yields
nli:"::l
z = a + y!W = a + V Iwl'
(
cos
arg{w}
n + t.sm
. arg{w})
n . (2.1.4)
Thus, the n distinct points z, defined by (2.1.4), are situated at the vertices
of a regular n-gon with center at z = a. The mapping (2.1.3) is conformal
at all points except z = a and z = 00, and every angle with the vertex at
one of these two points is enlarged n-times. In fact, since Iwl = Iz - al n ,
arg{ w} = n arg{ z - a}, the circle Iz - al = r is mapped into the circle
Iwl = rn. Also, as the point z traverses once around the circle Iz - al = r,
the image point w traverses the circle Iwl = r n n-times in the same direction,
since arg{ w } increases continuously by 2mr. Moreover, the ray arg{ z - a} =
eo + 2k7r (k an integer) in the z-plane going from a to 00 is mapped into the
ray arg{ w} = neo + 2m7l' (m an integer) in the w-plane going from a to 00.
Let D denote the region {z : eo + 2k7r < arg{ z - e

a} < 1 + 2k7r}, where
a
k is an integer and < e1 - eo ~ 27l'ln. This region D is called the interior of
2.1. POLYNOMIALS 43
the angle (fh - 80 ) which is bounded by the two rays arg{ z - a} = 80 + 2k7r
and arg{z - a} = 81 + 2k7r, k = 0, ±l, ±2, .... Then the function (2.1.3)
maps the region D onto the region G = {n8 0 +2m1r < arg{w} < n8 1 +2m1r,
°
m = 0, ± 1, ±2, ... , which is the interior of the angle n( 81 - 80 ) with vertex
at w = (Fig 2.1.1).
v
y
-::t- x _ _ _----"l'----_ _..-;u

o
Fig. 2.1.1. The mapping w = (z - a)n.
-+---t~I7-t---t- U
--.o;;o:+--il---+--t--+X
z-plane
w-plane
Fig. 2.1.2. The mapping w = Z2.
The mapping function (2.1.3) produces a conformal and one-to-one map

of the interior of an angle onto the interior of another angle which is n-times
wider. However, it does not map every circle into a circle (in the extended
sense). For example, for n = 2, a = 0, the map w = z2 maps every vertical
straight line z = b+it, where b =I=- 0 isreal and -00 < t < 00, onto the parabola
v 2 = 4b 2 (b 2 - u) which opens to the left, and every horizontal line z = t + ie,
where e =I=- 0 is real and -00 < t < 00, onto the parabola v 2 = 4e2 (e 2 + u)
which opens to the right (Fig. 2.1.2). This mapping is conformal except at
z = 0, but it is not one-to-one, since every point in the w-plane except w = 0
and w = 00 has two inverse images.
The multiple-valued function w = vrz has the inverse z = w n which

has been discussed above with reverse roles of wand z. Under the mapping
z = wn , the interiors G 1, ... ,Gn of the n angles, each equal to 27l'ln radians,
formed by the n rays emanating from the point w = 0 lead to the n single-
valued branches ( y'z) 1 ' . .. ,( y'z) n of the function w = vrz,all defined on
the region D. These branches, which have nonzero derivatives on D, map D
one-to-one and continuously onto Gk> k = 1, ... ,n (Fig. 2.1.3).
v
A Y
L D z - 'l.JZ
n=6
11410
_ _~"""""
o
--l>X
------::*"':-----u
Fig. 2.1.3. The map w = vrz.

The inverse image in the z-plane of the regions G k> whose boundaries are
marked by solid lines, is the single ray L emanating from z = 0 and inclined at an
angle n 1>0' The manner in which a branch ( vz) k changes into the next branch
( vz) k+l can be explained by letting a point Zo =I=- 0 in D make a complete
circle with center at z = O. We choose the value of vrz that is associated with
the branch ( y'z) k and represented by the value Wo = ~ eiOo / n E Gk.
Then, as the point Zo moves continuously along the circle Izl = Izol in the
positive direction, the value of w = ~ eiO / n varies continuously with ()
such that as the point z returns to its original value zo, the value of w goes to
2.2. BILINEAR TRANSFORMATIONS 45
the value WI = ~ ei (Oo+27l')/n, where WI is the value associated with the

branch ( v'Z) k+I on the adjacent region Gk+l, where G k n Gk+l = 0 for
k = 1, ... , n. Proceeding in this manner through n windings around z = 0,
the n branches ( v'Z) k undergo the following chain of transformations:
( v'Z) k ~ (v'Z) k+l ' ( viz) k+l ~ (v'Z) k+2' ... ,

(v'Z)n ~ (v'Z)I' (v'Z)I ~ (v'Z)2' ... (v'Z)k-I ~ (v'Z)k'
The points z = 0 and z = 00 are the algebraic branch points for the mapping
W = y'z.
2.2. Bilinear Transformations
The bilinear (linear-fractional, or Mobius) transformation has the form
W = f (z) = az + b, (2.2.1)
cz +d
where a, b, c, d are complex constants such that ad - be i- 0 (otherwise the

function f(z) would be identically constant). If c = 0 and d = 1, or if a = 0,
d = 0 and b = c, then the function (2.2.1) reduces to a linear transformation
W = az + b, or an inversion W = ~, respectively. The transformation (2.2.1)
z
can also be written as
a be - ad
W--+----,---",.. (2.2.2)
- c c(cz + d)'
that can be viewed as composed of the following three successive functions:
1 a be-ad
ZI = cz + d, Z2 =- , W = -C + C
Z2.
ZI
This shows that the mapping (2.2.1) is a linear transformation, followed by an

inversion followed by another linear transformation. The bilinear transforma-
tion (2.2.1) maps the extended z-plane conformally onto the extended w-plane
such that the pole at z = -die is mapped into the point w = 00. The inverse
transformation
z = r
1 (w) = b - dw (2.2.3)
-a+cw
is also bilinear defined on the extended w-plane, and maps it conformally onto
the extended z-plane such that the pole at w = alc is mapped into the point
I ad-bc -1 I -ad+bc
z = 00. Note that f (z) = ( d)2 f:. 0; also [f (w)] = ( )2 f:. o.
cz + cw - a
A bilinear transformation carries circles into circles (in the extended sense*).
A cross-ratio between four distinct finite points Z1, Z2, Z3, Z4 is defined by
(2.2.4)
If Z2, Z3, or Z4 is a point at infinity, then (2.2.4) reduces to
or
respectively. The cross-ratio (z, Z1, Z2, Z3) is invariant under bilinear transfor-
mations.
THEOREM 2.2.1. A bilinear transformation is uniquely defined by a

correspondence of the cross-ratios
(2.2.5)
which maps any three distinct points Z1, Z2, Z3 in the extended z-plane
into three prescribed points wI, W2, W3 in the extended w-plane.
The cross-ratio (z, Z1, Z2, Z3) is the image of z under a bilinear transforma-
tion that maps three distinct points ZI, Z2, Z3 into 0, 1, 00.
The points z and z* are said to be symmetric with respect to a circle C (in
the extended sense) through three distinct points Z1, Z2, Z3 iff
(2.2.6)
The mapping that carries z into z* is called a reflection with respect to C. Two
reflections obviously yield a bilinear transformation.
*There is no distinction between circles and straight lines in the theory of bilinear
transformations.
If C is a straight line, then we choose Z3 = 00, and the condition for

symmetry (2.2.6) gives
Z* - Zl
(2.2.7)
Zl - Zz
Let Z2 be any finite point on the line C. Then, since Iz* - zll = Iz - zll, the
points Z and z* are equidistant from the line C. Moreover, since 8'{ z* - Zl } =
Zl - Zz
-8'{ Z - Zl }, the line C is the perpendicular bisector of the line segment
Zl - Zz
joining Z and z*. If C is the circle Iz - al = R, then
(Z,Zl,Z2,Z3) = (z - a,Zl - a,Z2 - a,Z3 - a)

R2 R2 R2 )
= ( z-a,--,--,--
Zl - a Z2 - a Z3 - a
2
= (_ R _ ,Zl _ a, Zz - a, Z3 -
z-a
a)
2
= (_R _ + a, Zl, ZZ,Z3) .
z-a
RZ
Hence, in view of (2.2.6), we find that the points Z and z* = - - + a are
z-a
symmetric, i.e.,
(Z* - a)(z - a) = R2 . (2.2.8)
Note that Iz* - allz - al = R Z; also, since z* - a > 0, the points Z and z*
z-a
are on the same ray from the point a (Fig 2.2.1). Note that the point symmetric
to a is 00.
z*
Fig. 2.2.1. Symmetry with respect to a circle.

A generalization of this result is as follows: If r denotes an analytic Jordan

curve with parametric equation z = "(8), 81 < 8 < 82, then for any point z
sufficiently close to r, the point
z* =" ("r- 1(z))

l (2.2.9)
defines a symmetric point of z with respect to r (Sansone and Gerretsen, 1969,

p.103; Papamichael, Warby and Hough, 1986). Some examples are as follows:
(i) If r is the circle x 2 + y2 = a 2 /9, then z* = a2 /9z = z*.
(ii) If r is the ellipse (x + ~/2)2 + y2 = 1, then
a
* a (a 2 + l)(z + a/2) + 2iaJa l -1 - (z + a/2)

z = --
2
+ a2 - a
.
(iii) Ifr l is a cardioid defined by z = "(8) = (~ + cos ~) eis , -'Fr < 8 ~ 'Fr,
then from (2.2.9) we cannot write an explicit expression for symmetric points
with respect to r 1. However, for any real t
,,(±it) = (~+COSh~) e'f

t
defines two real symmetric points with respect to r 1, provided the parameter t
satisfies the equation ,,( it) ,,(-it) = a2 , i.e., ~ + cosh ~ = a which has the
roots
r-
l
t = 2 cosh- (a -1/2) = ±2 log p, (2.2.10)
where p = a- ~ + J(a - ~ 1.
a= 1.55 a= 1.8
Fig. 2.2.2. Cardioid inside a Circle.

(iv) If a doubly connected region n is bounded outside by the circle f z = {z :

Izl = a, a > 1.5} and inside by the cardioid f 1 defined above in (iii) (see
Fig. 2.2.2), then it follows from (2.2.10) that there is one pair of real common
symmetric points (1 E Int(f 1) and (z E Ext (f z ) such that (1 = a/pz and
(z = apz, where p is defined in (2.2.10).
The symmetry principle states that if a bilinear transformation maps a

circle 0 1 onto a circle Oz, then it maps any pair of symmetric points with
respect to 01 into a pair of symmetric points with respect to Oz. This means
that bilinear transformations preserve symmetry.
A practical application of the symmetry principle is finding bilinear trans-

formations which map a circle 0 1 onto a circle Oz. We already know that
the transformation (2.2.5) can always be determined by requiring that the three
points Zl, Zz, Z3 E 0 1 map onto three points Wi> wz, w3 E Oz. But a bilinear
transformation is also determined if a point Zl E 0 1 should map into a point
Wi E Oz and a point Zz ~ 0 1 should map into a point Wz ~ Oz. Then, by the
symmetry principle, the point Zz which is symmetric to Zz with respect to 0 1
is mapped into the point point Wz which is symmetric to Wz with respect to O2,
and then the desired bilinear transformation is given by
(2.2.11)
----1-->- X
o
Fig. 2.2.3.
Let U+ denote the region Izl < 1, U- the region jzj > 1, and let 0 =
{Izl = I} be their common boundary (Fig. 2.2.3). Let j(z) be a function
defined on U+. Then this function can be related to the function 1* (z) defined
in U- in the same manner as in (1.4.3) for half-planes, except that now the
conjugate complex points are replaced by points inverse with respect to the
circle C according to the relation (2.2.8). Thus,
I*(z) = 1 (~) = f(~). (2.2.12)
This relation is symmetrical, i.e.,
I(z) = f* (~). (/*(z))* = I(z). (2.2.13)
If 1(z) is regular or meromorphic in U+, then 1* (z) is regular or meromorphic

in U-. Also, if I(z) is a rational function defined by (1.4.4), then
I*(z) = an z-n + an-l z-n+l + + 0,0 . (2.2.14)

bm z-m + b-m+l zm-l + + bo
Moreover, if I(z) has the power series expansion
L
00
I(z) = ak zk z E U+,
k=-oo
then
L
00
I*(z) = ilk [k, z E U-. (2.2.15)

k=-oo
If I(z) has a zero (pole) of order k at z = 00 (z = 0), so does f*(z). Let us

assume that I(z) approaches a definite limit value I+(t) as z ........ tEe from
U+. Then f*-(t) exists and
(2.2.16)
because liz . . . . t from U+ as z ........ t from U-, and hence I*(z) = f (~) =
1 (~) . . . . I+(t). If I(z) is regular in U+ except possibly at infinity and
continuous on C from the left, then let F(z) be sectionally regular and be
defined by
I(z) for z E U+,
F (z) = { (2.2.17)
I*(z) for z E U-.
2.3. SCHWARZ-CHRISTOFFEL TRANSFORMATIONS 51
Then, F*(z) = F(z), and, as in (1.4.7),
(2.2.18)
Moreover, in view of the Schwarz reflection principle, if~{f+ (t)} = 0 on some

part of the circle C, then f* (z) is the analytic continuation of f (z) through this
part of C.
2.3. Schwarz-Christoffel Transformations
The Schwarz-Christoffel formula helps solve an important class of boundary

value problems that involve regions with polygonal boundaries. This formula
has an integral representation and is used to map conformally the upper half-
plane ~{z} > 0 or the unit disk U in the z-plane onto the interior of a given
n
n-gon (n 2 2) in the w-plane. Let r = U r~ in the w-plane with vertices Wi
i=l
n
andexteriorangles7rai,lail < 1,andL:an = 2 fori = 1, ... ,n(Fig. 2.3.1).
i=l
Then the conformal map from the upper half-plane D = {z : ~{z} > O} onto
G, which is the interior of r, is given by
l
z n
w=f(z)=A+B II((-xi)-aid(, (2.3.1)
ZO i=l
D
Tj r2
XI X2 r XJ ..• '-----------_u
z-plane IV-plane
Fig. 2.3.1. Schwarz-Christoffel transformation.

where A and B are constants that determine the size and position of the polygon
r, and integration is carried out along any path in D that joins Zo E D to z,
and the principal branch is used for the multiple-valued function (( - Xi) -<>i
in the integrand such that 0 < arg{( - xd < Jr, i = 1, ... ,n, for ~{z} > o.
These branches are a direct analytic continuation into the upper half-plane of
the real-valued functions (x - Xi)-<>i, where x > Xi. Then the integral (2.3.1)
is a single-valued analytic function in the upper half-plane ~{z} > 0, and the
points Xi lying on the x-axis are the singularities of the Schwarz-Christoffel
integral (2.3.1). This function w = j(z) defines a conformal mapping of D
onto G provided the points Xi are suitably chosen.
Thus, if all the numbers Xi are finite, then the function j(z) defined by
(2.3.1) remains bounded in the neighborhood of the singularities Xi. To see
that the Schwarz-Christoffel integral (2.3.1) remains bounded as z ~ 00, we
rewrite this integral as
w = j(z) = A +B l ZQ
z
(-(<>l+···+ a n)
(
1 _ ~1
)
-<>1
. ..
(
1 _ ~n
)-a n d(
l g
z
1 n ( )-<>i
=a +B ZQ (2 1- ~i d(,
(2.3.2)
which shows that the integral is convergent as z ~ 00. Hence, the Schwarz-
Christoffel integral (2.3.1) defines a univalent function of z in D that maps
D conformally onto the bounded polygon G in the w-plane. If G is convex,
then ai > o. Thus, there are (n - 1) independent numbers ai, and (n - 3)
independent numbers Xi.
In order to see the structure of the boundary r = U~=l r i of the polygon

G, consider the derivative of the Schwarz-Christoffel integral (2.3.1) , which
is
j'(z) = B(z - Xi)-<>l ... (z - xn)-a n .
Notice that l' =I- 0 everywhere in ~{z} ~ 0 except at the singularities Xi
where it vanishes or becomes unbounded. As z varies along every interval
Xk < X < Xk+1> k = 1, ... ,n - 1, the value of arg{f'(z)} does not change,
SInce
_} {-Jra k ' ifx<xk,
arg {( X - Xk ) <>k =
0, ifx>xk.
Geometrically, arg{f' (z)} determines the size of the angle through which the
tangent to a Jordan curve passing through Xk must be rotated in order to obtain
the tangent of the image of this curve at the point Wk = f(Xk). Hence the
segments Xk < x < Xk+l, k = 1, ... , n - 1, of the real axis are mapped by
f (z) into rectilinear segments in the w-plane. The points x k ofthe real axis are
transformed into the vertices Wk in the w-plane, where the polygon r is made
up of the polygonal lines through WI, W2, ... ,Wn with straight line segments
as sides. As the point z traverses the entire real x-axis in the positive sense,
the corresponding point W travels completely through the polygonal lines of r
which, in general, can have a self-intersecting point.
The size of the angle between adjacent segments of the polygon r can be
determined as follows: Consider the variation of arg{f' (z)} as z passes through
the point Xi in the positive direction (Fig. 2.3.2). Then the angle between the
vectors ~ and WiWi+i is equal to 11'ai· For Ili < 1, where Ili = 1 - ai,
the transition from the vector Wi-I w~ to the direction of the vector WiWi+i
occurs in the positive sense (Fig. 2.3.2a), whereas for Ili > 1 it occurs in the
negative sense (Fig. 2.3.2b), although the angle of transition in the positive
sense in both cases from the direction of the vector Wi-I w~ to the direction
of the vector WiWi+i is 11'ai. If a polygonal line does not intersect itself, it
becomes the boundary of a closed polygon. Also the sum of all interior angles
n
ofthis closed polygon is equal to L 11'Ili =L n11' (1 - ai) = (n - 2)11'.
i=1 i=1
Wi+l
Wi_l Wi+l
Wi-l
(a) (b)
Fig. 2.3.2.
For practical purposes, while constructing a conformal map of D onto G

one can specify any three points Xi, Xj, Xk of the real axis that go into the three
prescribed vertices Wi, Wj, Wk of the polygon r.
When one of the points Xi, say X n , coincides with the point at infinity, the
vertices of the polygon r correspond to the points x~, ... ,X~_l' 00, and the
formula (2.3.1) becomes
w=f(z)=A+B' 1 II((-x~)-ai
ZO
z n-l
i=l
d(. (2.3.3)
Note that formula (2.3.3) is similar to (2.3.1) except that the term corresponding
to the point X n = 00 has been dropped. The inverse Schwarz-Christoffel
transformation is given by
l
n
II (( -
w
Z = f(w) = Co + C Wi)-JLi d(. (2.3.4)
wo i=l
CASE STUDY 2.3.1 Tomaptheupperhalf-plane~{z}> oonto a triangle

A I A 2 A 3 with interior angles J.L17r, J.L27r and J.L37r, respectively, such that the
vertex Al corresponds to Xl = 0, A 2 to X2 = 1 and A 3 to X3 = 00, we get, by
using (2.3.1), the mapping function
(2.3.5)
~l 72 (+ DO)
0 I X3
z-plane
v
IV)
w-plane lV-plane
Fig. 2.3.3.
If the triangle is isosceles such that the vertices AI, A 2 , A 3 are at W = 0,

W = 1, and W = i, respectively (Fig 2.3.3), then from (2.3.5) the transformation
becomes
W = A +B lCz
1 2
/ (1 - ()-3/4 d(. (2.3.6)
Since W = 0 corresponds to z = 0, we get A = 0, and since W = 1 corresponds

to z = 1, we find from (2.3.6) that
B= 1 (2.3.7)
fo1 (-1/2 (1 - ()-3/4 d( .
For an equilateral triangle of side b (see Fig. 2.3.3), let Xl = -1 correspond to

the vertex WI, x2 = 1 to W = 0, and X3 = 00 to W3 = b. Since 0::1 = 0::2 =
0::3 = 2/3, the transformation is given by
(2.3.8)
When z = -1, we set ( = x. Then for -1 < X < 1, we have X + 1 > 0 and
arg{x + I} = 0, but Ix - 11 = 1 - x, and arg{x - I} = 1r. Thus,
WI =
/1 +
-1
(x 1)-2/3 e- 2i1r / 3 (1 - x)-2/3 dx = e i1r / 3 /1
-1
dx
(1 - x 2 )2/3
= 2e i1r / 3 B (~2' ~)
3
= be i1r / 3
'
(2'11)3 ' . .
(2.3.9)
where b = 2B and B denotes the beta functIOn of Its arguments. For
the vertex W3 note that it is on the positive u-axis, i.e.,
00 /00 dX
/
_ -2/3 -2/3 _
W3 - 1 (x + 11) (x - 1) dx - 1 (x2 _ 1)2/3'
But W3 is also defined by (2.3.8) when z goes to 00 along the negative u-axis,
i.e.,
W3 = Joo (x + 11)-2/3(x - 1)-2/3 dx
= WI + e- 4i1r / 3 /-00 [Ix + lllx -

-1
11]-2/3 dx using (2.3.9)
= b ei1r /3 + e -i1r/3/00 (x 1
dx
2 - 1)2/3
= b ei1r /3 + W3 e -i1r /3,
which yields W3 = b. A numerical evaluation of the improper integral in (2.3.7)

is given in Case Study 3.3.1. •
-Ilk
X3
,
A3
x4
.
-1 J V
B
xI
Az
Xz
Ilk
. X
-a+i b a+ib
A 4
---:..L-- -+ A
..L.......!I U
-G o G
Fig. 2.3.4.
CASE STUDY 2.3.2 To map the upper half-plane ~{z} > 0 onto a rec-
tangle Al A 2 A 3 A 4 with vertices at the points w = ±a, ±a + ib, where 2a
and b are the width and the height of the rectangle (Fig. 2.3.4), note that in the
formula (2.3.1) with n = 4, only three of the four points Xl> X2, X3, X4 may be
chosen arbitrarily. Since the rectangle is symmetric about the v-axis, we can
choose the x's symmetrically. Thus, for the right-half rectangle OA I A 2 B let
w = 0, a, ib correspond to z = 0,1,00, respectively, and let the preimage of
A 2 be z = 11k, 0 < k < 1. Similarly, for the left-half rectangle OBA 3A 4 let
w = 0, -a, ib correspond to z = 0, -1, -00, respectively, with the preimage
of A 3 as -11k. Then the formula (2.3.1) yields
J
z 4
w=A+B II((-Xi)-Ct;d(
i=1
l
z
d(
-B
J(=l V(TI J( -11k J( + 11k
l
- 0
l
z z
-B d( -B d(
- 0 ~J(2-1Ik2- 0 ~Jl-k2(2'
(2.3.10)
where A = 0 since z = 0 goes into w = 0, and B is an arbitrary constant.
Now, to evaluate the constants Band k, note that since z = 1 goes into w = a,
we get from (2.3.10)
1
1 d(
a-B (2.3.11)
- 0 J1=(2 }1 - k 2(2 .
Also, since z = 11k goes into w = a + ib, using (2.3.11) we find that
1
1/ k d(
a+ib =B
o J1=(2 }1 - k 2 (2
f1 j1/k) d(
= B ( Jo + 1 J1=(2 }1 - k 2 (2
1/k
j
d(
= a+iB
1 J1=(2 }1 - k 2(2 '
where B is an arbitrary complex constant. Thus, with B =1
1/k
j
d(
b- (2.3.12)
- 1 J1=(2 }1 - k 2 (2 .
Hence, we can determine Band k from (2.3.11) and (2.3.12) if a and b are
prescribed. But if k is preassigned and we take B = 1, then the values of a
and b are determined from (2.3.11) and (2.3.12). Then the mapping function
becomes
l
z
d(
w = ~' (2.3.13)
o V 1 - (2 k 2 (2 VI -
which is known as the elliptic integral of the first kind. The inverse function
is called the elliptic sine function z = sn w = sn (w; k) which is a Jacobian
elliptic function. The function sn is a 2a- and 2ib-periodic function. For more
material on the function sn and other Jacobian elliptic functions, see Whittaker
and Watson (1927, p.491 ff.), Nehari (1952, p.280 ff.), or Andersen et at.
(1962, p.159ff), and see Case Study 3.3.2 for numerical evaluation of improper
integrals in (2.3.13). If we denote the value of a, given by (2.3.11), by K(k),
where 0 < k < 1, then the ratio a I b of the sides of the rectangle is given by
~ _ 2 K(k)
(2.3.14)
b- K ( v'"f=k2)'
An evaluation of K(k) is accomplished by the Landen transformation: Set
k1 = v'"f=k2, and k' = 1 - kk
1
. Then
1+ 1
K(k) = ~k K(k'). (2.3.15)

1+ 1
In fact, by separating the integral (2.3.11) into two parts, we get
o~ T
_1_ K (1 - k
~ 1, and
I - 1)
1 - 1+ k1 + 1 k1 .
2 (1 -+
Hence,
It+h=2 It =--k- K k1 ) ,
--k-
1+ 1 1 1
which proves (2.3.15). Since
after n applications of the Landen transformation we find that the quantities

k,(n) ---> 0 very rapdly as n ---> 00. This means that we can use the Landen
transformation to evaluate the integral in (2.3.11) step-by-step with smaller
and smaller values of k, so that after a finite number of steps the value of the
integral (2.3.11) becomes equal to sin -1 (1) = 1r /2. However, for values of k
closer to 1, the convergence becomes very slow. In that case we can use the
inverse Landen transformation
K(k') = ~ (1 + k1 ) K(k), (2.3.16)
where k 1
l-k'
= --k-' k=
R 1 - k21
2v'i2 and then we can evaluate K (k)
= --k-'
1+ ' 1+ '
approximately with sufficient accuracy from the asymptotic formula
4 1 16
K(k) ~ In - =- In - -2 for k close to 1. (2.3.17)
k1 2 1- k
Further, if we set ( = sin ¢ in (2.3.11), then
K(k) =
io
1r
/2
\11 -
d¢
k2 sin 2 ¢
=--
2
1 + k1
i
0
1r
/2
VI -
d¢
k'2 sin 2 ¢'
or
i
1r
/
2 dA-.
K(k) =
Jcos
'I'
2 2
o ¢ + kr sin ¢
2 r/ 2
d¢
= 1+k 1 JO JCOS 2 ¢+(1+kI 2 )sin2 ¢
do
Let k 1 = - , where Co = 1 and do = k 1 . Then
CO
1 _ k'2 = 1 _ (~) 2 =
1 + k1
(2 +y'k; ) (2 ~) (d
1 k1
2=
Co + do
2= 1)
Cl
2,
where d 1 = .Jco do, and Cl = (co + do)/2, which yields
r/ 2
d¢
- J0 -Vr~=r=c=o=s2=¢===+=d=i=s=i=n=2=¢
Repeating this process n-times, we get
(2.3.18)
where Cn and dn are determined recursively from

Hence, as n -> 00, the sequences {Cj} and {d j } for j = 0, 1, 2, ... , and
CO > do, converge to the same limit, i.e.,
lim en = lim = M(k), (2.3.19)

n ....... oo n-+oo
which yields
7r
K(k) = 2M(k)"
Proof of this result is given in Andersen et al. (1962, p.163). Hence, from
(2.3.14) the ratio between the sides of the rectangle is given by
(2.3.20)
where M (kd is obtained in the same manner as (2.3.19) by applying the above
recursion to k l = ~. For more on elliptic integrals and Jacobian elliptic
functions, see Phillips (1957, Ch. 1 and 2). •
CASE STUDY 2.3.3 To map the upper half-plane ~{z} > 0 onto an
arbitrary quadrilateral A l A 2 A 3 A 4 with interior angles /-L17r, /-L27r, /-L37r, and
/-L47r, respectively, such that the angle /-L17r at Al is the smallest and the ratio of
the side A 4 A l to the side A l A 2 is.A. Without loss of generality, let the vertices
AI, A 2 , A 3 , A 4 correspond to the points Xl = -1, X2 = 1, X3 = k, X4 = 3.
Then, by (2.3.1), the transformation is given by
w = f(z) = A +B l z
(( + 1)-<:<1(( _1)-<:<2(( - k)-<:<3(( - 3)-<:<4 de.
(2.3.21 )
Newton's method for numerical evaluation of improper integrals in (2.3.21) and
an approximate value of k is given in §3.2.•
CASE STUDY 2.3.4 To map a horizontal parallel strip with a rectilinear

horizontal cut onto the upper half-plane (Fig. 2.3.5), note that the given strip is
equivalent to a quadrilateral with vertices at AI, A 2 and A 3 , all at infinity, and
A 4 at the origin. For the vertices at AI, A 2 , A 3 we have 01 = 02 = 03 = 1,
and for A 4 we have 04 = -1. We choose Xl = 1, X2 = 00, X4 = 0, and
let w = 0 correspond to z = O. Then A in (2.3.1) is zero. With this choice
the point A 3 would correspond to some point X3 = -k on the negative x-axis
(k > 0). Then from (2.3.1) the required transformation with proper choice of
principal values for the integrand is given by
t (d(
w = B Jo (1 - ()(k + ()
= B Jto 1 + k 1 (1 1_ ( - k
k) d(
+(
(2.3.22)
= B [IOg(l - z) + k log (1 + f)] ,

where B is an arbitrary constant. To determine Band k if a and b are given,
e
let z = 1 - € ei8 , -1r ::; < 0, € > 0 and small, and define a half-circle GE •
Then, as z moves along GE , its image point w varies from the ray A 4 Al to the
line Al A 2 • Thus, for the end point of the image curve the difference Aw in
the values of won A 4 Al and Al A 2 is
Aw = -ia + 0(1). (2.3.23)
v
A3 --; ia _
A A4
3========~.--:!.--------- U
AI a
A1-----------11----------
-ib
Fig. 2.3.5.
But the difference Aw given by the right side of (2.3.22) is given by
Aw = B (-i1r) + 0(1). (2.3.24)
Hence, comparing (2.3.23) and (2.3.24), we find that B = a/1r. Similarly, as

z moves on the other half-circle G~ with center at z = -k, i.e., z + k = € ei8 ,
the value of w varies from its value on A 2 A 3 to A 3 A 4 , and in this case the
difference is t::.w = -ib+o(l), which from (2.3.22) is equal to -Bki1l'+o(l).

This yields b = Bk7l' = ka, or k = b/a. Thus, the required mapping is given
by
a
w = :; log(l - z) a)Z .•
+ :;b log ( 1 + b (2.3.25)
v
f---------"...c
CASE STUDY 2.3.5 Map the semi-infinite strip u > 0,0 < v < a, as the
e
limit of the triangle OC A as ----t 1l' /2 onto the upper half-plane ~{z} > 0
(Fig. 2.3.6), such that z = -1, 1 correspond to w = ia,O, respectively. The
transformation (2.3.1) yields
w = f(z) = A + 0-+7f
hm/2 B
. J2 (( + 1)1/2((d( - 1 )1-0/ 7f
2 d( (2.3.26)
=B J(2=l = B cosh- 1
Z,
/ 1 (2 - 1
where A = 0 since w = 0 corresponds to z = 1. From (2.3.26) we have

w/ B = cosh- 1 z, which gives z = cosh(w/ B), and slncew = iacorresponds
to z = -1 = cosh i1l', we get B = a/1l'. Hence, from (2.3.26)
a -1
W = - cosh z,
1l'
and the required transformation is given by
1l'W
Z = cosh- .• (2.3.27)
a
CASE STUDY 2.3.6 To map the region in the z-plane, shown in Fig. 2.3.7,
onto the upper half-plane ~{w} > 0, formula (2.3.4) gives
fW d(
z=G Jl (l-k((_l)k' O<k<1. (2.3.28)
w-plane
z- plane
Fig. 2.3.7.
To determine the constant G, we integrate along the half-circle w = R eiO ,

o < () < 7r. Since the residue of the integrand at ( = 00 is -1, we find
from (2.3.28), as R --+ 00, that ai = -Gin, i.e., G = -a/n (see figure
on the right). The integral in (2.3.28) t1
can be evaluated in terms
of known functions if k is
rational. Let k = p/q, p < q,
Set ((_1)
where p, q E IR+.
-(-
l/q
= t.
qt q - 1
Then, d( = 2 dt,
(1 - t q )
and (2.3.28) becomes
z
aq
=-
n
it0
t- p + q - 1
tq - 1
dt. (2.3.29)
Now, the q poles ofthe integrand in (2.3.29) are the q-th zeros of (t q - 1), i.e.,
they are at t n = e2ni1r/q, n = 0,1, ... , q - 1. Thus,
and
Jort _1_
t - tn
dt = In It - tnl + const = In (1 - -!..) ,
tn
where the constant is zero because the integrand is zero at t = 0 and the principal
value of the logarithm is taken. Thus, the required transformation becomes
q-1
Z = 2: L
1T n=O
~
tn
In (1 - ~)
t
.
n
(2.3.30)
As a special case, if f..L = 1/2, i.e., if p = 1, q = 2, the transformation (2.3.30)

reduces to
z = 2:
1T
In 1- t
l+t'
t = ( ~ 1) 1/2 .•
."
(2.3.31 )
In the next chapter we shall discuss the problem of approximately computing the
values of the (2n + 2) parameters involved in the Schwarz-Christoffel formula
(2.3.1) or (2.3.4). This discussion involves numerical solution of improper
integrals known as Schwarz-Christoffel integrals.•
2.4. Catalog of Conformal Maps
The mathematical theory of conformal mapping has produced extensive in-

formation about mapping classes. Some classes of elementary mappings are
available in most textbooks on the subject of complex analysis. Although Math-
ematica can be used very effectively to produce cartesian, polar, parametric,
and conformal maps (see Maeder, 1991, and Wolfram, 1996), and there is a
DOS software conform (Ivanov and Trubetskov, 1995). Besices, there are two
catalog-style collections of conformal mappings, one by Koppenfels (1937) and
the other by Kober (1957), which are very practical and useful for applications
of computational conformal mapping. Both of these publications also contain
a number of formulas and properties of various different kinds of conformal
mappings; Kober's work, however, lists these formulas and properties, without
proof, although it expalins the method of Schwarz-Chritoffel transformations.
Although the choice of a collection of conformal mappings is a matter of

personal interest and each reader is directed in making this collection by the
type and scope of the study one is involved in, yet there are some very basic
and useful conformal mappings that need to be pointed out. A listing of some
such mappings can be found in Appendix E.
2.5. PROBLEMS 65
2.5. Problems
PROBLEM 2.5.1. Show that the Schwarzian derivative (also called Schwarz
differential operator)
{wz}- _(W- II
)' 1
-- (W- )2 -_ w"'
II
- - -3 (W
- )2
II
, w' 2 w' w' 2 w'

is invariant under bilinear transformations. (Nehari, 1952, p.199j Wen, 1992,
p.76.)
PROBLEM 2.5.2. Show that the Schwarzian derivative of the function

w = w{ z) that maps the upper half-plane ~ {z} > 0 conformally onto a
circular triangle with interior angles a1l", (311" and "(11" is given by
1 - a2 1 - (32 1 - a 2 - (32 + r 2
{w,z} = ~ + 2{1- z)2 + 2z{l- z)
(Nevanlinna and Paatero, 1969, p.342.)
y
v
u
o
z-plane w-plane
Fig. 2.5.1.
. Z - zl
PROBLEM 2.5.3. Show that the function j{z) = pe"1I.O - - maps the
z- Z2
region enclosed by two arcs onto the region enclosed by the sector shown in
Fig. 2.5.1, such that a fixed point ( i:- Zl,2 on r 1 goes into a point w on the
u-axis, where ao = arg { ((_- Zl
Z2
}
and p = w (_ I(- I (Pennisi, 1963, p.321.)
Zl
Z2 .
PROBLEM 2.5.4. Show that the bilinear transformation W = f(z) =

_z_ maps the region Int(f), where f = {Iz - 21 = 2} conformally onto
2z - 8
the region ~ {w} < 0, such that the point z = 2 goes into w = -1/2.
PROBLEM 2.5.5. Show that the bilinear transformation w = f(z) that

transforms the three points Zk into the points Wk, k = 1,2,3, can be expressed
as
1 z W zw
1 zl WI ZIWI
= O.
1 Z2 W2 Z2W2
1 z3 w3 z3 w3
PROBLEM 2.5.6. Show that the function w = f(z) = t .)

z - 1 +t
the lens-shaped region bounded by the circles Iz - 11 = 1 and Iz - i 1 = 1 onto
maps
the region bounded by the rays arg{ w} = 37r/ 4 and arg{ w} = 57r/ 4 such that
f(O) = 0 and f(1 + i) = 00.
PROBLEM 2.5.7. Show that the Schwarz-Christoffel transformation w =

2
f(z) = - sin- I zmapstheupperhalf-plane~{z} > oonto the semi-infinite
7r
strip 1~{w}1 < 1, ~{w} > 0, such that f(O) = 0 and f(l) = 1.
PROBLEM 2.5.8. Show that the Schwarz-Christoffel transformation
1 1 ~-J2
w = f(z) = -
7r
V2 (1 - z) +-
7r
log ~ J2
l-z+ 2
- i
maps the upper half-plane ~ {z} > 0 onto the region consisting of the fourth
quadrant plus the strip 0 < v < 1 in the w-plane, w = u + i v, such that
f(l) = O.
PROBLEM 2.5.9. Letzbefixedandlet~{z}~ O. Define a sequence of

bilinear transformations
ao
To(w) = , k = 1, ... ,n -1,
z + ao + bl + W
such that each aj > 0 is real and each bj is zero or purely imaginary, j =
0,1, ... ,n-1. Use induction to prove thatthechain of transformations defined
hy( = S(w) = (TooTn_2···oTn_I)(w)mapsthehalf-plane~{w} > 0
2.5. PROBLEMS 67
onto a region contained in the disk I( - 1/21 < 1/2. [Hint: Use Wallis
criterion: If P(z) = zn + CIZ n - 1 + C2Z n - 2 + ... + Cn is a polynomial of
degree n > 0 with complex coefficients Ck = Pk + i qk, k = 1,2, ... , n, if
Q(z) = PIZ n - 1 + i q2zn-2 + P3Zn-3 + i q4zn-4 + ... , and if Q(z)/P(z) can
be written as a continued fraction
Q(z) = a_o _
P(z) al
z + ao + b1 + - - - - - - - - - - -
z + b2 + - - - - - - - -
a2
an-l
z+b3 +···+--
z + bn
then Q(z)/P(z) = (To 0 Tn - 2 ··· 0 Tn-d (0).] (Saff and Snider, 1976, p.330.)
REFERENCES USED: Ahlfors (1966), Andersen et al. (1962), Betz (1964),

Boas (1987), Caratheodory (1969), Carrier, Krook and Pearson (1966), Gaier (1964),
Kantorovich and Krylov (1958), Koppenfels (1937, 1959), Ivanov and Trubetskov
(1995), Kober (1957), Mader (1991), Nehari (1952), Nevanlinna and Paatero (1969),
Pennisi (1963), Phillips (1943), Saff and Snider (1976), Sansone and Gerretsen (1969),
Silverman (1967), Wen (1992), Whittaker and Watson (1927), Wolfram (1996).
Chapter 3
Schwarz-Christoffel Integrals
In practical applications of conformal mapping of a standard region (the half-

plane or the unit disk) onto a problem region which is in the form of a poly-
gon, it becomes necessary to determine approximately the (2n + 2) param-
eters al, ... , an, Xl, ... , X n , and the constants A and B that appear in the
Schwarz-Christoffel formula (2.3.1). Evaluation of these quantities is known
as the parameter problem. We have seen in case studies in §2.3 that the map-
ping functions obtained by using the Schwarz-Christoffel formula involve
certain improper integrals which are known as Schwarz-Christoffel integrals.
We shall discuss methods for numerical solution of these integrals and present
Newton's method for the general case of mapping the upper half-plane onto
a quadrilateral.
3.1. Parameter Problem
If the values of all (2n + 2) parameters in the Schwarz-Christoffel formula

(2.3.1) or (2.3.4) are known, then the polygon is uniquely defined, and the
coordinates Wk of the vertices Wk are given by
k = 1, ... ,n. (3.1.1)
The length of the side joining the vertices Wk and Wk+l of the polygon is
determined by
(3.1.2)
68

3.1. PARAMETER PROBLEM 69
Therefore, it is very easy to figure out the behavior of the (2n + 2) parameters
in the formula (2.3.1) or (2.3.4). Thus, the parameters all ... , an are related
to the quantities /Ll,' .. , /Ln which are the ratios of the interior angles of the
polygon to 7r. The parameter A affects the location of the vertices of the
polygon. Any change in A changes the coordinates of every vertex by the same
amount through homothety and translation and displaces the polygon as a unit.
The parameter B that is a factor in the formula (3.1.2) affects the lengths of
the sides of the polygon. Any change in B changes all sides of the polygon
by the same amount and rotates the polygon as a unit. Any change in the
parameters Xl, ... ,Xn also produces relative change in the lengths of the sides
ofthe polygon.
Let the ratio of the second, third, ... , (n - 2)-th side of the polygon to the
first side be denoted by A2, A3, ... , An -2, i.e.,
j = 2,3, ... , n - 2. (3.1.3)
Consider the function
(3.1.4)
If we choose the numbers Xl, ... , X n in this function such that the relations
(3.1.3) are satisfied, where ai = 1 - J.Li' the functi<?.n W will define a conformal
mapping ~f the upper half-plane onto a polygon G. In order to pass from the
polygon G to G, we use the linear transformation
(3.1.5)
which does the following: It transfers any vertex, say Wk, of the polygon iJ.
to the corresponding vertex Wk of the polygon G, then rotates the polygon G
about the vertex Wk so that its sides become parallel to the sides of the polygon
G, and finally, without changing the position of the vertex Wk, it changes the
lengths of all sides of the polygon G such that the polygon Gcoincides with
G. The constants C l and C2 in (3.:...1.3) can be determined by comparing the
location and sides of the polygons G and G.
In order to determine the parameters Xl, ... ,Xn , there are only (n - 3)
equations, since three of these points can be chosen arbitrarily. We can use
a Mobius transformation to carry the upper half-plane onto itself such that
70 CHAPTER 3. SCHWARZ-CHRISTOFFEL INTEGRALS
the three points of the x-axis go into three preassigned points of the u-axis.
Let these three preassigned points be denoted by PI, Pz, P3. Let us denote
the improper integrals thus obtained from the formula (2.3.1) by 1m , m
1,2, ... , n - 2, and define them as
(3.1.6)
These integrals are known as Schwarz-Christoffel integrals. Then, in view
of (3.1.3),
1j (X3' X4, ... ,Xn-l) = Aj h (X3' X4, ... ,Xn-l), j = 2,3, ... ,n - 2.
(3.1. 7)
In view ofthe Riemann mapping theorem (§ 1.4), once the three points Xl, XZ, X3
are chosen arbitrarily, the system of equations (3.1.7) has a unique solution.
In order to solve the-system (3.1.7) numerically, we use Newton's method

which is as follows: Let X3, X4, ... , Xn-l denote the solution. Let us take
the initial guess for these values as x~O), x~O), ... , X~021 which are assumed
to be sufficiently close to X3, X4, ... ,Xn-l' If we expand each equation
of the system (3.1.7) in a Taylor series in powers of the difference Xv - xSO),
v = 3,4, ... ,n -1, and, as the first approximation, truncate these Taylor series
after the first power of the differences, then we obtain a system of equations
(3.1.8)
with a nonzero determinant, where
I
hS
), known as the corrections of the first
order, are the perturbed values of the differences Xv - xSO). Then the system
3.1. PARAMETER PROBLEM 71
(3.1.8) is reduced and solved for xS1

) , v = 3,4, ... ,n - 1. Then using the
initial values of hSO), the first approximations are given by
(3.1.9)
Next, the system (3.1.8) is expanded in Taylor series in powers of the differences
Xv - xS1), V = 3,4, ... ,n - 1, (3.1.10)

and these series are truncated after the first powers of these differences. Thus,
a new system, analogous to (3.1.8), is constructed with unknowns
2
), = xS v
3,4, ... ,n - 1, with the perturbed values of the differences (3.1.10), except
that now the new system is computed using the known values of
1
). It yieldsxS
the values of xS
2
) , and the second approximations are given by
(3.1.11)
This process is continued until we reach arbitrarily close to the solutions Xv,
such that the difference between two consecutive approximate values is within
a prescribed tolerance.
Once the system (3.1.8) is solved, we are able to approximate the coordinates
of the vertices of the polygon and the lengths of its sides. But in doing so, we
must compute the Schwarz-Christoffel integrals of the form
which are improper because the integrand of each integral becomes unbounded
at two points where ( = Xk, Xk+l which are the limits of integration.These
integrals exist because each O:k > O. The Kantorovich method to solve these
integrals is as follows: Let the integrand in (3.1.12) be denoted by F((). Then
F«() = «( - Pl)-a «( - P2)-a «( -
1 2 X3)-a 3 •.. «( - Xk)-a k
(Xk+l - o-ak+ 1 ••• (P3 - ()-a n

= «( - Xk)-a k
[¢(Xk) + ¢'(Xk)«( - kk)]
+ (Xk+l - O-ak+l [~(Xk+l) -~' (Xk+l) (Xk+l - 0]
+ {F«() - «( - Xk)-a [¢(Xk) + ¢'(Xk)« - kk)]
k
- (Xk+l - 0 [~ (Xk+l) - ~' (Xk+l) (Xk+l - 0] },

(3.1.13)
where
¢(() = (( - Pl)-<>1(( - PZ)-<>2(( - X3)-<>3 ... (( - Xk_l)-<>k-1

(Xk+l - ()-<>k+1 ... (P3 - O-<>n, (3.1.14)
'ljJ(() = (( - Pl)-<>1(( - PZ)-<>2(( - X3)-<>3 ... (( - Xk)-<>k
(Xk+Z - ()-<>k+ 2 ... (P3 - ()-<>n. (3.1.15)
Let E = E 1 + E z, where
Xk 1
l
E 1 = Xk + { (( - Xk)-<>k [¢(xd + ¢'(Xk)(( - Xk)]
+ (Xk+l - ()-<>k+I ['ljJ (Xk+l) - 'ljJ' (Xk+l) (Xk+l - 0] } d(
l
(3.1.16a)
Xk
Ez = +I {F(() - (( - Xk)-<>k [¢(Xk) + ¢'(Xk)(( - Xk)]
Xk (3.1.16b)
- (Xk+l - 0 [1/1 (xk+d -1/1' (xk+d (Xk+l - ()] } d(.
The integral E 1 can be evaluated directly in a finite form, and since E z has no
singularities it can be approximated by any formula for numerical integration
of definite integrals, like Simpson's rule.
3.2. Newton's Method
We shall discuss Newton's method for mapping the upper half-plane onto an
arbitrary quadrilateral (see Case Study 2.3.3). In fact, in order to determine k
in (2.3.14), we have only one equation from (3.1.8)
(3.2.1)
where
1
1:
00
I 4(k) = (( + 1)-<>1(( _1)-<>2(( - k)-<>3(( - 3)-<>4 d(
+ (-1 - ()-<>1 (1 - ()-<>2 (k - ()-<>3 (3 - ()-<>4 d(,

(3.2.2)
3.2. NEWTON'S METHOD 73
and
In 14 (k), set ( = 2 + 1 It in the first integral and ( = -1 It in the second. Then
I 4 (k) = 11 + (1 3t)-"1 (1 + t)-"2 (1 + (2 - k)t)-"3 (1 - t)-"4 dt
+1
1
(1- t)-"l(l + t)-"2(1 + kt)-"3(1 + 3t)-"4 de.
(3.2.4)
Let F(k) = I 4 (k) - >'h(k). Then F'(k) > O. To see this, we have from
(3.1.13)
F'(k) = CY311 t(l + 3t)-"1(1 + t)-"2 (1 + (2 - k)t)-"3- 1 (1- t)-"4 dt
- CY311 t(l - t)-"1 (1 + t)-"2 (1 + kt)-"3- 1 (1 + 3t)-"4 dt
+ >' CY3[11 (( + 1)-"1(1 - ()-"2(k - ()-"3- 1 (3 - ()-"4 de.

(3.2.5)
Note that F'(l) and F'(3) do not exist because they are not finite. Since
F"(k) = CY3 (1 + CY3) {1 1

t 2 (1 + 3t)-"1(1 + t)-"2
(1 + (2 - k)t)-"3- 2 (1- t)-"4 dt
+ 11 t 2 (1- t)-"1(1 + t)-"2 (1 + kt)-"3- 2 (1 + 3t)-"4 dt
- >.[11(( + 1)-"1(1- ()-"2(k - 0-"3- 2(3 - ()-"4 d(},

(3.2.6)
we find that F"(k) > 0 for 1 < k < 3, and F"(l) = -00 and F"(3) = +00.
Hence, as k varies from 1 to 3, both F(k) and F"(k) increase continuously
from -00 to +00 (for details, see Kantorovich and Krylov, 1958). Hence, the
function F(k) has a zero in (1,3). Let the zero of F(k) be denoted by k*,
and that of F"(k) by k**. Now we shall solve the equation F(k) = 0 by
Newton's method. Although the initial guess for the value of k is important in
this method, yet we shall show that it can be chosen arbitrarily from below or
from above in the interval (1,3). Let k o denote an arbitrary initial guess for the
value of k. Then
THEOREM 3.2.1. In order to solve the equation F(k) = 0 by New-

ton's method, the value of k o E (1,3) can be chosen arbitrarily, indepen-
dent of the values of k* and k**.
PROOF. There are three cases to analyze:
CASE 1: If K** = k*, then the initial value

F(k)
k o is any number in the interval
(1,3) (see figure on the right).
In fact, let ko be an
approximation of k from below. "'o+-----.c-------,.L------;...;,-- k
Then F(k) < 0, and F"(k) < 0
for all k E (k o, k*). Hence,
there is an M such that
F'(k) < M for all
ko ::; k ::; k*. The first
correction is given by
8(1) = _ F(k o) > O.
(3.2.7)
F'(k o)
Thus, the exact first correction hI is positive and satisfies the equation
F(k o) + hI F'(ko) + ~i F"(k) = 0, ko < k < k*. (3.2.8)
Hence, from (3.2.7) and (3.2.8), we have
which yields
h = h _ 8(1) = _ hi F"(k) (3.2.9)
2 1 2 F'(k o) '
which is positive in view of (3.2.8). Thus, the first approximation k 1 , like ko,
is an approximation from below, and, therefore, all subsequent approximations
will be very small. It is the basic property of Newton's method that the exact
corrections h are always positive and decreasing. In fact, we can show that if
the difference between h n and hn+l is sufficiently small, then h n is itself very
small. Assume that
(3.2.10)
wherec > ° is arbitrarily small. Since h n -hn+ I = 8(n), we have 8(n) < ~c.
Also,
8(n) = IF (kn-d I
F' (kn-d '
or
(3.2.11)
But h n can be evaluated from the equation
F (kn-d + h n F'(f), k n- I < k< k*. (3.2.12)
Hence, from (3.2.11) and (3.2.12)
h = IF (kn- I ) I = 8(n) F' (k n- I)
n F' (k) F'(k) ,
This shows that the sequence {hI, h 2 , ... ,hn , ... } '\. 0. This analysis leads
to the same conclusion if ko is an approximation from above.
CASE 2: If k** > k*, then for any initial guess ko < k* the convergence is the
°
same as in case 1. Also, F(k) < 0, and F'(k) < for ko < k < k* (as in case
1). Now, let ko be such that k* < k o < k**. Then F(k o) > 0, and F"(k) <
for all k :::; ko. Moreover, 8(1) < 0, and hI < 0. Now,
°
and thus, h 2 > 0. This means that the first approximation is an approximation
from below, and all subsequent approximations will converge from below to
, {<o
k* (as in case 1). In the case when ko > k**, we have F(k) > 0 for all k E
forallkE (k*,k**)
°
(1)
(k*, ko); F (k) () ;8 < 0, and hI < 0. Thus, k I
> for all k E k**, k o
can be in anyone ofthe three intervals (1, k*), (k*, k**), and (k**, ko). If k I is
in the first two intervals, then we have convergence from below to k*. Butif k I is
in the third interval, then all approximations, although decreasing continuously,
still remain greater than k** and approach some limiting value k I 2: k**. The
difference between hn and hn+l would be sufficiently small for sufficiently
K k** - k*
large n, and h n - h n + I < M ' where M = max F'(k). As in
2 kE(k· ,ko)
k** - k*
case 1, it can be shown that Ihnl < . But, by assumption, we have
2
Ihnl ;::: k** - k*. This contradiction shows that if Newton's method starts with
ko > k**, then the approximation will cross k** and fall in the interval where
the convergence is established as in case 1.
CASE 3: If k** < k8 , then this case can be analyzed by taking ko in anyone
of the intervals discussed in cases 1 and 2. We shall apply this method in Case
Study 3.3.3.
3.3. Numerical Computations
Now we shall solve some parameter problems.
CASE STUDY 3.3.1. Consider the integral in the denominator in (2.3.7)

Denoting it by E, we have B = 1/ E, where
E = (1 2+ 1~2) C
1
/
1 2
/ (1 - ()-3/4 d( = E 1 + E2 ·
Then, by (3.1.16),
1
1/2
E 1 = lim C 1/ 2(1- ()-3/4d(
t-+O t
= P!!61
1 2
/ {C + ~() + 1/ 2
1 2
/ (1 C [(1 - ()-3/4 - 1 - ~(]} d(
= Ell + E 12 ~ 1.59099 + 0.0708022 = 1.66179;
E 2 = lim r
t-+t}1/2
t
C 1/ 2 (1 _ ()-3/4 d(
J
t 1
= lim {(1- ()-3/4[1 + - (1- ()]
t-+1 1/2 2
+ (1- 0-3/4[C 1/ 2 -1- ~(1- O]} d(

= E 21 + E 22 ~ 3.53176 + 0.0505577 = 3.58232.
Hence E = E1 + E2 ~ 5.24412, and the constant B = 1/ E ~ 0.19069. Note

that the exact value of E v:;,~j~4) ~
= 0.19068994. The hypotenuse of the
3.3. NUMERICAL COMPUTATIONS 77
triangle is given by
with an error of 0 (10- 10 ) . •
CASE STUDY 3.3.2. In order to determine k in (2.3.10) such that the

ratio W2 - WI = 2, note that the system (3.1.7) reduces to only one equation
W3 -W2
h = 2h, i.e.,
1(1- (2)-1/2(1_ k 2(2) -1/2 d( = 2

jl/k ((2 _1)-1/2 (1- k 2(2) -1/2 dC
/-1 1
(3.3.1 )
which is solved by Newton's method as follows: We have only one correction
which we shall denote by h with subscript to denote the appropriate number of
approximation. Let us take the initial guess k = 1/2, and determine hI from
(3.1.8) which is
12 (
1)
2 + hI
dh (~) = h
dk
(1)2 + hI
dh (~)
dk (3.3.2)
The free term 12(1/2) = 1\1-()-1/2'1j;(()d(,where'lj;(() = (1+()-1/2+
(1 - (2/4) -1/2, which yields h(I/2) ~ 2.15652 on integration. Also,
d12 (~)
dk =2
1r (1 _ (2)1/2
Jo
1
(2 (
1-
(2)3/2
4 d( ~ 0.541732,
d11 (1/2) ~ -1. 79181.

dk
Substituting these values in (3.3.2) we find that hI :::::: 0.201739. Hence, the
first approximation for k :::::: 0.5 + 0.201739 :::::: 0.7. Now, for the second
approximation, first we compute h 2 from (3.1.8), i.e.,
dh(0.7) dh(0.7)
h(0.7) + h2 dk = h(0.7) + h2 dk ' (3.3.3)
which gives h 2 ~ 0.00668985, and the second approximation for k ~ 0.7 +

0.00668985 ~ 0.70669. However, the exact value of k can be determined in
this case by setting ( = 1 in 12 (k). Then
..)1 - (1 - k 2 )x 2
12(k) = 1 1
(1 - X 2)-1/2(1 - k,2 X 2)-1/2 dx,
where k'2 = 1 - k 2. Hence, Eq (3.3.1) gives
1 1
(1 - (2)-1/2(1 - k 2(2)-1/2 d( =
1
(1 - 1 X 2)-1/2(1 - k,2 x 2)-1/2 dx.
Thus, k = k', which gives k = 1/ V2 ~ 0.7071. A comparison of this valueof

k with the second approximation for k shows that the error is about 0.04% .•
CASE STUDY 3.3.3. Map the trapezoid Al A 2 A 3 A 4 in the z-plane onto

the upper half-plane (Fig. 3.3.1). From (2.3.14) the mapping function is given
by
w= A+B 1+ z
(( 1)-1/6(( _1)-1/3(( + k)-2/3(( - 3)-5/6 de. (3.3.4)
Fig. 3.3.1.
In order to determine the value k, we have from Eq (3.1.8)

14(k) = 3h(k), (3.3.5)
where from (3.2.4)
14(k) = 11
(1 + 3t)-1/6(1 + t)-1/3 (1 + (2 - k)t)-2/3 (1 - t)-5/6 dt
+ 1 1
(1 - t)-1/6(1 + t)-1/3(1 + kt)-2/3(1 + 3t)-5/6 dt
= 141(k) + 142 (k),
(3.3.6)
and from (3.2.3)
h(k) = [11( + 1)-1/6(1 - ()-1/3(k _ ()-2/3(3 _ ()-5/6 d(
= ({01 + 1 1
) ( + 1)-1/6(1 _ ()-1/3(k _ ()-2/3(3 _ 0- 5/ 6 d(
= Ill(k) + Idk).
Let k = 2 be the initial guess. Then we find that 8(1) ::;::j -0.650694, k 1 ~
1.34931,8(2) ::;::j 0.0150829, k2 ::;::j 1.36439, 8(3) ~ 0.0005, and k3 ~ 1.36554.
Now, in order to compute A and B, note that the function
(3.3.7)
maps the upper half-plane onto a trapezoid Ai A 2A 3A 4 similar to the given

A 1 A 2 A 3 A 4 . To determine the complex coordinate wi of the vertex Ai which
corresponds to z = -1, we have
wi = 1- 1
( + 1)-1/6(( - 1)-1/3( + k)-2/3(( - 3)-5/6 d(
::;::j - ( -1)1/6(0.24631) = -0.24631 i.
Similarly, w2 ::;::j 0.90311 i. Hence w = A + B w* yields
1 .
- ~2 = A - 0 .2463li B , 2= A + 0.90311 tB,
which gives A = -0.2851, and B = -0.87 i, and the required transformation

is given by
w::;::j -0.2851-0.87i l z
( +1)-1/6(( _1)-1/3(( +k)-2/3( _3)-5/6 d( .•
(3.3.8)
CASE STUDY 3.3.4 (SCHWARZ-CHRISTOFFEL INTEGRAL FOR THE

UNIT DISK). In view of (2.3.2), the integral in the Schwarz-Christoffel formula
(2.3.1) is approximately equal to (-2 when ( is close to infinity, whereas the
integrand in the formula (2.3.3) is approximately (-Qn. These quantities are
significant when the region is infinite. However, we can avoid an infinite region
by mapping the upper half-plane ~ {z} > 0 onto the unit disk by the chain of
. Z- i 1 + Zl 2i
mappmgs Zl = - - . and Z = i - - , where dz = ----,,2 dz 1 . Then the
z+t 1-z1 (1-z 1)
integrand in the formula (2.3.1) becomes
j
IIn
j=l
(z = Xj)
-a.
dz = II
J
j=l
(.1+Z 1
1 - Zl
n
t -- - Xj
)-a 2i
(1 - zI)
2 dZ1
j
=
2i rr(Zl(Xj+i)-(Xj-i))-a 1 2
dz1
j=l 1- Z 1 (1- Z 1)
n j
-2'II (+.)-a j ( Xj_i)-a dZ 1
- t x· t Zl- --
j=l J Xj + i (1 - zI)2-a j
= C1 II (Zl - bj)-aj dz 1 ,
(3.3.9)
where bj = (aj - i)/(aj + i), and the exponent 2 - exj is zero in the product.
Thus, formula (2.3.1) becomes
(3.3.10)
where the points bj lie on the unit circle IZ11 = 1. The lower limit ZlO may be
chosen as the center of this circle or a point on its circumference. Then there
are two cases to consider:
CASE 1. If ZlO = 0, then the integration is carried out along the ray Zl = r e i8 ,
o ~ r ~ R, B =const, and bj = ei4>j ,j = 1, ... ,n. Then the mapping function
(3.3.10) reduces to
W = C 2 ei8 1 II
R
o j=l
n
(re i8 - ei4>j) -aj rdr + Zo
= C 2 ei8 inrR IIn e- i8 4>j ( r - ei (4)j-8)

) -a·
J r dr + Zo (3.3.11)
o j=l
=C2 e- i8 inrR II n (
r_e i (<!>j-8)
) -a'
J rdr+zo·
o j=l
CASE 2. Ifz lO = 1, we choose the path of integration along the circumference.

Thus, for a point Zl = e iO the mapping function (3.3.10) becomes
1 II
0 n
w = C2 (e iO - ei</>j) -<>j ie iO dB + Zo
o j=l
1II
0 n
= iC2 e- C
£ </>j<>j)/2 (e i (0-</>j)/2 - e- i (0-</>j)/2) -<>j dB + Zo
o j=l
= i(2i)-2 C 2 in
r II TB _ ¢.)
O
n (
sin
-<>j
dB + Zo
o
B _ ¢.)
J=l
= K
r II
in
O
n (
sin T -<>j
dB + Zo·
o J=l
(3.3.12)
If sin T B - ¢.
< 0, then we choose the branch
sin (B ~ ¢j ) -<>j = exp { - (Xj log sin B I ~ ¢j I- i7f(Xj}'
There is no problem for sin B ~ ¢j 2': O. Thus, there exists a constant argument
in (3.3.12) in each interval ¢j-l j
4>,-1 j=l
IT I sin B ~ ¢j I-<>j dB, j = 1, ... , n. (3.3.13)
Therefore, the parameter problem for the unit circle is solved by carrying out
the integration in (3.3.13) over a finite interval. •
CASE STUDY 3.3.5 (REMOVAL OF SINGULARITIES). Thecomputation

of the improper integrals in the formulas (2.3.1) or (2.3.3) can be easily carried
out by removing the singularities in the integrand. There are two analytical
methods to do this.
METHOD 1. We shall consider the exponent -(Xl; others can be handled
similarly. If (Xl > 0 in (2.3.1), we set
Kl = (Xl - X2)-<>1 ... (Xl - xn)-<>n
and rewrite the integrand in (2.3.1) as
l
z
[(( - Xd-<>1 (( - X2)-<>2 ... (( - xn)-<>n - (( - Xt)-<>1 K l ] d(
Zo
+ -K-
l [(( - Xl )1-<>1 - (Zo - Xl )1-<>1] . (3.3.14)
1- (Xl
The integrand in (3.3.14) in the neighborhood of Z = Zl is approximately equal

to (( - Xl) -n 1 . In fact, it consists of two factors, the first of which goes to 00
and the second to 0 as ( -. Xl. Hence, the product is bounded or even goes to
zero.
If we use (3.3.14) for Case 2 of Case Study 3.3.4, then we get integrals of
the form
f} (
1= <PI sin--
i B - <P l ) -nl
2
dB=2 0
where we have set B = <PI + 2t. Now, let <PI

1(f}-<P Il/ 2 1
(sint)-a dt,
 0 in (2.3.1), then we remove the singularity at ( = Xl
by using the transformation Zl = (z - xd - n1 , i.e.,
l
1/(1-nIl
Z = Xl + Zl . (3.3.16)
zl-plane
Fig. 3.3.2.
Then the integral (2.3.1) becomes
J
w = Zo
C
Z
nr!(l-nIl + Xl - ) -n2 . •. (-nr!(l-n) 1) ) -n" d Zl,
+-
l (
Zl X2 Zl Xl - Xn
al ZlO (3.3.17)
3.4. KIRCHHOFF FLOW PROBLEM 83
which does not have infinity for al. The transformation (3.3.16) transforms the
half-plane ~ {z} > 0 onto an angular sector of argument (1 - aj) 71' = J..Lj 71'.
Note that the mapping (3.3.16) is not suitable for the case of the circle. However,
the transformation that maps the circle onto a region bounded by two circles is
given by
(3.3.18)
which is represented in Fig. 3.3.2.•
3.4. Kirchhoff Flow Problem
The classical Kirchhoff flow problem (Kirchhoff, 1869) deals with the flow of
an ideal incompressible fluid past an obstacle and around a stationary wake
bounded by free streamlines (see Problem 3.5.5). It is known that a plane
Kirchhoff flow past a solid polygonal obstacle composed of an open polygonal
line facing the flow, in theory, can be determined by constructing its conformal
mapping onto an n-gon in the log-hodograph plane and then onto the upper
half-plane by using the Schwarz-Christoffel transformation. In practice, how-
ever, this approach is fraught with computational difficulties as we shall soon
see.
The geometry of a plane Kirchhoff flow is as follows: A solid obstacle

with a open polygonal boundary r, composed of n straight line segments rk =
(Zk-l, Zk), k = 1, ... , n, lies in the region G z in the z-plane (physical plane)
as shown in Fig. 3.4.1 (a). The ideal incompressible fluid flow, flowing past the
obstacle r, is assumed to be irrotational. Let the complex velocity be denoted
by v(z) and normalized by v( (0) = 1. The flow divides between an upper and a
lower part at an unspecified stagnation point z* where the upper flow passes over
Zn and the lower over zoo Then the flow continues smoothly forward past Zn and
Zo with finite acceleration around a wake in which v == O. The two streamlines,
r+ and r- denote the curves of discontinuity which separate the wake from
the rest of the flow, and the stream function is zero on r±. The shape of r± is
not known but must be determined by using the condition that Iv( z) I = 1 along
them (this condition follows from Bernoulli's equation (p + O.51vl = const)
and the fact that the pressure must remain constant throughout the wake and
continuous on r±).
Now the Kirchhoff flow problem can be stated as follows: Given the obstacle
r in the physical region G z in the z-plane, determine the velocity field v( z),
the streamlines r±, and the location of the stagnation point z. for the above
flow. Also compute the associated lift and drag coefficients. A conformal
mapping solution of this problem can be stated as follows: Let T denote the
hodograph (or conjugate velocity) plane so that the complex conjugate velocity
is defined by T(Z) = v(z). Since the flow is incompressible and irrotational, the
velocity v = 'V¢, where ¢(z) is the real part of the complex velocity potential
(z) = ¢(z) + i'l/J(z) such that 'V 2 ¢ = 0 and 'l/J(z) is the stream function.
Thus,
d(
T(Z) = dz' (3.4.1 )
(a) z-plane
G~ GW
IV = lVeo
--~
~.=o ~o W=_I WI W
•
(b) ~-plane (c) IV-plane
Fig. 3.4.1.
The function ( z), regular in G z, maps the region G z conforrnally onto a slit
region Gc, in the (-plane, where the slit begins at (* = (z*) at which point
the flow separates to go around the polygonal obstacle (Fig. 3.4.1 (b». Without
loss of generality we take (* = O. Let a new complex variable w be defined by
(3.4.2)
where W is real and w* E (-1,1) (Fig. 3.4.1(b) in which Wk = W((k) is

marked), G w is the upper half-plane where [-1,1] is the image of r, and
(1,00) and -00, -1) that of r+ and r-, respectively. We shall discuss two
methods.
CLASSICAL HODOGRAPH METHOD. In this method the upper half-plane

G w in the w-plane is mapped onto the hodograph region G T = T (Gc,) in the
T-plane. The technique leads to a solution, at least theoretically, for the general
Kirchhoff flow problem, because although G z is unknown due to the presence
of unknown free streamlines, the region G T is almost known in the following
sense: Since the flow must be tangential on the solid boundary r, we know
arg{ T( z)} depending on whether the point on r is downstream of z* toward
Zo or upstream toward Zn, respectively, i.e.,
(3.4.3)
Also
IT(z)l=l forzEr±. (3.4.4)
The region G T is 'gearlike' , bounded by circular arcs and subsets of rays passing
through the origin. By introducing the log-hodograph variable
fl(z) = -log T(Z), (3.4.5)
the region G T is mapped onto a region Gn which is bounded by vertical and

horizontal line segments. Then we can use a Schwarz-Christoffel transforma-
tion to map G w onto Gn. This method establishes a relation between T and (
and then integrates (3.4.1) to obtain ( and T as functions of z.
However, in practice only a few simple cases involving a flat plate (Kirch-
hoff, 1869; for the classical case see Problem 3.5.5) and certain wedges (Birkhoff
and Zarantonello, 1957; Gurevich, 1965; Robertson, 1965; Elcrat, 1982) have
been solved by this method because the complexity of the conformal mappings
grows as the number of sides of the polygonal obstacle increases. Then the re-
sulting parameter problem inherent in the Schwarz-Christoffel integrals must
be determined numerically which is not an easy task. Another difficulty stems

from the fact that although the vertices in the w-plane are known, those in the
D-plane must be computed by integrating (3.4.1), which can be very time con-
suming. Finally, the most serious difficulty with this method is that, in general,
Gn is a Riemann surface with slits or branch points of unknown dimensions
rather than just a polygon.
ELCRAT-TREFETHEN METHOD. This method, developed by Elcrat and

Trefethen (1986), for computing flows past an arbitrary obstacle with a high
degree of accuracy, uses a modified Schwarz-Christoffel integral to map the
upper half-plane'S {w} > 0 directly onto the physical domain G z rather than
onto the log-hodograph plane. The Schwarz-Christoffel formula (2.3.5) is
n
modified as follows: Let the polygonal line f = U f k in the z-plane (the
k=l
obstacle in the physical plane) have vertices Zk, sides fk = (Zk-l, Zk), interior
angles ak 7r (counterclockwise), k = 1, ... ,n, and exterior angles J.Lk> where
J.Lk = ak-l - ak for k = 1, ... ,n - 1, and J.Ln = al - an + 2, thus
L~=l J.Lk = 2. Let Z = F(w) be the conformal mapping of the upper half-
plane'S {w} > 0 onto the G z such that the point w = 00 goes into a point on
fl' Let Wk = w (fk) denote the intervals (Wk-l, Wk), where Wk = F (Zk).
Then arg {:~} has a constant value a7l" on each W k and ajump of J.Lk7r at Wk,
i.e.,
dz
arg {dw} = al 7r for wE (w n , 00), (3.4.6)
dz
D.. arg {dw} = J.Lk 7r at w = Wk· (3.4.7)
Let gk(W) = (w - Wk)-/J-k denote the factors in the formula (2.3.5), where
the branch of gk(W) is chosen such that gk(W) > 0 for W > Wk. Then
arg{gk( w)} = const except for ajump of J.Lk7r at Wk. It means that the function
gk (w) maps 'S {w} > 0 onto the sector bounded by the rays e- i/J-k 71" jR+ and
jR+ in the z-plane (Fig. 3.4.2(a)). Hence,
C> 0, (3.4.7)
and the modified Schwarz-Christoffel formula is
(3.4.8)
where Co is a complex constant (cf. this formula with (2.3.4».
gk(W)
1m {w} >0 ~
-Ill-
IR+
Wk gk(wk )
(a) w- plane z- plane
hk(-I)
hk(w)
1m{W}>O ,
- Il kit
IR+
hk(wk ) hk(l)
(b) w- plane z- plane

Fig. 3.4.2.
Now, consider the function z = F 1 (() which maps the slit region G( in
the (-plane onto the region G z in the z-plane (Fig. 3.4.1). Since we know
arg { dz
d(} for w E [-1, 1] and IdZ\
d( elsewhere, we have
dz
arg { d(} = (Xk 1f for w = Wk,
dz
.6. arg {de} = J..Lk 1f atw = Wk for k= 1, ... ,n -1,
(3.4.9)
I:~ I = 1 for W E W±, i.e., Iwi > 1,
dz
arg {d(} = 0 at w = 00.
Thus, the Kirchhoff flow problem is a modification of the Schwarz-Christoffel

problem (3.4.6) in the sense that a constant-modulus condition, instead of the
constant-argument condition, is applied over the boundary. Define a function
hk(w) =[ W-Wk ]-J-Lk, (3.4.10)

1 - Wk W + J(1 - w 2 ) (1 - wn
where the branch is chosen such that hk(W) > 0 forw E (Wk, 1). The function
hk has singularity at Wk (like the function gk) and also at W = ±1, and it maps
the half-plane <;} {w} > 0 onto the closed circular sector bounded by the rays
e- i J-L k 1l' IR+, IR+ and the unit circle Izl = 1 (Fig. 3.4.2(b)). Since Ihk(W)! = 1
for Iwi > 1, we have
dz
d( = -:;.1 = e">n
.
11'
II hk(W), (3.4.11)
*
where II denotes the product over k = 1, ... , n - 1 and k = * (L* defined
*
analogously) and W* is the preimage of z*. Note that the function defined by
(3.4.11) satisfies all the conditions in (3.4.9) except the last one. In order to
satisfy this last condition we must choose W* properly. Thus, from (3.4.10) we
find that arg {hk(W)} = -JLk cos- 1 (-Wk), and hence,
iQ n
arg{e 1l' IIhk(W)} =a n 11'- LJLk cos- 1 (Wk).
* *
Hence, the last condition in (3.4.9) implies that
an 11' = a n11' - LJLk cos- 1 (Wk) = 0,

*
and thus,
n-l )
W* = - cos an 11' - L JLk cos- 1 (Wk) . (3.4.12)
(
k=l
Then Eq (3.4.11) yields the Kirchhoff flow as the Schwarz-Christoffel integral
z=C + ei c> n 1l' J( II hk(t) dt

*
= C + e c>i n
1l' W JW (t - w*) II hk(t) dt
* (3.4.13)
= C + We i n
c> 1l' JW (1 - W* t + J(1 - t 2) (1 - W;))
X II
n-l (
t - Wk
2
) -J-Lk
dt
k=11-w*t+J(1-t )(1-w;) ,
3.5. PROBLEMS 89
where we replaced the integration with respect to ( by that with respect to w by

setting d( = W (w - w*) dw and in the last step canceling the common factor
(t - w*). The above formula basically matches the formula derived by Mon-
akov (1983, Eq (5) on p.185, where he erroneously takes w* = 0). A Fortran
package for the Kirchhoff flow problem, containing the files scpack. exe and
kirch1 . exe, can be obtained from the second author in Elcrat and Trefethen
(1986).
3.5. Problems
PROBLEM 3.5.1. Let the field of an infinite two-dimensional capacitor

be as shown in the Fig. 3.5.1, with the values of the potential on the curves
r 1 and r z as 0 and 1. Show that the function w = f (z) that maps the upper
half-plane onto the triangle A 1 A zA 3 is given by
where C 1 = ih and Co = hlrr. (Sveshnikov and Tikhonov, 1978, pp.217-218.)
___v_=--'-o_--'- '-----+ X
z=o
Fig. 3.5.1.
l
z
dt
PROBLEM 3.5.2. Show that the function w = maps the
o Vt(1 - t Z )
upper half-plane ~{z} > 0 onto the interior of the square of side rZ~).
2 211"
(Phillips, 1966, p.65.)
PROBLEM3.5.3. Showthatthefunction7rw = cosh- 1 z-sin- 1 (l/z)+

7r /2 maps the region in the positive quadrant of the w-plane bounded by the
lines u = 0, v = 0, u = 1 (v > 1), and v = 1 (u > 1), onto the upper
half-plane ~{z} > 0. (Phillips, 1966, p.66.)
PROBLEM 3.5.4. Show that the Schwarz-Christoffel transformation that

maps the region (shaded) in Fig. 3.5.2 onto the upper half-plane ~ {w} > °
°
such that z = goes into w = is given by °
z =a- JW (-_1)J1.
- d(.
J17r 1 (
Let J1 = p/q, p < q (P, q E lR+), and set t =

((_1)
-(-
l/q
, as in Case Study
2.3.6. Show that the mapping function becomes
+ I>n
a 1 tP q-l P ( t )
z = - [- - -q log 1- - ],
7r J1 t - 1 n=O tn
where t n = e2i7rn/q, n = 0,1, ... ,q - 1, are the q zeros of t q - 1. [Use

1
tp-1
-- = -
tq - 1 q
L
q-l
n=O
tP
_n_.] In particular, for J1 = 1/2, P = 1, q = 2, show that
t - tn
. f unctIOn
the mapping . IS
. gIven
. by z = -a
7r
[2t--2
I-t
+ Iog -
1 --
l+t
t] ( . Koppenfels,
1959, pp.21O-211.)
o
z-plane w-plane
Fig. 3.5.2.
PROBLEM 3.5.5. The original Kirchhoff's problem deals with an irro-

tational flow of a weightless, ideal, incompressible fluid past a flat plate AB
with separation of the jet, such that the modulus of flow velocity is equal to
the modulus of the approaching stream Vo on the surfaces of the jeta AD and
BD (Fig. 3.5.3). The region of constant pressure behind the plate extends to
3.5. PROBLEMS 91
infinity and the plate AB is perpendicular to the flow. This problem deals with
the determination of the function (( w) such that Vo :~ = ((w). Show that the
function
where Uo is a real constant maps the w-plane cut along the positive real axis
from C to +00 onto the upper half-plane 8'{(} > 0 from which a semicircle
of unit radius is removed (Fig. 3.5.4). [Use the following conformal maps:
(-1 2
T = -,.--, T1 = T , t =
~ l+t
- , and t1 = - - . ] (Gurevich, 1965, pp.15-20.)
~+1 ~ 1-t
D D
X
A C B
Fig. 3.5.3.
v
aA AGa
D
u
0 c D
c C
-I
I ...... plane ~-plane
Fig. 3.5.4.
REFERENCES USED: Ahlfors (1966), Betz (1964), Birkoff and Zarantonello

(1957), Boas (1987), CaratModory (1969), Carrier, Krook and Pearson (1966), El-
crat (1982), Elcrat and Trefethen (1986), Gaier (1964), Gilbarg (1949, 1960), Gure-
vich (1965), Kantorovich and Krylov (1958), Koppenfels (1959), Monakov (1983),
Robertson (1965), Sveshnikov and Tikhonov (1978).
Chapter 4
Polynomial Approximations
Gaier's variational method is used to solve two extremal problems in the

theory of conformal mapping. The first deals with the conformal mapping
of a simply connected region onto a disk, and the second with that of the
boundary of the region onto the circumference of the disk. Both problems
use the Ritz method for approximating the minimal mapping function by
polynomials. This mapping function in the first problem is represented in
terms of the Bergman kernel function, and in the second problem in terms
of the Szego kernel. Another important problem deals with an investigation
into the nature and location of boundary singulaities and poles of the mapping
function close to the boundary, which is presented in Chapter 12.
4.1. Minimum Area Problem
An extremal property in the conformal mapping of a simply connected region

D onto a disk is connected with the minimum area problem. This problem,
which we shall denote as Problem I, is known as the Bieberbach minimizing
principle. The mapping function possesses an extremal property which pro-
vides a method to compute an approximate solution for the map. Let K}(D)
denote the class of all functions f E L 2 (D) with f(a) = 1, where a E D.
Similarly, let JCO(D) denote the class of all functions f E L 2 (D) such that
f(a) = O. Note that the classes JCl(D) and JCO(D) are, respectively, a closed
92
4.1. MINIMUM AREA PROBLEM 93
convex subset and a closed subspace of L 2 (D). Let the function

00
w=f(z)=Lan(Z-at,lz-a/<R, (4.1.1)
n=O
which is regular in D, map D onto the disk B(O, R) in the w-plane. Without
loss of generality, we shall sometimes take the point a as the origin.
PROBLEM I: In the class K l minimize the integral
(4.1.2)
The Riemann mapping theorem (Theorem 1.4.1) guarantees the existence and
uniqueness of the solution of this extremal problem.
THEOREM 4.1.1. Problem I (minimum area problem) has a unique

solution fo(z) = f'(z). The minimum is 1rR2.
PROOF. If fo(a) = 0 and fo(a) = 1, then in view of (1.4.9)
00
= 1rR2Ial/2 + 1r L n la n /2 R 2n .
n=2
The above result implies that in the problem of mapping by the function (4.1.1),
which is regular in B(O, R) and is such that f'(a) = aI, the area of the mapped
region D is always greater than 1rR21al12. It is exactly equal to this value
if the map f(z) is linear, i.e., if w = ao + alz. A particular case is when
al = 1. Then the mapping function is w = ao + z. If this linear transformation
is excluded, then the mapping function can be normalized by the conditions
f(a) = 0 and f'(a) = 1, by considering the function f(Z)jal' In either case
the minimum area of D is 1rR 2 . •
Before we solve Problem I, we shall examine the minimal function fo(z)

closely.
94 CHAPTER 4. POLYNOMIAL APPROXIMATIONS
THEOREM 4.1.2. The function fo(z) is orthogonal to every function

f E L 2(D) with g(a) = 0, i.e,
(4.1.4)
PROOF. For every c: > 0 and 0 :s e :s 27f, the function lo(z) + c:g(z)
belongs to the class }(,1. Then
f lllo(z)1 dS z
2
:s f f Ilo(z) + c:g(z)1 2 dSz
= fl1lo(z)12 dS z + 2c:~ { f l fo(z) g(z) dSz }
f
+ c: 2 llg(z)1 2 dS z ,
which implies that
If (4.1.4) were false, then the above expression would be negative for sufficiently
small c: > o.•
4.1.1. Bergman Kernel. If we take g(z) = lo(z) - lo(a), then from

(4.1.4) for IE }(,1(D) we have
and if I = 10, then
111011
2
=
2
f llfo(z)1 dSz = fo(a) f 1 fo(z) dS z ·
If we introduce the Bergman kernel
lo(z)
K(z,a) = 111011 2 ' (4.1.5)
4.2. NUMERICAL METHODS FOR PROBLEM I 95
where fo(z) minimizes Problem I, then for every f E /C1(D)
f 1 K(z, a) f(z) dSz = f(a). (4.1.6)
Hence, every function f E L 2 (D) is the eigenfunction ofthe integral equation

(4.1.6) with eigenvalue). = fo(a) = 1. Then the minimal function fo(z) is, in
view of (4.1.5), given by
i ( ) = K(z, a) (4.1.7)
JO z K(a,a)'
where
K(a, a) = fl,K(Z, a)1 dSz
2 = Ilf~112' (4.1.8)
Note that we cannot find fo (z) directly. We can find f' (z) since it appears in
the integrand in (4.1.2). Then the mapping function f is related to the Bergman
kernel of D by
f(z) = lID K(z, a) dS z . (4.1.9)
K(a, a)
4.2. Numerical Methods for Problem I
We shall study the Ritz method (RM) and the Bergman kernel method (BKM)
but postpone until Chapter 12 the investigation into the nature and location of
boundary singularities and poles of the mapping function that are close to the
boundary.
4.2.1. Ritz Method. The Ritz method is used to find the solution of the
above extremal problem approximately in the form of a polynomial. Consider
an arbitrary system of linearly independent functions uo(z), Ul (z), ... , which
are regular in D and are such that f lIUk(z), 2 dxdy < +00 for k = 0,1, ....
We assume that one of these functions, say uo(z), is such that uo(a) -I- 0.
Without loss of generality, we take a = O. Let {<Pn(z)} be a complete set of
L 2 (D), and denote by /C~ and /C~ the n-dimensional counterparts of /CO and
/Cl, respectively, i.e., if
n
<Pn(z) = L Ck Uk(Z), (4.2.1)
k=O
then 4>n E K~ if 4>(0) = 0, and 4>n E K~ if 4>n(O) = 1, where Ck = CXk + i{3k,

Co =I- 0, are complex numbers. Now we shall consider the integral
(4.2.2)
which is the same as (4.1.1) except that the integrand in (4.2.2) is 4>n (z) instead
of f'(z).
PROBLEM In: In the class K~ minimize the integral I(4)n), defined by (4.2.2).
Now we shall discuss the existence and uniqueness of the minimal polyno-
mial 4>n(z), determine <Pn(z), and approximate fo(z) by the minimal polyno-
mial 4>n (z). The numerical value of the integral (4.2.2) is equal to the area of
the image of the region D. Then the problem reduces to a choice of the coeffi-
cients Ck so that this value is a minimum among the values of the same integral
for any other linear combination 1/Jn of the functions Uk(Z), k = 0,1, ... ,n,
subject to the condition 1/Jn(O) = 1. Suppose that
(4.2.3)
where c; is a complex number and 'Yn (z) is a linear combination of Uk (z), k =

0, 1, ... ,n. The requirement that 1/Jn (0) = 1 for all c; implies that 'Yn (0) = O.
Now
I(1/Jn) = il'4>n,2dXdy+t i l 4>n1n dxd y

(4.2.4)
2
+c; i l ¢n4>n dxd y+\c;1 ilhn,2dxdy.
The sign of the difference I(1/Jn) - I( 4>n) for small c; will depend on the linear
terms in c; and t because the last term in (4.2.4) is of order 0(c;2). Thus,
I(1/Jn) - I(4)n) ~ 0 iff the following orthogonality relations hold:
(4.2.5)
for any linear combination 'Yn that satisfies the condition 'Yn (0) = O. Otherwise,
we can always choose an c; which will make I(1/Jn) < I(4)n), and this will
contradict the minimal properties of 4>n(z). Note that the integrands in (4.2.5)
are complex conjugates of each other, so we can use either as needed.
Conversely, if cPn (z) satisfies the orthogonality relations (4.2.5), then cPn (z)
imparts the integral I (cPn) its minimum value among the values imparted by all
linear combinations of 'l/Jn(z) with 'l/Jn(O) = 1. Thus, from (4.2.4) and (4.2.5)
we get
(4.2.6)
Hence, the polynomial cPn (z) will be unique if it exists, since the integral on
the right side of (4.2.6) is equal to zero only when Tn = 0, i.e., when 'l/Jn = cPn.
Thus, the orthogonality relations (4.2.5) constitute necessary and sufficient
conditions for cPn (z) to be the minimal polynomial.
We shall rewrite the conditions (4.2.5) in a different but equivalent form.

Let
Uk(O)
vdz) = Uk(Z) - uo(O) uo(z), k = 1, ... ,n. (4.2.7)
Then each of the functions Vk(Z) satisfies the requirements imposed on Tk(Z),
and the conditions (4.2.5) become
n
2::= A kj Cj = 0, (4.2.8)
j=O
where
(4.2.9)
Also, since
n
2::=Uk(O)Ck = 1, (4.2.10)
k=O
Eqs (4.2.8) and (4.2.10) provide us with a system of (n + 1) equations to

determine the numbers CO, Cl, ... ,Cn uniquely.
Note that the functions Uk(Z), though linearly independent, are still unde-
termined and are not subject to limitation. However, it becomes very easy to
determine the integrals in (4.2.9) if all Uk(Z) are suitably chosen beforehand
such that uo(O) = 1, and Uk(O) = 0 for k = 1, ... ,n. Then Vk(Z) = Uk(Z),
and the integral
Also, Eq (4.2.10) degenerates to CO = 1, which reduces the number of unknowns

by one. Moreover, if the system of functions uo(z), Ul(Z), ... is orthogonal in
J1
D, i.e., if
Uj Uk dx dy = 0 for k =I- j,
J1
then
Uj Vk dx dy =0 for j =I- 0 and k =I- j,
and
J"iDr Uovkdxdy = - Uk(O) J" r uouodxdy,

uo(O) iD
J1 Uk Vk dxdy = Jl Uk Uk dxdy.
Then Eqs (4.2.8)-{4.2.1O) simplify to
Ck
JJD
IUkl 2 dxdy - Co =Uk(O)
Uo(O)
JJ D
Iuol 2 dxdy = O. (4.2.11)
Thus, we have proved
THEOREM 4.2.1. The problem In has a unique solution. The coef-

ficients of the minimal polynomial 1Jn(z), defined by (4.2.1), are deter-
mined by the system of linear equations (4.2.8)-(4.2.9). The coefficients
A kj , defined by (4.2.9), are hermitian.
Let the system {Uk}k=O be taken as the complete set {I, z, z2, ... , zn}.
Then for the minimal polynomial
(4.2.12)
the system of equations (4.2.8)-(4.2.9), which determine the n coefficients Ck,

becomes
A 1n
1)
~::J : ~
Cl
(4.2.13)
1 {O},
where the coefficients A kj are given by
(4.2.14)
Note that if a =1= 0, then the coefficients A kj are determined from
(4.2.15)
The computation of the coefficients Akj depends on the region D, although it

may sometimes present difficulties.
CASE STUDY 4.2.1 Let the region D be starlike with respect to a point
a =1= 0 E D, i.e., every ray emanating from the point a intersects the boundary
in only one point. Let the equation of the boundary be r = r(B). Using the
polar coordinates z - a = r eiO , we get
r27r ((OJ j k
Akj = io i r + ei(j-k)() r dr dB
o 27r
=. 1 r rHk+2(B) ei(j-k)() dB
J+k +2 io
= 1
. 1 27r
rJ+k+2(8)
. casU - k)8d8
(4.2.16)
. 1
J+k+2 0
27r
+. t rHk+2(B) sinU - k)Bd8.
J+k+2 0
Note that ~ { A kj } and ~ { Akj } differ from the coefficients of the Fourier series
for r Hk + 2 (B) by a factor j + ~ + 2' and hence, they can be easily computed.•
DEFINITION 4.2.1. The system of polynomials {¢in} is said to be com-

plete in the Hilbert space L 2 (D) if for every function f E L 2 (D) and every
c > 0 there exists a polynomial ¢in such that Ilf - ¢in II < c.
Now the question arises, under what additional assumptions on D the poly-
nomials ¢in(z) form a complete system in L 2 (D). Naturally, lifo - ¢in II '\.. 0
must hold, and thus also ¢in(z) ----> fo(z) as n ----> 00 in any closed subset
(; cD.
THEOREM 4.2.2. Let the polynomial p(z) belong to the class K;.
JL
Then
lifo - pl12 = Ifo(z) - p(zW dSz (4.2.17)
is minimal only if p(z) = cPn(z).
PROOF. In view of (4.1.4)
(fo,p) = (fo,P - fo) + (fo,fo) = (fo, fo),

thus,
(fo - p, fo - p) = (fo, fo) - 2 (fo, fo) + (p,p) = IIpl12 -llfoI/ 2,

and the result follows since IIpl12 is minimal if p = cPn .•
Hence, among all p E K;

the polynomial cPn yields the minimum norm
1/ pll, and since K;+l J K;, then lifo - pll "\. 0 only if the system
fo -
of polynomial {cPn} is complete in the space L 2 (D). Moreover, in Problem
In there exists a polynomial p = cPn which satisfies the additional condition
p(a) = f(a). Also note that the Bieberbach polynomial 1Tn (Z) defined by
1Tn (Z) = l z
cPn-l(() d( --+ f(z) as n --+ 00 in G cD. (4.2.18)
y
A, :j 1 , B
, I I
I I I
, , ,,
,, ,
I , I
I I I
I I I \
I
: : : X j(z) I
\
\
f J __ • I
-I: 0: :1
I ~
I
\
\
o
,
I I I
I I I \
I , ,
I , I \
,,
I
,
I
I
,
, ,
D'--------~_I-------'
,
c
Fig. 4.2.1.
CASE STUDY 4.2.2. Determine the minimal polynomial cPn(z) and the
approximate mapping function f (z) that maps the square region D = {x, y :
-1 < x, Y < I} conformally onto Iwl < 1 (Fig. 4.2.1). From cs422.nb (see
Notes at the end of this chapter), the minimum polynomial is given by
97402305 4 68765697 s
¢s(z) = 1 + 266254834 z + 2130038672 z ,
which yields the approximate mapping function
j(z) ~ 1 z
¢s(t) dt
19480461 5 7640633 9 (4.2.19)
= z + 266254834 z + 2130038672 z
~ z + 0.0731647 Z5 + 0.00358709 Z9 . •
4.2.2. Bergman Kernel Method. In this method the mapping func-

tion j(z) is determined approximately from (4.1.19) by first approximating
the kernel K(z,O) by a finite Fourier sum. Let {¢j(z)} denote a complete
orthonormal set of L 2 (D). Consider the Fourier series expansion of K(z, 0).
Then, in view of (4.1.6),
(K,¢j) = /1 K(z,O)¢j(z)dSz = ¢j(O). (4.2.19)
Thus, the kernel has a series expansion
L ¢j(O) ¢j(z),
00
K(z,O) = (4.2.20)
j=l
which converges in the mean of L2 (D), i.e., the series (4.2.20) converges almost
uniformly in D.
Hence, if we have a complete set {¢j(z)} of L 2 (D), then by using (4.1.19)

and (4.2.20) we obtain an approximate mapping function j(z) as follows:
(i) Orthonormalize the set {¢j(z)};=l by using the Gram-Schmidt process
which yields the set of orthonormal functions {¢j(z)} 7=1' Note that the Gram-
Schmidt process requires evaluating ( ¢i, ¢j ), which, in view of Green's formula
(1.1.29), is given by
(4.2.21)
where 'l/;j (z) = ¢j (z). Then the integrals in (4.2.21) are computed by Gaussian
quadrature.
(ii) Truncate the series (4.2.20) after n terms to obtain the approximation
Kn(z,O) of K(z, 0) as
n
Kn(z,O) = L ¢j(O) ¢;(z). (4.2.22)
j=1
(iii) Use (4.1.19) to obtair the approximate mapping function fn(z) as
(4.2.23)
(iv) The approximate radius R n of the disk Iwl < R is given by
(4.2.24)
since II fa 11 2 = 1r R 2 (because fa E 1(1).

(v) Thus, from (4.2.23) and (4.2.24) the approximation of the mapping function
F(z) that maps D conformally onto the unit disk Iwl < 1 is given by
(4.2.25)
The maximum error estimate for IFn(z) I is given by
where Zj E r are the test points on the boundary and en(z) = 1 - IFn(z)l.
During the computation process the number n of the basis function is increased
by one each time and this process is terminated when the inequality En+! < En
no longer holds. Then such a number n is taken as the 'optimum number' for
the basis functions.
Note that in both RM and BKM we have obtained approximations of the

form
n
fn(z) = L aj Uj(z),
j=1
where uj(z) = cPj(z). In both methods the set of monomials zj-l, j

1,2, ... , which is a complete set in L 2 (D U r) is the best choice of basis
4.3. MINIMUM BOUNDARY PROBLEM 103
functions in computation. Then this basis gives the polynomials ¢j (z) defined
in (4.2.13).
CASE STUDY 4.2.3. The function w = fez) that maps the unit disk U
onto itself such that the point Zo E U goes into the origin of the w-plane is
given by
fez) = z - Zo .
z-l/zo
Thus, both fez) and the associated Bergman kernel function K(z, zo) have a
pole at z = 1/20. Since the polynomials ¢j(z) = Jj
/n zj-1, j = 1,2, ... ,
form a complete orthonormal basis set of U, the kernel K(z, zo) can, in view
of (4.2.20), be represented by the polynomial series
~ .-1
K(z, zo) = -1 LJJ (20 zr ,
n j-1
which converges rapidly when Izol is small, but the convergence becomes con-
siderably slower the faster Izol -+ 1, i.e., the closer the pole 1/20 gets to the
boundary of U.•
4.3. Minimum Boundary Problem
An analogous minimum problem in the conformal mapping of a region D onto

Iwl < R leads to another characterization of the mapping function fez) by
considering the line integral
1= i If(zW ds, (4.3.1)
where r is the boundary of D. This problem, studied by Julia (1931), is

known as the minimum boundary problem which we shall call Problem II. Let
r be a rectifiable Jordan curve, and let £1 (r) denote the class of all functions
f E L 2 (r) with f(a) = 1, where a E Dcan be taken as the origin.
PROBLEM II: In the class £l(r) minimize the integral (4.3.1).

THEOREM 4.3.1. Problem II has a unique solution lo(z), and it is

lo(z) = JI'(z). The minimum is 2-rrR.
PROOF. For every function FE L 2 (f)
rjF(zW ds = r-+RJrr IF(zW ds

Jr
lim
r
= r IF(g(w)) Jg'(w)1
lim
r-+RJ/wl=r
2
Idwl
r
27r
= lim r Ih (r ei &) 2
1 de,
r-+R Jo
where h(w) = F(g(w)) Jg'(w), h(O) = 1, and z = g(w) (see (1.1.26». If
00
h(w) = Lan wn , ao = 1, then

n=O
l lF
r
(zW ds = 2-rr f
n=O
la n J2 R 2n +l 2': 2-rrR,
where the equality holds only for an = 0, n > 0, which yields F( z) = J I' (z)
for h(w) = 1..
This theorem implies that in the class of all conformal mappings w = ¢( z)

oftheregion D with ¢(a) = 0, ¢'(a) = 1, the integral[I¢'(z)1 ds is minimum
only when ¢(z) = I(z). This is known as the principle of minimizing the image
boundary.
The conformal map I of D onto Iwl < R, normalized by I(a) = 0,

f'(a) = 1, is given by
I(z) = l z
[/o(()f de·
The theory for Problem II is developed exactly on the same lines as in §4.1
and 4.2. Thus, as in (4.1.4), we have
THEOREM 4.3.2. The lunction lo(z) is orthogonal to every function

9 E L 2 (f) with g(a) = 0, i.e.,
\!o,g) = /rfo(Z)9(Z)dS=0. (4.3.2)

4.4. RITZ METHOD FOR PROBLEM II 105
For the minimal function fo(z) of Problem II we introduce the Szego kernel
function
fo(z)
S(z,a) = Jr Ifo(z)1 2 ds' (4.3.3)
with the properties
S(z, a)
S(a,a) = [IS(z,a) 12 ds, and fo(z) = S(a, a) . (4.3.4)
For any f E L 2(r), we apply (4.3.2) to f(z) - f(a) and get
[fo(z) f(z) ds = f(a) [fo(z) ds,
and if f = fo, then, since h(w) = 1 implies f(a) = 1 = f(a), we have
[ Ifo(zW ds = [fo(z) ds.
Thus, analogous to (4.1.6) we have: For every function f E L 2 (r)
[ S(z, a) f(z) ds = f(a). (4.3.5)
Theorems 4.1.1 and 4.3.1 together with the definitions (4.1.5) and (4.3.3) yield
K(z, a) = 411' [S(z, a)]2 . (4.3.6)
Thus, S(z, a) can be evaluated by the method of §4.2. A relation between the
Cauchy and Szego kernels and their application to the problem of conformal
mapping is presented in §7.6.
4.4. Ritz Method for Problem II
Let .c; denote the class of all polynomials p( z) of degree::; n with p(a) = 1.
PROBLEM lIn: In the class .c; minimize the line integral h Ip(z)1 2 ds.
We shall discuss the existence and uniqueness of the minimal polynomial
l
<l>n(z), determine <l>n(z), and approximate fo(z) by the minimal polynomial
z
<l>n(z) and f(z) by the integral [n(()]2 d(, respectively. As in §4.2, it
can be shown that a minimal polynomial <l>n(z) exists for Problem II, that it is
unique, and that it is characterized by
(4.4.1)
for every polynomialg(z) of degree ~ n with g(a) = O. The proof is analogous

to that of (4.1.4).
In view of (4.4.1), the coefficients of the minimal polynomial

<l>n(z) = 1 + al (z - a) + ... + an (z - a)n (4.4.2)
are determined by
l (~adz - a)k) (z - a)j ds = 0, j = 1,2, ... ,n, (4.4.3)
and ao = 1. If we set
B kj = l (z - a)k (z - a)j ds, k,j = 0,1, ... , (4.4.4)
then the coefficients of the minimal polynomial <l>n(z) defined by (4.4.2) are
determined from the (consistent) system of equations
n
L Bkj ak = 0, ao = 1, j = 1, ... , n. (4.4.5)
k=O
For any arbitrary polynomial PEL;, we have, in view of (4.2.2),
(fa - p) = (fo,P - fa) + (fa, fa) = (fa, fa),

and hence
11lfo - pII 2 = IIpI1 2 -llfoI1 2 , (4.4.6)
which implies that p( z) has the minimal property:
llfo(z) - p(z)1 2 ds is minimum in L~ for p(z) = <l>n(z). (4.4.7)
Again, as in §4.2, lifo - <l>nll '\, 0 as n ~ 00, since L;+l ~ Then we ask, L;.
under what assumptions on D is the system of polynomials {n} complete in
the Hilbert space L 2 (f)? An answer was given by Smirnov (1928) as
4.4. RITZ METHOD FOR PROBLEM II 107
THEOREM 4.4.1. The system of polynomials in L 2 (f) is complete iff

r of the region D satisfies the condition
1
the boundary
2
1 R2 - r2
loglg'(w)I=-2
Ii 0
11"
log 19' (re t6 ) I R2 -
•
2R (
rcosa- B) +r 2 da ,
(4.4.8)
for all r < R, where z = g(w) is the inverse mapping function which
maps the region Iwl < R onto the region D in the z-plane.
A proof of this theorem can be found in Goluzin (1957, p.396; or 1969,

p.449). The condition (4.4.8) is known as the S-eondition. This condition
depends only on the region D and not on the normalization of g( w) or on
the choice of a E D. All such regions D whose mapping satisfy the S-
condition (4.4.8) are said to belong to the Smirnov class S. Not all regions
with a rectifiable boundary belong to the class S. Besides the rectifiability of r,
however, it is sufficient for DES if one of the following conditions is met: (i)
D is convex or starlike with respect to a point in D; (ii) r is piecewise smooth,
and its smooth arcs lk, k = 1, ... ,n, join one another with a nonzero interior
angle; (iii) the ratio of the length of any arc lk of r = u lk to the length of
its chord does not exceed a fixed limit; (iv) if D rf- S, then the behavior of
II¢n11 2 and of lifo - ¢n11 2 = II¢n11 2 - Ilfol1 2 is known; and (v) for any D with
a rectifiable boundary r, the integral t¢n (z) 12 ds \, 21i RfJ, where
l
{2~ 1
2
fJ = exp 11" log 19' (Re i6 ) I dB} . (4.4.9)
Thus, under the conditions (i), (ii), or (iii) we have lifo - ¢n II \, 0, and, in
view of Theorem 4.1.2, ¢n(z) --+ fo(z) as n --+ 00 for every region G c D.
Hence, for z E G c D the polynomial
li2n+l = l z
[¢n(()]2 d( --+ f(z) as n --+ 00. (4.4.10)
Note that the polynomialli2n+l (z)is different from the Bieberbach polynomial
(4.2.18).
THEOREM 4.4.2. For the system of polynomials to be complete in a

region D, it is necessary and sufficient that an arbitrary function F(z) E
L 2 (f) satisfies the condition
tlF(zW ds = ~ lak\2, (4.4.10)

where ak are the Fourier coefficients of F(z).
PROOF. Let p(z) be an arbitrary polynomial of degree n:

n
p(z) = L Ck Uk(Z). (4.4.11)
k=O
Then for F(z) E L 2 (f)
llF(z) - p(zW ds = lIF(z)!2 ds + 1,p(z)1 2 ds
-2SJ~{1 F(z)p(z)ds}
= lIF(z) 12 ds+ ~ickI2-2~{~akCk}

= llF(ZWdS- ~ICkI2+~Iak -ckI 2.
(4.4.12)
Thus, the polynomial p(z) defined by (4.4.11) with Ck = ak attains the mini-
mum of the integral
2 (4.4.13)
llF(z) - p(z)1 ds,
and, in view of (4.4.11), this minimum yields
(4.4.14)
Hence, a system of polynomials is complete iff the difference (4.4.14) ap-

proaches zero as n --. 00 for an arbitrary function F E L 2 (f) .•
THEOREM 4.4.3. If the S-condition (4.4.8) is satisfied, then an ar-

bitrary function F(z) E L 2(f) can be represented in D by a Fourier
series
L ak Uk(Z),
00
F(z) = (4.4.15)
k=O
which converges uniformly in D, where
(4.4.16)
4.5. ORTHOGONAL POLYNOMIALS 109
PROOF. Since the minimum of the integral (4.4.13) is attained out of

all polynomials Pn(z) by a polynomial defined by (4.4.11) with Ck = ak,
k = 0,1, ... , n, then, if the S-<:ondition is satisfied, we have
lim
n->oo Jrr IF(() - Pn(()1 2 ds = O. (4.4.17)
But by Cauchy's formula in D
F( Z ) - Pn ()
z = _1
2.
t1r
1
r
F(()I" - Pn(() dl".".
." - Z
If G is a closed subset of D and 8 is the distance from G to r, then for z E G
IF(z) - Pn(z)1 ~ ~8 r IF(() -

21r Jr
Pn(OI ds
r---------- (4.4.18)
~ 2~8 l [IF(() - Pn(()1 ds,
2
where l is the length of r, and this, in view of (4.4.17), implies that Pn (z) ~
F(z) uniformly on G as n ~ 00. But Pn(z) is a finite part of the Fourier series
(4.4.15), which proves the theorem.•
4.5. Orthogonal Polynomials
Functions analytic in a region D can always be represented by Taylor series

only if D is a circular disk. If the boundary r of D is not a circle, such a series
representation is not possible. Therefore, we must find a sequence of functions
which depend only on the region D so that any analytic function in D can be
expanded in the form of a series involving functions from this sequence. It turns
out that all such functions are certain polynomials of a special form which are
either orthogonal on the boundary r or in the region D. They not only provide
a series expansion for analytic functions on D but also playa significant role
in conformal mapping.
4.5.1. Polynomials orthogonal on the boundary. We shall

analyze the structure of polynomials that are orthogonal on a contour. Let
r be an arbitrary rectifiable curve (not necessarily closed) of length l. By

using the Gram-Schmidt orthogonalization process (see Gaier, 1964, p.132,
for an algorithm for this process) we construct a sequence of polynomials
{ Po (Z), PI (z), ... ,Pn (Z), ... } with the following properties:
(i) Pn(z) is a polynomial of degree n in z;
(ii) the coefficient of zn in Pn (z) is positive; and
(iii) the polynomials Pn (z) are orthonormal (orthogonal and normalized) along
the curve r, i.e.,
~ l Pn(z) Pm(z) ds = °nm, (4.5.1)
where onm is the Kronecker delta. We shall introduce the constants
hpq
1 r
= T Jr z z ds.
p-q
(4.5.2)
Note that ds = Idzl and h pq = hqp . Consider the positive-definite hermitian

quadratic forms
n
Hn(t) = L hpq t p tq
p,q=O (4.5.3)
=~ lito + tl z + ... ,+tn znJ2 ds,
with determinants D n defined by
h oo h lO
hOl h ll
Do = 1, Dn = (4.5.4)
Then the Szego polynomials <J n (z) are represented by
h oo h lO h no
1 hOI h ll h nl
<In(z) = (4.5.5)
JDn-IDn h On- l h ln - l h nn - l
1 z zn
It can be verified that these polynomials possess the above three properties. As
regards the question of expansion of an arbitrary analytic function in a series
involving Szego polynomials, the following result due to Smirnov (1928) holds:
4.5. ORTHOGONAL POLYNOMIALS III
THEOREM 4.5.1. Suppose that a junction j(z) is analytic inside a

region D bounded by a Jordan curve f, has almost everywhere boundary
values on f, and can be represented in terms oj these boundary values
by Cauchy integrals. Then j(z) can be expanded in a series involving
Szego polynomials:
00
j(z) = LAn (/n(Z), (4.5.6)

n=O
which is unijormly convergent everywhere within f,and the coefficients
An are determined by
1 r
An = T ir j(z) -(/n(z) ds. (4.5.7)
For a proof of this theorem see Smirnov (1928).
As an application of Szego polynomials to conformal mapping, let w =

F(z) map the region D in the z-plane onto the disk Iwl < R such that a point
a E D goes into the origin w = 0, F(a) = 0, and F'(a) = 1. In view of
Theorem 4.3.1, out of all functions F(z) analytic on D and normalized at a by
F(a) = 1, the function J j'(z) minimizes the integral (4.3.1), i.e.,
(4.5.8)
Using the series expansion (4.5.6) for the function F(z) in terms of Szego
polynomials (/n(Z), and using F(a) = 1, we find that
00
LAj(/j(a) = 1. (4.5.9)
j=O
Then, from (4.5.8)
1
T r -
ir F(z) F(z) ds = ~
L..,.
-
Aj Aj . (4.5.10)
r j=O
J
Then the system of coefficients corresponding to the function F( z) = j' (z)
attains the minimum value for the sum in (4.5.10) such that the condition (4.5.9)
is satisfied. We shall denote these coefficients by OJ, and set A j = OJ + cTJj.

Since, in view of (4.5.9), the condition
00
LOjO"j(a) = 1 (4.5.11)
j=O
still holds, the numbers TJj must be such that
00
L TJj O"j(a) = 0. (4.5.12)

j=O
Then, from (4.5.10)
00 00
(4.5.13)
+€ L iij 8j + Icl 2
L 7)j iij'
j=O j=O
L OJ 8
00
Since the expression on the right side in (4.5.13) must be less than j ,
j=O
which is the minimum value of the integral (4.5.8), it is necessary and sufficient
that the coefficients of c and € vanish for all TJj subject to the condition (4.5.12),
i.e.,
°= Lj=O iij 8j .
00 00
L TJj 8j = (4.5.14)
j=O
00
Now, from (4.5.12) we get TJo = - L TJj O"j(a), since O"o(z) = 1, and
j=O
00
L 7)j [8j - 80 0"j(a)] = 0, (4.5.15)

j=O
which is valid for arbitrary 7)1, 7)2, . .. only if
8j = 800"j(a). (4.5.16)
Substituting the values of 8j from (4.5.16) and (4.5.11) we get

00
80 L O"j(a) O"j(a) = 1. (4.5.17)

j=O
Set
00
S(z,a) = LiTj(a) iTj(z). (4.5.18)

j=O
Then we have
80 = 80 = _1_ = iTo(a)
S(a,a) S(a,a)'
8. _ iTj(a)
J - S(a,a)'
(4.5.19)
/( ) _ 1 ~-(-) () _ S(z,a)
V rp:'
J~z} - S-() LJiTj a iTj z - S-()'
a,a j=O a,a
1 a)
f(z) = S2(a, la
z
S2(z, a) dz,
where S(z, a) is the Szego kernel defined in §4.3. In order to derive an approx-
imate formula for f(z), we shall assume that only n Szego polynomials iTj(z)
are known. Then
n
Sn(z,a) ~ LiTj(a) iTj(z), (4.5.20)
j=O
l
and z
f(z) ~ S(a,a)
1 a S2(z,a)dz. (4.5.21)
The radius R of the disk Iwl < R is given exactly by
Then obviously the function g( z) that maps the region D onto the unit disk U
l
is given by
z
27T a)
g(z) = l S2(a, a S2(z, a) dz. (4.5.23)
CASE STUDY 4.5.1. We shall determine the mapping function F(z) that
maps the square {-1 ::; x, y ::; 1} onto the disk Iwl ::; R (see Fig. 4.2.1).
Sincez = x+ionAB, z = x-i on DC, z = l+iyon C B, andz = -l+iy

on D A, the numbers
h pq = ~ {jl [(x + i)P(x - i)q + (x - i)P(x + i)q] dx

8 -1 (4.5.24)
+ [(1 + iy)P(l - iy)q + (-1 + iy)P( -1 - iyF] dy}
are computed in cs422. nb for p, q = 0, 1, ... ,8. If Mathematica is not used,

then (4.5.24) can be written as
and then the numbers hpq can be evaluated with the same values as in (4.5.25).
Now, from (4.5.5)
Since all O'j(O) are zero except for j = 1,4, we find from (4.5.18) that
415
8(0,0) = 352' and thus, from (4.5.19)
1 t 2
f(z) ~ 8 2 (0,0) io 8 (z, 0) dz
63 5 441 9 (4.5.27)
= z + 830 z + 110224 z
~ z + 0.0759036 z5 + 0.004 z9,
which can be compared with (4.2.19). Let z = 1>(w) be the inverse function
of w = f(z) such that 1>(w) maps the circle Iwl = R onto the given square,
and 1>(0) = 0,1>'(0) = 1. By using the Schwarz-Christoffel transformation
analogous to Case Study 2.3.2, the function z = 1>(w) is represented by the
elliptic integral
(4.5.28)
where
1
1 d(
k = ~ ~ 0.927037. (4.5.29)
o Vi + (4
On inversion, (4.5.28) yields
k4 k8 11k 12
w= f( z ) =z+-z 5 + - z9 + - - z 13 + ... (4.5.30)
10 120 15600 '
(see, e.g., Gaier, 1964, p.148.) A comparison of (4.5.27) and (4.5.30) shows
k4 63 4(63 .
that 10 = 830' or k = V 83 ~ 0.933395, WhiCh, after comparing with the
value of kin (4.5.29) shows that the polynomial approximation of f(z) has an
error of 0.636%. This means that the polynomial f(z) maps the boundary ofthe
square onto some curve that does not quite coincide with the circle Iwl = R.
In order to determine the closeness of this curve to the circle Iwl = R, we
evaluate If(I)1 and If(1 + i)l, which are given by If(I)1 = 1.0799036, and
If(1 + i)1 = 1.075368896, which shows that the radius of the circle onto
which the square is mapped by the approximate polygon lies between these
two values. However, from (4.5.22) we find that R = 11k ~ 1.078705, which
gives a maximum error of at most 0.5% of the value of R.
The polynomial that maps the given square onto the unit disk U can be
determined from (4.5.23). The exact solution is given by the elliptic integral
,hW(1 + t 4 )-1/2 dt
z = .::...0"=-
I; (1 + t 4)-1/2 dt
_
(4.5.31)
1 5 1 9 5 11 )
~ 1.08 ( w - 10 w + 24 w - 208 w +... .•
The exact mapping function is known in terms of Jacobian elliptic functions

(see Case Study 2.3.2).
4.5.2. Polynomials orthogonal to a region. Let D be, as be-

fore, a simply connected region with a Jordan boundary r and area A. Using
the Schmidt orthogonalization process, we construct a system of polynomials
{Qo(z), Q1(Z), ... ,Qn(Z), ... }, with the following properties:
(i) Qn(z) is a polynomial of degree n in z;
(ii) the coefficient of zn in Qn(z) is positive; and
(iii) the polynomials Qn(z) are orthonormal (orthogonal and normalized) along
the curve r, i.e.,
(4.5.32)
These properties are similar to those in §4.5.1, except that the line integral is
now replaced by the surface integral. We introduce the constants
"Ipq -
- 1
A JrrJD zp-qd
z X dy, (4.5.33)
and, analogous to (4.5.4), define the determinants ~n by
)'00 )'10 )'nO

"101 "Ill )'n1
~o = 1, ~n = (4.5.34)
"IOn )'In "Inn
Then the polynomials
"100 "110 )'nO

1 "101 "Ill "In 1
lln(z) = (4.5.35)
.J~n-1 ~n )'On )'In
1 Z
are orthogonal in the region D, and form a complete closed system. Any
function f (z) analytic on D such that the integral
Jllf(ZW dxdy < +00

can be uniquely expanded in a series involving the polynomials IIn(z), i.e.,

00
f(z) = 2:>l!n IIn(z), (4.5.36)

n=O
where the coefficients an are determined by
(4.5.37)
As' an application, note that, in view of §4.1, the function F( z) = fo (z) E

K} maps the region D conformally onto the disk Iwl < R such that a point
a E D goes into w = 0 and f~ (a) = 1. Out of all analyticfunctions F (z) E K}
with F(a) = 0 and F'(a) = 1, the function fo(z) gives the minimum for
the integral (4.1.2). Analogous to Theorem 4.5.1 the function fo(z) can be
represented in a series expansion involving the polynomials IIj(z) as
fo(z) = K(a,1 a) l a
z
K(z, a) dz, (4.5.38)
where, as in (4.1.5),
00
K(z,a) = LIIj(a)IIj(z). (4.5.39)

j=O
Then the area of the circle Iwl = R is given by
(4.5.40)
whence
R= J 1T K1a, a) ,
(4.5.41)
and the mapping function is determined by
~() l
z
fo(z) = K(a, a) dz. (4.5.42)
Y~a
4.6. Problems
PROBLEM 4.6.1. Let w = f(z) map a finite, simply connected region D

onto the unit disk Iwl < 1. Show that the functions
cPn(Z) = In: 1 [f(z)t f'(z), n = 0, 1, ... ,
form a complete orthonormal set in D. (Nehari, 1952, p.247.)
PROBLEM 4.6.2. Show that the Bergman kernel K(z, a) can be expanded
into the infinite series (4.2.20) which converges absolutely and uniformly in
any closed region contained in D. (Nehari, 1952, p.251, 256.)
x2 y2
PROBLEM 4.6.3. Let E denote the ellipse a 2 + b2 = 1. Show that the
Bergman kernel for E has the form
where Un(z) = (1 - z2) -1/2 sin ((n + 1) cos- 1 z) are the Chebyshev poly-
nomials of the second kind and degree n. (Nehari, 1952, pp.258-259.)
PROBLEM 4.6.4. Prove that if r satisfies the S--condition, then
where the convergence of the series in the numerator is uniform in D. [Let

F(z) = J
f'(z). Use (4.4.16) and show that an = 27l' un(O) f'(O).] J
(Goluzil'\, 1969, p.453.)
PROBLEM 4.6.5. Let the arc r be defined by the ellipse E: (x - :c)2 +

a
(y _ yc)2
b2 = I, a > b, and let the parametric equation of r be z = ')'(8) =
4.6. PROBLEMS 119
Zc + ae COS(8 - iq), 0 ~ 81 < 8 < 82 < 271", where Zc = Xc + iyc is the

center C , e = y'1 - b2 / a 2 the eccentricity of the ellipse, cosh q = 1/e, and
82 - 81 < 271". Show that the function z = -y((), ( = 8 + it, is univalent in
the strip {(: (+ 8 + it, 81 < 8 < 82, -00 < t < q}, and the region G* is a
symmetric subregion of the rectangle {( : ( = 8 + it, 81 < 8 < 82, -q <
t < q}. (Papamichael, Warby and Hough, 1983, p.157.)
PROBLEM 4.6.6. Show that the function
w = f(z) = .Jk(q) sn (2: sin- 1 z, q) , a-

q= ( a+b
b)2
'
maps the interior of the ellipse :2 + ~2

2 2
= 1 onto the unit disk Iwl < 1, such
that the foci of the ellipse go into the points w = ±y'k(q), and
l' (0) = a ~ b (~qn(n+1») . (1 + 2 ~ qn)

1.0165984 for alb = 1.2,
= 1.2376223 for alb = 2,
{
2.372368 for alb = 5.
(Nehari, 1952, p. 296; Gaier, 1964, pp.16G--161.)
NOTES. cs442.nb:
A[j., k.] := Integrate[ Integrate[ (x+ I*y)" j * (x-I*y)"k,
{x, -l,l}], {y, -l,l}];
MatA = Tab1e[A[j,k], {j,l,8}, {k,l,a}];
MatrixForm[MatA];
B=Table[A[j,O] , {j,l,a}];
c=LinearSolve[MatA,· B];
(* These are the coefficients of phi.a[z] *)
phia[z.] := 1 + c. Table[z"i, {i, a}];
phia [z] ;
(* The mapping function is given by f' [z]=phi8[z] *)
f[z.] := Integrate [phi8[t], {t, O,z}];
Hz]
REFERENCES USED: Gaier (1964), Goluzin (1957, 1969), Kantorovich and

Krylov (1958), Nehari (1952), Papamichael, Warby and Hough (1983).
Chapter 5
Nearly Circular Regions
We shall investigate methods for constructing a mapping function for con-

formal mapping of a simply connected nearly circular region onto a disk. A
classical method that involves series expansion in powers of a small param-
eter for the interior and the exterior regions, known as the method of infinite
systems, is presented with case studies, in which successive approximations
are used to compute the approximate mapping function.
5.1. Small Parameter Expansions
Let a one-parameter family of Jordan curves fA in the z-plane be defined by
fA: z = Z(t,A), (5.1.1)
where t and A are real parameters (see §l.l). Let the origin z =
all of these curves. Further, let the function
° lie inside
w = f(Z,A) (5.1.2)
map the region D A bounded by the curve fA conformally onto a disk Iw I < R
in the w-plane, i.e. , the function (5.1.2) must also satisfy conditions
f(O, A) = 0, 1'(0, A) = 1. (5.1.3)
Note that if the function z = z(t, A) which defines the boundary curve fA'
where the parameter t defines the position of the point z on fA' is an analytic
120

5.1. SMALL PARAMETER EXPANSIONS 121
function of A in the neighborhood of some value of A, say A = 0, then the

mapping function (5.1.2) can also be regarded as an analytic function in that
neighborhood. Thus, the function f(z, A) for any zED>. can be expanded in
a power series in A as
00
f(Z,A) = fo(z) + LAn fn(z), (5.1.4)

n=1
which converges for sufficiently smalllAI. In order to compute the coefficients
fn(z), we know from (5.1.3) that fn(O) = 0 for n = 0,1,2, ... , fMO) = 1
and f~(O) = 0 for n = 1,2, .... Thus, fo(z) = f(z, 0), which implies that the
function w = fo(z) maps the region Do bounded by f o exactly onto the disk
Iwl <R.
Now, to compute fn(z) for n ~ 1, let us consider a system of functions

{Un(z)} :=1'which are analytic in a region D containing all D>. for sufficiently
smalllAI, such that un(O) = 0 for n = 1,2, ... , u~ (0) = 1, and u~(O) = 0 for
n = 2,3, .... Then any function f(z, A) analytic on D can be expanded in a
series involving Un (z). Thus, let
00
f(z, A) = U1(Z) +L On(A) un(z), (5.1.4)

n=2
where the coefficients On(A) depend only on A. Hence the problem of de-
termining the function f(z, A) reduces to that of computating the coefficients
On(A), which are, in fact, solutions of an infinite system of equations.
However, in practical problems, the function f(z, A) is represented ap-

proximately by taking a finite sum in (5.1.4). Then the coefficients On(A) are
determined by solving a finite system of equations. The details of computation
are as follows (Kantorovich and Krylov, 1958): Let
n
Un(z) = U1(Z) + L OJ(A) Uj(z) (5.1.5)
j=2
2
denote a partial sum of the series (5.1.4). Then IUn(z) 1 can be expanded in a
trigonometric series in t E [0,211") :
00
I
2
Un (z) 1 = aO +L (an COS nt + bn sin nt) . (5.1.6)
n=1
122 CHAPTER 5. NEARLY CIRCULAR REGIONS
This means that the coefficients an and bn in this expansion are quadratic
functions of aj(A). We can choose that all coefficients an(aj) and bn(aj) are
zero for n = 1,2, ... , or that only the first (n - 1) coefficients an (aj) and
bn (aj ) are zero. In theformer case we get an exact determination ofthe function
fez, A). But in the second case we obtain a system of (2n - 2) equations
(5.1.7)
which shall determine the (n - 1) unknown complex coefficients aj(A). The

form of the partial sum Un(z), defined by (5.1.5), depends on the choice of the
system of the functions Un (z). If we take the system {un (z)} = {zn}, then
Un(z) becomes a polynomial of degree n:
(5.1.8)
Let us assume that the boundary r of the region D is nearly circular and is
defined by
(5.1.9)
where F( T, A), T = eit , is an analytic function of its arguments for ITI close to
1 and A close to O. Then F( T, A) can be expanded in a Laurent series in T:
L L
00 00
F(T, A) = f3v(A) TV = f3v(A) eivt ,

v=-oo v=-oo
where the coefficients f3v(A) are analytic functions of A. Then from (5.1.9) the
boundary r is defined by
00
(5.1.10)
v=-oo
where
f3~l)(A) = {Af3v-l(A) for v =f 1, (5.1.11)
1 + Af30(A) for v = 1.
The k-th power of z(t), defined by (5.1.10), is given by
L
00
zk(t) = f3~k)(A)eivt, (5.1.12)

v=-oo
5.1. SMALL PARAMETER EXPANSIONS 123
where f3&k) (0) = 8vk . Thus, from (5.1.8) and (5.1.12)

2
[Pn(z)1 = Pn(Z),Pn(z)
n
= 2: CXk&j zk zj, (where CXl = 1)
k,j=l (5.1.13)
00 n 00
v=-oo k,j=l p,q=-oo

p-q=v
Note that the right side of (5.1.13) represents a trigonometric series, whose
coefficients depend on CXk. If we denote the free term (corresponding to v = 0)
on the right side of (5.1.13) by R 2 , we get
n
L
00
R2 = 2: cxkiij f3~k)().,)i3~j). (5.1.14)

k,j=l p=-oo
We will choose the coefficients CX2, ... , CXn, such that the coefficients of eit , e 2it ,
... , e(n-l)it are zero, i.e.,
n
L L
00
cxkiij f3~k)().,)i3~~m=O form=I,2, ... ,n-1. (5.1.15)

k,j=l n=-oo
Note thatthe coefficients of e-it, e- 2it , ... , e-(n-l)it are also zero. Hence, the
system (5.1.15) should determine the coefficients CX2, ... , CXn' Moreover, the
difference between f (z) and Un (z) decreases as n increases. The proof for the
convergence of Un (z) to f (z) through the method of successive approximations
is given in Kantorovich and Krylov (1958, p.435). We shall look into some
particular regions as case studies for which the system (5.1.15) provides simpler
solutions.
CASE STUDY 5.1.1. In order to determine the function f(z,).,) that maps
the interior of the ellipse x = (1 + ).,2) cos t, Y = (1 - ).,2) sin t conformally
onto the disk Iwl < R, first note that the equation of the ellipse can be written
as
z(t) = eit (1 +).,2 e- 2it ) . (5.1.16)
We shall find the approximate mapping function Un(z) = Pn(z) accurate to
)., 10. Then the last coefficient in (5.1.8) shall be cxu. Moreover, since the ellipse
has two axes of symmetry, all CXk are real and those with even indices shall be
zero, thus
and
IPn(zW = Pn(z) Pn(z)

= z Z + a3 (Z3 Z + Z3 Z) + [as (zS z + zS z) + a~z3 z3 ]
+ [a 7 (z7 z + z7 z) + aSa3 (zSz3 + zSz3)]
+ [a 9 (z9 z + z9z ) +a7a 3 (z7 z3 +z7z3) +a~zSzS]
+ [all (zllz + zllZ) + a9 a 3 (z9 z3 + z9 z3)
+ a7 a S (Z7 z S + z7 zS)] + [a 13 (Z13 z + z13 z )
+ an a 3 (zll z3 + zll z3) + a9a S (Z9 zS + z9 ZS) + a~z7 z7].

(5.1.17)
The combinations zk zj + zk zj that appear in (5.1.17) are determined by Mathe-
matica. Thus, substituting them into (5.1.17), equating the free term to R 2 , and
equating the coefficients of different cosines to zero, we obtain the following
system of equations:
1 +),4 + 6a3),2 (1 + ),4) + 20as),4 + a~ (1 + 9),4) + lOa3as),2 + a~

2
= R ,
),2+ a3 (1 + 6),4 + ),8) + 5a3),2 (1 + 4),4) + 3a~),2 (1 + 3),4)

+ 2Ia7),4 + a3aS (1 + 25),4) + 7a3a7),2 + 5a~),2 + a3a7 = 0,
a3),2 (1 + ),4) + as (1 + 5),4) + 3a~>.4 + 7a7),2 + 3a3as),2 + a3a7 = 0,
as),2 + a~),6 + a7 (1 + 7),4) + 3a3aS),4 + 9a9),2 + 3a3a7),2 + a3a9 = 0,
a7),2 + a9 = 0,
a9),2 + an = O.
(5.1.18)
These equations, except the first, will be solved by the method of successive
approximations. Thus, transposing a3, as, a7, a9 and all we get
a3 = - [),2 + a3 (6),4 + ),8) + 5a3),2 (1 + 4),4) + 3a~),2 (1 + 3),4)

+ 2Ia7),4 + a3aS (1 + 25),4) + 7a3a7),2 + 5a~),2 + a3 a7],
as = - [a3),2 (1 + ),4) + 5as),4 + 3a~),4 + 7a7),2 + 3a3as),2 + a3 a 7],
a7 = - [as),2 + a~),6 + 7a7),4 + 3a3aS),4 + 9a9),2 + 3a3a7),2 + a3a 9] ,
a9 = -a7),2,
all = -a9),2.
5.2. METHOD OF INFINITE SYSTEMS 125
The values of the coefficients a3, as, a7, a9 and all starting with initial val-
ues zero are computed up to the fifth successive approximation (see Table 1,
Appendix D). Hence,
p(z) = z - (>.2 + >.6 + 4>.10) z3 + (>.4 + 3>.8) zS _ (>.6 + 5>.10) Z7
+ >.8 z9 + >.10 zll,

which is accurate to >.10. The same fifth successive approximations for a3, as,
a7, a9, and all when substituted in the first equation in (5.1.18) yield
R 2 = 1 - 4>.4 + 10>.8.
= 1 + >.2 for t = 0, thus p(z) = P (1 + >.2) =
To check this result, note that z
1 - 2>.4 + 3>.8, which coincides with R = (1 - 4>. 4 + 10>.8) 1/2 = 1 - 2>.4 +
3>.8.•
5.2. Method of Infinite Systems
We shall present a general approach for the method of §5.1. This is known as
the method of infinite systems, initially developed by Kantorovich and Krylov
(1958) and later summarized in a systematic form with a computer program (in
ALGOL) by Andersen et al. (1962).
Let the boundary r>. of a nearly circular region D>. in the z-plane, denoted
by z = G (eit , >.), where>' is a small parameter, have a Laurent series expansion
in eit with suitable finite nand m as
n n m
z=G(eit,>.)=e it L Cp(>.)eiPt=eit L L(kq,p>.q)e iPt
p=-n p=-n q=[p[
= e it . { ... + e- 2it (0 + 0 + k 2,-2 >.2 + k 3,-2 >.3 + )

+ e- (0 + kl,-l >. + k 2,-1 >.2 + k 3,-1 >.3 + )
it
+ eO (1 + kl,o >. + k 2,0 >.2 + k 3,-0 >.3 + ...)

+ e''t ( 0 + kl,l A + k 2,1 A2 + k 3,1 A3 + ... )
+ e''t ( 0 + 0 + k 2,2 A2 + k 3,2 >. 3 +... ) + ... , }
(5.2.1 )
where the nonzero coefficients kp,q are known complex constants except ko,o
which takes the value 1 because>' = 0 must reduce the boundary r>. to the
circle z = e it , and small values of >. produce nearly circular boundaries. Note
that the expansion (5.2.1) is similar to (5.1.10).
Let the function w = f(z, >.) that maps the region D>. onto the disk Iwl <R
(or 1) have a series representation
L >.P [a~P)
00
= z + a~p) z2 + a~p) z3 + ... + a~1 zP+l]

p=o
= a~O) z + >. HI) z + a~l) z2] + >.2 [a~2) z + a~2) Z2 + a~2) Z3] + ....
(5.2.2)
Since>' = 0 gives the identity mapping, we have a~O) = 1. Also, all coefficients
a~p) are real since 1'(0, >.)= a~O) + a~l) >. + a~2) >.2 + ... is real for all
>.. The problem of approximating f(z, >.) reduces to that of determining the
unknown coefficients a~p) from the fact that after z = G (e it , >.) from (5.2.1)
is substituted, we should have Iwl 2 = w . W = R 2 for all t and every>. up to
the desired accuracy in powers of >..
First, we determine zq. Thus, from (5.2.1)
zq = e iqt . { ... + e- 2it (0 + 0+ k~~~2 >.2 + k1~~2 >.3 + ...)

+ e- it(0 + 1,-1 >. + 2,-1 >.2 + 3,-1 >.3 + ...)
k(q) k(q) k(q)
+ eO (1 + 1,0 >. + 2,0 >.2 + 3,-0 >.3 + ...)

k(q) k(q) k(q) (5.2.3)
+ (0 + 1,1 >. + 2,1 >.2 + 3,1 >.3 + ...)
e it k(q) k(q) k(q)
+ e (0 + 0+ k~~~ >.2 + k~;~ >.3 + ...) + .... }

it
Note that the coefficients k~;~ in (5.2.3), though known, are different from kp,q
of (5.2.1).
Next, while computing Iw1 2, we shall group together terms that have same
powers of >.. Thus, the coefficients of >.P includes only a~s), s = 0,1,2, ... , p
(8 > p does not occur). This feature provides an application of the method
of successive approximations to determine a~p) under the condition that the
coefficient of AP in Iwl 2 must vanish for p = 0, 1,2, ... (which yields a system
of equations) and the free term (all terms without A) must be equal to R 2 (or 1
if the region D).. is mapped onto the unit disk).
Since a~q) does not depend on a~s) for 8 > p, any subsequent revision of
(5.2.2) to include higher powers of A than previously taken shall only entail
determination of new terms that should be added to the free term and to each
equation of the above system to be solved by successive approximations.
An algorithm to compute (5.2.2) up to AN is as follows:

Use all terms in (5.2.1) corresponding to A0 , )..l, A2 , ... , AN.
Use all terms in z2 corresponding to A0, AI , A2 , . .. , AN -1 .
Continue for z3, ... , zP, i.e., use all terms in zP corresponding to A0 , AI , A2 ,
... , AN - p + 1 .
Continue until zN +1, i.e., use all terms in zN corresponding to A0 , AI .
Use all terms in zN+1 corresponding to AO.
°
Note that the sum of all coefficients of A yields the free term, and the coef-
ficients of AI , A2 , . .. ,AN equated to zero yield a system of N equations to
determine a~s), 8 = 1, ... , N, by successive approximations. This method
will produce an approximate function that maps every single boundary (for a
fixed A) by a power series in z, whose accuracy will depend on the manner in
which a~p) are computed.
CASE STUDY 5.2.1. Details for computation of a~O), a~l), a~l) and a~2),
a~2), a~2) are given below. Without using any computational tools, we shall
determine f(z, A) up to A2 .
p °
= yields a~O) = 1. Then
z = eit , w = a~O) z = Z,
Iwl 2 = a~O) e it . a~O) e- it = R2 .
p = 1 yields
z = eit {I + A [k(l)
1,-1 e- it + k(l)
1,0 + k(l)
it
1,1 e ] } ,
W = Z + >. [ail) z + a~l) z2]

= e it {I + >. [k(l) e- it
1,-1
+ k(l)
1,0
+ k(l)
1,1
e it ] }
+ >. [ail) eit + a~l) e2it ]

= e it {I + >. [k(l)
1,-1 e-
it + (k(l)a(l))
1,0 1 + (k(l)
1,1 + a(l))
2 e it ] } ,
IwI = 1 + >. [(k(l) + k(l) + a(l)) e- it + (k(l) + a(l) + k(l)

2
1,-1 1,1 2 1,0 1 1,0
+ a(l))
1
+ (k(l)
1,1
+ a(l)
2
+ kfil)
1,-1
eit ]
= 1 + >. [ (M~l) + a~l)) e- it + (M~l) + M~l) + ail) + ail))

+ (M?) + a~l)) e it ] = R2 .
p = 2 yields
z = e
it
{ 1 + >. [ki~~l e-
it
+ k 1,0 + k 1,1 e it ]
+ >.2 [k(l) e- 2it + k(l) e- it + k(l) + k(l) e it + k(l) e 2it ] }
2,-2 2,-1 2,0 2,1 2,2 ,
Z2 = e 2it {I + >. [k(2)

1,-1 e-
it + k(2)
1,0 + k(2)
it
2,1 e ] } ,
z3 = e 3it ,
W = Z + >. [ail) z + a~l) z2] + >.2 2

[ai ) z + a~2) Z2 + a~2) z3] ,
which gives
W = e it {I + >. [L~f e- it + L~l) + Li1) eit ]

+ >.2 [k(l)
2,-2
e- 2it + k(l) e- it + k(l) + k(l)e it + k(l) e 2it
2,-1 2,0 2,1 2,2
(1) k(l) -it + (1) k(l) + (1) k(l) it
+ a 1 1,-1 e a1 1,0 a1 2,1 e
(1) k(2) + (1) k(2) it + (1) k(2) 2it
+ a 1,-1 a 1,0 e a 2,1 e
2 2 2
+ ai ) + a~2)
2
e it + a~2) e it ] }
= eit 1 +), [L~~ e- it + L61) + Li1) eit ]

+),2 [K(2)
-2
e- 2it + K(2)
-1
e- it + (K(2)
0
+ a(2»)
1
+ ( K?) + a~I») eit + (K~2) + a~2») e2it ]}

where
K(2) - k(l)
-2 - 2,-2'
K(2) _ k(l) + a(l) k(l)
-1 - 2,-1 1 1,-1'
(2) _ k(l) + (1) k(l) + (1) k(2) + (2)
K o - 2,0 a 1 1,0 a2 1,-1 aI'
K 1(2) -_ k(l) + (1) k(l) + (1) k(2) + (2)
2,1 a 1 1,1 a2 1,0 a , 2
K (2) _ k(l)+ (1) k(2) + (2)
2 - 2,2 a2 1,1 a3'
and K~q] = Kjq) = Kjq). Hence,
Iwl 2 = 1 +),2 [ (MJ2) + af») e- 2it + (M~2) + a~2)) eit

+ 2 (M~2) + a~2») + (M?) + a~2)) eit
+ (MJ2) + a~2)) e 2it ] = R 2,
where M(q) = M(q) = M(q), and

-J J J
2
My) = K9J + K~2) + L~~ L~I) = 2 (Ki ) + L~I)) ,
M(2) - K(2)
1 - 1
+ K(2)
-1
+ L(I)
0
L(I) + L(1) L(1)
-1 0 1
- 2 (K(2)
- 1
+ L(I)
0
L(I»)
1 ,
M~2) = K~2) + L~~ L~~ + L61) L61) = K~2) + 2 (L6 1 ) + L~I») .

Hence a(2) - _M(2) a(2) - _M(2) a(2) - _M(2) and
'1- 0'2- 1'3- 2'
CASE STUDY 5.2.2. We shall consider the same problem as in Case Study
5.1.1, where the contour r-x
is given by (5.1.16), i.e.,
z(t,),) = eit (1 + ),[0· e- it + 0 + O· eit ]

+ ),2[1. e- 2it + O. e- it + 0 + O· eit + O· e2it ]).
The coefficients a<,J') for N = 8 are given by
p\q 1 2 3 4 5 6 7 8 9
0 1
1 0 0
2 0 0 -1
3 0 0 0 0
4 2 0 0 0 1
5 0 0 0 0 0 0
6 0 0 -3 0 0 0 -1
7 0 0 0 0 0 0 0 0
8 1 0 0 0 5 0 0 0 1
Hence, the mapping function is given by
w = f(z,'x) = z_,X2z3+,X4 (2z + Z5)_,X6 (3z 3 + Z7)+>.8 (z + 5z 5 + Z9) .•

(5.2.4)
This method can be improved as follows: Since
(5.2.5)
we can redefine the contour r A by
(5.2.6)
where J.L = ,X2 and T = 2t. The new contour r I-' yields the mapping function
w = ¢(Z, J.L) such that ¢(z2, ,X2) = [fez, ,X)]2, since f(z,'x) has the form
z· g(z2,,X2), i.e.,
Also, since f (z(t, ,X),,X) = eiO implies that [f (z(t, ,X), ,X)]2 = e2iO , we find
that
Z(T,J.L)' [g(Z(T,,X),,X)]2 = e2iO ,
and for z2 = Z = 0 we get [fez, ,X)]2 = 0, where :z [f(z,'xW is real. Hence,

the functions ¢(Z, J.L) and Z . [g(Z, J.LW yield the same mapping function of
the contour r I-' onto the unit circle for 0 ::; T < 2n for every fixed J.L = ,X 2 as
the function [z(t,'x)]· [g(z2(t,,X),,X2)]2 forO::; t < n. An advantage of this
technique is that the computation of ¢(Z2, ,\2) up to the power I-£N provides the
function f(z2, A2 ) up to the power ,\2N, which while computing J¢(z2, A2 )
will yield a mapping function with power up to AI6 instead of A8 as in (5.2.4).
CASE STUDY 5.2.3. We shall again consider the contour (5.1.16) for the
ellipse, which we rewrite as
Z = eiT [1 + 1-£ [2. e- iT + 0 + O· eiT ]

+ 1-£2 [1. e- 2iT + O· eitT + 0 + O· eiT + O. e2iT ] ].
Taking N = 8 we obtain the coefficients a~p) as follows:
p\q 1 2 3 4 5 6 7 8
0 1
1 0 -2
2 4 0 3
3 o -10 0 -4
4 6 0 20 0 5
5 o -28 0 -34 0 -6
6 8 0 77 0 52 0 7
7 o -62 0-164 0 -74 0 -8
Thus, after replacing 1-£ by A2 and T by 2t, we have
[f(z, A)F = z2 +,\2 (-2z 4) + A4 (4z 2 + 3z6) +,\6 (-10z 4 - 4z 8 )

+ A8 (6z 2 + 20z 6 + 5z 10 ) + AlO (-28z 4 - 34z 8 _ 6z 12 )
+ A12 (8z 2 + 77z 6 + 52z 10 + 7Z 14 )
+ A14 ( -62z 4 - 164z 8 - 74z 12 - 8z 16 ) .
An application of the binomial expansion to the above expression yields the

mapping function as
f(z, A) = z + A2 (_z3) + A4 (2z + z5) + A6 (-3z 3 - z7)

+ A8 (z + 5z 5 + z9) + AID (-7z 3 _ 7z 7 - zll)
+ A12 (2z + 16z 5 + 9z 9 + z13)
+ A14 (-12z 3 - 29z 7 -llzll - Z15) .• (5.2.7)
5.3. Three Special Cases
In the case of conformal mapping of the unit disk Izl < 1 onto a simply
connected region D in the w-plane, the mapping function z = F( w) can be
represented by a Taylor series
L
00
z=x+iy= cnw n , cn=an+ibn . (5.3.1)

n=O
In particular when the region D is also the unit disk, bounded by Iwl = I, we
set w = e i cP in (5.3.1) and obtain
L
00
z= incP (5.3.2)
C ne ,
n=O
which, when separated into real and imaginary parts, gives
L (an cos n¢ - bn sin n¢) ,

00
x = j (¢) =
n=O (5.3.3)
= g(¢) = L
00
y (b n sin n¢ + an cos n¢).

n=O
Thus, the mapping function (5.3.1) can be represented in the form of a Cauchy
integral as
Z
= ~
21r
1
0
2
11' j(¢) +.'"ig(¢) eicPdA-.'/'.
e t 'f'
(5.3.4)
The method of infinite systems is useful when the boundary r is defined by an

implicit function.
CASE 1. Let r be defined implicitly by
,(x, y) = 0, (5.3.5)
where ')'(x, y) = ')'(z) is an analytic function for z E r = 8B(0, 1). Then, by
substituting x and y from (5.3.4) in (5.3.5) and expanding the result in a Fourier
series, we obtain
')'(x,y) = ')'(I(¢),g(¢))
L
00
= ,o(aj, bj ) + {')'n(aj, bj ) cos n¢ + ,~(aj, bj ) sin n¢}, (5.3.6)

n=l
5.3. THREE SPECIAL CASES 133
of, and ,*
where ,0, ,n, ,~ are the Fourier coefficients, and aj, bj denote the dependence
on ao, bo, aI, bl , .... Now, by equating the coefficients of cos n¢
and sin n¢ to zero, we obtain an infinite system of equations for aj, bj :
CASE 2. Let f be defined by an implicit complex function
7/J(z, z) = 0. (5.3.8)
Then, by substituting the series (5.3.2) for z in (5.3.8), we find that
7/J (~ Cn e
in
</> , ~ c e- in
n </» = n~oo 7/Jn(aj, bj ) einO , (5.3.9)
which leads to an infinite system
7/Jn(aj,bj ) =0, n= ... ,-I,O,I, ... (5.3.10)
In order to obtain a solution for the unknowns ao, bo, aI, bl, ... from the sys-
tem (5.3.7) and (5.3.10) it is necessary that all related series converge and the
· . -d
denvatlve dz I
W Iwl=l
> . °
METHOD. Let a nearly circular boundary fA be represented by
x 2 +y2+>'P(x,y)=1 inCase 1, (5.3.11)
or by
zz + >. TI(z, 2) = 1 in Case 2, (5.3.12)
where>' is a small real parameter and P(x, y) and II(x, y) satisfy the same
conditions as the functions ,(x, y) and 7/J(x, y), respectively. Note that both
equations (5.3.11) and (5.3.12) are equivalent since 2x = z+z and 2i y = z-z.
Hence, we can use either equation. Suppose we consider Eq (5.3.12). Then
zz = ... + (COCI + Cl C2 + ) e -iO + (COco + Cl Cl + ... )
(5.3.13)
+ (CICO + C2 Cl + ) eiO + ...
In II(z, z) the coefficients of conjugate quantities einO and e- inO must also be
conjugate. Also, TI(z, 2) must be real since in Eq (5.3.12) both zz and>. are
real. Thus, after substituting the values of z and 2 from (5.3.2) , we get
L Tn(aj, bj ) einO + L -Tn-(""'a-j-,b"'-j"-) e- inO ,

00 00
II(z, z) = (5.3.14)
n=O n=O
where
(5.3.15)
Substituting the quantities (5.3.13) and (5.3.14) in (5.3.12) and comparing the
coefficients of positive powers of eiO , we obtain the infinite system of equations
COCo + C1C1 + C2C2 + + )do(aj, bj ) = R 2 ,

C1CO + C2C1 + C3C2 + + AT1(aj, bj ) = 0,
(5.3.16)
C2CO + C3C1 + C4C2 + + AT2(Zj, bj ) = 0,
Note that we will obtain an infinite system conjugate to (5.3.16) if we compare

the coefficients of negative powers of eiO in Eq (5.3.12). The system (5.3.16),
except for the first equation, can be rewritten as
(5.3.17)
CASE STUDY 5.3.1. We shall use the above method to obtain the function
that maps the unit disk onto the interior of the ellipse
(1 + A) x 2 + (1 - A)y2 = 1, (5.3.18)
with semimajor and semiminor axes as (1 + A)-1/2 and (1 - A)-1/2, respec-

° °
tively, such that center w = goes into the center Z = and the real axis into
the real axis. Then Co = 0, and C1 shall be real. Because of the symmetry of
the ellipse about the x-axis, all bj , j = 1,2, ... will be zero. The symmetry
of the y-axis also implies that all aj with even j shall be zero. The equation of
the ellipse (5.3.18) in the complex form is
Z2 + z2
ZZ +A 2 = 1.
Hence,
2 -2
II(z,z) = Z ; Z
Z2 = (al e i IJ + a3 e3i IJ + ... )2 ,

= ai e
2iIJ + 2ala3 e 4iIJ + (2alaS + a~) e6iIJ
+2(al a7+ a3a5)e8'1J
t + ( 2ala9+2a3a7+aS2) e 10'1J
t + ... ,
which yields (7n = einlJ )
70 = 71 = 73 = 75 = ... = 0,
1 2 1 2
72 = -2 a1 ' 74 = -al a3, 76 = -(alaS + 2a3),
1 2
78 = -(al a7 + a3 a5), 710 = al a9 + a3 a7 + 2as,
and so on. Thus, the system (5.3.17) becomes
(5.3.19)
. . a2j + 1 .
If we Introduce the notatIOn A o = aI, A j = - - - for J = 1, 2, ... , then the
al
system (5.3.19) reduces to
L
00
A 2 = -'\A 1 - A j Aj+2,
j=1
L
00
A3 = -A (A z + Ai) - A j Aj+3'
j=1
L
00
A 4 = -A (A 3 + Al A z ) - A j Aj+4,
j=1
As = -A ( A 4 + Al A3 + ~A~) - f
J=1
A j A j +5'
Holding the first equation of the above system, thus treating A o as undetermined,
we shall use the method of §5.2 and solve the remaining equations in this system
by successive approximations. Let the initial values be taken as A j = 0 for
j = 1, 2, .... The fifth approximations for AI, A z, A 3 , A 4 , and As are available
in Table 2 (Appendix D). Then the first equation of the above system yields
1 3
Ao = 1 -"21 (2
Al + A 2 ) 2·2 (2 z
2 + ... + 2! Al + A z + ...
)2 + ...
(5.3.20)
1 2 3 4
= 1- -A +-A
8 128'
which is accurate to AS. Hence, the mapping function is given by

00
z = Ao w [1 + LAn w zn ]
n=1
which is accurate to AS. •
CASE STUDY 5.3.2. We shall compute the function that maps the unit
circle Iwl = 1 onto the family of squares
Z 24 z-2) Z=1
-
zz+k ( (5.3.22)
in the z-plane such that the point w = 0 goes into the point z = O. Note that
for k = 1, Eq (5.3.22) reduces to (x 2 - 1) (y2 -1) = 0 which represents the
sides of the square of Fig. 4.2.1. Obviously, CO = 0, and since the real axes are
preserved, arg{ Cl} = O. The square (5.3.22) is symmetric about the x and y
axes and also about the lines y = ± x. Hence, bj = 0, a2j = 0, and a4j -1 = °
for j = 1,2, .... Then, from (5.3.2)
00
Z -- '"
~
a4n-3 ei (4n-3)1I , (5.3.23)
n=1
which gives
II(z, z) =(
Z 24 z-2) 2
-
= ~
16
[ai e2ill + 2al as e6ill + (2al ag + a~) e
lOili
+ ...
- ai e- 2ill - 2al as e- 6ill + (2al ag + a~) e- 10ill + ... ] 2.
Thus,
which yields the system (5.3.19) as

+ (aia5a13 + ala~a9) + ... ],

3 22 22
k
a13 = - a5;17 _ ... + 2al [ - a 12a
g
_ a 1a 5 _ a~a5
2
+~ (ar a 13 + ai a5a9) + ... ].
If we take the initial values as al = 1, a5 = a9 = ... = 0, then computing up
to the third successive approximations (see Table 3, Appendix D), we get the
approximate mapping function as
(5.3.24)
For k = 1, this becomes
z = 1.077w - 0.1006w 5 + 0.0288w 9 - 0.0054w 13 ,
which compares with the exact solution (4.5.31) with a maximum error of the
order of 10- 3 . •
CASE 3. If the boundary f,\ is defined by the parametric equations
x = get, >.), y = h(t, >.), (5.3.25)
then we can expand the functions g and h as trigonometric series
+ L (an cos nt + t3n sin nt) ,

00
g(x, >') = ao
n=l
(5.3.26)
+L
00
h(t, >') = 'Yo (/'n cos nt - Dn sin nt).

n=l
Since the series in (5.3.26) are conjugate, we obtain the complex form for the
equation of f,\ as
+ iy = 7l'0 + L (7l'nCosnt + Pnsinnt)

00
z=x
n=l
(5.3.27)
_ + LJ ~ {7l'n -2 ipn eint + 7l'n +2 ipn e -int} ,
- 7l'0
n=l
where 1rn = an + i fn , and Pn = f3n - iOn. We shall assume that the curve
(5.3.27) has the same form as (5.2.1), where the coefficients 1rn and Pn depend
on A. If we take A = 0 in (5.3.27), then this equation for fA reduces to
L
00 .
z = G (e it , A) = 1ro + n ~ tpn eint .

1r (5.3.28)
n=1
The function w = G (e it , A), where eit is a point on the unit circle is assumed
to be analytic in wand A near the values A = O. The parameter t represents the
polar angle of the point in the w-plane such that w = Iwl eit for any w E U.
However, the parameter t, in general, does not coincide with the argument B
taken for the values of x and y in (5.3.3). Therefore, we shall substitute in
(5.3.28)
(5.3.29)
where 1/Jj(B), j = 1,2, ... , are real, periodic functions, yet to be determined.
Note that for A = 0, the series (5.3.29) reduces to t = B. Now, the functions
1/Jj (B) must be determined for j = 1,2,... such that the coefficient of Aj in
the series (5.3.27) does not contain any term in negative powers of ei9 . This
process is explained in the next case study.
CASE STUDY 5.3.3. Let the boundary fA be defined in the parametric
form by x = cos t + ~ cos 2t, Y = sin t + ~ sin 2t (Fig. 5.3.1), which in the
complex form is
z = eit +~ [ei(n+l)t + e- i(n-l)t] . (5.3.30)
Substituting for t from (5.3.29), we obtain
z = ei9 + A [i1/Jl ei9 + ~ { e i (n+l)9 + e- i(n+l)9} ]
+ A2 [ (i1/J2 - ~ 1/Jf) ei9 + ~ i(n + 1)1/Jl e i (n+l)9 (5.3.31 )
- '1.(
2t n- 1)./'
<pI e i(n-l)9] + ... .
It is obvious from this expression that z will contain only positive powers of
ei9 if
i1/Jl ei9 +~ {e i (n+l)9 + e-i(n+l)()} = ei(n+l)(),
(i1/J2 - ~ 1/Jf) ei9 + ~ 1/Jl [en + 1) e i (n+l)9 - (n -1) ei (n-l)9] = e i (2n+l)9,

which yields
i
'l/Jl = 2 (e- in9 - ein9 ) = sin nO,
'l/J2 = i
2n - 1
--8- (e- 2in9 - e2in9 ) = T
2n 1
sin2nO.
y
1
-1
Fig. 5.3.1.
Substituting these values in (5.3.31), we get
(5.3.32)
which is accurate up to 0(>,2). Hence, the approximate mapping function is

given by
Z=
2n + 1 \
( 1--- -"
2) w+"w
\ n+l 2n + 1 \ 2 2n+l
+---" w .• (5.3.33)
4 4
5.4. Exterior Regions
We shall approximate the function that maps the region exterior to the boundary
(5.1.9) or (5.2.1) onto the exterior or interior of the circle Iwl = R, assuming
5.4. EXTERIOR REGIONS 141
that the point at infinity z = 00 goes into the point w = 00 or into the origin
w = 0, respectively. There are two cases to consider:
CASE 1. In the case of mapping onto the exterior {Iwl > R}, the mapping
function w = f(z) with f(oo) = 1 has an expansion
(5.4.1 )
{Iwl <
°
CASE 2. In this case of mapping onto the interior R}, the mapping
function with f( 00) = has the expansion
1 a2 a3
w = - + -2 + -3 + ... . (5.4.2)
z z z
In both cases we shall use the method of infinite systems to approximate w =
f(z).
In Case 1, the mapping function can be approximated by taking the first n

terms in (5.4.1) :
a1 an-2
w = z + ao + - + ... + --. (5.4.3)
Z zn-2
In order to find Iwl on the boundary (5.1.9), we represent 1/ z in the form of

2
the series (5.1.10). Thus,
2:=
00
1
j3~-1}(>\) eillt , (5.4.4)
z v=-oo
where the coefficients j3~-1) (>,) are regular functions ofthe parameter A, such
that
for v #-1
for v = -1.
First we compute Iwl from (5.4.1) and substitute in it the value of z from
2
(5.1.10) and the value of l/z from (5.4.4). This will yield a trigonometric
series, in which we equate the free terms to R 2 (i.e., those terms which are
independent of trigonometric functions), and set the coefficients of the first n
terms of this series to zero. This will yield a system of equations exactly as
in §5.l, which can be solved by the method of successive approximations to
determine approximate values of ao, a1, ... 1 an, and R.
Case Study 5.4.1. We shall consider the mapping of the region exterior to the
ellipse
onto the exterior Iwl > R. Since the region is symmetric about the coordinate
axes, the mapping function has an expansion about the point at infinity
al a3
w=z+-+-+···,
z z3
(5.4.5)
where all aj are real and w( 00) = 00. Let us approximate w by a polynomial
of the form
al a3 a5 a7 a9
w =z+- +- + - + - +-.
Z z3 z5 Z7 z9
Then
(5.4.6)
z2 = 1 + .>.
2 + 2'>' cos 2t,
~ +~ = -2 [(.>. - .>.3) - cos 2t - (.>. - .>.3) cos 4t] ,

z z
:3 + 2 = 2 [6.>.2 - 3'>' cos 2t + (1 - 3.>.2) cos 4t + '>'cos 6t] ,
z z3
--=
1
zz
= (1 + .>.2) - 2'>' cos 2t + 2.>.2 cos 4t,
z 2
-5
z
+"5
z
= -2 (5'>' cos 4t - cos 6t - '>'cos 8t),
1 1
"""3=
z z
+ -=3
zz
= -2 (3'>' - cos 2t + .>. cos 4t) ,
z 2
-7
z
+ -;;
z
= 2 cos 8t,
1 1
5"=
z z
+ -=s
zz
= 2 cos 4t,
1
z 3-3
z = 1.
Hence, substituting these values in (5.4.6), equating the terms independent of

e it to R 2 , and equating the coefficient of cosines to zero, we obtain
1 + A2 - + 12A2a3 + (1 + A2 ) ai - 6Aala3 + a~ = R 2,
2 (A - A3 ) al
A + al - 3Aa3 - Aai + ala3 = 0,
(A - A3 ) al + (1- 3A2 ) a3 + A2ai - 5Aa5 - Aala3 + ala5 = 0,
Aa3 + a5 = 0,
Aa5 + a7 = 0,
(5.4.7)
which except for the first equation is rewritten as
al = -A + 3Aa3 + Aai - ala3,

a3 = 3A2a3 - (A - A3 ) al - A2ai + 5Aa5 + Aala3 - ala5,
a5 = -Aa3,
a7 = -Aa5'
Choosing the initial values for al, a3, a5, a7 as zero, the successive approxi-
mations for these coefficients are available in Table 4 (Appendix D), where we
have retained the values up to the fourth approximation. Hence, the mapping
function accurate up to A4 is given by
and the approximate value of R 2 from the first equation in (5.4.7) is R 2

1 - 4A 2 - 2A 8 , which yields the radius R = 1 + 2A 2 - 3A4 (compare this value
of R with that obtained in Case Study 5.1.1).•
Case 2 can be analyzed analogously by taking the approximate function as

al a2 an
W= -+-+"'+-,
z z2 zn
(5.4.8)
and following the above method step-by-step, where al = 1 for a nearly circular
boundary of the type (5.1.9) or (5.2.1).
CASE STUDY 5.4.2. We shall map the exterior of the square {-I ::;
x, y ::; I} onto the disk Iwl < R. The equation of the square in complex form
is
Z2 _ z2
ZZ + 4
= 1.
We shall, however, analyze the family of curves

z2 - z2
ZZ +A 4 = 1, (5.4.9)
where A = 1 gives the above square. Since the squares are symmetric about
the coordinate axes and about the diagonals y = ± x, the function w has, from
(5.4.8), the form
Z = ale-ill + ase 3ill + age7ill + ... , (5.4.10)
where aj are real. Now,
z2 ~ z2 = ~{_ (~a~ + ... ) e-l4ill _ (aSag + ... ) e-lOill
- (ala g + ~a~) e- ~ai) e- 2ill

6ill
- (alas -
+ (alas - ~ai) e2ill - (ala g - ~a~) e6ill

+ ( aSag+ ... ) e lOi II - ( 1 2 ... ) e l4i II} . (5.4.11)
2ag+
After substituting (5.4.10) and (5.4.11) in Eq (5.4.9) and comparing the coeffi-
cients of different exponential powers, we get
5.5. PROBLEMS 145
If we set ,\ = 1, we obtain
11251 203 3 1 7
W = 1024:; - 2048 z + 2048 z .•
5.5. Problems
PROBLEM 5.5.1. Let E denote the nearly circular ellipse b2 u 2 + a 2 v 2 =

2 2
a b, where b = 1 and a = 1 + c. Show that the function j(z) = z +
~ z (1 + z2) +o(c) maps the unit disk Izi < 1 onto the region Int (E). (Nehari,
1952, p.265.)
PROBLEM 5.5.2. Show that the function
w = j(z) = z + -cZ
21l'
1 0
2
11' e
it
-'t-
e' - z
+ z p(t) dt + o(c)
maps the unit disk Izl < 1 onto a nearly circular region whose boundary has the
polar equation r = 1 + c p( B), where p( B) is bounded and piecewise continuous
and E: > 0 is a small parameter. (Nehari, 1952, p.263.)
PROBLEM 5.5.3. Let the boundary f of a simply connected region be

defined in polar coordinates by r = 1 + c g( B), c > 0, where g( B) has a finite
Fourier series expansion of the form
n
g(B) = ao + L (aj cosjB + bj sinjB) .
j=l
Show that the function

n
j(z) = Z + €Z + [a o + L (aj - ib j ) zj] + o(c)
j=l
maps the unit disk Izi < 1 onto the nearly circular region Int (f). (Nehari, 1952,
p.265.)
PROBLEM 5.5.4. The ellipse defined by (5.1.16) can be written as z =

a cos t + i b sin t, where a = 1 + >. 2 and b = 1 - >. 2. Determine the mapping
function !(z, >.) in the following cases: (i) >.2 = 1/11, i.e., alb = 1.2; (ii)
>.2 = 1/5, i.e., alb = 1.5; and (iii) >.2 = 1/3, i.e., alb = 2. What does the
value of 1'(0, >.) signify in each case? [Hint: Show that the mapping function
is
(i) !(z, >.) = 1.0165984 z - 0.0932071 z3 + 0.008615 z5 - 0.0007963 z7 +
0.0000734 z9 - 0.0000068 zll + 0.0000006 z13 + 0.00000005 z15;
(ii) !(z, >.) = 1.081728 z - 0.226394 z3 + 0.049024 z5 - 0.010611 z7 +
0.002176 z9 - 0.000461 zll + 0.000064 z13 + 0.000013 z15;
(iii) !(z, >.) = 1.2373 z - 0.4787 z3 + 0.1948 z5 - 0.0791 z7 + 0.0247 z9 -
0.0091 zll + 0.0014 z13 + 0.00046 z15.] (Andersen et aI., 1962, pp.254-255.)
REFERENCES USED: Andersen et al. (1962), Goluzin (1937), Kantorovich

and Krylov (1958), Nehari (1952).
Chapter 6
Green's Functions
Green's functions are useful in solving the first boundary value problem
(Dirichlet problem) of potential theory in itself and in the case of confor-
mal mapping of a region onto a disk. In the latter case a relationship is needed
between the conformal map and Green's function for the region. An approx-
imate determination of Green's functions is an important numerical tool in
solving both the Dirichlet problem for different types of regions and the re-
lated mapping problem. An integral representation of Green's function for
the disk leads to the Poisson integral. The Dirichlet problem is a special case
of the Riemann-Hilbert problem which is discussed in Appendix C. Analo-
gous to Green's functions, the solution of the second boundary value problem
(Neumann problem) of potential theory is the Neumann function which also
possesses an integral representation.
6.1. Mean-Value Theorem
The mean-value theorem for harmonic functions is discussed in §1.3. Thus,

if we integrate in (1.3.2) on the circle Iz - zol = ro and if the function u(z)
is harmonic on Iz - zol ::; ro, then this integral must vanish, i.e., in polar
1 -aau
coordinates
21T
-21l"1 0 r
r o d8=O. (6.1.1)
After multiplying the integral (6.1.1) by dr and integrating from rl and r2,
ro
147

148 CHAPTER 6. GREEN'S FUNCTIONS
we get
(6.1.2)
or
-1
21r
10
2
71"
u(r1, 8)d8 = -1
21r
1 0
2
71"
u(r2,8)d8. (6.1.3)
These are the mean values of u(z) taken on both circles Iz - zol = rl and
Iz I = r2, where rl < r2. So long as u( z) is harmonic on the larger circle, these
mean values are equal. For rl --t 0 the left side of (6.1.3) takes the value u(zo)
at the center, so that we finally obtain
u(zo) = - 1
21r
10
2
71"
u(r,8)d8. (6.1.4)
This is the mean-value theorem of potential theory which states that for every
function harmonic on a circle the value at the center is equal to the mean value
of the function on the circumference (see (1.3.3)). An important consequence
of the mean-value theorem is that a nonconstant function u(z) harmonic in a
region D takes neither a maximum nor a minimum value in the interior of D.
Since the real and imaginary part of a regular analytic function w = fez) =
u(z) + iv(z) are harmonic functions, the mean-value theorem also holds for
analytic functions, and Cauchy's integral formula (1.2.2) remains valid, i.e.,
f(zo) = ~ J fez) dz. (6.1.5)

2Z1r J!zl=r Z - Zo
6.2. Dirichlet Problem
Let Zo E D be a fixed point (known as the source point). Green's function for
the Dirichlet problem in the region D with a logarithmic singularity at Zo is the
function Q(z, zo) with the following properties:
(i) As a function of z, Q(z, zo) is harmonic everywhere in D except at the point
ZOo
(ii) At the point Zo the function Q(z, zo) is defined by
1 1
Q(z, zo) = -2 log -
1r r
+ Q(z, zo), r = Iz - zol, (6.2.1)
6.2. DIRICHLET PROBLEM 149
where 9(z, zo) is harmonic everywhere in D.

(iii) 9(z, zo) = 0 if the point Zo lies on the boundary r.
The Dirichlet problem for the region D can be solved in an explicit form
by using Green's function. Since, in view of the property (ii), the function 9
becomes unbounded at z = zo, we can indent the point Zo by a circle r c of small
radius E: (see Fig.B.l(a)). Then the functions u and 9 become continuous in
the region Dc bounded by rand r c' An application of Green's second identity
(B.5) (with 9 = 9) yields
J(
r+r.
- - 98U)
u89
8n
- ds=O,
8n
(6.2.2)
where n denotes the outward normal to the boundary which will be normal to
r, exterior to Dc, and interior to r c' Separating the integral (6.2.2) over two
contours rand r c and using (6.2.1), we find that
1
r(u 8n
Jr
89 _ 9 8U)
8n
ds = ~
27r Jr.
r U 8n 8 log -:;: ds + r U8n ds
Jr.
89
-~ r log ~r 8n
27l' Jr.
8u ds - r 9 au ds == h + II + 1 + 1 + h
Jr. an
2 3 (6.2.3)
The interior normal to r" is along the radius r with its direction opposite to that
of increasing r, and hence, :n -:r' which yields
1 1
810g -r
___ alog -
r 1
an Or r
Also, since r = E: on the circle r c' we get, in view of (6.1.4),
h = _1_
27l'E:
r
Jr.
uds = u(zo)
for any E: by the mean value theorem. In h, since r = E: and u is harmonic, we

have
h = -1
27r
log - 11
E:
-8u ds = 0
r. 8n
for any E:.
The remaining two integrals 12 and 14 tend to zero as E: ----> O. In fact, since
u, 9, ~~ and :~ are bounded in the neighborhood of the point zo, we find that
Ihl ~ IIf~ lu :~ I· 27rE:, and 114 1~ IIf~ 19 ~~ I· 27rE:.

Hence, as € -4 0, the relation (6.2.3) yields
-1
21T
1
r
(09 ou)
u - - 9 - ds=u(zo)·
on on (6.2.4)
Moreover, by condition (iii) the function 9 vanishes on r, and hence, from

(6.2.5) we obtain
u(zo) = - 1
21T r
09 ds.
U!.l
un
1 (6.2.5)
Note that this equation is also a consequence of Green's third identity (B.8). If
Green's function is known for the region D, then formula (6.2.5) can be used
to solve the Dirichlet problem for any continuous or piecewise continuous,
boundary values of the harmonic function u( z). An alternate form of formula
1
(6.2.5) is
u(zo) = 2~ ud9(z, zo). (6.2.6)
The relationship between Green's function and conformal mapping is es-

tablished as follows: Let w = j(z) = pe iIJ map a simply connected region D
conformally onto the unit disk Iwl < 1 such that the point Zo E D goes into
the point w = 0, i.e., j(zo) = O. If j(z) has a simple zero at zo, the function
F(z) = j(z) '" 0 for all zED and is regular everywhere in D. Thus,
z - Zo
log F(z) is also regular analytic in D. Let us denote log F(z) = U + i V.
Then
f(z) = (z - zo) eU + i v,
which yields, with z - Zo = r ei 4>,
1 1 1 ( 1 ) (6.2.7)
21T log If(z)1 = 21T log;;: - U .
It can easily be verified that the function 2~ log Ij!z) I satisfies all three prop-
erties for the Green's function 9(z, zo), namely, this function with a simple
pole at Zo is harmonic in D except at zo, where it has a logarithmic singularity,
and it is equal to
1 1
-log - + g(z),
21T r
1
where g(z) = - - U. Moreover, If(z)1 = Ion the boundary r, and, there-
21T
1 1
fore, 21T log Ij(z)1 = 0 there. Thus, we have shown that Green's function
Q(z, zo) and the mapping function j(z) which produces the conformal map of
the region D onto the unit disk are related to each other by
1 1
Q(z, zo) = 21r log Ij(z)I' (6.2.8)
If Green's function for a region D is known, we can use (6.2.8) and construct
the function j(z) which maps the region D conformally onto the unit disk.
The method of accomplishing this is as follows: For each term in (6.2.7), we
determine the respective conjugate harmonic function. The conjugate harmonic
function for ~ log ~ is _1-, where ¢ = arg{z - zo}. Let h(z) be conjugate
21r r 21r
to the function g(z). Then, in view of (1.3.2),
l (au au) +
z
h(z) = Zo ax dy - ay dx C, (6.2.9)
where C is an arbitrary constant, which corresponds to the rotation of the unit

disk about w = O. Hence, the required mapping function is given by
(6.2.10)
Note that the construction of the Green's function Q(z, zo) involves determining
the harmonic function g(z), whose boundary values are determined from the
third property, namely, that Q(z, zo) = 0 on the boundary f. This means that
1 1
g(z) must take the values --log - on f. Hence,the conformal mapping
21r r
problem of transforming the region D onto the unit disk reduces to the solution
of the Dirichlet problem with the boundary condition
1
g(x,y)lr = 21r log r. (6.2.11)
If D can be mapped conformally onto the unit disk, then the dependence
of Green's function on Zo can be given explicitly. Thus, if w = j(z) is any
function that maps D onto the unit disk, then we can use the mapping
i-y w - Wo
zl--4e ,
1-wwo
where, is an arbitrary real constant, which maps the unit disk onto itself such
that the point w(zo) goes into the origin and Wo = j(zo). Hence,
w -Wo
Q(z, zo) = log _. (6.2.12)
1-wwo
For practical applications, once Green's function 9(z, zo) is known for a region,
we can use formula (6.2.5) and determine the solution of the Dirichlet problem
at the point Zo0 Now, to compute the solution of the Dirichlet problem at another
point, say Zl, we must recompute Green's function with a pole at Zl. Let this
new Green's function be denoted by 9(z, zt}. Now we shall establish a relation
which connects the two Green's functions 9(z, zo) and 9(z, zt}. Assuming
that 9(z, zo) is known, the other Green's function 9(z, Zl) can be determined
as follows: By the method presented above, first we determine the conformal
mapping function W = j(z) from the known Green's function 9(z, zo). This
mapping function transforms the region D into the unit disk such that the point
z = Zo goes into the origin W = O. Under this mapping, the point Zl is
transformed into some point, say, WI = j(ZI), which is known because the
function j(z) is known. Next, to determine 9(z, Zl) by formula (6.2.6), we
must find the function W = it (z) that would map the region D onto the unit
disk such that the point Z = Zl goes into the origin W = O. If j(z) is known,
then it (z) can be determined by mapping the unit disk onto itself, i.e., we use
W-WI
the transformation W ........ . Hence,
1 - wtlh
j(z) - WI
it(z) = 1 - wd(z)' (6.2.13)
which yields the required Green's function

1 1 1 11 - W WI I
9(z, Zl) = 271" log lit(z)1 = 271" log Ij(z) - wII' (6.2.14)
6.3. Numerical Computation
Green's function Q(z, zo) with a pole at Zo can be determined from (6.2.1),
provided that the harmonic function g(z) = ~ log r, defined by (6.2.11), is
271"
determined for z E r. An interpolation method for numerically computing
the function g(z) is as follows: In the polar coordinate system let Zo be taken
as the pole, and let the polar axis be directed parallel to the x-axis. Thus,
z - Zo = r eiO . Let g( z) have the series representation
L
00
g(z) = Ck(Z - zo)k, Ck = ak + ibk· (6.3.1)

k=O
6.3. NUMERICAL COMPUTATION 153
n
Then we take the harmonic polynomial Pn(r, 0) =~ {2: Ck(Z - zo)k}, i.e.,
k=O
n
Pn(r,O) = ao + 2: r k (ak cos kO - bk sin kO) . (6.3.2)
k=l
Since this polynomial has (2n + 1) coefficients, we take (2n + 1) arbitrary

points zl, ... , z2n+l on r and choose the coefficients ak> bk such that at the
points Zj, j = 1, ... ,2n + 1, the polynomial Pn(r, 0) has the same values
as g(Zj), respectively. Noting that g(z) has the boundary value defined by
(6.2.11), the coefficients ak, bk are determined from the system of equations
ao + t r~ (ak cos kO l - bk sin ked = .2.-

21f
log rl,
t r~
k=l
ao + (ak cos k0 2 - bk sin k0 2) = .2.- log r2,

k=l 21f (6.3.3)
where Zj - Zo = rj ei()j, j = 1, ... ,2n + 1. The determinant of the system

(6.3.3) depends on the choice of the points Zj. Let us assume that these points
lie on an equipotential line of a harmonic polynomial Qn(r, 0) of degree at
most n:
n
Qn(r, e) = ao + 2: r k (ak cos kO - bk sin kO) = 0, (6.3.4)
k=l
where the coefficients ak and bk are not zero at the same time. Then this line
exists iff the homogeneous system of (2n + 1) equations
n
ao + 2: rj (ak cos kOj - bk sin kOj ) = 0, j = 1, ... , 2n + 1, (6.3.5)
k=l
has a nonzero solution for ao, al, bl , . . .. Hence, if the determinant of the
system (6.3.5) is zero, there exists a nonzero solution, which implies that the
points Zj lie on an equipotential line. However, if this determinant is not zero,
then the system (6.3.5) has a trivial solution, and there is no curve that passes
through the points Zj. Since this determinant of the system (6.3.5) is the same
as that of the system (6.3.3), a zero determinant of the latter system implies that
all (2n + 1) chosen points Zj lie on the said equipotential line. Thus, first we
solve the system (6.3.3) for the unknown coefficients ao, al, b1 , .... Then their
values are substituted in Pn(r, B). This harmonic polynomial so constructed
has the property that at the (2n + 1) points Zj it takes the same values as the
harmonic function g(z). Hence, in view of (6.2.1), Green's function has the
approximate representation
lin
Q(z, zo) ~ -log - + ao + '" r k (ak cos kB - h sin kB). (6.3.6)
27l" r L....-
k=l
It is assumed here that the difference between Pn(r, B) and g(z) decreases as
n increases and the above interpolation method becomes justified, provided
lim Pn(r, B) = g(z).

n-->oo
CASE STUDY 6.3.1. We shall approximate Green's function for the square
{(x,y) : -1 ::; x,y ::; I} in the z-plane with the pole Zo at the origin (see
Fig. 4.2.1). In view of the symmetry about both coordinate axes, we shall
approximate g(z) by the harmonic polynomial Pn(r, B) defined by (6.3.2).
Since the values of g(z) = g(x, y) are arranged symmetric to the y-axis, all
bk = O. Also, since the values of g(x, y) are symmetric to x-axis and bisectors
of coordinate angles, the series (6.3.2) will contain cosine terms of angles 4nB,
n = 0,1,2, .... Hence, the series expansion for g(x, y) becomes
g(x, y) ~ ao + a4 r 4 cos 4B + as r S cos 8B + ... . (6.3.7)

We shall consider the points Zj on the boundary defined as follows: Zl = 1,
Z2 = ~ei1r/6, Z3 = V2e i1r 4
/ . Then the coefficients aO,a4, and as are
determined from the system
1
S
-9
-4
whose solution is as follows:
0.075578
ao = 0.12648 = ,
27l"
0.0740122
a4 = -0.012244 = - 27l" '
0.00156583
as = -0.000404057 = 27l"
Note that additional points Z4 = ~ eirr / 3 and Z4 = eirr / 2 add nothing to the
solution. The approximate Green's function is given by
1 1
Q(z,O) = -log - + g(z)
211" r
:::::: - 1 [ log -1 + 0.075578 - 0.0740122r 4 cos40
211" r
- 0.00156583r 8 cos 80]
= ~
211"
~ {lOg ~Z + 0.075578 - 0.0740122 Z4 - 0.00156583 z8} .
Then, in view of (6.2.8) the mapping function w = f(z) that maps the square
onto the unit disk is found from
log If(z)1 = -211"Q(z, 0)

= ~ {log z - 0.075578 + 0.0740122 z4 + 0.00156583 z8}
up to a purely imaginary additive constant which we shall ignore. Hence,
f(z) = z e-O.075578+0.0740l22 z4+0.00 l5 6583 z8. (6.3.7)
Note that
f'(0) = e-O.075578 = 0.927207.
Jor
l
The exact value from (4.5.29) is f'(0) = ~::::::
1 + (4
0.927037 which
shows that the error in the approximate value for f(z) is about 0.017
CASE STUDY 6.3.2. In order to approximate Green's function for the

ellipse (5.1.16), let the points Zj be taken for t = 0, 11"/8, 11"/4, 311" /8, and 11"/2.
Then, we have
g(z) :::::: ~{ - 0.0025 + 0.063 Z2 - 0.0038 Z4 + 0.0034 z6 - 0.000035 z8}
= ~ ~{ - 0.0158 + 0.396 z2 - 0.0238 Z4 + 0.00215 Z6 - 0.0002 Z8}.

211"
Thus,
Q(z,O) = - 1 ~ { log -1 - 0.0158 + 0.396 z2 - 0.0238 z4

211" z
+ 0.00215 Z6 - 0.0002 z8},
which gives
f(z) = z eO.01S8-0.396 z2+0.0238 z4-0.0021S z6+0.0002 z8.
Note that l' (0) = eO.01S8 ~ 1.0159, which matches with the value of p' (0) in
Case Study 5.1.1 .•
CASE STUDY 6.3.3. We shall consider the curve

F(x, y, a) = [(x - 0.5)2 + (y - a)2] [1 - y2 - (x - 0.5)2] = 0.1.
The graphs for a = 00, 1,0.5, 0.3,and 0.2746687749 can be plotted as sug-
gested in cs633. nb (Notes, end of chapter). For a = 00 it represents the unit
circle with center at (0.5, 0). The region does does not remain simply connected
for a = 0.2746687749. We shall approximate Green's function for the region
at a = 1 which is a nearly circular cardioid (see Fig. 6.3.1).
0.5
~ _ _-+ ~ _ _~ _ _-+x
·0.5 0 0.5 J.5
-0.5
-I
Fig. 6.3.1. The curve for a = 1.
Since the region is symmetric about the line x = 0.5, we shall choose seven
boundary points Zj as follows: Zl = 0.5 - 0.98725 i, Z2 = 0.8 - 0.92024 i.
Z3 = 1.2 - 0.69296 i, Z4 = 1.474, Zs = 1.2 + 0.5972 i, Z6 = 0.8 + 0.65589 i,
and Z7 = 0.5 + 0.60343 i. Unlike previous case studies, we have computed
the radii and arguments for each of these points (see Notes, cs633. nb, end of
chapter). Green's function is given by
1 1
Q(z,0.5) ~- log - - 0.0431944 + 0.0404596r cosO
21f r
+ 0.0107145 r 2 cos 20 - 0.0114393 r 3 cos 38 + 0.0569173 r sin 0
- 0.039137r 2 sin 20 + 0.00477567r 3 sin 30,
where r = Iz - 0.51. Then the mapping function f(z) can easily be obtained
with 1'(0.5) = e-O.0431944 = 0.957725.•
CASE STUDY 6.3.4. We shall consider the Cassini's ovals
The graphs for a = 0,0.5,0.9,0.99, and 1 are plotted in Fig. 6.3.2. For a = 0
the curve becomes the unit circle. The region does does not remain simply
connected for a = 1. We shall approximate Green's function for the region at
a = 0.5 which is bounded by a nearly circular ellipse (see also Fig. 9.2.1).
Fig. 6.3.2. Cassini's ovals.
Because of the symmetry of the region about the axes of coordinates, the
function g( z) has a series expansion of the form
g(z) = ao + a2 r 2 cos 28 + a4 r 4 cos 48 + ....
Letthe points Zj on the boundary be chosen as Zl = 1.11803, Z2 = 1.04942 e i7r /6,

Z3 = 0.9839ge i7r / 4, z4 = 0.92265e i7r / 3, and Zs = 0.866e i7r / 2 . Then, as in
the previous Case Study, we can compute ao = -0.00513, a2 = 0.02124,
a4 = -0.00282, a6 = 0.0005, a8 = -0.000096. Thus, Green's function is
given by
Q(z, 0) ~ - 1 ~ { log -1 - 0.0322236 + 0.133465 z2 - 0.0177425 z4

271' z
+ 0.0031616 z6 - 0.000605 z8},
and the mapping function by
j(z) ~ z eO.0322236-0.133465 z2+0.0177425 z4-0.0031616 z6+0.000605 Z8,
with 1'(0) ~ 1.03276.•
6.4. Schwarz Formula
We shall consider specially the case when D is a circle with center at the origin
and radius R. In this case we carry out the mapping onto the unit disk by
z
W= R' (6.4.1)
and Green's function (6.2.12) becomes
R(z - zo)
Q(z, zo) = log R2
- ZZo
_. (6.4.2)
Then
1
dQ(z, zo) = (- - + R2 f- OZzo )
z - Zo
dz. (6.4.3)
2
Since I z 1 = ZZ = R2 on the boundary of the circle and dz = iz dB, so
dQ(z,zo) = .(Z
t -- +~
Z -1 ) dB. (6.4.4)
z - Zo z - Zo
Setting z = R ei8 and Zo = Pei ¢, we find that

. R2 _ p2
d Q( Z, zo) = (B cP) dB. (6.4.5)
t R2
+p 2 R
- 2 peos -
If we substitute (6.4.5) in (6.2.5), we obtain the Poisson integral
1
1 211" R2 2
i¢ _ i¢ - P
-2 (B cP) dB. (6.4.6)
u(pe ) -
1f 0
u(Re ) R2
+P - 2 2R
peas -
By a similar integral representation we can determine the harmonic function

v(z) which is conjugate to u(z). In view of (1.3.2)
v(z) - v(O) =i
tau
an ds. (6.4.7)
o
6.5. NEUMANN PROBLEM 159
When we apply this operation on (6.4.6) and follow through the corresponding
integrations and differentiations, we get
v(peic/» - v(O)
2
1 1 11' • 1pei<l> a [ R 2 + p2 ]
= - u(Retc/» - dsdB
27r 0 0 au R2 + p2 - 2Rpcos (B - ¢) (6.4.8)
where the inner integral is taken on an arbitrary path that lies entirely in the
interior of the circle. Note that
R 2 - p2 Z Z
= --
----=-----=----'------,------,- +-- - 1
R2 + p2 - 2Rp cos (B - ¢) Z - Zo Z - fo
(6.4.9)
= ~ {Z :Zzo -I} = ~ { ; ~ ;~}.
Thus,
~ { Z + Zo } = - 2Rp sin (B - ¢)
(6.4.10)
Z - Zo R2 + p2 - 2Rp cos (B - ¢) ,
and hence
ic/> _
v(pe ) - v(O) - -2
7r
1 10
2
11' iO
u(Re ) R2
2Rsin(B - ¢)
+ P2 - 2R (B ¢) dB.
P cos -
(6.4.11)
If we combine (6.4.6) and (6.4.11), we obtain the Schwarz formula:
.
w(petc/» = i v(O) + -1
27r
1 0
211' • ReiO + peic/>
u(Retc/» ROc/> dB,
e' - pet
(6.4.12)
which allows us to determine the value of a complex potential function f (z) =

u(z)+ i v(z) in a circle with prescribed boundary values u(z) and v(O).
6.5. Neumann Problem
With the help of the above formulas, we can find an explicit solution for the
Neumann problem, for a disk, and thereby for all regions whose conformal
mapping onto the disk is known. In this case the value of the normal derivative
~~ on the boundary of the region is prescribed, and we seek a function harmonic
on D whose normal derivative takes this boundary value. This problem has a
solution and, as in (6.1.1), we require that
1 au
Jr an ds = O. (6.5.1)
Since, in view of (1.3.2),
au
1
Z2
V(Z2) = v(zI) - -a ds, (6.5.2)

Z1 n
the value of v(z) is determined on the boundary up to an additive constant.

With this boundary value v(z) the value of u(z), which is a harmonic function
conjugate to v( z), can be determined in the interior of D by the formula (6.4.11).
Thus,
i¢ _
u(pe ) - u(O) + -2
1r
1 1 0
2
71" iO
v(Re ) R2
2Rpsin(B - ¢)
+ P2 - 2RP cos (B - ¢) dB
= u(O) +~ [v(Re iO ) log (R 2 + p2 - 2Rpcos (B - ¢))]~71"
21r
ior
1
271" av(Re iO )
2
- 21r aB log (R + p2 - 2Rpcos (B - ¢)) dB.
(6.5.3)
Since v(Re iO ) and log (R 2 + p2 - 2Rpcos (B - ¢)) are periodic, the second
term on the right side of (6.5.3) vanishes. Also,
av au au
aB dB = dv = an ds = an RdB, (6.5.4)
so that we finally obtain the following integral formula:
u(Re iO ) = u(O) - -1
21r
1 au
0
2
71" -a R log (R 2 + p2 - 2Rpcos (B - ¢)) dB,
n
(6.5.5)
which establishes a relationship between the boundary values of ~~ and the
values of u(z). We shall denote the expression
log (R 2 + p2 - 2Rpcos (B - ¢)) = log {(z - zo)(z - fo)} = N(z, zo)

(6.5.6)
and call N(z, zo) the Neumann function. It plays the same role for the
second boundary value problem as Green's function does for the first. This
6.5. NEUMANN PROBLEM 161
function represents a regular analytic function of z in D except for logarithmic

2
.
smgu Iarities.
" R , we have
S'mce z- = -
z
(z - zo)(R2 - zzo)
log {(z - zo)(z - zon = log
z
. (6.5.7)
This function can be regarded as a

complex potential function for a flow,
r
which has a source at Zo and a sink ~
at zand for which the circle Izi = R

acts as an impermeable wall
(see figure on the right).
As in the case of Green's function,

for general domains the Neumann function
N (z, zo) can be characterized by the
following conditions:
(i) N(z, zo) is a regular analytic function of z on a region D except for a

logarithmic singularity at z = Zo and at another fixed point z = z*, i.e.,
z - Zo
N(z, zo) = log - -
z - z*
+ N(z, zo), (6.5.8)
where N(z, zo) is a regular analytic function on D.

(ii) N(z, zo) as a function of z has a boundary value which is continuous
everywhere on the boundary f of D, and
r;s N(z, zo) = 0 on f. (6.5.9)
If these conditions are satisfied, then
u(zo) = u(z*) - ~ j N(z, zo) ~u ds. (6.5.10)

271" Jr uu
If the region D is mapped conformally onto the unit disk by the function
w = f(z), then N(z, zo) can also be defined by
-1 [f(z) - f(zo)]~
N( z, Zo ) - og f(z) , (6.5.11)
where z* under this map goes into f(z*) = O.

Note that the Neumann problem is solvable only if the condition (6.5.1) is
satisfied. For a multiply connected region the contour r in (6.5.1) must include
the exterior and all interior paths. The reason why this condition does not hold
for each individual path is that the function u may be multiple-valued, which
does not let the integral of ou/on around r vanish.
6.6. Series Representations
To obtain a series representation for Schwarz formula (6.4.12), we shall develop

a series expansion for the kernel function
Re iB + pe i </> z + Zo
(6.6.1)
Re iB - pe i </> z - Zo
into a geometric series in Zo by using

Z
= 1+2 L L
2 00 kook
--.,..- -1 (ZO) = 1+2 (ZO) eik(</>-B). (6.6.2)
1 - zo/ Z k=l Z k=l Z
This series converges absolutely and uniformly for I~ I < 1, i.e., for Izol < R,
and thus, we can integrate (6.4.12) term-by-term, which yields
(6.6.3)
If we separate this expression into real and imaginary parts and set for k =
1,2, ...
ao = - 1 1 2
71" iB
u(Re ) dB, bo = -v(O),
1 1
211" 0
2 2
ak = -1 71" i9
u(Re ) cos kBdB, bk = -
1 71"
u(Re tB ) sin kBdB,
•
11" 0
11" ° (6.6.4)
6.6. SERIES REPRESENTATIONS 163
which are the Fourier coefficients for the function u( R eill ), then
k
L
00
u(peit/» = (~) (ak cos k¢ + bk sin k¢),

k=O
(6.6.5)
k
L
00
V (peit/» = (~) (-bk cos k¢ + ak sin k¢).

k=O
Since the Fourier coefficients (6.6.4) can easily be computed and the series
(6.6.5) converges inside the circle, this series expansion represents the most
convenient way to solve the first boundary value problem for the circle. The
second boundary value problem can also be handled analogously. Thus,
~u = ~ul
un up p=R
= f
k=O
k(akcos kB+b k sin kB), (6.6.6)
where the Fourier coefficients a k and bk are defined by
1 {21r au
ak = k7r io an cos kB dB,
(6.6.7)
1 (21r au
bk = k7r io an sin kBdB, for k = 1,2, ... ,
and the coefficients corresponding to ao, bo are not needed in this case.
These series expansions show that every harmonic function on a disk and,
hence an analytic function, can be represented by a series (6.6.5). Since u(zo)
and v(zo) are real and imaginary parts of a complex function w = J(zo),
respectively, we obtain from (6.6.5) the Taylor series
00
J(zo) = L Ck z~, (6.6.8)

k=O
where
ak - ibk
Ck = Rk (6.6.9)
This representation leads to the Liouville theorem: Every analytic function that
is regular and bounded in the entire plane is constant. In fact, if f (z) is such
a function, then its real part u(z) must be bounded and so must the Fourier
coefficients ak and bk for any large circle. On the other hand since R can be
chosen arbitrarily large, the quantity ICk I for every k can be made, in view of
(6.6.9), smaller than an arbitrarily small positive number, so it must vanish,

which implies that j(z) = CO = constant, and the theorem is proved.
We obtain yet another integral representation for the Taylor coefficients Ck

by using Cauchy's integral formula (1.2.2) which we write as
j(zo) =~
2t7r
f j(z) dz,
Z - Zo
(6.6.10)
where the path of integration is arbitrary and can be fixed so long as Zo lies
inside a region bounded by it. Now we use the series expansion
(6.6.11)
and integrate term-by-term. If we equate the result to the representation (6.6.8),

we obtain
Ck = 2~7r f :~:~ dz. (6.6.12)
On the other hand, by differentiating (6.6.8) term-by-term, we can show that
Ck = ;, [:~] z=O
. (6.6.13)
Thus, we get the series expansion
L Ck(Z -
(Xl
j(z) = zo)k, (6.6.14)

k=O
where the coefficients Ck can be determined from (6.6.12) or from (6.6.4) and
(6.6.9) with the path of integration as the circle Iz- zol = R. The function j(z)
is regular on the disk Iz - zol < R. Thus, we can expand every regular analytic
function at a point z = Zo into the series (6.6.14) and this series converges in
every neighborhood B(zo, R) about Zo, where the function is regular analytic.
Conversely, every potential series (6.6.14), which converges in a neighbor-

hood B(zo, R). represents a regular analytic function. Moreover, we define
two potential series of the same function j (z):
L L
(Xl (Xl
ck(z - zo)k = ck*(z - zo) = j(z), (6.6.15)

k=O k=O
6.7. PROBLEMS 165
if cic = cic* for all k. In fact, if we compute the coefficients of f (z) by using
(6.6.12) or (6.6.4)-(6.6.9), and substitute them in the series (6.6.15), then we
obtain Ck = cic = ck* (this proves the uniqueness of Ck)'
6.7. Problems
PROBLEM 6.7.1. Approximate Green's function for the regions bounded

by the following curves:
(a) Cassini's oval [(x + 1)2 + y2] [(x - 1)2 + y2] = a4 (Fig. 9.2.1) for a =
1.1, 1.2, and 1.5.
(b) z = 2 cost +i (sint +2cos 3 t),0::; t < 2Jr(Fig. 9.4.1).
(c) Square with round corners x 4 + y4 = 1 (Fig. 9.4.2).
(d) Lima<;on r =a- cos fJ, a = 1.5,2,0 ::; fJ < 2Jr (Fig. 9.6.1).
(e) Bean-shaped curve (Fig. 12.5.4)
z = 2.25 [0.2 coss + 0.1 cos2s - 0.1

+ i (0.35 sin s + 0.1 sin 2s - 0.02 sin4s)], -Jr::; s ::; Jr.
(f) Two hyperbolic arcs of Problem 12.7.4 (Fig. 12.7.2).
(h) Quadrilateral of Problem 12.7.5 (Fig. 12.7.3).
(i) Octagonal curve of Problem 12.7.6 (Fig. 12.7.4).
PROBLEM 6.7.2. Show that Green's function for the upper half-plane
D = {'S{z} ~ O} is given by
z +-
Q(z, zo) = log - io. I
Iz - Zo
(Wayland, 1970, p.333.)

NOTES. cs633.nb
«Graphics' ImplicitPlot'
f[x_,y_,a.] := «x-l/2)- 2+(y-a)- 2)*(1-y- 2-(x-l/2)- 2)-1/10;
ImplicitPlot[f[x,y,l]== 0, {x, -1, 2}];
ImplicitPlot[f[x,y,0.5]== 0, {x, -1, 2}];
ImplicitPlot[f[x,y, 0.3]== 0, {x, -1, 2}];
ImplicitPlot[f[x,y,0.2746687749]== 0, {x, -1, 2}];
(* Plot at a=infinity *)
ImplicitPlot[1-y-2-(x-l/2)-2 == 0, {x, -1, 2}]
(* To find Green's function for a=l, by A. Buzing *)

N[NSolve[f[0.5,y,1]== O,y], 10];
N[NSolve[f[0.8,y,1]== 0, y] ,10];
N[NSolve[f[1.2,y,1]== 0, y] ,10];
N[NSolve[f[x,O,l]== 0, x],10]
(* Seven points on the boundary *)

points = {
{0.5, -0. 9872580108} , {0.5, 0.6034331769},
{0.8, -0. 9402428315}, {0.8, 0.6558861255},
{1.2, -0. 6929674261}, {1.2, 0.5972261642},
{1.474003746, O} };
r [L] : = Sqrt [ (l[ [1] ] -0 . 5) - 2+ l[ [2]] - 2]
radii = Map [r, points] th [L] : = ArcTan [ 1 [[1]] -0.5, l[ [2]] ];
theta = Map[th, points];
rhs = Map[(Log[#]/2(2*Pi»&, radii];
lhs = Table[ If[j==l,
1,
If[j>l && j<.5,
radii[[i]r (j-l)*Cos[theta[[i]]*(j-l)],
radii [[i]r (j-4)*Sin[theta[[i]]*(j-4)] ]
], {i,1,7},{j,l,7}];
coeff = LinearSolve[lhs, rhs);
GreensFunction[rr_,tt_) :=
(Log[1/rr]/(2*Pi»+
coeff[[l)]+Sum[rr-k * coeff[[k+l]) * Cos[k*tt],{k,1.3}]
- Sum[rr-k * coeff[[k+4]] * Sin [k*tt] , {k,1,3});
GreensFunction[rr,tt]
(* fp=f'(O)=e- (-0.0431944) *)
6.7. PROBLEMS 167
fp = Exp[-O.0431944]
cs634.nb
«Graphics' ImplicitPlot,
f [a_] := «x+a)· 2+y· 2) * «x-a)· 2) +y. 2== 1;
ImplicitPlot[f[O] , {x, -1.5, 1.5}];
ImplicitPlot[f[O.5] , {x, -1.5, 1.5}];
ImplicitP1ot[f[l] , {x, -1.5, 1.5}];
REFERENCES USED: Andersen et al. (1962), Carrier, Krook and Pearson

(1966), Gaier (1964), Kantorovich and Krylov (1958), Goluzin (1969), Lawrentjew
and Schabat (1965), Muskhelishvili (1948, 1992), Nehari (1952), Koppenfels (1959),
Wayland (1970).
Chapter 7
Integral Equation Methods
We shall discuss certain integral equations that arise in the problem of comput-
ing the function w = f (z) that maps a simply connected region D, with bound-
ary r and containing the origin, conformally onto the interior or exterior of the
unit circle Iwl = 1. In the case when r is a Jordan contour, we obtain Fredholm
integral equations of the second kind </>(8) = ± l N(8, t) </>(t) dt + g(8),
where </>(8), known as the boundary correspondence function, is to be de-
termined and N(8, t) is the Neumann kernel. We shall discuss an iterative
method for numerical computation of the Lichtenstein-Gershgorin equation
and present the case of a degenerate kernel and also of the Szego kernel. The
case when r has a corner yields Stieltjes integral equations and is presented
in Chapter 12.
7.1. Neumann Kernel
Let r: z = ,(8), 0 ::; 8 ::::: L, be a Jordan contour with continuously turning

tangent and positive orientation with respect to a simply connected region
r
D = Int (r). We say that belongs to the class r~ (and write E r~) if r
r
z' (8) satisfies a HOlder condition of order 0:, 0 < 0: ::::: 1. Similarly, E r~ if
Zll (8) satisfies a Holder condition of order 0:. We assume that w = f (z) maps
the region D univalently onto the unit disk Iwl < 1, such that a point Zo E D
goes into w = 0 and a boundary point z = e iIJ goes into a point w = ei ¢. The
168

7.1. NEUMANN KERNEL 169
Neumann kernel N(s, t) is defined for t =I s by

N(s, t) = sin (7 - Os) = ~ ~Os(t) = _~ or st , (7.1.1)
1r r st 1r ot 1r ont
where 7 = 7(t) is the tangent angle, Os = Os(t) = arg{-y(t) - ,(s)}, rst =

Ir(t) - ,(s)l, and nt is the interior normal at ,(t) (Fig. 7.1.1).
-t-------,+---+- U
o
r
D
Fig. 7.1.1.
This kernel first appeared in the solution of the Dirichlet problem by Carl
Neumann (1877). Some of its properties are as follows:
1. N(s, t) is continuous for t =I s, but, in general, it is not bounded as t ---; s.

2. If r E r~, then N(s, t) Is -
o
til-a and os N(s, t) Is - tl 2- a are bounded
for 0 ~ s, t ~ L.
3. If r E r~, then :s N(s, t) Is - til-a is bounded for 0 :s s, t :s L.
4. If r has a continuous curvature, then for every So
1
lim N(s, t) = -I\;(so), (7.1.2)
s,t-+so,So 2
where I\;(so) is the curvature of r at the point ,(so).

5. The kernel is normalized by
l N(s, t) dt = 1, or l N(t, s) dt = -1, (7.1.3)
since an application of the formula (1.2.9) to j(z) == 1 gives
1 = ~ {sin (7 - Os) dt _ i:.. {cos (7 - Os) dt. (7.1.4)

1r Jr rst 1r Jr rst
170 CHAPTER 7. INTEGRAL EQUATION METHODS
The integral in (7.1.3) takes a Cauchy p.v.

6. From (7.1.1) we have
a 1aa 1aa
-a N(s, t) = - -a -a arg{T(t) - ,(s)} = - -a -a arg{T(t) - ,(s)}
s 7rst 7rts
1 a a a'
= :; at as arg{T(s) - ,(t)} = at N(t, s),
(7.1.5)
h
were
a at
as a = at a as . permltte
a IS . dbecause these mlxe
. dd . . .
envatIves eXist an
d
are continuous for t =J s.
The proofs of these properties can be constructed as in Gaier (1964, pA).

This kernel plays an important role in certain integral equations that arise in
conformal mapping.
Let Ddenote the length of r. Then for all functions I(s) E C[O, L], the
quadratic functional
(I, I) = 11 r r
I(s) I(t) log i-
rst
ds dt, (7.1.6)
is positive-definite. For a fixed D we introduce the Hilbert space H which is

obtained by completing the set of all functions I(s) E C[O, L] with the norm
11/(s)11 2 defined by (7.1.6). Let T denote an operator on H such that for a
continuous function f(t) E H
T1= l N(s, t) I(t) dt. (7.1. 7)
The kernel N(s, t) can be made symmetric by log i-.

rst
Thus,
M(s, t) = 1 r
N(s, x) log i-
rst
dx (7.1.8)
is symmetric, i.e., M(s, t) = M(t, s). This implies that the operator T is
hermitian: (T I, g) = (I, T g). It can also be shown that T is a completely
continuous operator on H. This means that if .Ai, i = 1,2, ... , denote all
eigenvalues of an equation cP = .A T cP, where each .Ai is counted according
to its multiplicity, and if hi is a set of associated eigenfunctions such that
(cPi' cPj) = Dij, where Dij is the Kronecker delta, then
(7.1.9)
7.2. INTERIOR REGIONS 171
and
1 (T</;, </;)
T:Q ::; sup 11</;11 (7.1.10)
for all </; E H with (</;, </;1) = 0, which implies that [ </;( s) ds = 0 since
</;1(S) = const.
7.2. Interior Regions
First, we shall derive the three following integral equations when the function
w = fez) maps Int (f) conformally onto the unit disk Iwl < 1.
7.2.1. Lichtenstein's Integral Equation. Let w = fez), f(O) =

0, map a simply connected region D conformally onto the unit disk Iwl < 1
such that a boundary point z = e ilJ goes into a boundary point w = ei</J. We
shall consider the function
F(z) = log f(z), (7.2.1)

z
which is analytic in D and continuous on fJ, where
arg{F(z)} = arg {J (e ilJ ) e- ilJ } = ¢(s) - 8(s).
An application of Cauchy's formula (1.2.9) on F(z) yields
log fez) =
Z
2.
~1r Jr
rlog f(()~, z
(( - z
E f. (7.2.2)
d( ei{</J{t)-lJs{t))
If we set ( = 'Y(t), z = 'Y(s), then -(- = dt. Thus, equating
- z rst
the imaginary parts on both sides of (7.2.2), we obtain
¢(s) - 8(s) =~ [[¢(s) - 8(s)] sin [</;(t;s~ 8s(t)] dt
+ -1 [1 ()t
ogr
cos [</;(t) - 8s(t)] d
t,
1r r r~
which gives Lichtenstein's integral equation
¢(s) - B(s) =.!:. Jrr N(s, t)

7f
[¢(t) - B(t)] dt + g(s), (7.2.3)
where
s = -1
g ()
7f
1 r
1og r (t ) cos [¢(t) - Bs(t)] dt,
rst
(7.2.4)
and r(t) = 1,(t)l. The integrals in (7.2.3) and (7.2.4) take Cauchy p.v.'s.
The integral equation (7.2.3), derived by Lichtenstein (1917), is periodic in the
angular deformation ¢(s) - B(s).
7.2.2. Gershgorin's Integral Equation. Let D' denote the region

obtained from D by indenting a disk B(O, c) whose boundary is denoted by r'
(Fig. 7.2.1). The function F(z), with F(r(O+)) = i ¢(O), is single-valued on
D'. Then, in view of (1.2.9), for z = ,(s), s =1= 0, we have
log f(z) = ~ r log f(() d(. (7.2.5)

Z7f Jr+r' (- z
D'
Fig. 7.2.1.
:s
Since If(() I A I(I, where A > 0 is a constant, the contribution of the integral
over r' is of order O(e log c) = 0(1). Also, along the cut the integral has the
value log f(z+) = log f(z-) - 2i7f. Thus, as e -* 0, the integral along
1 1(="Y(0) d(
the cut approaches -2i7f -:- -".- , whose imaginary part is equal
Z7f (=0 ., - z
7.2. INTERIOR REGIONS 173
to -2 arg {1'~0~~~~S)} = -2 {3(s) (see Fig. 7.2.1). Hence, equating the
1
imaginary parts on both sides of (7.2.5) we get
I i ¢(t) . ei(¢(t)-es(t» }
¢(s) = 8' { -:- dt - 2 {3(s)
~7l' r r st
or
¢(s) = [N(s,t)¢(t)dt-2{3(S). (7.2.6)
This is known as Gershgorin's integral equation (Gershgorin, 1933).
7.2.3. Carrier's Integral Equation. Carrier (1947) considered the

problem when w = j(z) maps the region D conformally onto Iwl < 1 such
that two interior points P and Q in D go into two points w = ±a, 0 < a < 1,
respectively, i.e., j(P) = a and j(Q) = -a. The function j(z) and the
quantity a are uniquely determined. In fact, if we consider the function
j(z) - a j(z) - a- 1 } .
F(z) = log { j(z) + a - j(z) + a- 1 - ~7l' (7.2.7)
Yls)
D'
Fig. 7.2.2.
in the region D' bounded by r, r' and r" (Fig. 7.2.2), then, in the case when
the boundary of D' is a Jordan contour (i.e., it has no corners), we find by
(1.2.9) that
F(z) = ~ r F(() d(, z = 1'(s), s 1= 0,
1
~7l' Jr+p+ru ( - z
1 F(() l(='Y(O) d( l(='Y(O) d( (7.2.8)
= - --d(-2 -- +2 --
i7l' r ( - z (=p ( - z (=Q ( - z
=h+h+h,
because the value of F(z) is given by F(z+) = F(z-) - 2iJr along the cut
from P to /,(0), and by F(z+) = F(z-) + 2 iJr along the cut from Q to /,(0).
Since h + 12 = 2 log I~ =~~;~ I which is real, we find that ~ {F(z)} = 0

for z E r. In fact, on Iwl = 1, we have from (7.2.7)
w-a w-a-l}
arg { - - . = Jr.
w+a w+a- 1
Hence, taking <1>(8) = ~ {F(z)}, we find from (7.2.8) that
<1>(8) = t N(8, t) <1>(t) dt + 2 log I~ =~~:~ I, 8 -:J 0, (7.2.9)
which is known as Carrier's integral equation. This equation describes the

problem of the potential flow of an inviscid fluid past a periodic array of airfoils
of arbitrary shape (more on this problem in Chapter 10).
Although the problem of conformally mapping the region Ext (r) onto IwI > 1
can be reduced to that of the interior regions of §7.2 by applying the reflection
principle (§ 1.4), the following direct method produces faster converging results
in numerical computations. As before, we assume that r is a Jordan contour,
the arc length 8 is measured in the positive sense, and at z = 00 the mapping
function has the series representation
h(z)=Az+ao+al+a22+···' A>O. (7.3.1)

z z
First, as in §7.2.1, by considering the function F(z) = log h(z) and

z
applying the formula (1.2.11), we obtain the integral equation
¢E(8) - 0(8) = - t N(8, t) [¢E(t) - O(t)] dt - g(8). (7.3.2)
Secondly, as in §7.2.2, we consider the function F(z) = log h(z). Then,

for a fixed z = /'(8), 8 -:J 0, and sufficiently large R > 0, we consider the
region between r and the circle I( - zl = R with a cut from the point z = a
to z = ,(s) + R = (R (Fig. 7.3.1). Let r* denote the boundary of the
resulting simply connected region. Obviously, F(z-) = F(z+) + 2i7r, so that
by (1.2.11) we have
log h(z) = -~
m
r log h(() d(
Jr- (- z
= _~ r log fE(() d( _ 2j(=(R --.!!:L
m Jr ( - z (=a ( - z (7.3.3)
+~ r log h(() d(
27r J1(-zl=R (- Z
== h +I2 +h
Note that ~{I2} = 2 [arg{(z - a) - 7l'}]. Since, in view of (7.3.1), with
( - z = Rei<PE,
~ r
i7rJ1(_zl=R
log h(()
(-z(-z
--.!!:L = 2 log A
'
which is real, we get
~ {I3} = ~ {~r log (( - z) d(} = 2 7r.

27r J1(-zl=R ( - z
Hence, equating the imaginary parts on both sides of (7.3.3), we obtain the
integral equation
<PE(S) = -l N(s, t) <PE(t) dt + 2 arg{r(s) - a}. (7.3.4)
~R
Fig. 7.3.1.
If fE(ZO) = 00 for a finite point Zo E D and fE(OO) = Poo ei'Poo, then the
mapping of D onto Iwl > 1 is univalent only if h(a) = 1. In this case we
obtain the integral equation
cPE(S) = -l N(s, t) cPE(t) dt + 2 arg{,B(s) - <Pool· (7.3.5)
For the external regions under consideration, Eqs (7.3.2) and (7.3.5) are ana-
logues of the integral equations (7.2.3) and (7.2.6).
Now we shall derive two integral equations for cP'e(s), one from (7.3.4)
and the other from (7.2.6). These integral equations will involve the kernel
N(t, s) which is conjugate to the Neumann kernel N(s, t). These equations
are interesting from a numerical standpoint. As opposed to cP E (s), the function
cP'e (s) is periodic with period L, and hence, an application of the quadrature
formula to cP'e (s) increases computational precision.
7.3.1. Banin's Integral Equation. Let r E r~ be a Jordan contour.

If the function h(z), with the series expansion (7.3.1) at Z = 00, maps Ext (f)
conformally onto Iwl > 1, then we can rewrite (7.3.4) as
cPE(S) = -l N(s, t) [cPE(t) - cPe(s)] dt - cPE(S) + 2 arg{r(s) - a}.

(7.3.6)
Since IcPE(t) - cPE(s)1 = 0 (It - sl), and It - sl2-a I:s N(s, t)1 is bounded
by property 2 of §7.1, we find after differentiating (7.3.6) with respect to s that
cP'e(s) = -l :s N(s, t) [cPE(t) - cPE(S)] dt + 2 arg{r(s) - a}, (7.3.7)
where the differentiation under the integral sign is justified in view of the
Lebesgue convergence theorem. Then, using (7.1.5) and integrating (7.3.7),
we get
cP'e(s) = -l %t N(t, s) [cPE(t) - cPE(S)] dt + 2 :s arg{r(s) - a}
= -N(O, s) [cPE(L) - cPE(O)] + l N(t, s) cP'e(t) dt

8
+ 2 8s arg{r(s) - a}
= ~ 80 (s) . 27l' +
_.!.7l' uS 1
r
N(t, s) cP'e(t) dt + 2 ~
uS
arg{r(s) - a},
which yields Banin's integral equation
¢'p;(s) = 1 N(t, s) ¢'p;(t) dt, (7.3.8)
since :s Oo(s) = arg{-y(s) - ,(On·
7.3.2. Warschawski-Stiefel's Integral Equation. If we apply

the method outlined above in §7.3.1 to Gershgorin's integral equation (7.2.6),
we obtain the following integral equation:
¢'p;(s) = -1 N(t, s) ¢'p;(t) dt + 2 d~~) , (7.3.9)
which was established independently by Warschawski (1955) and Stiefel (1956).

If we set r(s) = ¢ E (s) - O( s) in (7.3.9), then the integral equation for r' (s) is
r'(s)=-lN(t,s)r'(t)dt+k(s), (7.3.10)
where
k(s) = -1 N(t,s)O'(t)dt+O'(s). (7.3.11)
7.3.3. Interior and Exterior Maps. In §7.2 we have considered the

problem of determining the mapping function w = f (z) which maps the region
D = Int (r) univalently onto the unit disk U = {IwI < I} such that f (0) = 0
and 1'(0) > O. In §7.3.1 and 7.3.2 we have considered the problemoffindingthe
mapping function w = fE(Z) which maps the region D* = Ext (r) univalently
onto the region U* = {Iwl > I} such that fE(oo) = 00 and lim f;';(z) > O.
z-+oo
These two problems are related to each other by the inversion transformation
z ~ z-l, which transforms the boundary r into a Jordan contour and maps r
D onto D* = Ext (t) and D* onto b = Int (t). Let J and JE be the interior
and exterior univalent maps associated with Then r.
fE(z) = {J (Z-l)} -1,
(7.3.12)
f(z) = {JE (Z-l)}-l
Hence, there is no need to consider the interior and exterior mappings as sep-
arate problems. From the computational point of view, it is convenient first
to determine f(z) and then use the relations (7.3.12) to compute fE(Z). But
in integral equation methods it is advantageous to determine f(z) and fE(Z)
separately.
In each case the conformal maps are determined from the respective bound-
ary correspondence function ¢>( s) and ¢> E (s).
7.4. Iterative Method
As seen from §7.2 and 7.3, the function ¢>(s) = arg{j (r(s))}, 0 ::; s ::; L, in
general, satisfies the integral equation
¢>(s) = A l L
N(s, t) ¢>(t) dt + g(s), (7.4.1)
where A = 1 corresponds formally to the integral equations (7.2.3), (7.2.6)

and (7.2.9) for the interior regions, whereas A = -1 corresponds to equations
(7.3.8) and (7.3.10) for the exteriorregions. We shall present an iterative scheme
for the numerical solution of Eq (7.4.1) for A = 1; the case A = -1 can be
handled by similar iterations. Note that A = 1 is the smallest eigenvalue of the
kernel N(s, t). The associated eigenfunction for the homogeneous equation
£
(7.4.1) is a constant. The eigenfunction for the conjugate kernel N(t, s) is the
equilibrium distribution j.l(t) with j.l(t) dt = 1. Since A = 1 is the only
simple pole of N(s, t) on IAI = 1 and its principal part at this pole is j.l(t)"
I-A
the function
= 2: Ai
00
'Y(s, t; A) [Ni+l(s, t) - j.l(t)], (7.4.2)

i=l
where Ni+l (s, t) denote the iterated kernels with N 1 (s, t) = N(s, t), is analytic
for IAI < IA21, where A2 is the next eigenvalue close to 1 (IA21 > 1). Then the
series
2: Ai [Ni+l(S, t) -
00
j.l(t)]
i=l
converges for IAI < IA21. Since l L

g(t) j.l(t) dt = 0, by the Fredholm theory,
it follows that the Neumann series
2: Ai [Ni+l(S, t) -
00
j.l(t)] + g(s) (7.4.3)

i=l
7.4. ITERATIVE METHOD 179
converges for IAI < IA21, and for A = 1 it represents a solution ¢(s) of Eq
(7.4.1). The main result due to Warschawski (1956) is the following:
THEOREM 7.4.1. Let r E r~, °

< a ~ 1. Suppose that ¢o(t) E
C[O, L] and that ¢o(L) - ¢0(0) = 21r. Then the iterations ¢n(s) defined
by
¢n+l(s) = i N(s, t) ¢n(t) dt + g(s), n = 0, 1,2, ... , (7.4.4)
i i
converge uniformly to the solution ¢(s) of Eq (7.4.1) with A = 1, such
that ¢(s) pes) ds = ¢o(s) pes) ds. More precisely,
!¢n+l(s) - ¢n(s)! ~ 1~21n II N (s, t)III1¢~ - ¢/II JA{~ l'

1r (7.4.5)
In fact, by the Schwarz inequality,
II¢~ - ¢/II ~ ii (lOg r~J 2 dsdt· i (¢b - ¢/)2 dt.
The factor II¢~ - ¢/II in the error estimate (7.4.5) must be small which happens
I¢~ -
when max
0$8$L
¢/I is small or when (
Jr (¢~(t) - ¢/(t)) dt is small. Thus,
the factor lI¢b - ¢/II in the error estimate (7.4.5) plays a useful role.
To prove the uniform convergence of the iterations ¢n (s) to the solu-
i
tion ¢(s), it suffices to assume that ¢(s) exists and satisfies the condition
g(s) ¢(s) ds = 0, and that ¢n(s) E C[O, L] for all n = 0,1,2, ... , where
¢o(L) - ¢0(0) = 21r. Then
I¢n+l(s) - ¢n(s)1 ~ IA~ln-l IIN(s, t)IIII¢~ - ¢bll JA2A~ l'

1r (7.4.6)
which leads to the solution
+ I.: [¢n+l(s) -
00
¢(s) = ¢o(s) ¢n(s)]. (7.4.7)

n=O
ESTIMATES FOR IA21: The inequality (7.4.6) gives an estimate for the rate
of convergence. A result on the convergence of the derivatives ¢~ (s), which is
due to Warschawski (1956), is as follows:
THEOREM 7.4.2. If r E r~, a< a ::; 1, then the derivatives 4>~(s)

converge uniformly to 4>'(s), a ::; s ::; L. More precisely, the following
estimate holds:
(7.4.8)
Let r o be a Jordan contour and No(s, t) denote the associated Neumann

kernel. Suppose that the second eigenvalue A2 of No(s, t) is known. For
example, for an ellipse b2x 2+a 2y2 = a2b2, A2 = a + bb; for a circle x 2+y2 =
a-
a 2 , A2 = 00. Then estimates for >'2 can be given in terms of A2 in the following
cases:
(a) Let r o be close to r E r~, a < a ::; 1, in the sense that either r o c Int r or
r c Int r. The former situation corresponds to the case of the interior regions
(§7.2) when w = f(z) maps Int(r) onto the unit disk Iwl < 1, whereas the
latter corresponds to the case of the exterior regions (§7.3) when w = f (z)
maps Ext (r) onto the unit disk Iwl > 1 such that z = 00 goes into w = 00.
Let
max If'(z)1 2
Iwl=l and M = f &N~:, t) I dt. II
q = min If'(z)I' Jfr
Iwl=l
If d is the Frechet distance* of rand r 0, then
1 1 2
Cl ::; 1>'21 ::; A2 + ad>'2, (7.4.9)
where a = 2 qM/7l', andcl is the real root of the cubic equation d a x 3 +x/ A2 =
1.
(b) Let r E r~ and roE r~, and let contours rand r 0 have the same length
8. Suppose that for some choice of the points corresponding to s = 0 on each
contour
ii (N(s, t) - No(s, t))2 dsdt = €2,
f f (log ..!- _log ..!-) 2 ds dt = 1/2,

Jr Jr rst Pst
*See (1.1.20) for its definition.

where Pst = Izo(s) - zo(t)l, No(s, t) is the Neumann kernel associated with
r 0, and Zo (s)
is the parametric representation of r 0 in terms of the arc length
parameter s, °::;
s ::; L. Then
(7.4.10)
where B = rr
Jr Jr
(log ~)
Pst
2 ds dt, C2 is the real root of the cubic equation
a8 ) x 3 +-x
a8 2 +-=1,
x
( --+E:B
27r A2 27r A2
and a is the same as in (7.4.9).
(c) If r E r~ for 1/2 < a < 1, i.e., r is the boundary of a nearly circular
region, then
;2 : ; ~ [l N (s,t)ds -1],
2 (7.4.11)
where, by the Schwarz inequality,
N 2 (s,t) = II N(s,x)N(x,s)dxds::; II 2
N (s,x)dxds,
Jr
and
;~ : ; ~ [ii N 2
(s,x)dxds -1] ~ [ii
= (N(S,X) - dXdS].
(7.4.12)
Since the kernel No(s, t) = J for a circle of circumference 8, the condition
that the last integral in (7.4.12) be less than unity implies that r is a near circle.
°stant,
(d)
°
Neumann's lemma states thatifr is a convex Jordan contour, then N(s, t) ~
and there exists a constant K,. < K, < 1, known as the Neumann con-
which depends only on r and has the following property: Let g*(s) =
l N(s, t) get) dt, where the function get) is bounded and integrable on r,
°::;
Then
t ::; L. Let m ::; get) ::; M on [0, LJ and m* ::; g*(s) ::; M* on [0, LJ.
M* - m* ::; (M - m) (1 - 4 (7.4.13)
A Jordan contour is said to be nearly convex if it satisfies the following cri-

terion: There exists a convex Jordan contour f o such that (i) f o has the same
length as f, and (ii) if N (s, t) and No (s, t) are the kernels associated with f
and f o, respectively, then for all s E [0, L]
lIN(s, t) - No(S, t)/ dt :::; € < K,.
If f is nearly convex, and if ¢o(s) E e[o, L] is an arbitrary function, then the

iterations (7.4.4) satisfy the inequality
(7.4.14)
where V is a constant; V :::; Wo + 21f, and Wo is the oscillation of ¢o (s) in

[0, L]. If ¢o(s) is nondecreasing and Wo = 21f, then V = 21f. Finally, if ¢o(s)
I¢o(s) - ¢(s)1 :::; "1, °:: ;

is an approximation of the solution ofEq (7.4.1), i.e., if it is known a priori that
s :::; L, for some solution of (7.4.1), then V :::; 2"1.
If f 0 is a circle, then K, = 1. The Neumann constant K, characterizes a nearly
circular region in a manner different from that presented in Chapter 5.
For computational purposes, the best method for numerically solving the
integral equation (7.4.1) is to discretizethe integral and replace the equation by
a matrix equation. Thus, the problem becomes one of matrix inversion. To do
this, we partition the boundary f into n parts at the points
and obtain for the values ¢(tj) of ¢(s) at the n points tk = kL/n, k =
1,2, ... ,n, the following system of n linear equations where the integral in
(7.4.1) is replaced by a sum:
n L
¢ (tk) = L N (tk' tj ) ¢ (t j ) ;: + 9 (tk),
j=l
or
n
L[Ojk - N(tk,tj)] ¢(tj) = L9(tk)' (7.4.15)
j=l
In practical applications, since the boundary f cannot be divided into partitions

of equal length, it is useful to take more partition points on those portions of f
where the curvature is positive and larger. This is accomplished by transforming
the arc length parameter t into an integration variable 7 such that t = 'I/; (7),
S = 'I/;(a), '1/;'(7) > 0, a ~ s, t ~ L, a ~ 7 ~ l, and '1/;(7) is small (large)
according as the curvature is large (small). This substitution transforms Eq
(704.1) into
(7.4.16)
which, after discretization with partitions of equal length in 7, yields the matrix
equation
n
2: [Ojk - N* (ak' 7j)] ¢* {Tj) = ~ g* (7k) , (7.4.15')
j=l
which is similar to (704.15).
In this method the matrix inversion is of O(n 3 ), which becomes consid-

erably large if n is large. To overcome this difficulty, an iterative method is
used where the computations are of order O(mn 2 ), m being the number of
iterations. This iteration starts with a function ¢( s) = ¢o (s), called the initial
guess, that is taken close to the correct solution. Then, with this function the
right side of Eq (704.1) is computed, which yields
¢l(S) = 1£ N(s, t) ¢o(t) dt + g(s),
and the iterative process is repeated n times. It leads to Eq (70404) for n =

1,2, ... , which, by Theorem 7.4.1, converges uniformly to the solution ¢(s).
In numerical computation, the rate of convergence is fast only if the region

D is nearly circular, i.e., if it can be approximated in polar coordinates by the
function r = r(e), a ~ e ~ 27r, r(O) = r(27r), which belongs to the class
C 2 [0, 27r], is almost constant for all e, i.e., r( e) ~ const, and has a small first
derivative r' (e) « 1. Thus, the algorithm for solving Eq (704.1) is as follows:
1. Check if r is a Jordan contour with no corners. In case r has corners, they
should be first analyzed by the methods of Chapter 12.
2. Use elementary conformal mapping (like, log, exp, sin, cos functions) to
make the region D "circular" (see Fig. 704.1).
3. Carry out the iterations (704.4) using the discretized formula (7.1.15) or
(7.1.16).
4. Stop the iterations when the difference I¢m(s) - ¢m-1(s)1 < c, where
c > 0 is a preassigned quantity (called the tolerance).
,,
,, ,,
, ,,
,
I \
,
I
r 6 ,
I
I
,
I I
I
,
I I
, I
I
\ I
\ I
\
,, ,I
, ,
, ,,
Fig. 7.4.1.
If c denotes the rate of convergence, i.e., if it is the largest number greater

than unity such that
for all 8 and n, then an upper bound for the error made by taking 1>m (8) ~ 1>(8)
is given by
c c c
11>(8) - 1>m(8)1 :S - + - 2 + ... = - . (7.4.17)
C c c-l
The value of c may be approximately estimated during the iteration process.
Since c > 1, we find from (7.4.17) that the error is smaller than c. However, in
the entire computation, besides this error, we have the discretization as well as
round-off errors.
The eigenvalue Al is important in numerical computations. Ahlfors (1952)

has given a simple estimate for 1/ AI which is called the convergence factor. If
the boundary r is defined in polar coordinates, as above, by r = r(B) and if
Vo = max v, where v is the angle between the radius vector and the normal
(Fig. 7.1.1), then the Ahlfors estimate is given by
1
)'1 ~ - . - . (7.4.18)
sm Vo
This estimates for an ellipse E, defined in the z-plane by
x=acost, y=bsint, O:St:S21r, (7.4.19)
with foci ±1, semi-axes a and b, a> b, a 2 - b2 = 1, and axes-ratio k = alb,

and for Cassini's oval Iz2 - 11 = k (see §6.3) are given below (Andersen et al.,
1962, p.190):
FOR THE ELLIPSE:
k = 1.2 : esc Vo = 5545, and C = -11.0,
k = 1.6 : esc Vo = 2282, and C = -4.4,
k = 2.0 : esc Vo = 1667, and C = -3.0.
FOR CASSINI'S OVAL:
k = 1.2 : esc Vo = 1.2, and C = -2.0,
k = 2.0 : esc Vo = 2.0, and C = -3.8,
k = 5.0 : esc Vo = 5.0, and C = -10.0,
where ICI denotes the rate of convergence which is sufficiently large for nearly
circular regions. The negative sign for C indicates that the iterations "oscillate".
CASE STUDY 7.4.1. Let r be the boundary of an ellipse E defined in the

z-plane by (7.4.19). It is well known that the function 2 z = w + w- 1 maps
the circle Iwl = R = a + b, R > 1, conformally onto E. Then the function
f (z), which is univalent in Int (E) and If (z) I = 1 on E, is regular in Int (E)
(real at z = 0) and maps the ellipse E conformally onto the circle jwl = R
such that it satisfies (Szego, 1950)
f(z) 2 (-I)n 2R- 2n

+ L -n-
00
log -z- = log R R2n + R-2n T 2n (z), (7.4.20)

n=l
wn + w- n
where Tn (z) = 2 are the Chebyshev polynomials of the first kind.
(a) To prove that the function w = f(z) maps Int (E) univalently onto Iwl < R,
we shall use the argument principle and show that as z goes around the ellipse
E, the point w describes the circle Iwl = R exactly once and in the same
direction, i.e., d~ ~ {log f(z)} > O. We find from (7.4.20), with w = Rei<f>,
that
Thus,
d
d¢ ~ {log f(z)} = ~
{i (Re i¢ - R- I e- i¢) }
Rei¢ + R-I e-i¢
R2n _ R- 2n
+ 2 2:) _1)n
00
2n
R2n + R-2n R- cos2n¢
n=1
== Al + A 2 ·
1 _ R- 2 e-2i¢
= ~ {I + R-2 e- 2i ¢} = 1 + 2
00
2n
Since Al 2:)-I)n R- cos2n¢, we get
n=1
d
= Al + A 2 = 1 + 2 2:) _1)n R- 2n
00
d¢ ~ {log f(z)} cos 2n¢

n=1
R2n _ R- 2n
2:) _1)n R2n + R-2n R- 2n cos 2n¢
00
+2
n=1
2
2) _1)n R2n + R-2n
00
= 1+2 cos2n¢
n=1
2 n
p~~_ {I + 2 2:) _1)n
00
= R2n : R-2n cos 2n¢}.
.1
n=1
00
cos ax 7r
If we define R 2 = eO"", then Since h( j ) dx = h ' we have
o cos x 2 cos a7r
for n = 0,1, ...
2 = 2 = 1 =.!. roo cos nax dx
R2n + R-2n en",.". + e- n",.". cosh na7r 7r Jo cosh(xj2) .
Hence,
2
I) -It R2n:R-2n cos2n¢
00 n
1 +2
100
n=1
= -1 h(1 j) (00
1 + 2 2)-p)n cosnax cos2n¢ ) dx
7r 0 cos x 2 n=1
1 roo 1 1 _ p2
= 27r Jo cosh(xj2) C+ 2pcos(ax + 2¢) + p2
+ 1 - p2 ) dx > 0,
1 + 2pcos(ax - 2¢) + p2
by (6.4.9) and (6.4.10).
(b). We shall determine the exact solution for the boundary map fez). Let
w = ei (1T/2-x) = ie- ix , and z = cos(1f/2 - x) = sinx. Then
fez) 2 1 2 R- 2n
+ 2::; R2n + R-2n cos2nx.
00
log -z- = log R
n=l
Since
1 2R- 2n (2K )
~:;
00
R2n+R-2n cos2nx = log sn 1f x,k -log (2 q 1/4)
+~ log k -log sinx,
(see Whittaker and Watson, 1927, p.509, Ex. 3), where q = R-4, sn is a
Jacobian elliptic function of modulus k, 0 < k < 1 (see §2.3), we find that
2KX ) = ~sn (2K

fez) = ~sn ( -1f-,k . 1 z,k ) ,
--:;;:- sm- (7.4.21 )
which is the Schwarz formula.
(c) For the interior regions (§7.2) the boundary correspondence function is given
by
00 (_l)n R4n_1
<p(t) = arg{j(z)} = arg{z} +~
~
-n- R- 2n R4n+1 sin2nt
n=l
2 2
= tan -1 {(R _1)2 - 2 cos2 t tan t } - 2 2: -00
(_l)n -2n
-R
sin2nt
.
(R2 + 1)2 - 2 sin t n=l n R4n + 1
(7.4.22)
(-1) n 2 sin 2nt
2:
00
IfwesetL· = 2 --R- n then
J . n R4n + l'
n=J
2R6 1
ILj I ~ R6 _ 1 j R6j , (7.4.23)
and the function <p(t), defined by (7.4.22), can be approximated by

<p(t) ~ <Pj(t)
= tan
-1 {(R 2 _1)2 - 2 cos 2 t } 8(_1)n R- 2n sin2nt
tan t - 2 ~ - - ,
(R2 + 1) 2 - 2 sin2 t n= 1 n R4n + 1
(7.4.24)
with an error (uniform in t, 0 ~ t ~ 2?T) given by (7.4.23). In computations j

in (7.4.24) is usually estimated so that
6
2R 1 -12
R6 _ 1j R6j < 10 .
Thus, j becomes larger, the closer R is taken to 1. Todd and Warschawski

(1955) have taken the minimum value of R in their computations as R = -/1.5;
then j = 22. The function actually computed is ¢22 - ?T for the values of
t = 0° (3°) 90°. Since these values are symmetric about the y-axis, they are
extended in other quadrants. Then the results obtained by computation from
(7.4.24) are compared with those from (7.4.21).
(d) In the case of exterior regions (§7.3), ¢(t) = t. Note that ),2 = : ~~ for
the ellipse E. The function
Q(z,O) = ~ {lOg ftz)} = log IftZ)1
is Green's function of Int (E) with respect to the origin.•
CASE STUDY 7.4.2. We shall determine the correspondence between the

boundaries in the case of mapping the ellipse E, defined by (7.4.19), onto the
unit circle. The boundary correspondence function ¢ is normalized so that the
origin is preserved and ¢(O) = -?T. Using the parameter t instead of the arc
length s, we find from (7.4.19) that
~: = J a 2 sin2 t + b2 cos 2 t = b J k 2 sin2 t + cos 2 t.

Then, Gershgorin's integral equation (7.2.6) can be written as
¢(7) = -1
7r
10
27T
N(7,t)¢(t)dt+g(7), (7.2.25)
where
g(7) = -2,8 (S(7)) = 2 tan- 1 { 2 k sin 7 . 2 }, (7.4.26)

k (cos 7 - cos 2 7) - sm 7
and
N(r, t) = -r========k=/2=======
t- r [ t- r t - r] 2
k 2 sin 2 - - + k 2 sin t sin - - + cos t cos - -
2 2 2
x 2t-r t-r
k 2 sin - - + cos 2 - -
2 2
k
(k 2 + 1) - (k 2 -1) cos(t + T)·
(7.4.27)
The branch of arctan in (7.4.26) is chosen such that g(O) = T-+O+
lim g(T) = -1r
and g(21r) = lim g(T) = 1r. Since ¢(21r - T) = -¢(r), we can write Eq
7"-+27r-
(7.4.25) as
¢(r) = ~
1r Jo
r [1 - k k1cos(t
¢(t) _ k 1 ¢(t - T) ] dt
+ r) 1 + k cos(t + T)
2 2
(7.4.28)
+ 2 tan- 1 { ~nT } ,
k(l - cosr) (k 3 COST - k- 2)
where
1 k2 - 1 1
k1 = k2 + 1' k2 = k 2 + 1 ' k3 = 1 - k 2'
Eq (7.4.28) can be represented in the operator form as
¢(T) =T¢+g(T), (7.4.29)
where T is the integral operator in (7.4.28) and g(T) is the arctan term. If the
initial guess is taken as ¢o = g(r), then the iterations
¢n+!(T) = T ¢n + g(T), (7.4.30)
lead to the equation
¢n+1 = go + g1 + ... + gn, go = g, T gn = gn+!· (7.4.31 )
Thus, first we must compute ¢o = g(r), and then compute the integrals T gn
which are replaced by an appropriate approximate quadrature. For k = 1.2
Todd and Warschawski (1955) found that at a fixed t the functions g( r) and
N(r,t) behave approximately like g(T) ~ 1.5(1 + cosr) and N(r,t) ~
0.5 + 0.1 COST. Weddle's rule was used for quadrature (see Birkoff et al.,
1950, 1951). The results for <Pn for k = 1.2 are computed against t at a step-
size of 3° from t = 0° to t = 180° and presented in Fig. 7.4.2 ( t along the
horizontal axis). Similar computations can also be carried out for k = 2 and
5. The results match the exact value given by (7.4.22). These computations,
though carried out for an ellipse, are well suited for other types of nearly circular
regions.
2.5
1.5
0.5
o 30 60 90 120 150 180
Fig. 7.4.2.
CASE STUDY 7.4.3. The integral equation for the dipole distribution is
1
given by
L
p,(8) = N(8, t) p,(t) dt + g(8), (7.4.32)
where the dipole strength on the boundary f is 211" p,(8), g(8) denotes the bound-
ary value of the potential function, and t, 0 ~ t ~ L, is the arc length parameter
along the positive direction of f. The integral equation (7.4.32) has the same
form as (7.4.1). Let us assume that the distribution p,(8) has already been de-
termined by solving Eq (7.4.25). Then the Dirichlet problem and hence the
problem of conformal mapping is reduced to that of quadratures. To see this,
note that the potential u(P) at a point P = (x, Y) E D is given by
u(P) = 1 L
N(P, t) p,(t) dt, (7.4.33)
where N(P, t) = - f:38 TPt (Fig. 7.4.3). The kernel N(P, t) becomes un-
Unt
bounded as P approaches the boundary f. To avoid this difficulty, we proceed
as follows: Consider a point P' near r (Fig. 7.4.3). Let t' be a point on r such
that the normal nt' passes through the point P'. Then, we can rewrite (7.4.33)
as
u(P) = tt(t') + 1 L
N(P, t) [tt(t) - tt(t')] dt
(7.4.34)
l
L/2
= tt(t') + N(P, t + t') [tt(t + t') - tt(t')] dt,
-L/2
"1((') P'
D
r
Fig. 7.4.3.
where t is replaced by t' + t, and N (P, t' + t) = - - 88 r Pt. The integral in

nt'+t
(7.4.34) is finite for all t, and tt(t' + t) - tt(t') ~ t tt'(t') for numerically small
values oft. Then (7.4.34) can be written as
l
L/ 2 8
u(P) = tt(t') - {[tt(t' + t) - tt(t')] -8- rpt
-L/2 nt'+t
(7.4.35)
+ [tt(t' - t) - tt(t')] - 88 r pt } dt,
nt'-t
since
1r
o
L
8 rpt =
-8
nt
1 0
27T
dw = 211', and l
0
L
-8
8 log
nt
rpt = 1
0
7T
dw = 11'
for ')'(s) E (Fig. 7.4.3).•
CASE STUDY 7.4.4. If U and U* denote the region Izi < 1 and Izl > 1,
respectively, z = eiO , and if D = Int (r), as before, then for certain curves the
following data is useful (Gaier, 1964, p.264):
1. Eccentric circle in the w-plane: Let D ; Iw - al < b, b > a > O. Then
r : p = p(¢) = a cos ¢ + Jb 2 - a 2 sin 2 ¢;

. D
Mappmg ~
U :z = bw
b 2;
aw+ 2 -a
. sin¢ a
Boundary correspondence functIOn: tan B = () ,c= - b2 .
cp ¢ + cos¢ 2 - a
2. Inverted ellipses: The region D is formed by inverting the exterior of the

ellipse with half-axes lip and 1 into the unit circle.
. U D 2pz
Mappmg ~ :w = (l+p+l-pz
) ( ) 2;
r : p = p(¢) = Jl - (1 - p)2 cos 2 ¢, 0 < p < 1;
Boundary correspondence function: tan ¢ = p tan B.
3. Ellipses, with boundary r : z = a cos t + i b sin t, a 2 - b2 = 1;

For the mapping Int (r) ~ U, the boundary correspondence function is defined
by (7.4.22);
For the mapping Ext (r) ~ U*, the boundary correspondence function is
defined by ¢(t) = t.
The boundary correspondence function determines the correspondence be-

tween the boundaries of the two regions where one is the problem region and the
other the image region. It is denoted by ¢ when a given simply connected region
in the z-plane is mapped onto a region in the w-plane (z = r ei () ~ w = p eicf>,
which is the notation used in this book), but by B if w = p eicf> ~ z = r ei ().
This function is usually normalized in so that the origin is preserved. •
7.5. Degenerate Kernel
A kernel is said to be degenerate if it can be represented in the semi-discrete

form of a finite sum ofproducts. A degenerate Neumann kernel can be expressed
as
n
N(s, t) = L Dk(S) fA(t), (7.5.1)
k=l
where it is assumed that the functions Dk (s) are linearly independent. Otherwise
the number of terms in the expression for the kernel in (7.5.1) would reduce.
For such a kernel we can determine a complete solution of the Fredholm integral
7.5. DEGENERATE KERNEL 193
equation of the form
</>(s) = A 1£ N(s, t) </>(t) dt + g(s). (7.5.2)
Let us assume the required solution ofEq (7.5.2) with A = 1 in the form
n
</>(s) = g(s) + L Ai Qi(S), (7.5.3)
i=1
where Ai are constants yet to be determined. We shall use the notation:
(7.5.4)
Then substituting (7.5.3) for </>(s) in Eq (7.5.2), we get
~AiQi(S) - A 1£ (~Qi(S)/3i(t)) g(t)dt
1£ t t
- A
o i=1 j=1
A j Qi(S) /3i(t) Qj(t) dt = O. (7.5.5)
Equating the coefficients of Qi (s) and using the notation (7.5.4), we find that
n
Ai - A LA j bi,j = A Ii. i = 1,2" .. ,n. (7.5.6)
j=1
The determinant of the system (7.5.6) in the unknowns Ai is
D(A) = (7.5.7)
-Abn,1 -Abn,2 1- Abn,n
Then, by Cramer's rule,
n
L: Di,k fk
Ai = A k=~(A) , i = 1,2, ... ,n, (7.5.8)
where Di,k is the algebraic complement of the k-th row and i-th column.
Hence, from (7.5.3) the approximate solution ¢( s) of Eq (7.5.2) is given by
n
n L: Di,k!k
¢(s) = g(s) + A ~ k=~(A) ai(s)
n(n
L: L: Di,k ai(s) Io£ ) f3k(t) g(t) dt
= g(s) + A k=1 .=1 D(A)
(7.5.9)
n n
= g(s) +A 1 £ L: L: Di,k ai(s) f3k(t)

k=1 i=1 D(A) g(t) dt
= 9 s()+ A ir£o D(s,D(A)t, A) 9 ()t dt,

where
n n
D(s, t, A) = L L Di,k ai(s) f3k(t). (7.5.10)
i=1 k=1
Thus, the resolvent,( s, t, A) of Eq (7.5.2) is given by
D(S,t,A)
D (S,t,A ) = D(A) . (7.5.11)
In practical applications an arbitrary kernel N (s, t) can be replaced ap-

proximately by a degenerate kernel N(S, t), which will help solve the resulting
approximate equation
¢(s) - A 1£ N(s, t) ¢(t) dt = g(s), (7.5.12)
rather than the original equation (7.5.2). A suitable degenerate kernel close to
the one given in an integral equation can always be found by taking it as a finite
part of its Taylor series (as in Case Study 6.5.1 and 6.5.2), of its Fourier series,
or by an trigonometric interpolation scheme (see §8.7). An error estimate in
replacing the given kernel by an approximate degenerate kernel is contained in
the following result:
THEOREM 7.5.1. Given two kernels N(s,t) and N(s,t) such that
1£ IN(s, t) - N(s, t)1 dt < h,

let the -resolvent ')'(s, t, A) of Eq {7.5.2} satisfy the inequality
If the conditions Ig( s) - g( s) I < c and 1 -IAI h (1 + A B) > 0 are satisfied

where g(s) is an approximation of g(s), then Eq {7.5.2} has a unique
solution <1>( s), and
- I M IAI (1 + IAI B)2

I¢(s)-(s) < l-IAlh(1+AB) +c(1+IAIB), (7.5.13)
where M is the maximum modulus of g( s).
A proof of this result can be found in Kantorovich and Krylov (1958, p.143),
or Berezin and Zhidkoy (1965, p.64?).
CASE STUDY 7.5.1. (a) Solve the integral equation
fl/2
(s) = J sinst¢(t)dt+g(s), (7.5.14)
o
with no assumptions on g( s) at this time. Expanding sin st in its Taylor series
we get
s3t 3 s5t 5
sin st = st - - +- - ....
3! 5!
If we replace sin st in (7.5.14) by the first two terms of this series expansion,
then Eq (7.5.14) reduces to
¢(s) =J
fl/2 (
st -
s3 t 3) ¢(t) dt + g(s),
6 (7.5.15)
o
which has an algebraic kernel. We shall assume a solution of the form
¢(s) = as + bi + g(s).
Then substituting it in (7.5.12) we get
3 _ _ 3 _ as _ ~ as 3 bs 3 _ 0
as+bs sft s h 24 160 + 768 + 5376 - ,
where
II =
1 1/2
t g(t) dt, 12 = --
1
6
1 1 2
/
t 3 g(t) dt.
0 0
Equating the coefficients of a and b to zero, we obtain
23a b
24 - 160 - II = 0,
a 5377b
160 + 5376 - 12 = 0,
and
¢(8) = g(8) + a8 + b8 3 = g(8) + Jo

r1 ,2
,(8, t, 1) dt, (7.5.16)
where the resolvent ,(8, t, 1) is given by
,(s, t, 1) = 1.043277 (1.000186 s t - 0.0010416 s3 t - 0.0010416 S t 3

- 0.15972228 3 t 3 ).
1
1/2 1
Since I,(s, t, 1)1 dt < -, we can take B = 1/12 in the above estimate
o 12
(7.5.16). Also, since
r
Jo
1 2
/ 1 _
N(8, t) - N(s, t)
1 r
dt ~ Jo
1 2
/
5 5
8 t 1 (1) 5
120 dt ~ 46080 '2
1
< 1474560'
1 3
we can take h = 1474560 ~ 4.10 6 ' Then
3
_ --6 (1 + 1/12) M
14>(8) - 4>(8)1 < M 4.10 < 106 '
3
1 - - - 6 (1 + 1/12)
4·10
In particular ifg(8) = 1 + ~ (cos:' - 1) = 1-:' + ~ _... then M = 1

, s 2 8 384' ,
and the approximate solution is
4>(8) ~ 1 + 0.0000009 8 - 0.00000028 3 ,
which has an error of 0 (10 6 ).

(b) To solve the equation
¢(s) = 1 1
sin st ¢(t) dt + g(s), (7.5.17)
we take ¢(s) = as + bs3 + g(s). Then
s3 t 3) (as + bs 3 + g(s)) dt
as + bs 3 = J( ( st - -6-
o
. d f' 2a
ylel s the system 0 equatIOns 3" - 5"b - f 1 = 0, a
30 + 43 b f
42 - 2 = O. The
resolvent is
Since1 1
b(s, t, 1)1 dt ::; ::~ < 1, we take B = 1. Also, since
Jr IN(s, t) - N(s, t) I dt ::; 720'

1
1
o
we take h = 1/720. Then
1
-(2) M
I¢(s) - ¢(s)! < M 7201 < 103 '
1- - (2)
720
In particular, if g(s) = 1 + ~ (1- coss) = 1- ~ + ;: - ... , then M = 1,

and the approximate solution is
¢(s) ~ 1 +0.001545s,
which has an error of 0 (10 3 ) . •
CASE STUDY 7.5.2. To solve
¢(s) + 11
s (est -1) ¢(t)dt = eS - s,
we take
- 1 1
N(8, t) = 8 (est - 1) = 82t + 2' 83t 2 + 684t3.
Then we shall solve the equation
1>(8) + 1 1
N(8, t) 1>(t) dt = e
S
- 8,
where we have
1>(8) = e S - 8 + a8 2 + b8 3 + C8 4.
Substituting this expression for 1>(8) into the above equation, we obtain the
system
5 1 1 2
- a - - b+ -
C = --
4 5 6 3'
1 13 1 9
-a--b+-c=--e
5 6 7 4 '
1 1 49 29
- a - - b + - c = 2e - -
6 7 8 5'
whose solution is a = -0.501019, b = -0.167126, c = 0.0418054. Thus
1>(8) = e S - 8 - 0.5010198 2 - 0.1671268 3 - 0.04180548 4 .
The exact solution of the given equation is ¢>(8) == 1. Note that 1>(0) = 1,
1>(1/2) = 0.999963, and 1>(1) = 1.00833.•
7.6. Szego Kernel
Let D be a bounded simply connected region with a Jordan boundary r. The

Cauchy kernel
1 1
H(z,a) = -.- - - a E D, (7.6.1)
2m z - a
defined by (1.2.3), represents an analytic function f(z) on D with
f(a) = ~
221l'
r f(z) dz = ~ Jor
Jr z - a 221l'
L
f(z) ,'(8) d8,
z- a
a E D, (7.6.2)
7.6. SZEGO KERNEL 199
wherez = -y(s),dz = -y'(s)ds,O ~ s ~ L,istheparametrizationofrinterms

of the boundary arc s. The Szego kernel, defined by (4.3.3), also represents an
analytic function f on D with
f(a) = lr S(z, a) f(z), dz, a ED. (7.6.3)
Let 1i2 (r) denote a closed subspace of L 2 (r), containing boundary values of
analytic functions on D, and let S : LCr, ds) I--> 1i2 (r) denote the orthogonal
projection. Then for any f E L 2 (r)
S f(a) = lr S(z, a) f(z), dz, a ED. (7.6.4)
This relation implies (7.6.3) because f = Sf in (7.6.3). The Szego kernel

coincides with the Cauchy kernel
1 1
H(z,a) = -2' ---y'(s), (7.6.5)
21r z - a
iff D is a disk. In fact, the Szego kernel for the unit disk U, denoted by Su (z, a),
is given by
1 1
Su(z,a) = -2 - - - , lal < 1, Izi = 1. (7.6.6)
1r 1 - az
The following result holds for an analytic function f : D I--> U such that
° °
f(a) = and f'(a) > real, where a E D. Such a function is called the
Riemann mapping function (see Theorem 1.4.1).
THEOREM 7.6.1. For a given a E D let an analytic function w =

f (z) be the Riemann mapping function. Then
f'(z) = S(~~ a) S2(z, a), zED. (7.6.7)
PROOF. The following transformation connects S to Su:
S(z, () = vi f'() Su (f(), f(z)) vi f'(z) for z, ( E D. (7.6.8)
Set ( = a. Then f(a) = 0, and (7.6.6) yields

1
S(z, a) = 21r vi f'(z) vi f'(a),
which implies (7.6.7).•
We use (7.6.7) to compute the Riemann mapping function by the formula

1 f'(z) ,
f(z) = i If'(z)[ , (8). (7.6.9)
This formula is used by Kerzman and Trumrner (1986) for numerical computa-
tion of f(z). Now we shall define the Kerzman-Stein kernel A(z, a) in terms
of the Cauchy kernel by
A(z,a) = H(z,a) - H(z,a) E COO(f x r)
1
= - 2i1r ~
{,"(8)}
,'(8) .
(7.6.10)
Note that A(z, z) = 0 since ,"(8) is orthogonal to ,'(8). Let A : L 2 (f) t--+
L(f) define the integral operator
Af(a) = l A(z, a) f(z) ds, a, z E r, z = ,(s). (7.6.11)
This operator is compact, and iA is self-adjoint. The operator (1 - A) is

bijective on L 2 (f) t--+ L 2 (f), and hence it has a bounded inverse (1 - A)-l :
L 2 (f) t--+ L 2 (f). The geometric interpretation of the kernel A(z, a) is as
follows: Let both z and a lie on a closed contour in f, with tangent vectors
(complex numbers}f'(s) and ,'(t), z = ,(s) and a = ,(t), respectively. Then
the vector (complex number) ,'( s) is the reflection of ,'( a) in the chord joining
z and a (Fig. 7.6.1). Then
A(z,a) = ~ _1_ [,'(s) -,'(t)].

2~1r Z - a
/-----------'---l. a ='Y (t)
Fig. 7.6.1.
This relation is obvious if arg {z - a} = 0 or 'Jr, i.e., if the chord joining z

and a is horizontal. Otherwise, it is proved by rotation. Since a circle is the
only closed contour where a chord joining any two boundary points meets the
circumference at the same angle at both points, we have A(z, a) == 0 for all
z, a on the boundary iff the boundary is a circle. If we expand z = ,(8) in a
Taylor series at a = ,(t), we obtain
1
z = ,(8) = ,(t) + ,'(t) (8 - t) + 2 ,"(t) (8 - t)2 + O(h 3 ), h = 8 - t,
(7.6.12)
which yields
and
_1_ = 1 [1 _ ~ ,"(t) (8 _ t) O(h2)]
Z - a ,'(t)(8 - t) 2 ,'(t) + .
By differentiating (7.6.12), we get
,'(8) = ,'(t) + ,"(t) (8 - t) + O(h 2 ),

which together with (7.6.13) gives
,'(8) = _1_ + ~ ,"(t) + O(h). (7.6.13)

z-a 8-t 2,'(t)
Similarly,
,'(t) = _1__ ~ ,"(t) + O(h). (7.6.14)
z- a 8 - t 2 ,'(t)
Subtracting (7.6.13) and (7.6.14) we find that the singularities (which are real)
of 1/ (8 - t) cancel, and then using (7.6.5), we obtain the Kerzman-Stein kernel
A(z,a) defined by (7.7.10). This kernel is continuous and skew-symmetric,
i.e., A(z, a) = -A(z, a). The following result (Kerzman and Stein, 1986) is
useful:
THEOREM 7.6.2. The Szego kernel S(z, a), as a function of z, is the

unique solution of the integral equation
S(z, a) + l A(z, a) S(z, a) d,( = H(a, z), z, ( E r, (7.6.15)

where, denotes the arc length on r.
If we use a parametrization z = ,(s) on r, 0 :::; s :::; L, the integral equation

(7.6.15) becomes
B(s) + 1£ k(s, t) B(t) dt = g(s), 0:::; s :::; L, (7.6.16)
where
B(s) = 1,'(s)1 1/ 2 S (/(s), a),
g(s) = 1,'(s)1 / H(a,,(s)),
1 2
(7.6.17)
k(s,t) = 1,'(s)1 / A (/(s),,(t)) 1,'(t)1 / .

1 2 1 2
This integral equation is solved by Nystrom method (see Atkinson, 1976; Delves
and Mohamed, 1985) which is as follows: Since all functions in this equation
are periodic, we take n equispaced collocation points Sj = (j - l)L/n, j =
1, . .. ,n, and use the trapezoidal rule. This gives
(7.6.18)
Let B jm = !:..n k (Sj, sm) define the skew-hermitian matrix B, and let Xj =
e(s j), be written in matrix form as
(1 +B) x = y, (7.6.19)
which is solved by an iterative method based on the generalized conjugate

gradient method (GeM), the details of which can be found in Trummer (1986).
The discretized form (7.6.18) of Eq (7.6.16) gives the interpolation formula
L
B(s) = g(s) - -
n
L k (s,
n
Sm) X m · (7.6.20)
m=1
Once the solution B of Eq (7.6.16) is computed, the boundary correspondence

function ¢(s), defined by f(z) = f (/(s)) = ei</>(s), z = r, e iB , can be com-
puted from the formula
¢(s) = arg { - i B2 (sh'(s)}. (7.6.21 )

Trummer (1986) has applied this method to the following six conformal map-
ping problems:
1. r is the inverted ellipse, defined by z(s) = eis )1 - (1 - p2) cos 2 s,
0(s) (see §7.4).
2. r is the ellipse z(s) = eis - ee- is , 0 ::; e < 1, with eccentricity=
1 en .
(1- e)/(1 + e), where 4>(s) = s + 2
00
L;
1 + e2n sm2ns.
n=l
3. r is the epitrochoid ('apple') z(s) = eis + ~ e2is , 0 ::; a < 1, where
4>(s) = s.
4. r is Cassini's ovallz - allz + al = 1,0::; a < 1, or
where 4>(s) = s - 0.5 arg{h(s)}, h(s) = )1 - a 4 sin 2 2s.

5. r is the unit square, where cos 4>(s) = dn(Ky).
6. r is the stadium with boundary composed of two semicircles joined by two
line segments, all of the same length.
This method requires some programming. For details see Trummer (1986).
CASE STUDY 7.6.1. The results for the ellipse with eccentricity 0.5 are
presented in Fig. 7.6.2.•
Fig. 7.6.2.
7.7. Problems
PROBLEM 7.7.1. Find the approximate solution of the integral equation
Jr /
1 2
¢>(x) - sinxy¢>(y)dy = f(x),
o
3 5
X y3 X y5 1 ( x )
sinxy ~ xy - -6- + 120' Then taking f(x) = 1+ ~ cos '2 - 1 ,show
that the approximate solution is ¢>( x) ~ 1+0.0000009 x -0.0000002 x 3 . [Note
that the exact solution is ¢>(x) = 1.] (KantoroYich and KryloY, 1958, p.145.)
PROBLEM 7.7.2. Show that the approximate solution of the integral equa-
1
tion
1
y(x) + x (e XS -1) y(s)ds = eX-x
is y(x) ~ eX - x - 0.501x 2 - 0.1671x 3 - 0.0422x 4 . Compare it with the
exact solution y(x) == 1. (Berezin and Zhidkoy, 1965, p.653.)
PROBLEM 7.7.3. Show that the first two eigenvalues and corresponding
eigenfunctions of the homogeneous integral equation
u(x) -.A 1 1
K(x, s) u(s) ds = 0,
where
x(l - s) 0 < x < s < 1,
K(x,s) = - - -
{ s(l-x) O~s~x~l.
are.Al = 7l'2,.A2 = 47l'2 and Ul(X) = y'2 sin7l'x, U2(X) = y'2 sin 27l'x, and
the approximate solutions are .AI ~ 9.8751, .A2 ~ 40, Ul(X) ~ -0.0684 +
5.817x(1- x), U2(X) ~ 14.49x(1- x)(l- 2x). (Berezin and ZhidkoY, 1965,
p.657.)
PROBLEM 7.7.4. Consider the function u = log 11r, Izi = r, which is a

potential function, regular for r =J 0, Show that
(i) this function yields the force flux of 27l' at r = O. [Hint: Evaluate lim
£--->0
aaun ds,
where n is the outward normal.]
7.7. PROBLEMS 205
(ii) Define a source of strength q at a point ( = (~, TJ) on the boundary f of a

simply connected region D by
u(~, TJ) = .2.... log --r;=~~I==;==~

21f J(x - ~)2 + (y - TJ)2
A source of strength -q is called a sink of strength q. By combining a source

and a sink of the same strength q and letting the distance between them tend to
zero while keeping the moment constant, show that the potential
qx q cosr
u(() = 21f (x2 + y2) 21fr
Note that this potential is known as a dipole of strength q which is also the
moment along the x-axis. (Andersen et al., 1962, p.I73.)
PROBLEM 7.7.5. If s denotes the arc length on the boundary f of a simply

connected region D, 0::; s ::; L, and if the dipole density off is 21fv(s), then
the potential u(z) is defined in D and in D* = Ext (f) by
u(z) = rv(s) un»0

Jr s
(log~)
r sz
ds, zED, (7.7.1)
where n s is the inward normal at a point ( E ((s) and r sz = Iz - (I. The

function u(z) is regular in D and D* but is discontinuous on f. If the unknown
potential function u(z) is determined by the boundary values ur(s) = 9(S)
from (7.7.1), then show that the dipole distribution v(s) satisfies the integral
equation
v(s) =.!.1f [9(S) - rL

J v(t) »0
o Unt
(log~)
rtz
dt] ,
where nt is the inward normal at a point ( = (( t), 0 ::; t :S L, t =/= s. (Andersen

et al., 1962, p.173.)
PROBLEM 7.7.6. Let f(z) = u(z) + iv(z), u and v real, be continuous

and regular in Izi < 1. Show that
v (r i 4» = v(O) + -1
21f
10
27f
¢ --
u (e iO ) cot -
2
0 dO.
For () = ¢, the integral takes the Cauchy p.v. If f(z) = zn, n = 0, 1, ... , then
show that
-
1
271"
1
0
2
71" {Sinn{)}
cosn{)
¢-()
cot--d{) =
2
{-cosn¢
sinn¢
forn= 1,2, ... ,
forn=O,I, ....
(Gaier, 1964, p.62.)
PROBLEM 7.7.7. Prove the transformation relation (7.6.8) between S( z, a)

and Su(z, a). (Kerzman and Stein, 1978, p.87; 1986, p.ll1.)
PROBLEM 7.7.8. Develop the details of the conjugate gradient method.

(Trummer, 1986, p.857.)
REFERENCES USED: Ahlfors (1952), Andersen et al. (1962), Atkinson (1976),

Banin (1943), Berezin and Zhidkoy (1965), Birkoff et al. (1950, 1951), Carrier (1947),
Carrier, Krook and Pearson (1966), Gaier (1964), Gershgorin (1933), Kerzman and
Stein (1978, 1986), Lichtenstein (1917), Kantorovich and Krylov (1958), Neumann
(1877), Stiefel (1956), Szegii (1950), Trummer (1986), Todd and Warschawski (1955),
Warschawski (1955, 1956), Whittaker and Watson (1927).
Chapter 8
Theodorsen's Integral Equation
The problem of conformally mapping an almost circular region onto the in-
terior or exterior of a circular disk by the the method of infinite systems was
discussed in Chapter 5. In this chapter we shall present Theodorsen's integral
equation and establish the convergence of the related iterative method for the
standard case of mapping the unit circle onto the interior (or exterior) of almost
circular and starlike regions, both containing the origin. A trigonometric inter-
polation scheme is presented, and Wegmann's iterative and Newton's method
for numerically solving this equation are discussed. The last two methods are
based on a certain Riemann-Hilbert problem, which turns out to be a linearized
form of a singular integral equation of the second kind. Unlike the classical
iterative method, the solution of the linearized problem in Wegmann's method
for the conformal map of the unit circle can be represented explicitly in terms
of integral transforms, which leads to a quadratic convergent Newton-like
method that avoids the numerical solution of a system of linear equations and
thus becomes more economical. Theodorsen's integral equation has specific
significance in the theory of airfoils.
8.1. Classical Iterative Method
Let r denote a Jordan curve defined in the polar coordinate system by p =

p( </»,0 ::; </> ::; 27r, where p( </» E 0 1 , such that for some € (0 < € < 1) and a
constant a > 0
a
1+€ ::; p(</» ::; a(l + e), (8.1.1)
207

208 CHAPTER 8. THEODORSEN'S INTEGRAL EQUATION
and
p'(¢) I < c. (8.1.2)
I p(¢) -
Any Jordan curve f that satisfies these two conditions is called a nearly circular
contour or a near circle. Let us assume that a function w = f(z) = pe i ¢,
z = eiO , with f(O) = 0 and 1'(0) > 0, maps the unit disk U = {Izl <
I} onto Jnt (f) which is starlike with respect to the origin. Then f (e iO ) =
p (¢( 0)) ei¢(O) defines the boundary correspondence function ¢ : [0, 27r]1--+ R.
If ¢ is known, then f is known. Consider the function
F(z) = log f~Z) = log If~Z) 1+ i arg { f~Z)} , (8.1.3)
which, defined as real-valued log 1'(0) for z = 0, is single-valued and analytic

in U, and continuous on U u au. If we set z = eiO , then arg { f ~z) }
arg {J (e
iO
) e-
iO
} = ¢(O) - O. Thus,
F (e
iO
) = log p[¢(O)] + i [¢(O) - OJ. (8.1.4)
Then, in view of the Schwarz formula (6.4.12) with p = R = 1 and v(O) = 0,

we obtain
¢(B) - B = - 1
27r
10
21r
log p (¢(B)) cot -¢ - 0 dO.
2
(8.1.5)
Note that not only the mapping function f(z) is defined in the form p = p(¢)
on f, but the relation (8.1.5) represents an integral equation for the unknown
function ¢(B), i.e.,
1
¢(B) - 0 = - -
27r
l0
1r
[log p (¢(O
2
+ t)) -log p (¢(O - t))]
cot -t dt.
(8.1.6)
This is known as Theodorsen's integral equation. Once the function ¢( B)
is determined, the function F(z) and then the mapping function f(z) can be
computed. The term arg {f(z)} = arg {J'(O)} is not added to the right
z z=o
side of (8.1.6) because it is zero. Theodorsen (1931) showed that ¢ is a solution
of Eq (8.1.6). Gaier (1964, p.66) proved that this equation has exactly one
solution which is continuous and strongly monotone. The Riemann mapping
theorem (§ 1.4) guarantees the existence of a continuous solution of this integral
equation. We shall also show that this solution is unique.
8.2. CONVERGENCE 209
Theodorsen's method for solving the integral equation (8.1.6) for nearly
circular regions is based on the iterations
1>o((~) = 0,
1
1>n(B) - B = --2
rr
1 0
w
{log p [1>n-l (B + t)] - log p [1>n-l (e - t)]} cot -t dt,
2
n = 1,2, ....
(8.1.7)
The functions 1>n(B) are absolutely continuous, and 1>~ (B) E £2[0, 2rr]. In fact,
this is obviously true for n = 0. Suppose that this statement is true for some
n ~ 0. Since, in view of (8.1.2), the function log p( 1» has a bounded difference
quotient and 1>n (B) is absolutely continuous, it follows that log p (1)n (B) is also
absolutely continuous. Also, since
p' (1)n (B» -I.' (B») 2 < e2 [-I.' (B)]2 (8.1.8)

( p (1)n (B» 'f'n - 'f'n ,
the integral12W [~(~:(~N ¢~(B) r dB exists. Hence, the function 1>n(B)-B
°
which is the conjugate of log p (1)n(B» exists and is absolutely continuous. The
integrands in (8.1.7) are singular at t = where the integrals take the Cauchy
p.v.'s. In what follows we shall use the notation
p' (1)( B)
a (1)(B)) = p(1)(e)) ' and
d
p (1)) = d1>
[p'p((1))]
¢) . (8.1.9)
Both a and p are HOlder-<:ontinuous.
8.2. Convergence
The following result holds for the convergence of Theodorsen's iterative method
(8.1.7).
THEOREM 8.2.1. The sequences {¢n (Bn and {1>~ (en converge uni-
formly to ¢( B) and 1>' (B), respectively, as n ...... 00.
This result will in turn establish that log p[4>n (8)] converges uniformly to
log p[4>(8)] as n -+ 00, so that the functions
Fo (e ill ) = log a, Fn (e ill ) = log P[4>n-I(8)] +i (4)n(8) - 8), n ~ 1,
(8.2.1)
will compute f (e ill )) to any desired accuracy. Let the functions Fn(z) be
analytic on U and assume boundary values Fn (e ill ) on Izi = 1. In view
of the maximum modulus principle, the uniform convergence of Fn (e ill ) to
F (e ill ) implies the uniform convergence of Fn(z) to F(z) on Izl ~ 1. Hence,
the functions fn(z) = z eFn(z) converge uniformly to the mapping function
f(z) = zeF(z).
To prove the above theorem, we shall first derive the estimates for the
differences l4>n(8) - 4>(8)1 and 14>~(8) - 4>'(8)1 in terms of e and n, and show
that these differences approach zero as n -+ 00.
THEOREM 8.2.2. If r is a near circle and if 4>n(8) and 4>(8)

arg {f (r eill)} are defined by (8.1.7) and (8.1.4), then
1/4
l4>n(8) - 4>(8)1 ~ 2 (
1 ~ e2 ) e(n+2)/2. (8.2.2)
Note that the bound in (8.2.2) goes to zero as n -+ 00 since 0 < e < 1.
This will establish the convergence of 4>n (8) to 4>(8).
THEOREM 8.2.3. If r is a near circle and if a E HI, then

l4>n(8) - 4>(8)1 ~ J27l'A(n + l)e n + 1 , (8.2.3)
This result provides a bound that converges to zero more rapidly as n -+ 00.
THEOREM 8.2.4. Ifr is a near circle and if a E HI andp(4)) E HI,

then
14>~(8) - 4>'(8)1 ~ v'27l'cn [A(n + 1)]3/2 en+l, (8.2.4)
where
n
CI = 1 + e, Cn = (1 + e) II (1 + e k
v'27l' Ak) , (8.2.5)
k=2
8.3. PROOFS 211
and for all n

(8.2.6)
The last inequality shows that en is bounded if a < c < 1. The proofs for
these theorem are given in Warschawski (1945). We shall outline these proofs
in the next section. It should be noted that the estimates for the difference
IFn(z) - F(z)1 for Izl ~ 1 are obtained from those for l¢n(B) - ¢(B)I given
above. Thus, in view of (8.1.2), we have
IFn (e ill ) - F (e ill ) I ~ Jc 2 [¢n-l(B) - ¢(B)]2 + [¢n(B) - ¢(B)]2,
and
IFn(z) - F(z)1 <
-
max
II
IFn (e ill ) - f (e ill ) I for Izi ~ 1.
In the case of Theorem 8.2.2 this yields
IFn(z) - f(z)1 ~ 2(1 ~ c) 1/4 c(n+2)/2 v'l'+"€,

and in the case of Theorem 8.2.3
IFn(z) - f(z)1 ~ 2c n
+l J7rA (n +~).
Hence, for a< c < 1 the iterations Fn(z) converge uniformly to F(z)
log fez) for Izi ~ 1. The above theorems constitute the classical theory for
z
the convergence of the numerical method for solving Theodorsen's integral
equation by iterations in terms of the boundary correspondence function ¢,
which provides the required boundary map f.
8.3. Proofs
Theodorsen's method converges strongly like a geometric series. First, we

obtain the bounds for the square means
M; = 2~ 1 21f
[¢n(8) - ¢(8)]2 dB, M~ 2 = 2~ 1 [¢~(B)21f
- ¢'(B)]2 dB,
M"2 = ~
n 27r io(21f[¢II(B)
n
_ ¢"(8)]2d8.
(8.3.1)
Then the results of these theorems are obtained by using the inequalities
The following results are also needed:
LEMMA 8.3.1. If the function g(B) E L 2[0,21r] is real-valued, 21r-

periodic, and square-integrable (in Lebesgue' sense) on 0 ~ B ~ 21r, and
if g(B) is a conjugate function of g(B), then
1 {271" 1 (271"
21r Jo [g(BW dB + 0:2 = 21r Jo [g(B)]2 dB + ,82, (8.3.3)
where
0:
2
= -1
21r
1 0
2
71" g(B) dB, ,82 = 2- (271" g(B) dB.
21r Jo
LEMMA 8.3.2. If g(B) E L 2[0, 21r] is a real-valued, absolutely contin-

uous, 21r-periodic and square integrable function, then for any Bo
[g(B)]2 - [g(BoW ~ 21rMM', (8.3.4)
where
M2 = 2- {271" [g(BW dB,

21r Jo
The factor 21r in (8.3.4) is the best possible.
LEMMA 8.3.3. If r is nearly circular, then the function ¢(B)

arg{f (re i9 )}, defined by (8.1.4), is absolutely continuous and [¢'(B)jZ E
L 2[0, 21r] in Lebesgue's sense such that
2- (271"[¢'(BWdB ~ _1_. (8.3.5)

21r Jo 1 - e2
LEMMA 8.3.4. If g(B) E HC>., 0 < 0: ~ 1, is a 21r-periodic function,

then any conjugate function of g(B) also satisfies a Holder condition.
8.3 PROOFS 213
LEMMA 8.3.5. Ifr is a near circle and if (J(<jJ) E HI, then
~ ~ <jJ'(8) ~ A,
A 1 +.0 2
(8.3.6)
ior
21'
1
21l' [<jJ1I(8)]2 d8 ~ A 3/ 2 E min (1 + E; h) . (8.3.7)
LEMMA 8.3.6. If u(t) E Gl[O,21l'] is a 21l'-periodic function and if

v(t) E Gl[O, 21l'] is a function conjugate to u(t), then for every 8
21r 21'
r (u(~) ~~~)) 2 dt = r (v(~) ~~~)) 2 dt.
io sm--
2
io sm--
2
Proof of Lemma 8.3.1 is available in Zygmund, 1935, p.76 (Eq (4)); of

Lemma 8.3.4 in Privaloff (1916) or Zygmund (1935, p.156); of Lemma 8.3.5 in
Warschawski (1950); and ofLemma 8.3.6 (on conjugate functions) in Warschawski
(1945).
PROOF OF LEMMA 8.3.2. Note that for 0 ~ 8 :S 21l', 0 ~ 80 :S 21l',
0 10-21'
g2(8) - g2(80 ) = 2 g(t) g'(t) dt = 2 g(t) g'(t) dt. (8.3.8)
1(Jo 00
Since
O21' O 21r
rO Ig g'l dtl + I r - Ig g'l dtl = r Ig g'l dt = r Ig g'l dt,
I~o ~o ~-21' io
one of the two integrals in (8.3.8) does not exceed -1
2
1 Ig g'l
0
2
1' dt. Hence, by
the Schwarz inequality,
W ~ ior
21'
[g(8)f - [g(8 0 Igg'ldt:S 21l'MM'.
Also, applying (8.3.4) with g( 8) = cos n 8 (80 = 1l'/2) and letting n ~ 00, we
find that the constant 21l' cannot be replaced by any smaller one. -
PROOF OF LEMMA 8.3.3. Since f is rectifiable, the function F (e iO ) is

absolutely continuous (this follows from a theorem ofF. and M. Riesz, 1923).
Hence,
:BF(eiO) -i=cr(¢(B)) ¢'(B)+i¢'(B)
existsa.e. forO ~ B ~ 27r and is integrable. Moreover,thefunction :BF(z)-

i = u(z) + i v(z), z = r eiO , has the Poisson integral representation in the unit
disk as
r
27r 2
u(z) + iv(z) = ~ [u (re it ) + iv (re it )] 1- r dt
27r Jo 1+r 2 - 2r cos(t - B) ,
(8.3.9)
(see (6.4.6». For almost all B E [0,27r] we have
lim u
r-+l
(r eiO ) = cr (¢(B)) ¢'(B) = u (e iO ) ,
lim v (re iO ) = ¢'(B) = v (e iO ),
r-+l
and since Int (f) is starlike (see §8.5), ¢'(B) 2': O. Then by (8.1.2)
Thus, in view of (8.3.9) we have v(z) + u(z) 2': 0 and v(z) - u(z) 2': 0 for
Izi < 1. Hence, v 2 (z) - u 2 (z) 2': 0 for jzl < 1. Also,
Then, taking the limit as r --4 1 and using Fatou's lemma, we get
which, in view of (8.1.2), yields (8.3.5).•
PROOF OF THEOREM 8.2.2. (a) First we determine an estimate for M n .

In view of (8.1.4) and (8.1.7), we have
Jor 27r
[¢(B) - B] dB = 0, Jro 21r
[¢n(B) - B] dB = O. (8.3.10)
8.3. PROOFS 215
Now, applying Lemma 8.3.1 with g( 0) + i g( 0) = Fn (e iO ) - F (e ill ) and

noting that f3 = 0 because of (8.3.10), we get
M~ = :11" 1 27r
[</>n(O) - </>(0)]2 dO
1 r27r
~ 211" io [log P(</>n-l(O)) -log p(</>(0))]2 dO.
(8.3.11)
Since by (8.1.2)
Ilog P(</>n-l(O)) -log p(</>(O))I ~ e l</>n-l(O) - </>(0)1,
from (8.3.11) we obtain
M~ ~ e 2~ 2
1 27r
[</>n-l(O) - </>(0)]2 dO ~ e M~_l>
2
or
For n = 0 we find from (8.3.3) by using (8.1.1) that
M5 =.2-
211" i
r27r
[</>(0) _ Oj2 dO ~.2- r [log p (</>( 0))] 2 dO ~ e2,
211" i
27r
a
o o
which yields the estimate
(8.3.12)
(b) Now we determine an estimatefor M~. SinceFn (e ill ) andF (e ill ) areabso-
lutely continuous and :0 Fn (e and :OF (e belong to the class L 2[0, 27l'],
ill ill
) )
the imaginary part of :0 [Fn (e ill ) - F (e ill )] is a conjugate function of the

real part. Then
i
r27r dOd F (e ill
) dO
27r
= F (e iO ) 111=0 = 0, io
r27r dOFn
d
(e
ill
) dO = O.
o
(8.3.13)
Hence, using Lemma 8.3.1 with g(0) + i g(0) = :0 [Fn (e ) - F
ill
(e ill )] , we
obtain
27r
M,2
n
=.2- r
27l' io
[</>' (0) - </>'(0)]2 dO
n
1 27r (8.3.14)
= 2~ {o-(</>n-l(O)) </>~_1(0) - a (</>(0)) </>'(O)} 2 dO.
In view of (8.1.2), then
10r
21r
M~2 ::; 2 1
2c 21T [4>/~-l(B) - 4> /2(B) ] 2 dB. (8.3.15)
By Lemma 8.3.3, we have - 1

21T
1 0
21r
2
4>/ (B) dB ::; - -12 .
1 - E;
Moreover, again
using Lemma 8.3.1 with g( B) + i g( B) = d~ Fn (e i9 ), and (8.1.2), we get

21r 21r
~
21T 1
r Wn(B) _1]2 dB = ~ r
21T 10
[0" (4)n-l(B)) 4>~_1(B)]2 dB
1 4>/~_l(B)dB,
0
21r
::; E;2 2~
or, after suppressing the argument B in the integrands,
21r 21r 21r
~ {r
21T 10
Wn _ 1]2 dB _ 2 r
10
4>~ dB + 21T} ::; E;2 ~ r
21T 10
4>/~-1 dB.
Since 1 21r
4>'n dB = 21T, we have
1 ,,/2n dB _ <- E;2 ~ 1r ,,/n-l dB.

r 21r 21r
~ 1 2
21T If' 21T If'
0 0
10r 4>/~ dB, then m; ::; 1 + m;_l' and hence, m~ ::;

21r
1
If we set m~ = 21T
1 + E;2 + E;4 + ... + E;2n m6. Since m~ = - 1

21T
1 0
21r
dB = 1, we get
m~::; (1_E;2)-1. (8.3.16)
Thus, from (8.3.15), (8.3.5) and (8.3.16) we find that
MI2<~.
n - 1 _ E;2 (8.3.17)
(c) Finally, we determine an estimate for l4>n(B) - 4>(B)I. We set g(B) =
4>n(B) - 4>(B) in Lemma 8.3.2. Then, since 1 r21r g(B) dB = 0, there exists a
0
value Bo such that g(Bo) = 0, which yields
8.3. PROOFS 217
This gives (8.2.2) after using (8.3.12) and (8.3.17), which completes the proof
of Theorem 8.2.2.•
PROOF OF THEOREM 8.2.3. (a) First, we determine an estimate for M~.

Using (8.3.14), we have by Minkowsky's inequality for n ~ 1
M~= 1
2; 1 0
2
1l'
[(O"(¢n-d-O"(¢»¢'+O"(¢n-l) (¢~_I-¢,)]2dB
< -!; 12 [0" (¢n-I) - 0"(¢)]2 </l,2 dB

1l'
+ 2..1
211' 0
2
1l' (¢~-I - ¢,)2 [0" (¢n_I)]2 dB.
Since 0 < ¢' (B) ::; A (Lemma 8.3.5) and 0"( ¢) E HI, from (8.1.2) we obtain
and, then by (8.3.12)
(8.3.18)
For n = 0 from (8.1.2) and Lemma 8.3.1 we get
M'~ = 2..1 2
1l' (<Ii - 1)2 dB = 2..1 2
2
" [O"(¢) </l'J dB
1
211' 0 211' 0
2
(8.3.19)
::; f;2 - A 1l' ¢' dB = f;2 A.
211' 0
Now we shall prove by induction that
(8.3.20)
In fact, since (8.3.19) holds for n = 0, we assume that (8.3.20) is true for some
n > O. Then from (8.3.18)
M~+I::;C [Ae n+l+A(n+1)cn+l] =A(n+2)cn+2,
which proves (8.3.20) for (n + 1).

(b) Next, we determine an estimate for l1>n (B) -1>( B) I. In fact, by using Lemma
8.3.2, (8.3.12), and (8.3.20)
l1>n(B) -1>(B)1 ::; J21rMnM~ ::; €n+l J21rA(n + 1).. (8.3.21)
PROOF OF THEOREM 8.2.4. Since P the function F (e iO ) E

E HI,
eZ[O, 21r] (see Warschawski, 1935, Theorem III). Similarly, all functions F n (e iO ) E
e z[0, 21r], where Fn (e iO ) are defined by (8.2.1). Also,
dF ( )1>' +21>
dB =0"1> . (' -1,
)
~:~ = p( 1» 1>'z + 0"( 1» 1>" + i 1>",

(8.3.22)
dF n
dB = 0" ( 1>n-1 )1>n-1
'
+ 2 1>n -
. ('
1 ,)
Z
Fn = P ("-
ddBZ 'l-'n-l
) ,,-'
'l-'n-l
Z
+ 0" ("-'l-'n-l ) ,,-"
'l-'n-l + t'l-'n'
. ,,-"
(a) Now we shall estimate
Jr
Z1r
1
M~ = 21r (1)~ -1>") dB,
o
and prove that
M~ ~ AZ(n + 1)z en €n+l for n 2:: 1. (8.3.23)
Proof of (8.3.23) is as follows: Since, by (8.3.13),
Z1r
1 r dZ
21r J dBz [Fn (e )
iO iO
- F (e )] dB = 0,
o
we find by applying Lemma 8.3.1 that
M~+lz = 2~ 1 Z1r
{(P(1>n) -p(1))) 1>'z +p(1)n) (1)~Z _1>'Z)
+ (O"(1>n) - 0"(1))) 1>" + O"(1)n) (1)~ -1>")} ZdB. (8.3.24)
Since 0"(1)) E HI, we have Ip( 1» 1::; €, and since 1>' (B) ::; A (Lemma 8.3.5)
and P E HI, we find from (8.1.2) that
Z rZ1r
M~+/ ::; ;1r J [A z l1>n -1>1 + 11>~z _1>,zl
o
+ 11>"II1>n -1>1 + 11>~ -1>"I]zdB,
8.3. PROOFS 219
which, by Minkowski's inequality, yields
M::+ 1 :::: € [A2Mn + ~ 1 (¢~2

2
7< _ ¢,2)2 dO
1 (¢~)2
(8.3.25)
2
+ 2~ 7< (¢n - ¢)2 dO + M::],
where M n and M~ are defined in (8.3.1). Since -1

27r
10
2
7<
(¢//(0))2 dO ::::
€ A 3 / 2 min(l + €; V2) (Lemma 8.3.5), we have
and since l¢n(O) - ¢(0)1 :::: €n+! J27r A(n + 1) by (8.2.3), we get
2~ 1 2
7< (¢n - ¢)2 (¢//)2 dO:::: V2€n+2 A 3 / 2 J27rA(n + 1)
(8.3.26)
= 2€n+2 A 2 J7r(n + 1).
Thus, using Lemma 8.3.2 with g(O) = ¢~(O) - ¢'(O), we find that
-/..' _ 'I'-/..,)2 <
('l'n _
27r M'n M//
n'
which, after taking the square root and using ¢'(O) :::: A, yields I¢~ + ¢I ::::
I¢~ - ¢'I + 2WI :::: J 27rMri M:i + 2A. Hence,
(8.3.27)
Now, integration by parts gives
;7r 1 2
7< (g'(0))2 dO = Ig(O) g'(O) 1:7< _1 2
7< g(O) g//(O) dol
::1 2
7< Ig(O) g//(O)I dO
< 1 2
7< (g(0))2 dO .1 2
7< (g//(0))2 dO,
which, aftersettingg(B) = ¢n(B)-¢(B), yields the inequality M~ ~ y' M~ M::'.

Applying this inequality to (8.3.27), we get
~
27r
12
0
11" (A.-/
2
'l-'n 2_ 'I-'A.-/ ) 2 dB <
-
2AM'n + Miln y'27rMn M'n' (8.3.28)
Hence, from (8.3.25) we find, after using (8.3.12), (8.3.28), (8.3.26), and
(8.3.20), that for n ~ 2
M::'+1 ~ c; { A 2 c;n+l + 2A 2(n + 1)c;n+1 + 2A2c;n+2 y'7r(n + 1)
+ (1 + c;n+1 y'27rA(n + 1)) M::' }
= A 2c;n+2 { 1 + 2(n + 1) + 2c; y'7r(n + 1)

Mil }
+ (1+c;n+1y'27rA(n+l)) A2c;:+1
::; A 2 c;n+2 { 1 + 2(n + 1) + 2c; y'7r(n + 1)
+ (1+c;n+1y'27rA(n+l)) (n+l)2 Cn } by (8.3.23)
~ A 2 c;n+2 Cn+l {I + 2(n + 1) + (n + 1)2}

= A 2(n + 2)2 Cn+I c;n+2,
(8.3.29)
because
1 + 2(n + 1) + 2c; y'7r(n + 1) < [1 + 2(n + 1)](1 + c;) < [1 + 2(n + 1)] cn+I
for n 2': 2, where we have assumed that (8.3.20) is true for n ~ 2. To show
that (8.3.23) is true for n = 1, note that from (8.3.24) for n = 0
M lil-
-
2~ 12 11" [p(O) - p(1)(0))]1>/2 + p(e) (1 - 1>/2) + a (1)(0)) 1>'f de
:5£ [A2MO+ 2~t' (1_¢'2)2 dO + M;]'

8.3. PROOFS 221
by Minkowski's inequality and (8.1.2), where <t>o = B, a(<t» E Hl andp E Hl.

Since M6' ~ A2 e (1 + e), and <t>'(B) ~ A by Lemma 8.3.5, we get, in view of
(8.3.12) and (8.3.20),
1 A
M{'~e2 2
[A +(1+A)A+A (I+e)]=A e (2+
2 2 2
: +e).
Since A 2: 1 and 1 : A ~ 2, we find that

M{' ~ A 2 c2 (4 + c) < 4A 2 e 2 (1 + c). (8.3.30)
We shall establish that (8.3.23) holds for n = 2. In fact, by applying (8.3.29)

with n = 1 and replacing M{' by A 2 e 2 (4 + c) from (8.3.30), we get
M~' ~ A 2 c3 [5 + 2c .J2; + (1 + 2e 2 v;A) (4 + c)] . (8.3.31 )
Since 2V27r < 6 and 1 + 2c 2 ;;A > 1, we find from (8.3.31) that
M~' ~ A 2 c 3 (5 + 6c + 4 + c) (1 + 2c 2 v;A)
< 9(1 + c) (1 + 2c 2 v;A) A 2 c 3 (8.3.32)
= 9A 2 C2 c 3 .
(b) Next, we determine an estimate for 1<t>~(B) - <t>'(B)I. By applying Lemma
8.3.1 with 9(B) = <t>~(B) - <t>'(B), we find from (8.3.20) and (8.3.23) that
(8.3.33)
(c) Finally, we estimate Cn. Note that
II (1 + c
n
k \/2ll-Ak) ~ e .,'211' A L~=2 gk v'k.
k=2
Then, by the Schwarz inequality
Lc
n n
k -Ik = e Lc(k-l)/2 (-Ikc(k-l)/2)
k=2 k=2
n n
~c Lc k- 1 L kc k- 1
.
k=2 k=2
Hence,
n
I:
k=2
c k vIkf < c
-
V- - (1
- - - - ) < ..,-----,..."...,.".
C
l-c l-c 2
1V2 c
2
(l-c)3/2'
which yields (8.2.6). This completes the proof of Theorem 8.2.4.•
PROOF OF THE UNIQUENESS OF THE SOLUTION OF THEODORSEN'S

INTEGRAL EQUATION: We shall show that if r is a near circle, then the integral
equation (8.1.6) has at most one continuous solution. Let us assume that there
exist two such solutions (h(B) and Ih(B). Since
= 0, Jor [lh(B) - B] dB =
27T
t
Jo
TT
[lh(B) - B] dB 0,
we have by Lemma 8.3.1

27T
M2= r
-.!..-
27f Jo
[<1>1 (B) - <l>2(B)]2 dB
1
~ 27f r27T
J [log p(I(B)) -log p(2(B))]2 dB.
o
Since, in view of (8.1.2),
we find that M 2 ~ c 2 M 2 , which yields M = 0 because 0 < c < 1. Hence,

<l>1(B) = <l>2(B) .•
8.4. Integral Representation
An integral representation for the function ~ (B) is given by the following
result.
THEOREM 8.4.1. If r is a near circle and if the function p( <1» which

defines r satisfies the condition HI, then <1>' (B) is continuous and
1
~(B) -1 = --2
llJ+7T [O'(t) - O'(B)] cot -2-B dt,
t -
(8.4.1)
7f ()-7T
8.4. INTEGRAL REPRESENTATION 223
and for n ~ 1
¢~+I (e) - 1 = --2

-rr
1 111
+71'
11_71' [a (¢n-l(t)) - a (¢n-I(e))] ¢~_I(t)
t-e
x cot -2- dt - a (¢n-I(e)) a (¢n-2(e)) ¢~_2(e). (8.4.2)
PROOF. The integrand in (8.4.1) is continuous in both t and except e,

at t = e, and is bounded because a E HI. Hence, the integral (8.4.1) is a
continuous function of (), represents a conjugate function of a(e), and is equal
to (¢~ (e) - 1) at least for almost all e and, because of continuity, for all e. This
proves (8.4.1).
We shall prove (8.4.2) by induction. Let us assume that ¢k (e) is a continuous

function for k = 1,2, ... ,n. We shall prove that (8.4.2) holds for (n + 1) and
that ¢~+I (e) is continuous.
The absolute continuity of Fn+I (e ill ) = log P(¢n(e)) + i (¢n+I(e) - e)

implies that ( ¢~+ 1(e) - 1) is conjugate of a (¢n (e)) ¢~ (e), and for almost all
e we have
¢~+I (e)
1 171'
- 1 = -- a (¢n(T)) IIIH cot -2t dt
2-rr 0 r=lI-t
= --
1 171' [a (¢n(e + t)) - a (¢n(e))] ¢~(e + t) cot -t dt
2-rr 0 2
+ -1 171' [a (¢n(e - t)) - a (¢n(()))] ¢~(e - t) cot -t dt
r
2-rr 0 2
1 t
- a(¢n(e)) 2-rr Jo [¢~(() + t) - ¢~(e - t)] cot "2 dt
== h + 12 + 13 .
Note that, since 0'( ¢) E HI and ¢'(e) is continuous, the integrals hand
12 represent continuous functions of e. The integral h, without the factor
0'(¢n(e)), is equal to a (¢~-I (e)) ¢~-I (t), since (¢~ (e) - 1) is conjugate to
e
this function. If we set T = + t in hand T = () - t in 12 , we get
which defines a continuous function of B.•
Theorem 8.4.1 establishes conditions under which the images r n of the

unit circle underthe mapping function w = inez) = Z eF,,(z) are starlike with
respect to the origin. We shall discuss starlike regions in the next section. The
advantage in assuming starlike contours is that ¢n (B) becomes a monotone
increasing function and, therefore, possesses a unique inverse function B =
B(¢). This helps us compute the approximate inverse mapping function Z =
ei9 ,,(q,) that maps the unit circle onto r n'
8.5. Starlike Regions
A region D is said to be starlike with respect to a point Zo E D if every ray from

the point Zo intersects aD in exactly one point. The functions inez) = Z eF,,(z)
map the unit circle Izi = 1 onto Jordan curves r n' Since the functions inez)
are approximations of the mapping function i(z), it is important to assume
that the contours r n enclose regions that are starlike with respect to the origin,
as we have done in §8.3 in the proof of Lemma 8.2.3 and remarked about the
advantage of such an assumption at the end of §8.4. The functions inez) are
starlike with respect to the origin if W{ ZI~~~)} > 0, which, in view of
(8.3.22), leads to the condition (¢~ - 1) < 1.
CASE STUDY 8.5.1. We shall determine the conditions on € under which

the regions bounded by r n are starlike. In fact, we shall show that if r is a near
circle and if <7( ¢) E HI, then the region bounded by the contour r 1 is starlike
with respect to the origin if € ~ (2 log 2) -1 ::::: 0.72, by r z if € ~ 0.34, by r 3
if € ~ 0.31, and by r n for n 2:: 4 if € ~ 0.295.
These results are obtained by evaluating the values of € for which I¢~ (8) -
11 ~1, so that ¢'(8) 2:: 0, which implies that ¢n(8) is a monotone increasing
function. Thus, since <7(¢) E HI, we have, in view of (8.4.1),
1¢I(B) -11 ~ 1
211"
9
- 1 +11' Idt) - dB)1
9-11'
It
81 dt
cot -2-
~ -
€ 10+11' (t - B) t - B
cot--dt = 2€ log 2,
211" 9-11' 2
8.5. STARLIKE REGIONS 225
where (J is defined by (8.1.9). Then the region bounded by r 1 is starlike if

¢i (8) ~ 0, i.e., if 2 clog 2 ~ 1, which gives c ~ (2 log 2) -1.
Assuming that ¢~(8) ~ 0 for some n ~ 1, provided that c ~ co < 1, we
shall evaluate ¢~+l (8). Thus, since (J( ¢) E HI, we have by (8.4.2) and (8.1.2),
7r
t- 8
1¢~+l(8) - 1\ ~ 2c1r 19+
(J-7r
[¢n(t) - ¢n(8)] ¢~(t) cot -
2
dt
(8.5.1)
+ c I¢~-1 (8) I == € m~ + €2 I¢~-l (8) I.
2
Note that ¢n(t) - ¢n(8) has the same sign as (t - 8) since ¢~(t) ~ O. Now,
integrating by parts, we find that
1 19+7r t- 8
m~ = -2 [¢n(t) - ¢n(O)] ¢~(t) cot -2- dt
1r 0-7r
= ~ r9+ 7r
(¢n(t) -¢n(O))2 dt
21r } 0-7r 2. t - 0
sm--
2
=~ 7r
r9+ (¢n(t) - t - [¢n(O) - 0] +t - 8)2 dt,
21r } (J-7r 2 . t- 8
sm--
2
which, by Minkowski's inequality, yields
7r
~ r9+ (¢n(t) - t - [¢n(O) - 0]) 2 dt
21r }0-7r . t- 0
2 sm--
2
l
(8.5.2)
O 7r
1 + ( t- 8 )2
+ - t - 8 dt == jJ; + VJ;.
21r 0-7r 2·
sm--
2
Now, by Lemma 8.3.6,

7r
J = ~ r9+ (lOg P (¢n-l (t)) - log P(¢n-l (8))) 2 dt
1
21r }(J-7r 2 . t- 0
sm--
2
7r
~ ~ r9+ (¢n-l(t) - ¢n_l(O))2 dt
21r }(J-7r 2 . t- 0
sm--
2
2 2
= € mn - 1·
Also, integrating by parts, we get J 2 = 2 log 2 == c2 . Hence, from (8.5.2)
(8.5.3)
Since c = mo, we find that
Thus, by (8.5.1),
I<p~+ 1 (B) - 11 : :; c; m~ + c;2 I<P~-1 (B) I

1 c;n+l) 2 (8.5.4)
:::; 2 c; ( ~_ c; log 2 + C;2 I<P~-1 (B) I.
Set n = 1 in (8.5.4). Then, since <Po(t) = 1, we find that
(8.5.5)
which is less than 1 if c; :::; 0.34. If we set n = 2 in (8.5.4), then, since

<PI (B) :::; 1 + 2 € log 2, and <PI (B) > 0 for c; < (2 log 2) -1, we find that
1<p;(B) - 11:::; 2c; (1 + € + C;2)2 log 2 + c; (1 + 2c; log 2),

which is less than 1 if c; :::; 0.31. Note that, by (8.5.5), I<p~(el ::; 1.7927 if
c; = 0.3. Hence, setting n = 3 in (8.5.4) and using this estimate for <p~(B), we
find that 1<P4 (B) :::; 1 if c; :::; 0.3. Let us assume that 0 < <P~-1 (B) :::; 2 for some
n 2: 1. Then, from (8.5.4),
if c; :::; 0.295. Since this assumption is obviously valid for n = 1 and n = 2 if

€ :::; I
0.295, it follows that <P~+l (B) - 11 < lfor all n ;::: 4 if c; :::; 0.295.•
Theodorsen (1931) considered the case where the exterior of the circle I(I = R
is mapped onto the exterior of a nearly circular region D bounded by r. In this
case the mapping function w = g(() is normalized such that lim ~ = 1. This
(->00 ."
case reduces to that analyzed in §8.1 if we use the transformations w = l/w
and z = R/(. Let the boundary r be defined by r = r (<1», 0 :S <1> :S 21r,
where, analogous to (8.1.1) and (8.1.2), for some b > 0 and 0 < c < 1,
b
1 + c :S r(<1» :S b (1 + c), (8.6.1)
and
r' (<1» I <- c. (8.6.2)
I r(<1»
Thus, the function w = 1(z) = gto' ( = R/z, maps the unit disk Izi < 1
onto Int (f), where r is represented by the equation
such that ¢ = -<1>, p( ¢) satisfies the conditions (8.1.1) and (8.1.2), and a = l/b.
Then, for ( = R ei 1{J, we have
arg {g(()/O = <1>(1/1) -1/1,
where we take arg {g( () /(} (=00 = O. Thus, for ( = R ei 1{J and z = eilJ ,
where 8 = -1/1, we have
log g~() = log r [<1>(1/1)] + i [<1>('l/J) - 'l/J] -log R
(8.6.3)
= -log -1(z) - log R
z
= -log P [¢(8)] - i [¢(8) - 8] -log R.
Thus, <1>n(1/I) = -¢(8), 1/1 = -8, and we can form the iterations in this case,
analogous to (8.1.7), as
<1>0(1/1) = 0,
<1>n(1/I) -1/1 = -1 111" {log r (¢n-l(1/I + t)) -log r (¢n-l(1/I - t))}

21r 0
t
x cot "2 dt, n = 1,2, ....
(8.6.4)
Hence, the estimates for I¢n (8) - ¢( 8) I and for derivatives of these differences
obtained in §8.3 also hold for jepn('Ij!) - ep('Ij!)j and derivatives of these dif-
ferences. See §IOA for more on Theodorsen's method and its convergence
problems.
8.7. Trigonometric Interpolation
In practice, computations are specially simple if functions are expanded in

Fourier series and then the method of trigonometric interpolation is applied.
We fix 2N points on the unit circle in the z-plane by
Ok = k7r, k = 0,1, ... , 2N - 1,

n
and denote
¢kV ) = ¢(v)(Ok), wkv) = 10gp(¢(v)(Ok)). (8.7.1)
Using the Fourier polynomial
(v) N
v
wk ) = a~ + L (a~) cos nOk + f3~) sin nOk )' (8.7.2)
n=l
where
2N-l
a(v)
n
= 2. """ w(v) cos n8k ' n -"00'"
NL...J k
- 0 1 N
k=O
2N-l
(8.7.3)
(3~) = ~ L v
wk ) sinn8k , n = 0,1, ... ,N-1.
k=O
Then, in view of (504.17), the harmonic function
L (a~) sin n8k - f3~) cos n8k)

N
¢k
V
+1) - 8k = (8.7.4)
n=l
v
is conjugate to wk ). If we substitute the values of a~) and (3~) from (8.7.3),
then we obtain the formula
N
",(v+1) _ 8 __ 2. """ ( (v) _ (V)) t 8n (8.7.5)
'f'k k- N L...J wk+n wk- n co 2'
n=l
8.8. WEGMANN'S METHOD 229
where we set W1±2N = WI. Taking the initial value ¢kO) = ()k> formula (8.7.5)
together with (8.7.1) provides an iterative technique to compute the quantities
wk
¢f) and v ). Note thatthe finite sum (8.7.5) has a form that corresponds to the
integral in (8.1.6). Since the method converges, we can determine the Fourier
v
coefficients an and (3n by taking the limiting processes ¢k = lim v -+ oo ¢k )
and Wk = lim v -+ oo Wk
v
), which yields an approximation F(z) for the function
F(z) as
N-I
F(z) = ~O +L (an -i(3n)zn+ aNZ N, (8.7.6)
n=1
and an approximation j(z) for the mapping function as
j(z) = z eF(z). (8.7.7)
The function j(z) maps the unit disk onto a region b with a boundary which t
cuts the curve r in 2N points determined by z = ei1h . The quality of the
approximation depends on the closeness of the two curves rand between t
these 2N points. For better closeness, even 4N points can be chosen where the
values of ¢k and Wk already computed can be used as the first approximation.
This approximation method with Fourier series is suitable for regions with
smooth boundaries.
8.8. Wegmann's Method
This quadratic convergence method deals with the general problem of conformal
mapping between any two simply connected regions D and G bounded by
Jordan contours r and ~,respectively. Let this conformal map be denoted
by f : D ....... G such that all such maps f can be extended to fj together
with the homeomorphism f : r ....... ~. The function f is uniquely defined
if f(zo) = ao for Zo E D and f() = al for ( E r. Since the mapping
function is fully determined by its boundary values, the problem of computing
f can be reduced to that of finding the boundary map f : r ....... ~. Thus, if a
parametrization ')'( 8) of ~ is prescribed, then there exists a real function S(()
which satisfies
f(() = ')' (S()) , (8.8.1)
where ( = ((t) is a parametric representation of the boundary r, ')'(8) is
assumed to be a 27T-periodic function, and S is a continuous, multiple-valued
function that changes by 27r while winding around r once, i.e.,

[Sh)]r = 27r.
This mapping problem can be generalized to the case when the boundary ~ is not
necessarily a Jordan contour. In that case, let )'(8) E C 1[0, 27r] with )"( 8) =I- 0,
where prime denotes differentiation with respect to 8, and let", (2 1) be the
winding number of )" with respect to the origin. Then a function j analytic
in the region D and continuous on tJ and satisfying (8.8.1) and (8.8.2) maps
D onto a ",-sheeted Riemann surface with '" - 1 branch points. Since these
branch points cannot be determined by the boundary r alone, we can fix '" - 1
additional parameters through '" - 1 interpolation functions. Then j can be
determined if the following additional conditions are satisfied:
(8.8.3)
where Zo E r, ao E ~, and ZI,'" 'ZK. E D.
i
The iterative method to determine the approximate function that maps the
boundary r conformally onto the boundary ~ requires that ill() = )' (SlI())
prior to the v-th iteration. Then at the v-th step this map is updated by shifting
the function values along the tangent to the boundary curve, i.e., by determining
a real-valued function U ll () such that
(8.8.4)
where the function hll+l is analytic in D and continuous on tJ with hll+l (ZlI) =
all, v = 1,2, ... ,n. For the boundary point Zo E r (the case v = 0) the
prescribed value ao lies on~, i.e., ao = )'(80)' But since h ll +1(zo) lies on the
tangent through the point)' (SlI(ZO)), the condition hll+l (zo) = ao, in general,
is not satisfied. Therefore, we replace this condition by
(8.8.5)
Note that the function hll+1 is an approximation of the boundary map j, al-
though the values that it takes on r, in general, do not lie on~. The function
h ll + 1, however, yields a new approximation for the parameter mapping function
Sas
(8.8.6)
Thus, after starting with an arbitrary function SI (), the conditions SlI(ZO) = 80
and Ull (zo) = 0 are satisfied for all v 2 2, and hence, the condition hll + 1 (zo) =
ao is satisfied.
If we multiply (8.8.4) by -2~ r d( and integrate over r, we get

211" ." - Z
for z E r, (8.8.7)
Note that the formula (8.8.7) is a linearization of the equation
0 for z (j. D,
1
-2'
211"
1(
r
'Y (S(())
- z
d( =
{ 2
1
- 'Y (S(z)) for z E r, (8.8.8)
av for z = zv, v = 1, ... ,"'.
The middle part of Eq (8.8.8) is a nonlinear singular integral equation of the

second kind for the parameter function S associated with the boundary map f.
The method of solution for this equation is linearization and the use of Newton 's
method. This fact is used to prove convergence which is locally quadratic. A
proof is available in Wegmann (1984).
The numerical computation of the iterative method begins with the dis-
cretization (8.8.7) of Eq (8.8.8) for the boundary r which is used to compute
the updates U v ' However, in the particular case when r is the unit circle, explicit
representations of the solution in terms of integrals can be obtained. Details for
this case are as follows: All functions are discretized at n = 2 N equidistant
points (i = ei<t>i E r, rPi = rPo + i1l"/N, i = 1, ... ,no The integrals are
evaluated by trigonometric interpolation. Thus, for example, if
F(z) =~
2211"
1 rP(() d(,
r ( - z
(8.8.9)
then the function rP is represented as a polynomial in ( and (-1, i.e.,
N
¢((i) = L <Xj (f = F+((i) + F- ((i), (8.8.10)
i=-N
where
N
F-((i) = L <Xj (f· (8.8.11)
i=-N
If we define
eF(z)-F(O)/2 for Izl < 1
G(z) ={ ' (8.8.12)
Z2K e F (z)-F(O)/2 for Izl > 1,
g(()~ {j(()}
7jJ(z) = G(z)
2i7f Jr
r G+(()
g(() ~
(-z'
(8.8.13)
where the function j(() = "( (S(()) and g(() = "(' (S(()) are H61der- contin-
uous,and
ho(z) = ~ [7jJ+(() + 7jJ-(()] , (8.8.14)
then Wegmann (1984) has proved the following:
THEOREM 8.8.1. Let ( and"( be Holder-continuous functions with

((t) -1= 0 and "('(s) -1= 0 for all sand t, where ( = ((t) and"( = "((s)
are parametric representations of the boundaries rand D.., respectively,
of simply connected regions D and G. Let S(() be a Holder-continuous
junction satisfying (8.8.1). Let the winding number", of "(' be positive,
and let the points Zl, ... , ZK E D and Zo E r and the complex numbers
al, ... , a K and a real number uo be given. Then there exists a unique
Holder-continuous real function u(() defined on r and a unique complex
function h analytic in D and continuous on [) such that
"( (S(()) + u(() "(' (S(()) = h((), (8.8.15)
satisfying the conditions
h(zv) = a v for v = 1,2, ... , "', (8.8.16)
and
u(zo) = ao, (8.8.17)
In the case when r is the unit circle, the solution of the linearized inte-
gral equation in (8.8.7) can be represented explicitly in terms of integral
transforms.
Thus, the integrals are computed by using the discrete Fourier transform.
Note that if FFf is used instead, then the transformation of <p to p± takes
o (n log n) operations. The foregoing outline of the method and this theorem
show how this iterative method differs from that discussed in §8.1 and 8.2.
The iterative process is carried out in the following two steps:

STEP 1: Assuming that the functions ')'(s), ')"(s) and ¢o(s) = arg {I'(s)}
are defined explicitly and the initial value of S1 ((0) is available (initial guess),
once Sv has been computed for some 1/ ~ 1, then compute
iv ((i) = ')' (Sv((i)), 9v((i) = ')" (Sv((i)),

(8.8.18)
¢((i) = 2¢0 (Sv((i)) - 2",Oi.
STEP 2: Use one of the methods described below.

METHOD 1: Determine'I/J and ho from (8.8.13) and (8.8.14), and set
hv+1 (z) = ho(z) + P(z) G(z), (8.8.19)
where the polynomial P is chosen such that
P(zo) G+(zo) = (so - Sv(zo)) 9v(ZO) + lv(zo) - ho(zo),

(8.8.20)
P(Zi) G(Zi) = ai - hO(Zi) for i = 1, ... , "'.
Note that in this discretization Zo is equal to one of the (i. Then use Cauchy's
formula (1.2.2) and integral representation (8.8.9) to compute hO(Zi) and F(Zi)
for i ~ 1. Thus, G(Zi) = eF (Zi)-F(0)/2 :f; 0, and P is taken as an interpolation
polynomial. Finally, set
(8.8.21 )
METHOD 2: Instead of (8.8.13), compute 'I/J for 1/ ~ 2 by the formula
(8.8.22)
Then h o is computed from (8.8.14), which leads to
hV+l(z) = hv(z) + ho(z) + P(z) G(z), (8.8.23)
where the polynomial P is computed from
(8.8.24)
As noted by Wegmann (1986), the first method gives more accurate results than
the second.
CASE STUDY 8.8.1. The result of the conformal mapping of the unit
circle ((t) = eit onto the inverted ellipses (see Case Study 7.4.4), defined by
p(s) = Jl - (1 - p2) cos 2 S, 0 < p < 1, are shown in Wegmann (1986)
for p = 0.4,0.6, and 0.8, where the error after the 100th iteration is 11811 -
811 ::::; 1.8 x 10- 2 , 5.7 x 10-5, and 10- 8 , respectively, for the above values
of p with n = 40. The exact boundary correspondence function is given by
tan s = p tan 0, where 0 = arg {z}, and the boundary map is defined by
2pz
w- •
- (l+p)+(I-p)z2'
8.9. Newton's Method
As explained in §8.1 and 8.8, the function f (r e iO )) = p (¢(O» ei¢(O) defines

the boundary correspondence function ¢ : [0, 21r]1--t lR for the boundary map
f. Let K denote the conjugation operator
K [h] (0) = -
21r
1 1
0
2
11" 0- t
h(O) cot -2- dt, (8.9.1)
where the integral takes a Cauchy p.v. Then Theodorsen's integral equation
(8.1.5) can be written as
¢(O) = 0 + K [log p(¢(O»] (0), (8.9.2)
or, if we set 'l'(0) = ¢(O) - 0, then as
'l'(0) = K [log p('l'(O) + 8)] (8), (8.9.3)
which, according to Gaier (1964) and Wegmann (1984), has a unique 21r-
periodic solution 'l'*(0). Wegmann (1984) has also proved that for f E L 2 and
Kf E L OO
(8.9.4)
We shall assume that the 21r-periodic function p is absolutely continuous on lR

and 10"1 =< c a.e. (we say that p satisfies the c-<:ondition), where 0" is defined
in (8.1.9). The solution 'l'* (O)ofEq (8.9.3) is given explicitly as a consequence
of the Riemann-Hilbert problem (see Appendix C). As we have seen in §8.1

and 8.3, a numerical solution of the integral equation (8.9.3) can be obtained
by first discretizing this equation and then applying an iterative method. The
convergence of this method depends on the c-condition (10 < 1), except in the
case of certain symmetric curves with corners and pole singularities. Thus,
it can be shown, as in Gaier (1964), Warschawski (1955), Hiibner (1979) and
Gutknecht (1981,1983), that Ilw n - w*112 ---4 0 and IIwn - w*lloo ---4 0 as
n ---4 00.
Let F : W f-t W, where W is the Sobolev space of 2n-periodic and

absolutely continuous functions f with l' E £2 (see §1.1). Define the operator
F (W(O)) = '1'(8) - (K [log p(w(O) +8)]) (0). (8.9.5)
Then Newton's method for solving the equation (8.9.3) is
WoE W,
(8.9.6)
F' (w n ) ['1'n+1 - wnl = -F (w n ) , n = 0,1, ... ,
where F' (w n ) is the F -derivative * of F in the Banach space (W, II . II) at'1'n.
The operator F(w) has an inverse for any W E W. Using the solution of a
Riemann-Hilbert problem (see §C.3), Hiibner (1986) has shown that
(8.9.7)
Thus, to determine Wn +1 from wn in formula (8.9.7), we require two applica-

tions of the conjugation operator !C. For numerical computation, we discretize
(8.9.7) instead of (8.9.3). Then by using FFr we first approximate n by a w
vector in jR2n and then compute an approximate value of Wn +1 in the same
space. This method, however, is different from Wegmann's (§8.8) as shown in
Case Study 8.9.1.
Hiibner (1986) has proved the following two results on the convergence of
Newton's method.
*If P is Holder-continuous and 'It E W, then the F-derivative of :F at 'It in

(W, II . II) exists and is given by
(:F' ('It)O) (0) = 0(0) - (K. [a('It(8) + 8) . 0(8)]) (0)

for some 0 E W.
THEOREM 8.9.1. Ifp is 21f-periodic and Lipschitz-continuous, then

Newton's method for Theodorsen's integral equation {8.9.3} converges
locally and quadratically in (W, II . II).
THEOREM 8.9.2. If 10'1 < 1/3 and p E Loo, then Newton's method
for Theodorsen's equation {8.9.3} converges globally in (W, I . 112)'
Proofs of these theorem are given in Hubner (1986).
CASE STUDY 8.9.1. Wegmann's and Newton's methods are not identical
for boundary curves defined in polar coordinates. As an example, take p(¢;) == 1
and'li o = sin 8. Then.r ('li o(8)) = 'li o - K[I] = sin 8. Hence, 'li 1 = 1, which
is the exact solution of Eq (8.9.3).•
8.10. Problems
PROBLEM 8.10.1. Prove Theorem 8.8.1. (For a proof see Wegman, 1986,
p.9.)
PROBLEM 8.10.2. Prove Theorems 8.9.1 and 8.9.2. (For a proof see
Hiibner, 1986, p.25.)
PROBLEM 8.10.3. Develop an algorithm for Newton's method as devel-

oped in Hubner (1986), pp.19-30.
REFERENCES USED: Carrier, Krook and Pearson (1966), Gaier (1964), Gutknecht
(1981, 1983), Hiibner (1979, 1986), Koppenfels (1959), Privaloff (1916), F. and
M. Riesz (1923), Theodorsen (1931), Theodorsen and Garrick (1934), Warschawski
(1935, 1945, 1950, 1955), Wegmann (1984, 1986), Zygmund (1935).
Chapter 9
Symm's Integral Equation
A potential-theoretic formulation of the problem of conformally mapping a

simply connected region (or its complement) onto the unit disk leads to a
Fredholm integral equation of the first kind, known as Symm's integral equa-
tion, which has a kernel with a logarithmic singularity. Unlike Fredholm
integral equations of the second kind, e.g., Theodorsen's equation, in which
the singularity of the kernel at points near but not on the boundary creates
computational difficulties, Symm's integral equation is found easily solvable
by numerical methods, such as the orthonormal polynomials method or its
modified form, Lagrange's interpolation method, and spline approximations
which are discussed in this chapter. Numerical evaluation of Green's func-
tions, as developed in Chapter 6, is another viable alternative to obtain the
approximate mapping function.
9.1. Symm's Integral Equation
We shall derive Symm's integral equation for both interior and exterior regions.
9.1.1. Interior Regions. The function w = f(z) that maps a given

simply connected region D with a Jordan boundary r onto the unit disk U
such that f(zo) = 0 is given by
w = f(z) = e10g (z-zo)+g(z-zo)+i h(z-zo), (9.1.1)
237

238 CHAPTER 9. SYMM'S INTEGRAL EQUATION
(see (6.2.10); also Gram, 1962), such that \12 g = 0, zED, which yields
g(z) = -log Iz - zol, z E f, and h is the conjugate of g. Without loss of
generality, we shall take Zo as the origin. Then f(z) = e10g z+g(x,y)+ih(x,y),
z = x + i y, where
\12 g =0, ZED,
1 (9.1.2)
g(x,y)=-2Iog(x2+y2), zEf.
If w = u + i v, then from (9.1.1), we get
u(x,y) = e10g Izl+g cos (arg{z} + h),

(9.1.3)
v(x, y) = e 10g
Izl+g sin (arg{z} + h).
Thus, the conformal mapping problem reduces to that of determining the har-
monic functions 9 and h. We shall represent the harmonic function g(x, y) as
a single-layer logarithmic potential
g(x,y) = Ir log Ix - (I J.l(() de (9.1.4)
where J.l(() is a suitable source density function on the boundary f (see Kythe,
1996, p.21; Maiti, 1968). The harmonic conjugate ofthe representation (9.1.4)
l
is given by
h(x, y) = B(z - () J.l(() de (9.1.5)
(see Jawson, 1963), where B(z - () = arg{z - (}. Hence, the problem further
reduces to that of finding the density function J.l(() such that 9 satisfies (9.1.2) on
f. Once J.l(() is known, the functions 9 and h can be determined by quadrature
at any point in D. Since the function J.l(() is continuous in D, it satisfies the
integral equation
Ir log Iz - (I J.l(()d( = -log Izl, zED, (E f, (9.1.6)
which is known as Symm's integral equation for interior regions. This equa-
tion has a unique solution provided cap (r) i- 1 (for the existence of the
solution, see Jawson, 1963).
Note that although the function g( x, y) is single-valued, the function h( x, y),

in general, is multiple-valued. In fact, suppose that some Jordan contour f* is
9.1. SYMM'S INTEGRAL EQUATION 239
contained in D such that 0 E Int (f*) and r

Jr- un
~g ds = A, where n denotes the
inward normal to f*, and A =I- 0 is a constant. Then, in view of the Cauchy-
Riemann equations, the value of h will increase by an amount A whenever z
traverses f* in a clockwise direction. Hence, h will be single-valued on the
contour f* only if A = 0, i.e., for any such f* we require that
1 r- un
8g
-;::;- ds = 0, (9.1. 7)
as a condition for the function h to be single-valued.
9.1.2. Exterior Regions. For conformal mapping of the region D* =

{w : Iwl > I} such that f(O) = 1 and f(oo) =
Ext (f) onto the region U* =
00, the mapping function w = fE(Z) is unique up to a rotation. Let C =
diam (D) = lim If'(z)I- 1 > 0 denote the transfinite diameter of D (see
z---+oo
§1.1). Then the required mapping function h(z) must satisfy the condition
1
f~(z) -t C as z - t 00. (9.1.8)
Assuming that 0 ED, the function f E (z) is regular in D* , including the point
z
z = 00. Hence, we take
h(z) = e10g z+1/;(z) , (9.1.9)
where 'l/J(z) is regular in D*, and in view of (9.1.8), 'lj;(z) - t -log C =

, as z - t 00, where, is Robin's constant (§1.3). Let'l/J(z) = (3(z) + ,.
Then (3(z) is regular in D*, and (3(z) - t 0 as z - t 00. Hence, (3(z) can be
represented as
(3(z)=g(x,y)+ih(x,y), z=x+iy,
where 9 and h are conjugate harmonic functions in D* such that 9 -+ 0 and

h -+ 0 as z - t 00. Thus,
(9.1.10)
where the boundary condition IfE(Z)1 = 1 for z E f becomes
, + g(x, y) = -log Izl, z E f. (9.1.11)

As in §9.1.1, the harmonic functions fJ(x, y)and h(x, y) can be represented in

the form (9.1.4) and (9.1.5), respectively. Then the boundary condition (9.1.11)
for the density function J.L( () becomes
1 log (z - () J.L«() d( + "f = -log Izl, z E f, (9.1.12)
and the condition g( 00) = 0 reduces to
lJ.L«()d( = O. (9.1.13)
Eqs (9.1.12) and (9.1.13) are coupled integral equations for J.L( () and "f, and
they have a unique solution (see lawson, 1963, and Symm, 1967). Once J.L
and "f are computed from (9.1.12)-(9.1.13), the mapping function fe(z) can
be determined from (9.1.10).
As explained in §7.3.3, the region D* may be mapped onto U by using the

inverse transformation z 1-+ z-l such that the point z = 00 goes into w = o.
9.2. Orthonormal Polynomial Method
The orthonormal polynomial (ONP) method, developed by Rabinowitz (1966),

is used to compute the density function J.L«() numerically. The basic idea is to
approximate J.L( () by a step-function. To do this, we partition the boundary
f into N sections f l , ,f N , and assume that J.L«() == J.Lj(= const) for any
point ( E f j , j = 1, ,N.
(a) INTERIOR REGIONS. Then Eq (9.1.6) reduces to

N
I : {1 10g Iz - (1Id(l} J.Lj = -log Izl, z E f. (9.2.1)
j=l rj
If we take z E f with each of the N nodes Zk = Xk + i Yk> k = 1, ... ,N, then

we obtain a system of N linear equations in N unknowns J.Lj:
N
I:{110g IZk - (1Id(l} J.Lj = -~ log (x~ +yD, k = 1'00' ,N.
j=l rJ
(9.2.2)
9.2. ORTHONORMAL POLYNOMIAL METHOD 241
The solution J1j, j = 1, ... ,N, of this system gives the approximate values of
9 and h from (9.1.4) and (9.1.5), respectively, as
G(x, y) = L
N
{1
j=l rj
10g Iz - (1Id(l} J1j, zED,
(9.2.3)
L {1
N
H(x, y) = O(z - () Id(l} Mj, ZED, (9.2.4)
j=l rj
and then the approximate mapping function f(z) can be determined from
(9.1.3).
The details for selecting the nodes Zj and evaluating the integrals in (9.2.3)
and (9.2.4) are as follows: A convenient way to partition r is to take the sections
r j with end points Zj-l/2 and Zj+1/2 and to take the nodes Zk as any point in
each r j. Then, for any Z E D\ {r j }, we take
r log Iz - (1Id(1 = !i6 [log

Jr j
IZk - Zj-l/21 + 4 log IZk - Zjl
+ log IZk - Zj+1/2IJ, (9.2.5)
r O(z -
Jrj () Id(1 = !i
6
[0 (Zk - Zj-l/2) + 40 (Zk - Zj)
+ 0 (Zk - Zj+l/2) ], (9.2.6)

where lj is the length of r j . If r is a simple Jordan contour, the length lj for
each r j is the arc length which can be easily evaluated analytically, and if Zj is
the mid-point of r j , the formulas (9.2.5) and (9.2.6) correspond to Simpson's
rule. However, if the boundary r, in general, is analytic, the length lj can be
approximated by
lj = IZj - Zj-l/21 + IZj - Zj+1/21, (9.2.7)

where Zj need not be a midpoint of r j '
When Z E r j , formulas (9.2.5)-(9.2.6) cannot be used because log Iz - (\

has a singularity at Z = ( and O(z - () is undefined at Z = (. However, in each
case the integrals exist and can be computed approximately. In particular, for
Z = Zj we take
[log IZj - (1Id(1 = IZj - Zj-l/21 (log IZj - Zj-l/2!-1)

J
+ IZj - Zj+1/21 (log IZj - Zj+1/2\-1), (9.2.8)

and for z = Zj±l/2 we take
liog IZj±1/2 - (1Id(1 = IZj+l/2 + Zj-l/21 (log IZj+l/2 - Zj-l/21- 1) ,

1
(9.2.9)
and
(9.2.10)
Thus, formula (9.2.5) with Z = Zk for j =J k, and formula (9.2.8) for j = k

give the coefficients of /1j in (9.2.2), whereas the coefficients of /1j in (9.2.3)
and (9.2.4) are given by (9.2.5) and (9.2.6), respectively.
To estimate the error involved in this method, let F(z) denote the ap-
proximate mapping function. Then, by hypothesis, !F(z)1 = 1 at each node
Zj E r j . The maximum error EM can be estimated by computing the values
of IIF(z)l- 11, z E r, such that
EM = sup 1!F(z)I-11· (9.2.11)

zEr
In view of the maximum modulus theorem (§1.2), !F(z) - f(z)1 assumes its
maximum value somewhere on the boundary r. Also for any z,
!F(z) - f(z)1 ~ IIF(z)1 -If(z)ll + If(z)ll arg{F(z)} - arg{f(z)}j

= 1!F(z)I-lf(z)11 + If(z)IIH(x,y) - h(x,y)1
~ max{I!F(z)I-11
zEr
+ IH(x,y) - h(x,y)l}
~ EM + max IH(x, y) - hex, y)l,
zEr
(9.2.12)
where we have used (9.1.5) and (9.2.4). This inequality implies that if his
known on r, then the absolute error in the approximate mapping function F(z)
can be determined at any point. But h, in general, is not known for any z, except
when the region D has symmetry. In that case h is known at some points of
r. In fact, if B(z - () is so defined that the axes of symmetry are mapped onto
themselves, then h = 0 on such axes. Also, we may expect maximum error
at some of these points on r, which are, e.g., the end points of the major and
minor axes of an ellipse or comers of a rectangle or square. Therefore, we take
the largest value attained by IHI at such points as an estimate of maximum
error in IH(z) - h(z)1 for z E r, which, in view of (9.2.12), accounts for
maximum error in arg{F(z)}, z E r. We denote this estimate by EA. Then,
from (9.2.12), the sum EM + EA gives an estimate for the upper bound on
absolute error in the GNP method.
CASE STUDY 9.2.1. Consider Cassini's oval
4
a = 1.5 a = 1.4
a =5 -4
Fig. 9.2.1. Cassini's oval for a = 1,1.1,1.2,1.3,1.4,1.5,2,2.5,3,5.
which is represented in Fig. 9.2.1 for different values of a (also see Fig 6.3.1).
The region D is not univalent for a = 1. Each contour r is symmetric about
both coordinate axes, so we shaH consider the first quadrant. The exact mapping
function is given by (Rabinowitz, 1966)
az
f(z) = va4 _ 1 + z2
Tabular data for the approximate mapping function, using the GNP method,
are given in Symm (1966) for a = 1.2. The functions U(x, y) and V(x, y)
computed for a = 1.3 are given below in Table l. •
Table 1. Cassini's oval, a = 1.3
x y U(x,y) V(x,y)
0.00 0.830662 0.00000 0.99999

0.05 0.830937 0.09541 0.99491
0.10 0.831741 0.18887 0.98201
0.15 0.833019 0.27838 0.96047
0.20 0.834681 0.36248 0.93199
0.25 0.836608 0.44009 0.89795
0.30 0.838659 0.51061 0.85981
0.35 0.840675 0.57384 0.81897
0.40 0.842488 0.62991 0.77667
0.45 0.843923 0.67921 0.73395
0.50 0.844805 0.72227 0.69161
0.55 0.844960 0.75971 0.65026
0.60 0.844218 0.79217 0.61029
0.65 0.842412 0.82026 0.57199
0.70 0.839382 0.84456 0.53546
0.75 0.834966 0.86559 0.50075
0.80 0.829006 0.88381 0.46783
0.85 0.821342 0.89963 0.43665
0.90 0.811805 0.91338 0.40709
0.95 0.800220 0.92538 0.37903
1.00 0.786394 0.93587 0.35234
1.05 0.770111 0.94507 0.32686
1.10 0.751122 0.95317 0.30245
1.15 0.729135 0.96031 0.27895
1.20 0.703789 0.96662 0.25619
1.25 0.674634 0.97223 0.23401
1.30 0.641080 0.97723 0.21219
1.35 0.602324 0.98169 0.19049
1.40 0.557216 0.98568 0.16862
1.45 0.503985 0.98973 0.14612
1.50 0.439623 0.99251 0.12225
1.55 0.358101 0.99542 0.09562
1.60 0.242597 0.99806 0.06228
1.6401219 0.000000 1.00001 0.00000
(b) EXTERIOR REGIONS. In this case we solve the coupled equations

(9.1.12)-(9.1.13) by the GNP method as follows: With the partitions r j ,
j = 1, ... , N, Eq (9.1.12) reduces to the system of linear equations
k= 1, ... ,N,
(9.2.13)
where..y approximates, and Eq (9.1.13) becomes
N
L: {lld(l} ILj = O. (9.2.14)

j=l rj
Thus, Eqs (9.2.13)-(9.2.14) form a system of (N + 1) linear equations which

are solved to determine the (N + 1) unknowns ILl, ... , ILN and..y. The solution
for..y determines the approximate transfinite diameter (} of the region D, where
(} = e -"'I. The approximations Gand H for the functions 9 and h are given by
G(x,y) = L:{110g Iz-(lld(I}ILj,

j=l rj
N
H(x,y) = L:{1rj arg{z-Old(I}ILj,

j=l
where z = x + i Y E D* U r. Then the approximate mapping function FE(z)

is computed from (9.1.10) as
z· ..
h(z) ~ FE(z) = U + i V = "eG(x,y)+tH(x,y). (9.2.15)
C
The choice of the nodes and evaluation of integrals is performed in the same
manner as in part (a) above, except for the integral in (9.2.14) which is evaluated
by
(9.2.16)
where Zj ± 1/2 denote the end points of r j '
ERROR ESTIMATE. Since IE(oo) = 00 and fE(z) =I- 0 for zED, the
relative error is given by
IFE(z) - fE(z)1 = FE(z) - I 11.

IfE(z)1 fE(z)
By the maximum modulus theorem, the maximum value of this error is attained
on the boundary r or at Z = 00. In view of Symm (1967), the error at Z = 00
is considerably less than the maximum error on r. Thus, for Z E r
- 11 = IIFE(z)1 11 ::; II FE(z) 1- 11 + IiI - hi·

IFE(Z)
fE(Z)
ei(fI-h) -
fE(Z)
SinceIiI - h I is generally not known on r, although it is of the same order as

16 - gl, and since, by hypothesis, IIFE (Zj)l-l! = 0, the maximum error E
in FE(z) can be measured by
(9.2.17)
CASE STUDY 9.2.2. Consider the ellipse x 2 ja 2 + y2 = 1, a = 1.5, and

N = 100 in the first quadrant because of the symmetry about both axes. The
values of U(x, y) and V(x, y) are given in Table 2 on page 247. The transfinite
diameter C = (a + 1)/2 (P6lya and Szego, 1951), and the exact mapping
function is given by (Phillips, 1966)
z+.Jz2-a2+1
f E ()
Z =
a+l
.•
9.3. Modified ONP Method
The GNP method, discussed in §9.2, for numerically solving Eqs (9.1.4), (9.1.5),
and (9.1.6) has been modified by Hayes, Kahanerand Kellner (1972) as follows:
Letthe parametric representation ofthe boundary r be ( = ((t) = x(t)+i y(t),
0::; t ::; L. Then Eq (9.1.4) and (9.1.5) can be written as
g(z) = 1 L
p,(t) log Iz - ((t)1 dt, (9.3.1)
h(z) = 1 L
p,(t) arg {z - ((tn dt, (9.3.3)
where p,(t) = p,(((t)) and zED. We shall assume that p,(t) E C 3 ( -00, +00)
and the boundary r has no corners (for corner singularities, see §9.5 and 12.2).
9.3. MODIFIED ONP METHOD 247
Table 2. Ellipse, a = 1.5
x y U(X, y) V(X,y)
0.00 1.000000 0.00000 0.88989

0.05 0.999444 0.03633 0.88936
0.10 0.997775 0.07267 0.88774
0.15 0.994987 0.10903 0.88503
0.20 0.991071 0.14542 0.88123
0.25 0.986013 0.18184 0.87633
0.30 0.979796 0.21831 0.87032
0.35 0.972397 0.25483 0.86317
0.40 0.963789 0.29142 0.85487
0.45 0.953939 0.32808 0.84540
0.50 0.942809 0.36483 0.83472
0.55 0.930352 0.40167 0.82278
0.60 0.916515 0.43862 0.80961
0.65 0.901234 0.47567 0.79509
0.70 0.884433 0.51285 0.77918
0.75 0.866025 0.55019 0.76184
0.80 0.845905 0.58761 0.74296
0.85 0.823947 0.62521 0.72248
0.90 0.800000 0.66296 0.70025
0.95 0.773879 0.70088 0.67614
1.00 0.745356 0.73896 0.64997
1.05 0.714143 0.77722 0.62152
1.10 0.679869 0.81565 0.59048
1.15 0.642045 0.85425 0.55647
1.20 0.600000 0.89303 0.51892
1.25 0.552771 0.93198 0.47703
1.30 0.498888 0.97110 0.42959
1.35 0.435890 1.01039 0.37452
1.40 0.359011 1.04983 0.30778
1.45 0.256038 1.08942 0.21902
1.50 0.000000 1.12915 0.00000
We partition r into an even number n of uniform sections r j, j = 1, . .. , n, of

length a = L / n each. Then the nodes on the section r j are uniform as regards
the arc length of each rj. First, we define a set of piecewise polynomials Pj(t),
j = 1, ... ,n, by
2:2 (t + a)(t + 2a), -2a:S t :S 0,

P1(t) = 2:2 (t - a)(t - 2a), O:S t :S 2a,
{
0, otherwise,
P2(t) = { - 2:2 t(t - 2a), O:S t :S 2a, (9.3.3)

0, otherwise,
n
P2k+1(t) = P1(t - 2ka), k = 1,2, ... , 2 -1,
n
P2m(t) = P2 (t - 2(m -l)a), m = 1,2, ... , 2'
The graphs of some of these polynomials are presented in Fig. 9.3.1 for a = 0.1.
t
.6
Fig. 9.3.1.
Next, we define the function

n
jj,(t) = L jl(ja) Pj(t). (9.3.4)
j=l
The following results hold:
(i) jj,(t) is a polynomial of degree two on [ja, (j + 2)a], j = 0,2,4, ... ,n - 2,
(ii) ji,(t) = jl(t) at t = ja,j = 0,1,2, ... ,n.
n
(iii) jl(t) = ji,(t) + 0 (a 3 ) = LjljPj(t) + 0 (a 3 ), (9.3.5)
j=l
9.3. MODIFIED ONP METHOD 249
where J.-Lj = J.-L(ja) for j = 1, ... ,n. Using the approximation (9.3.5) for J.-L(t)
in (9.3.1), we obtain an approximation for gas
n
LJ.-Lj
j=l
10
L
Pj(t) log Iz - (t)1 dt = g(z) + 0 (a 3 ). (9.3.6)
The function g(z) = -log Iz - zol for z E r. Thus, we can evaluate Eq

(9.3.6) at the points z = ja for j = 1, ... ,n, which yields a system of n
linear equations with constant coefficients for the unknowns J.-L1, J.-Lz, ... ,J.-Ln.
Set A = (ajk) and B = (b j ), where
ajk = 1 L
Pj(t) log I((ja) - (t)1 dt for j, k = 1, ... ,n,
(9.3.7)
bj = -log I(ja) - zol, for j = 1, ... ,n.
Using this matrix notation Eq (9.3.6) becomes the linear system
(9.3.8)
where 0 (a 3 ) is a vector (column matrix). Thus, each component ofEq (9.3.8)

is bounded by 0 (a 3 ), and J.-L = (J.-L 1, ... ,J.-Ln) T, where T denotes the transpose
of a matrix. Hence, we solve the matrix equation
Ajj = B, (9.3.9)
where A = (ajk) is the approximation of the matrix A and jj that of the vector
J.-L. To evaluate A, however, we must compute the elements ajk which contain
integrals of the form
1
j O:
t j log Iz - (t)1 dt (9.3.10)
(j -1)0:
for j = 1, ... ,n. Also, for each fixed x, y, and j, we approximate Iz - (t)1
by a polynomial q(t) of degree two on the interval ((j - l)a, ja), where q(t)
is chosen such that q(t) = Iz - (tW for t = (j - l)a, (j -1/2)a, ja. Then
l
j
(j-l)o:
t j log Iz - (t)1 dt ~ -
O: 1
2
l j
O:
(j-l)o:
t j log [q(t)] dt, (9.3.11)
which can be evaluated explicitly. In some particular cases, e.g., when Iz -

(t)1 = 0 on [(j -l)a, ja], the integral (9.3.10) is computed with a polynomial
of higher order.
Then the matrix equation (9.3.9) is solved for the vector tL = (tL1, ... ,tLn)T
which is the approximation for the density function j-L, with an error estimate
(9.3.12)
where the error due to the term A. -l_A -1 seldom dominates the IIA -1110 (a3 )
term, an observation made by Hayes et al. (1972) from certain examples. Once
tL is computed, the functions g(z) and h(z) are obtained from
g(z) = 1 L
j-L(t) log Iz - ((t)1 dt
= L tLk
k=l
n
1 0
L
pk(t) log Iz - ((t)1 dt,
(9.3.13)
and
h(z) = 1 L
j-L(t) arg {z - ((tn dt
~ tp,(ja)
j=l
1 0
L
Pj(t) arg{z - ((tn dt (9.3.14)
n
= I>(ja) [Pj(t) 7/1(t) - pj(t) 7/2(t) + p'j(t) 7/3 (t)] ,
j+1
by integration by parts, where
7/1(t) = 1 L
arg{z - ((tndt, 7/2(t) = 1
L
arg{z - ((tndt,
7/3(t) = 1 L
arg {z - ((tn dt.
Note that p'j(t) is constant. Thus, (9.3.14) involves integrals of the form
(9.3.10).
Hayes et al. (1972) have a Fortran IV computer program for this method.
This program can easily be adapted to any modem operating system. As ex-
amples, they investigated cases of Cassini's oval (Case Study 9.2.1; Problem
9.6.3), an ellipse (Problem 9.6.2), a rectangle (Problem 9.6.4), and an isosceles
triangle (Case Study 9.3.1).
9.4. LAGRANGE INTERPOLATION 251
CASE STUDY 9.3.1. Consider an isosceles triangle with corners at the
°
points (0,1), (2, -1) and (-2,1) such that the point (0,0) goes into the point
w = under the conformal mapping w = f(z). The partition is taken with
equal number of nodes on each side. The error EM, defined by (9.2.11) is
found as follows:
2 X 10- 4 forn = 17,

Em = 2 X 10- 5
forn = 33,
{
10- 6 for n = 65.•
9.4. Lagrange Interpolation
The choice of a suitable basis set used in the RM, BKM or ONP methods is
an important aspect of any numerical technique for polynomial approximation
ofthe mapping function. Gautschi (1977, 1978, 1979) investigated numerical
conditions of various bases for polynomial approximation on the real axis.
Reichel (1985) observed that a well-conditioned basis depends on the shape of
the simply connected region D bounded by a Jordan contour r. For example,
the monomials cPj(z) = (zlr)j, j = 0,1, ... , are well-conditioned for disks
B(O, r), butthis basis becomes ill-conditioned for ellipses (see Case Study 9.4.1
below where a criterion for well-conditioned bases is developed and applied to
these cases).
CASE STUDY 9.4.1 (Gautschi criterion). Letpj(z), j = 0,1, ... ,n, de-
note polynomials such that span {Pj} = span {zj} 7=0 7=0'
We shall determine
the sensitivity of the functions
n
Pn(z) =L aj Pj(Z), ZED, (9.4.1)
j=O
subject to the perturbations in the coefficients aj. Let M n : en+! ....... II n denote
the mapping of the coefficient space en + 1 onto the space II n of polynomials
of degree::::: n. Let a = (ao, aI, ... , an) E en+! denote a vector. Then we
define
n
(Mna) (z) = L ak Pk(Z), zED. (9.4.2)
k=O
Note that M;;1 Pn(z) = a. The maximum norm in en+l and II n is defined,
respectively, by Iiall oo = max lanl, and IIIInll r = sup IPn(z)l· Let IIMnl1
O~k~n zEr
and II M;; 111 be the induced operator norms. We are interested in determining
how the condition of the map M n defined by
(9.4.3)
grows with n for different choices of the polynomial Pj(z). We shall examine
two choices:
(a) Let D be the unit disk U and Pj(z) = zj, j = 0,1, .... Then
n
IIMnl1 = max II2:ajzjll
lIall oo = l .)=0 r
=n+l,
and
Since
we find that
Hence, cond (Mn ) = n + 1, which shows that the monomial basis {zj} ;=0 is
well-{;onditioned for the unit disk. Thus, this basis is well-{;onditioned for the
disk B(O, r).
(b) Let r denote the ellipse E(a, b) = {(x, y) : x 2/a 2 + y2/b 2 = I}, z =
x + i y, and a ~ b. Using the scaling factor 1/ a, this ellipse is transformed
into the ellipse E(I, b/a) = {Z : Z = z/a} which has foci at ±~, where
~ = Jl - (b/a)2. We use the Chebyshev polynomials Tn of the first kind
Tn(Z,~) = 2~:1 Tn (Z/~) = 2~ [( Z + JZ2 - ef

+ ( Z - J Z2 - ~2 f] , n = 0,1, .. .. (9.4.4)
Let the points Z E E (1, b/a) be taken as
Z = ~ [(1 + ~) eiO + (1 - ~) e- iO ] , OS e S27f.
r r
Substituting them in (9.4.4), we obtain
Tn(Z,~) = [~ (1 +~) inB

+[~ (1-~) inO
r
e e-
(9.4.5)
~ [~ (1 + ~) e
inO
as n ~ 00.
Let.B(n) = (.Ban), .Bin), ... ,.B~J2])T denote the coefficient vectorofTn(Z, ~),
i.e.,
[n/2]
Tn(Z,~) = L .Bkn ) zn-2k, (9.4.6)
k=O
where [n/2] is the greatest integer 5 n/2. If ~ = 0 (the case of the circle
Izl = a), then .Ban) = 1 and .Bkn) = 0 for all k ~ 1. We shall not discuss this
case because it has been examined in part (a) above. Let ~ > O. Then
k ] -1 1I.B(n) 1100
II Lak Z
-1 [ .
IIMnl1 = mm
lIall oo =l
IIE(lb/a)
'
~ liTn II E(l,b/a) . (9.4.7)
From Gautschi (1979) we have
1I.B(n) II ~ (2 + e)3/4
V~
(1 ~
(1 + ~)n 2
as n ~ 00. (9.4.8)
Hence, from (9.4.5), (9.4.7) and (9.4.8) we find that
Since IIMn l1 = n + 1, we have
f§2n ( + ( l+~)n
2)3/4
1 ~
cond (Mn ) ~ -:;- ~ as n ~ 00.
1+-
a
If we set p = b/ a and
F( ) = 1 + ~ = 1 + J2=P2
p 1+~ l+p'
a
then
F(O) = 1 + v'2, which corresponds to r = [-a, a],

F(I) = 1,
F'(p) < 0, 0 S pSI,
which imply that in the case of a nondegenerate ellipse (~ > 0) the condition
number cond (Mn ) increases exponentially with n for the monomial basis and
the growth rate increases with p. Hence, if the region is not circular or nearly
circular, the choice of the monomial basis is ill-conditioned for the ellipse and
may produce computational inaccuracy. •
Reichel (1985) has shown that Lagrange's interpolation functions
lk (z)
= II
n
Z - Zj, k = 0,1, ... ,n, (9.4.9)
j=O Zk - Zj
j#
where Zk are Fejer points on the boundary r, provide a well-conditioned basis

which is simple to compute for boundaries that are not circular or nearly circular.
In fact, he has proved the following:
THEOREM 9.4.1. For Lagrange's functions lk(z), defined by (9.4.9),

2
cond (Mn ) S - log n + 0:, (9.4.10)
7r
where the constant 0: depends on the shape of the analytic boundary r.
Let w = fE(z) map the region Ext (r) conformally onto U* = {w : Iwl >
I} such that fE(oo) = 00 and h(zt} = 1, where Zl is an arbitrary point on
r. By analytic continuation the function h(z) can be continued to a bijective
map from D* u r onto Iwl 2: 1. The points Zk> k = 1, ... , n, are called the
Fejer points if
fE (Zk) = e2i (k-l)1r/n, k = 1, ... ,n. (9.4.11)

To determine a set of Fejer points {Zk} :=1

to be used in (9.4.9), we must
restrict f E to r. This restriction, however, leads to a modified Symm '8 integral
equation which has a unique solution, is solvable for all scalings ofthe boundary
r, and differs only in their righthand sides from Eqs (9.1.12)-(9.1.13) in the
case of exterior regions (and from Eq (9.1.6) in the case of interior regions).
For exterior regions we have the following (Reichel, 1985):
THEOREM 9.4.2. Let ¢(z) = arg {fE (z)} , and let, = cap(r). Then
the unique solution (C*, J..L*) of the modified Symm's integral equations
hlogjz-(IJ..L(()ld(l+q=o, zEr,
h J..L(() Id(1 = 1,
(9.4.12)
where J..L* = J..L* (s) E L 2 (r) and C* is a constant, satisfies
¢(z) = 21r 1~ J..L*(() Id(l, C* = -log " (9.4.13)
where integration in (9.4.13) is carried out in the positive direction.
PROOF. Let the boundary r have the parametric representation s ~ ((s),

° S s < L, ((0) = Z1, where s is the arc length of r. Then the boundary
correspondence function ¢(() may be regarded as a function of s, i.e., we write
¢(s) = ¢(((s)). It is shown in Gaier (1976) that for a rectifiable curve rand
'Y i= 1, the integral equation
h iog Iz - (I J..L(() Id(1 = 1 (9.4.14)
has a.e. a unique, integrable solution
1 1
J..L(() = 271" log 'Y ¢' (s), (= (( s), (9.4.15)
where the prime denotes differentiation with respect to s. Thus, Eq (9.4.12)

has a unique solution
J..L*(() = 2~ ¢'(s), (= ((s),

(9.4.16)
C* = -log 'Y.
It is also known (Reichel, 1984) that Eq (9.4.12) has a unique solution for any
scaling of f, and that J..L * is invariant under scaling and C* varies continuously
with scaling. Hence, (9.4.16) is also a unique solution for 1= 1. •
Let w = f(z) map the region Int (f) conformally onto the unit disk U.
Then for interior regions we have the following (Reichel, 1985):
THEOREM 9.4.3. Let ¢>j(z) = arg {f(z)} for z E f with ¢>j(Zl) = O.

Then the unique solution (C;, J..Li) of the system of modified Symm's
equations
[log Iz - (I J..Lj(() IdCI + C; = log Izl, z E r,
l J..Lj(C) IdCI = 1,
(9.4.17)
where J..Li = J..Lj(z) E L 2 (f) and C; are constants, satisfies
¢>j(z) = 271' l Zl
z
J..Lj(() IdCI, C; = 0, (9.4.18)
where the integration in (9.4.18) is taken in the positive direction ofr.
PROOF. As in the above proof, we shall use Gaier's result. Let I = cap(f),
and assume that I =j:. 1. Then
[log Iz - (I J..L(C) IdCI = log Izl, z E f, (9.4.19)
has a unique solution J..L*(() such that [J..L*(() IdCI = 1. Thus, J..Li = J..L*, and
C* = 0 is a solution of (9.4.17), and, as in the proof of the previous theorem,
this solution is unique and invariant under scaling. Hence, (9.4.18) holds for
boundaries for which I = 1. •
To approximate the mappings f(z) and fe(z) numerically, the following

algorithm is used:
1. Compute ¢>(z) and ¢>j(z) as defined in Theorem 9.4.2 and 9.4.3.
2. Determine the (n + 1) Fejer points Zk by solving the system of equations
2k7l'
¢> (Zk) = n + l' k = 0,1, ... ,n.
3. Compute the images Wk of Zk under f by evaluating
4. Determine a polynomial approximation Pk(Z) of f(z) of degree S n by

Z
.
mterpoIatmg
' - f(z)
- at the FeJer
. ,pOints
. Zk,
k = 0, 1, ... , n.
Z
Then an approximation of f(z) is given by
(9.4.20)
The accuracy of fn+! depends on that of ¢j(z), ¢(z) and the interpolation
error. Reichel (1985) has found no computational problems with polynomials
Pj (z) of degree 80-100, since the basis with Lagrange's interpolation functions
is well-conditioned.
CASE STUDY 9.4.2. Consider the region D bounded by the contour

r = {z = 2cost+i(sint+2cos3 t), 0 S t < 21l'}(seeFig. 9.4.1),
where the Fejer points are marked with dots. The following data for the error
E = Illfn(z)l- 11k is from Reichel (1985):
n Basis E
16 Monomial 3 x 10- 2
32 Monomial 3 x 10- 3
64 Lagrange 9 x 10- 6
andcond(M32 ) = 5 x 109 . •
CASE STUDY 9.4.3. Consider the ellipse r = {z = cost + ibsint, 0 S

t < 21l'}. The mapping function f (z) onto U with f (0) = 0 is known in terms
of the elliptic sine function (see Kober, 1957, p.I77)
f(z) = v'ksn (2K1l' sin' n)

1- b
, 2
which shows that the singularities of f(z) close to the boundary are the poles
at the points 1"12 = ± ~. We use a Mobius transformation fM (§2.2),
" , 1 - b2
with f M (0) = 0, such that f M (z) maps the circle of curvature through ib onto
1 - b2 /2
the unit circle. It has a pole at G = 2ib b ' Note that G = (1 + 0 (b 4 )
1- 2
as b ---. O. Hence, we approximate the function f (z) (z - (i) (z - (i) by the

polynomials defined with Lagrange's interpolation functions as the basis. The
nature and location of singularities adjacent to the boundary are studied in detail
in Chapter 12. As shown above, the use of the Mobius transformation in such
cases generally provides good approximations of the singularities by locally
approximating f by f M· •
---I---~ol---~I--- X
Fig. 9.4.1. Fig. 9.4.2.
CASE STUDY 9.4.4. Consider the region (square with round comers)
bounded by r = {z = x + i y, x 4 + y4 = I} (see Fig. 9.4.2, where 16 Fejer
points are marked). In this case the error E = Illh6(Z)1 - lllr = 6 x 10- 4
(see Reichel, 1985).•
9.5. Spline Approximations
Spline functions of various degrees can be used effectively to approximate the

source density J.l( () at corner singularities of the boundary r. In fact, the
singular functions are combined with splines and together they approximate
J.l over the entire boundary. In previous sections we have evaluated J.l and the
functions 9 and h to approximate the mapping function w = f (z) for any point
z E fJ. If f (z) is required only at the boundary points, then the function h can
be determined by simple integration of J.l as follows: Let ¢ denote the boundary
9.5. SPLINE APPROXIMATIONS 259
correspondence function defined by
1
¢(t) = arg {f (((t)) = 2i (log w -log w), z E r. (9.5.1)
Then, since Iwl = 1 for z E r, we have (see Gautschi, 1976)

1 d¢
J.L(t) = - 27l' dt = 4i7l'
1 (ow
w
_ow) .
-w (9.5.2)
at at
Let Zo be a corner of the polygonal boundary r with interior angle a7l', 0 < a <
2. Then the Schwarz-Christoffel formula (2.3.1) implies that in a neighborhood
of Zo
00
J(z) = J(zo) + L A j (z - zo)j/a . (9.5.3)
j=l
For any point z = ((t) on adjacent sides of the polygon at a corner point
Zo = ((to), we may take
t-t o, ift~to,
z - Zo = { (t _ to) e imr , ift < to. (9.5.4)
Then (9.5.2)-(9.5.4) give
L..j=l aj ( t - to )-1+j/a , t > to,

'\'00
J.L~)= . (9.5.5)
{ '\'00 (-l)j+l a. (t _ t)-1+J/a t < to,
L..J=l J 0 ,
where
1
8'{ L k A
j -
aj = 2a7l' j- k Ak }, (9.5.6)
k=l
and A o = J(zo). Hence, we conclude that
(i) If a = l/q, where q 2': 1 is an integer, then (9.5.5) has no fractional powers
of (t - to), and
(a) if q is odd, then J.L has no singularity at t = to;
(b) if q is even, then ~:~~~, in general, has a finite jump discontinuity at
t = to;
q
(ii) If _1_ < a < ~,q ~ 1 an integer, then dd J.qL is unbounded at t = to; and
q+1 q t
(iii) If 1 < a < 2 (re-entrant corner), then J.L is unbounded at t = to. For more
on corner singularities, see §12.2.
The spline approximation of J.L is generally carried out by using spline

functions of degree n for most boundaries. However, in a neighborhood of a
corner point Zo with interior angle 0:1r where J.L exhibits the behavior mentioned
in (i)-(iii) above, we use either of the two schemes: (a) continue the spline
approximation through Zo or (b) use a special function which reflects the known
asymptotic behavior of J.L near Zo to approximate J.L in the neighborhood of this
point. In the latter scheme the special functions are combined with the spline
approximation subject to appropriate continuity conditions through the point
ZOo This yields an augmented spline of degree n. The choice between scheme
(a) or (b) depends on the nature and location of the singularity of J.L and on the
degree of the splines used to approximate J.L over the rest of the boundary. Thus,
for example, scheme (b) should always be used if 1 < 0: < 2 (case (iii) above).
However, if 0: = 1/2 (case (i)(b) above) where there is a jump discontinuity
in d:' scheme (a) for splines of degree 0 or 1 and scheme (b) for splines of
degree n ~ 2 should be used.
The details of the numerical method for augmented spline approximations

are as follows (Hough and Papamichael, 1981): Assume that J.L is an L-periodic
function of the real parameter t. Let the corner point Zo, where scheme (b) is
to be used, be designated as a singular corner point, and assume that there
are N ~ 1 such points with arc lengths t = T m , m = 1, ... ,N, where
o < T1 < ... < t n < L. Let each [Tm , Tm +1] for m = 1, ... ,N be divided
into (k m + 3) intervals, m = 1, ... ,N, where km ~ 0 and TN+! = T1 + L.
The end points of these intervals are denoted by t mj , where
T m = tmo < t m 1 < ... < t m ,k=+3 = T m +1, m = 1, ... ,N. (9.5.7)
We shall also use the notation
t1,-1 = tN,kN+2 - L, t m ,-1 = t m -1,k m _l+ 2 , m = 1, ... ,N.

(9.5.8)
We shall assume that every corner point on the boundary r is an end point in the
partition (9.5.7) which may not be uniform (i.e., may have unequal arc lengths).
Then the source density J.L is approximated by
[L(t) = { Tm(t), t m ,-1 < t < t m 1,

(9.5.9)
8 m (t), t m 1 < t < t m ,k m +2,
where Tm(t) is an appropriate singular function which depends on the nature
of the boundary singularity, and 8 m (t) is a spline of degree n with knots tmj,
j = 1, ... ,km + 2. Thus, for example, if the singular corner point at T m = tmo
has an interior angle am 1r, 0 < am < 2, then the series (9.5.5) for /l- is truncated,
and we use
(9.5.10)
and
n k m +1
Sm(t) = I)mj (t - tmd + L Cmj (t - tmjt X (t - tmj) , (9.5.11)
j=O j=2
where
0, ift tmj,
~
X (t - t mj ) = { (9.5.12)
1, ift > t mj .
The total number of unknown parameters needed to compute p, is determined
N
from (9.5.9)-(9.4.10) to be equalto M o = (n+1)N+ L (k m +nm ). These
m=1
parameters are determined by the collocation method at a number of points on
r and also by subjecting p, to certain continuity conditions.
The collocation equations are formulated as follows: For splines sm(t) of

odd degree n we collocate at the end points (9.5.7) of each interval. For splines
of even degree n we collocate at the midpoints of each of these intervals. We can
always reduce the number collocation equations in the case of any symmetry
of the boundary. Thus, let Zi, i = 1, ... ,Mb denote the chosen collocation
N
points. Then from (9.5.7) we find that M 1 = 3N + L k m , and Symm's
m=1
integral equation (9.1.6) yields the following collocation equations:
(9.5.13)
which in view of (9.5.9)-(9.4.11) reduce to
N nm n k m +1
L [LAmijamj+LBmijbmj+ L CmijCmj] =-loglzil,
m=1 j=1 j=O j=2
i = 1, ... ,MI , (9.5.14)
where
(9.5.15)
The continuity conditions to be imposed on jl are based on the assumption that

the first lm derivatives of jl must be continuous at the points tm,-I and tmi
where the type of approximation changes. This leads to
(k) _ {8~~1 at t = tm,-l> where m = 2, ... , N,

rm - (k) _ _
8m at t - tml, where m - 1, ... , N, (9.5.16)
r~k) (ft,-d = 8~) (tN,kN+2), k = 1, ... , lm,
which, in view of (9.5.10) and (9.5.11) yield the continuity equations
~amj (-1+ 1m ) (-2+ 1 m

) .. ·(-k+ :m)
X (tmi - tmo)-I-k+j/O: m = k!bmk,
~amj (_I)Hk-1 (-1 +1 (-2 +1m

)
m
)'" (-k + :m)
X (t mO - t m,-I )-I-k+j/O: m
n
= r)m-I,j j(j - 1)··· (j - k + 1) (tm,-I - tm_l,j)j-k
j=O
k m -l+1
+ L Cm-I,j n(n -1)·.· (n - k +1) (tm,-I - tm_l,jt-
k
,
j=2
bOj = bNj , Coj = CNj, tOj = tNj - L,
(9.5.17)
where m = 1, ... , N, and k = 0,1, ... , lm. The total number of these
N
continuity equations is M 2 = 2 L (1 + lm). Combining the collocation
m=1
equations (9.5.14) and the continuity equations (9.5.17), we obtain a linear

system of (Ml + M 2 ) equations which is solved for the unknown parameters
amj, bmj , and emj, which in turn determines
g(z) + i h(z) = 1£ log (z - ((t)) ji,(t) dt, (9.5.18)
so that the approximate mapping function fez) can be evaluated from (9.1.1)
(with Zo = 0). In fact, substituting (9.5.10)-(9.5.12) in (9.5.18) we get
where
j
t",O
Amj(z) = (_I)m+ j (tmo - t)-Hj/a", log (z - ((t)) dt
t,n,-l
j
t m1
+ (t - tmo)-Hj/a m
log (z - ((t)) dt,
tmo
(9.5.20)
Bmij(z) = jt
t"'l
m km
• +
2
(t - tmd log (z - ((t)) dt,
j
tm 'k m +2
Cmij(z) = (t - tmjt log (z - ((t)) dt.
trn,j
Note that the coefficients in (9.1.14) are related to(9.1.19) by A mij = ~ {Amj(z)},
B mij = ~ {B mj (z)} and Cmij = ~ {Cmj (z) }. The error E is defined by
E= j=r.~~Ml III (Zj+l/2) 1-11, (9.5.21 )
where M l is the number of collocation points used, and Zj+l/2 E r denotes

the end point or midpoint of the interval depending on whether the degree n of
splines is even or odd.
Certain reduction in the number of splines occurs in some particular cases.

For example, if a corner has interior angle 1r / q, where q is an odd positive
integer, then the corner is treated as a singular corner with (i) n m = 2, when
n = 0; (ii) n m = 2, when n = 2,4 and q ~ 2; (iii) n m =.4, when n = 2,4
and q < 2; and (iv) n m = 3, when n = 1,3,5. Symmetry of any kind always
reduces the size of the linear system.
CASE STUDY 9.5.1. Consider the rectangles fa = {(X,y) : Ixl <

1,lyl < a} for a = 1,0.5,0.2,0.1. The exact mapping function is known
(see Case Study 2.3.2). Take M 1 = 128. The boundary is partitioned into
sections of uniform length on each side with 32, 20, 10, and 6 intervals on the
side x = ±1. Because of the symmetry about both coordinate axes, the total
number of equations in the linear system reduces to (17 + n) (n even) and
(18 + n) (n odd) when a = 1, and (32 + 2n) (n even) and (33 + 2n) (n odd)
when a =f:. 1. Hough and Papamichael (1981) have computed the error E as
follows:
n a=1 a = 0.5 a = 0.2 a = 0.1
°
2 x 10- 4 1X 10- 4 4 X 10- 4 3 X 10- 3
1 9 X 10- 6 2X 10- 5 2 X 10- 4 1 X 10- 3
2 4 X 10- 8 3X 10- 7 1 X 10- 5 3 X 10- 4
3 9 X 10- 9 6X 10- 8 3 X 10-6 7 X 10- 5
4 3 X 10- 9 3X 10- 9 6 X 10- 7 8 X 10- 5
5 3 X 10- 11 5X 10- 10 1 X 10- 7 2 X 10- 2
Symm 4 X 10- 4 2 X 10- 4 4 X 10- 4 3 X 10- 3

(1966)
Hayes et al. 1 X 10- 6 2 X 10- 5 5 X 10- 4 5 X 10- 3

(1972)
Global cubic 9 X 10- 5 6 X 10- 5 3 X 10-6 7 X 10- 5

spline
Their conclusions on the effectiveness of this method are as follows:

(i) For splines and augmented splines ofdegree n ::; 3 there is no ill-conditioning
effect, as is found in other polynomial basis methods, in solving the linear sys-
tem for parameters used in the approximation.
(ii) If n ~ 4, there is a possible loss of significance in computations, which
indicates that for higher degree splines the use of B-splines instead of (9.5 .11)
may be necessary if the knot spacing is not uniform.
(iii) Augmented cubic splines with three singular terms for each singular corner
provide the best results in most problems.
(iv) Every corner of a polygonal boundary must be treated like a singular cor-
ner. This makes the method easy to implement because it introduces appropriate
terms from (9.5.10) at every corner.
(v) This method can be generalized to regions with curved boundaries. In such
cases, the integrals in (9.5.20) must be evaluated by numerical quadrature, like
9.6. PROBLEMS 265
Gauss quadrature, which gives good approximations for nonsingular integrals,

but any logarithmic and fractional power singularities need special treatment.
(vi) Besides the corner singularities, any poles of f(z) that lie very close to
the boundary may also introduce inaccuracies in numerical computations (see
§12.3). In such cases a suitable choice of knot spacing which may not be equi-
spaced is recommended.
(vii) The corner singularity in narrow rectangles (a « 1) does not produce any
computational difficulty. The difficulty arises from the location of the poles of
the mapping function f(z).
9.6. Problems
PROBLEM 9.6.1. Using complex variable methods, prove that either the
h
equation
iog Iz - (I J.L(() d( = 1
h
or the equation
iog Iz - (I J.L(() d( = 0
has a solution for any Jordan contour r, where z and ( are distinct points on r.
(Muskhelishvili, 1953, p.184.)
PROBLEM 9.6.2. Consider the Lima<;on r = a - cos e,

where r = 0
corresponds to z = 1/2 for a = 1, and to z = 0 for a > 1 (Fig. 9.6.1). The
e
ray = 0 is the direction of the polar axis which is the axis of symmetry. The
e
partitions r j, j = 1, ... ,N, are taken for = 0 (1l'/ N) 21l', N = 2n. Show
that the errors in the OPN method are:
For n = 16 For n = 32
a EM EA EM EA
1.0 0.0723 0.0029 0.0408 0.0014
1.1 0.0800 0.0002 0.0215 0.0003
1.2 0.0261 0.0004 0.0048 0.0004
1.3 0.0116 0.0006 0.0019 0.0004
1.4 0.0064 0.0006 0.0011 0.0004
Note that E A « EM and the maximum error in each case occurs at the
interval point onthe polar axis when r = a - I and the boundary is concave.
For a = 1, there is a singularity at the point, but the concavity of the boundary
decreases with an increase in a , and so does the error. An increase in N results
in a decrease in error. (Symm, 1966.)
::+---+--+H:------';------;~f_+_-+._x
-3
a=2
Fig. 9.6.1. Limayons for a = 1, 1.1, 1.2, 1.5, and 2.
PROBLEM 9.6.3. Considertheellipsex 2 ja 2 +y2 = 1. Take the partitions

r j , j = 1, ... ,N for e = 0 (1rjN) 21r, N = 4n. Show that the errors in the
OPN method are as follows:
For n = 16 For n = 32
a EM EA EM EA
1.25 0.00002 0.0024 0.0001 0.0012

2.5 0.0010 0.0005 0.0002 0.0003
5 0.0079 0.0003 0.0012 0.0001
10 0.0525 0.0003 0.0083 0.0001
20 0.2135 0.0018 0.0530 0.0001
Note that E A > EM for small a, but as a increases, the error on the whole
9.6. PROBLEMS 267
increases, and EM becomes prominent. The maximum error occurs at the end
points of the minor axis where the nodes are widely spread. The error decreases
as N increases. (Symm, 1966.)
PROBLEM 9.6.4. Consider Cassini's oval
of Fig. 6.3.1. Take a = 5/6 (the near circle case), and determine the errors in
the OPN method for N = 16 and N = 32 for both interior and exterior regions.
PROBLEM 9.6.5. Consider the rectangle -1 ::; x ::; 1, -a ::; y ::; a (see
Case Study 2.3.2 and 3.3.2 and compare it with Case Study 4.5.1). Determine
the error in mapping problems for the interior and exterior regions. Note that
C= ~ (E' - k 2K') = ~ (E - k,2 K) , where E and E' are complete elliptic
integrals and k' = .Jl="k2, k being the modulus of elliptic functions (Bickley,
1932), such that E(k, ¢» =J
('/2.
o
VI - k2 2
sin ¢> d¢>
rK
= Jo dn u duo
PROBLEM 9.6.6. Consider the ellipse r : z(t) = eit (1 + >,2 e- 2it ),

where 1>'1 is taken sufficiently small for the region D to be nearly circular (see
Case Study 5.1.1). The approximate mapping function obtained by the method
of infinite systems is given by (5.2.7). Take A = 0.1, obtain the approximate
solution by the ONP method, and compare it with the solution (5.2.7).
PROBLEM 9.6.7. Develop the numerical method for evaluating Green's

functions (of §6.3) by using the spline approximations discussed in §9.5.
PROBLEM 9.6.8. It is known that the unique solution j.L of Symm's integral
equations generates an interior harmonic function
9int(Z) = hiog Iz - (I j.L(() de, zED,
and an exterior harmonic function
gext(Z) = hiog Iz - (I j.L(() de, z E D*,
each of which reduces to g(z), defined by (9.1.4), on r.

(a) Show that g~nt + 9~xt = 27rj.L, where the prime denotes the normal derivative
into the respective regions.

(b) Take gint = r sin ~ and gext = a2 cos ~ I r, where (r, ~) are polar coordi-
nates, and, by using the result of part (a), show that J1, = - cos ~ In.
(c) If g( z) is constant in D, then show that its complex conjugate function h( z),
defined by (9.1.5), is also constant in D. (Maiti, 1968.)
REFERENCES USED: Bickley (1932), Gaier (1976), Gram (1962), Gautschi

(1977, 1978, 1979), Gutknecht (1986), Hayes, Kahaner and Kellner (1972), Jawson
(1963), Hough and Papamichael (1981), Kober (1957), Kythe (1996), Maiti (1968),
Muskhelishvili (1953), Phillips (1966), P61ya and Szego (1951), Rabinowitz (1966),
Reichel (1984, 1985), Symm (1963, 1966, 1967).
Chapter 10
Airfoils
The research developed by Joukowski to determine the force exerted by a flow

on a body around which it is flowing eventually led to the theoretical foundation
for practical aircraft construction, and the methods of conformal mapping
played an important role in modern aviation. During the 1930s Theodorsen's
iterative method became a pioneer in transforming the exterior of the unit
circle onto the exterior of an almost circular contour. But in the potential flow
analysis of airfoils, James's method developed in 1971 turns out to be more
successful for all types of contours which do not have corner singularity. We
shall develop Joukowski mapping functions, compare numerical solutions of
single-element airfoils by both Theodorsen's and James's iterative methods,
unfold the mechanism of divergence of Theodorsen's method in the cases
where the image boundary is not almost circular, and finally analyze multiple-
element airfoils by using von Karman-Trefftz transformations and FFf with
Garrick's method of conjugate functions.
10.1. Joukowski Function
The Joukowski function
(10.1.1)
is a second order rational function which satisfies the condition f (z) =

f (1/ z). It means that every point of the w-plane except w = ± 1 has only
269

270 CHAPTER 10. AIRFOILS
two distinct inverse images Zl and Z2 such that
(10.1.2)
since the two points Z + 1 ~ Z2 are transformed by (10.1.1) into one and the
. 10
same pomt . th
e w-p iane,'l.e., Zl +-1 = Z2 +-,
1 an d then Zl - Z2 = -
Zl - Z2
--
Zl Z2 Zl Z2
only if Zl Z2 = 1. Thefunction (10.1.1) is analytic in Coo except at Z = 0 which
is a simple pole for this function. The derivative
f'(Z) = ~2 (1 - ~)z2
(10.1.3.)
is nonzero at all points except Z = ±1. Thus, the mapping by this function
is conformal everywhere except at Z = ±1. The regions of univalence for the
Joukowski function are U = {izi < I} and U* = {izi > I}, both of which
are mapped conformally by this function onto one and the same region in the
w-plane which is determined as follows: Consider the mapping of the circle
Izi = r by the function (10.1.1). With Z = r eiO 0 :::; 0 < 27[, we find that
1 1
u(r,O) = '2 (r+r- 1 ) cosO, v(r, 0) = '2 (r-r- 1) sinO. (10.1.4)
If we eliminate 0 in (10.1.4), we get
4u 2 4v 2
------n2 + 2 = 1. (10.1.5)
+
(r r- 1 ) (r - r- 1 )
Hence, the function (10.1.1) maps concentric circles Izi = r conformally onto
confocal ellipses with semiaxes (r ± r- 1 ) /2 and foci at ±1 on the u-axis (Fig.
10.1.1).
y v
_+-+-+-_~'--_l-+-+-_X
Fig. 10.1.1.
10.1. JOUKOWSKI FUNCTION 271
Moreover, as r --+ 1, the ellipse (10.1.5) reduces to the segment [-1, IJ of

the u-axis traversed twice. As r --+ 0, the ellipse is transformed into a circle of
infinitely large radius. Hence, the Joukowski function (l 0.1.1) maps the region
U in the z-plane conformally onto the w-plane cut along the segment [-1,1]
of the u-axis. The boundary Izi = 1 is mapped onto this segment such that
the upper semicircle is mapped onto the lower edge of the cut and the lower
semicircle onto the upper edge of the cut. The region U* in the z-plane is
mapped onto the second sheet of the w-plane cut along the segment [-1, 1J of
the u-axis, the upper semicircle Izl = 1, ~{z} > 0 onto the upper edge, and
the lower semicircle Izi = 1, ~{z} < 0 onto the lower edge of the cut. Thus,
the Joukowski function (10.1.1) maps the extended z-plane conformally onto
the Riemann surface of the inverse function
z =g(w) =w+~, (10.1.6)
which is a two-sheeted surface made up of two sheets of the w-plane cut along
the segment [-1, 1J of the real axis.
To determine the image of a rays arg{ z} 80 , we eliminate r in Eqs

(10.1.4) and replace 8 by 80 . This gives
u2 v2
-2- - -2- = 1, (10.1.7)
cos 80 sin 80
which shows that the rays arg{ z} = 80 are transformed into branches of the
hyperbola (10.1.7) with foci at ±1 (Fig. 10.1.1). The Joukowski function
defines the orthogonal system of polar coordinates in the z-plane in terms of an
orthogonal system in the w-plane such that the confocal families of ellipses and
hyperbolas in the w-plane are orthogonal. For 8 = 0, we find from (10.1.4)
that u = (r + r- 1 ) /2, v = 0, 0 S r < 1, which represents the interval
1 < u S +00. The infinite interval -00 S u < 1 is the image of the ray
8 = 1r. For 8 = 1f /2, we have u = 0, v = - (r - r- 1 ) /2, 0 S r < 1, which
represents the negative imaginary axis -00 S v < O. The positive imaginary
axis 0 < v S +00 is the image of the ray 8 = -1f /2. Thus, the horizontal
diameter of the unit disk U is mapped onto the real axis going from -1 to +1
through the point at infinity and excluding the points ±1. The vertical diameter
of U is mapped onto the entire imaginary axis including the point at infinity but
excluding the origin.
The geometrical interpretation of the mapping
w = z + Z-l (10.1.8)
is as follows: If r is a circle in the z-plane passing through the point z = -1,

such that the point z = 1 lies inside r, then the function (10.1.8) conformally
maps the region exterior to r onto the region exterior to the Joukowski profile
C (Fig. 10.1.2). The shape of the curve C is obtained from the circle r by
making the point z trace out the circle r and adding the vectors z and 1/ z.
-I
Fig. 10.1.2.
CASE STUDY 10.1.1. Consider the irrotational plane flow of an ideal

fluid around an infinite plate. Let the (x, y)-plane intersect the plate along the
segment -a :::; x :::; a and the velocity vector of the flow lie in the (x, y)-plane
and have a prescribed complex value Woo' Since the Joukowski function
z = F(() = ~ (( + Z) (10.1.9)
conformally maps the exterior of the unit circle in the (-plane onto the z-plane
cut along the segment -a :::; x :::; a, we have F( 00) = 00, and F' (00) = a/2.
Hence, the flow problem reduces to that of an irrotational flow around a circular
cylinder of unit radius in the (-plane with complex velocity Woo = ~ Woo at
infinity. The complex potential of this latter problem is given by
(10.1.10)
z+vz 2 -a2 1 z-vz 2 -a 2

If we set ( = and - = , which are obtained
a ( a
from (10.1.9) such that VZ2 - a 2 > 0 for z = x > a, and separate Woo
into real and imaginary parts (woo = (vx)oo + i (vY)oo)' where the vector
10.1. JOUKOWSKI FUNCTION 273
v 00 = (V x ) i + (V y ) j + iv by the Cauchy-
and Vx and v y are related to w = u
. . au av au av
Riemann equatIOns V x = ax = ay and v y = ay = - ox' then the complex
potential of the original problem from (10.1.10) is given by
f(z) = (vx)oo z - i (vY)oo J z2 - a 2. (10.1.11)
Now, since the force of pressure acting on an element ds of any contour C
is proportional to the hydrodynamic pressure p at the given point of flux and
is directed along the inward normal n such that -dn = j dx - i dy, we find
that the components of the force acting on the contour C are given by R =
R x + i R y , where R x = - r pdy and R
lc y = r
lc
pdx. Since p = A _ pv
2
2
from Bernoulli's integral, where A is a constant and p is the fluid density, we

obtain the pressure force with which the flow acts on the plate as
21rc v 21rc v
R=!!. 2 (dx-idy)=-!!. 2 dz, (10.1.12)
where the integral of the constant A around the contour C is zero. Since
the velocity acts tangentially to C at points of C , the complex velocity w of
the flow is related to v by w = v eiq" where ¢ is the angle that the tangent
makes with the x-axis. Then, since in view of the Cauchy-Riemann equations
_ . _ ou .ou _ ou .ov _ f'() h iq, - f'( )
w - V x + ~ v y - ox + ~ oy - ox - ~ ox - z, we ave v e - z,
and dz = e-iq, ds. Thus, v 2 dz = v 2 e- 2i q, eiq, ds = f'2(z) dz, and (10.1.12)
becomes
R = _!!.
2
r f'2(z) dz,
lc
(10.1.13)
which is known as Chaplygin's formula that expresses the force exerted by a

flow on a body around which it flows in terms of the derivative of the complex
potential.
If we use the formula

roo
+ L-'
00
n C
f(z) = Woo z + -.- logz
2~n n=O zn
which defines the complex potential in the neighborhood of the point at infinity,
where roo is the circulation of the flow at infinity, we find that
, ) roo -1 + ""cn
00
f (z = Woo z + -.- L.J - ,

2~n z n=2 zn
f ,2 () Woo roo 1 ~ bn
z = -+L.J-'
in z n=2 zn
and thus, fc f'2(z) dz = 2w oo roo. Then (1O.1.l3) yields R = p (vY)oo r 00-
p (vX)oo roo, or
(10.1.14)
known as Joukowski's theorem on lifting force, which states that the force of
pressure of an irrotational flow with velocity v 00 at infinity and flowing around
a contour C with circulation r is given by the formula R = p Ivoollrl. The
direction of the force is obtained by rotating the vector v 00 through an angle
7r/2 in the direction of the circulation.•
10.2. Generalized Joukowski Mappings
A generalized Joukowski mapping has the form

00
W = f (z) = z 'LJ
" Cj ,
zJ
(10.2.1)
j=O
where Cj may be complex and Co :F 0 (CO is usually taken as 1). Assume that the
series in (10.2.1) converges for Izi ~ a and f' (z) has no zero outside the circle
r: {izi = a}. Then the mapping (10.2.1) is conformal in the region Ext(r)
and maps the circle r onto a Jordan contour C. But if f'(z) has a simple zero
at a point Zl E r, then the contour C will have a cusp at WI = f(ZI) E C.
The inverse mapping z = F( w) = W ~ bj

LJwJ
is a single-valued function for
j=O
sufficiently large w. If we set z = ae(, a real, in (10.2.1), then the region
Ext (C) is mapped onto the semi-strip S : {~ ~ 0, 0 ::; 'f] < 27r, ( = ~ + i 'f]}
of the (-plane, and the contour corresponding to ~ = 0 will have the parametric
equations
(10.2.2)
For example, consider
W = ~
2
(a +a bz+ a(a z- b») . (10.2.3)
e
Then for z = a e i (} on the circle r we have u + i v = a cos + i b sin i.e., e,
the contour C is an ellipse with semi-axes a and b and eccentric angle e. If the
10.2. GENERALIZED JOUKOWSKI MAPPINGS 275
mapping is taken as
aZ
W = z+-,
z
(10.2.4)
which is obtained by setting b = 0 in (10.2.3), then u = a cosO, v = 0, i.e.,

the circle r is mapped onto the two sides of the straight line from (2a,O) to
(-2a,O) and back. This mapping can be written as
W + 2a = (z + a) Z , (10.2.5)
w-2a z-a
· h Yle
whlC dw = 1 - 2'
. ld s -d zZ N
ow conSl'der
z a
dw =
dz
IT (1 - Zk) ,Z
(10.2.6)
k=l
where 2::>k = 0, IZl! = a, IZk I < a if k =I- 1. The mapping (10.2.5) is a special
caseof(IO.2.6)forn = 2,Zl = a,andzz = -a. Then the contour C has a cusp
at WI, and in the neighborhood of this point we have ~: = (z - Zl) g( z), where
g(z) is regular, g(zt} =I- O. Thus, W - WI = ~ g(Zl) (z - zt}z +"', which

means that, as z traverses the circle r and passes through Zl> W approaches
WI and then recedes along a curve with the same tangent. If instead of r we
take a larger circle that passes through z = a but slightly beyond z = -a, and
transform this circle, then we have the mapping in Fig. 10.1.2 (in which a is at
x = -1), with a cusp at W = 2a and a rounded end at w a little less than - 2a.
10.2.1. Glauert's Modification. Since Joukowski airfoils are map-

pings of the type (10.2.1), they all have cusps. But an airplane wing is not a
cusp. Glauert (1948) removed this problem as follows: Consider the mapping
w-(2-n)acosf3 = (z_ae- if3 )z-n

(10.2.7)
w+(2-n)acosf3 z+ae if3 '
where nand f3 are positive and small. Set
arg {z - a e- if3 } = 01 , arg {w - (2 - n) a cos f3} = (1)1,

arg {z + a eif3 } = Oz, arg {w + (2 - n) a cos f3} = <Pz,
where all arguments are defined to be zero when z is on AB (produced) but

vary continuously as z traverses a contour outside the circle (see Fig. 10.2.1).
P'
Fig. 10.2.1.
Then at P, fit - 92 = ~ - (3, and ¢h - ¢2 = (2 - n)(9 1

2) for points
- (
outside f. The image point pI in the w-plane is on a circular arc through the
points ±(2 - n) a cos (3, subtending an angle>.. = (2 - n) (~ - (3). In the
case when P moves near B and then travels around a small semicircle about B,
then 92 increases by 1T, and in this case ¢1 - ¢2 = (2 - n) [( ~ - (3) - 1T] =
-(2 - n) (~ + (3) < O. We add 21T to this angle to make it positive. Then
the lower arc of f is mapped onto the lower circular arc in the w-plane that
+
subtends an angle J.l = 21T (<PI - ¢2) = 21T - (2 - n) (~+ (3). If J.l < 1T,
the lower arc in the w-plane is concave downward. The two circular arcs in
the w-plane intersect at an angle n1T (see Fig. 10.2.1). If instead of f, we
take a circle passing through A but a little beyond B, then we obtain a rounded
leading edge. For large z the Glauert mapping (10.2.7) can be approximated
by
(l-n)(3-n) a2
w = z + i a sin (3 + cos 2 (3 - + .... (10.2.8)
3 z
10.2.2. Symmetric Joukowski Airfoil. Let a > O. Consider the

circles
f: {lzl=a}, fl : {lz+al=a-c,-oo<c<a,c~O,~},
10.3. NEARLY CIRCULAR APPROXIMATIONS 277
(See Fig. 10.2.2; Kober, 1957, p.67). Then the mapping (10.2.4) for a> 0,
(i) maps the circle r 1 onto the symmetric airfoil r~ : A' G' F' D' A' with cusp
at A';
(ii) maps the circle r 2 and the line x = -c onto the circle r 2 : 0' B'A' C'0'
with a cusp at A' and the line u = -a as its asymptote, respectively; and
(iii) maps the region Int (r 1) bijectively, with 0 < c < a/2, onto the region
Ext (rI).
O·
v
I2 r r
y 12 Ii -2a
2a
O· -'_'--~~ _ _"'"
Fig. 10.2.2.
10.3. Nearly Circular Approximations
In practical applications airfoils are not exactly defined by the loukowski

function (10.1.1). They are often approximated by nearly circular regions.
This idea can be explained as follows: Let D denote the region bounded by two
e e
circular arcs II and 12 with end points a and b. Let (0 < < 211") denote the
angle between the arcs II and 12. The region D has a central arc Ie connecting
points a and b and bisecting the angle () (see Fig. 10.3.1 (a) if D is the interior
region and Fig. 1O.3.1(b) if D is the exterior region). The region D is mapped
onto the unit disk U by a chain of mappings 1-1 0 9 0 I (Fig 10.3.2), where
I(z) = eia z - a , (10.3.1 )

z- b
,b,
,,
,, , I
I I
I
, I
,,,
I
I
,,
,
,,
,
, , ,,
,
Fig. 10.3.1.
a is a real number chosen such that the central arc Ie is mapped onto the positive
real axis in Db and
g(z) = z7f/(). (10.3.2)
--
D
Yc f
a b
g
u·
]1 D2
0
0
Fig. 10.3.2.
10.4. JAMES'S METHOD 279
Note that the function j-l maps the right half-plane D 2 onto a region similar
to D except that the angle between the arcs now becomes 7r. Thus, j-l maps
D 2 onto U which is bounded by the unit circle through a and b and orthogonal
to the central arc. Similarly, the chain of mappings j-l 0 (-g) 0 j will map
the region D onto the region U* exterior to the unit circle. These mappings
have points a and b as fixed points. In the case of Fig. 1O.3.1(b), where D is
the exterior region, a circle slightly larger than the boundary of U and touching
U at point b is the map of the curve that looks like an airfoil (the curve on the
left in Fig. 10.3.3).
Although airfoils are shaped slightly differently than regions like D con-
sidered above, yet they can be approximated by such regions. The idea is to
map an airfoil approximately onto a nearly circular region such as D which can
then be mapped onto the unit disk.
-----
,
,
I
b ,
I
I
I
\
\, a
,
Fig. 10.3.3.
10.4. James's Method
As we have seen in §8.6, Theodorsen's method starts with the conformal map
of the the exterior of the unit circle Izi > 1 onto the exterior of a physical
boundary in the (-plane. The iterations in this method are convergent only if
the image boundary is a near circle. An analysis of two-element airfoils using
Theodorsen's method is given in Case Study 1004.1. However, the limitation
ofTheodorsen's method do not apply in James's method (James, 1971) which
will be presented below in §10.4.1. Note that every multi-element potential
flow analysis makes use of the single-element mapping methods at some stage
during the computations.
The single-element mapping deals with the problem of conformally map-

ping the exterior of the unit circle onto the region outside (or inside) a Jordan
contour f. Contours f with corners or boundary curves that do not close can
be considered by using preliminary transformations or by including singular
functions in the series for the mapping function with augmented bases, as in
the RM, BKM or GNP method. A further assumption often made during the
approximation process is that the image curve maintains its proper scale and
orientation whereas the distant regions are left undisturbed. Mappings are gen-
erally made, however, without any prior knowledge as to the scale factor or
rotation of the boundary.
10.4.1. Single-Element Airfoils. Let the mapping be represented

by a function of the form
Cl a2
z=l"+ao+-+-+'"
'> ((2 '
(10.4.1)
where z = eiO and ( = P ei ¢ denote the complex coordinates on the unit

circle and the physical plane, respectively, and aj, j = 0,1, ... , are complex
constants. Both Theodorsen (1931) and James (1971) use series closely related
to (10.4.1). Their methods can be called the method of successive conjugates
(Garrick, 1949).
THEODORSEN'S METHOD. This method uses the truncated series of the
(-(z)
form
log = bo + -bl( (+2
-b2 + ... + -
bN -
- 1
(N-l'
(10.4.2)
which is applied at equispaced points j on the unit circle (clockwise along the
perimeter). This yields
N-l
log ~ .=
()
L bk ei2 -rrjk/N. (10.4.3)
(J k=O
The terms on the left side of (10.4.3) are related to the geometric variables by
the relation
log (Z)j =logpj+i(</>j-Oj), (10.4.4)
where Pj is the radial coordinate of the point j in the (-plane and </>j and OJ
are the arguments (positive clockwise) of the points in the (- and z-plane,
respectively. The real and imaginary parts of log (z / () are conjugate harmonic
functions, so that if one is known the other can be computed efficiently by

Fourier transforms. Thus, the iterations are carried out step-by-step as fol-
lows:
STEP 1. Compute the values of Bj at the defining points in the z-plane.
STEP 2. Approximate the values of ¢>j at the points in the (-plane correspond-
ing to the equispaced points in the z-plane (¢>j = Bj is often assumed).
STEP 3. Use the curve-fit coefficients to determine the values of log Pj corre-
sponding to the estimated values of ¢>j'
STEP 4. Compute the conjugate harmonic function corresponding to the latest
values of log Pj, and use them to update the estimated values of ¢>j.
STEP 5. Repeat steps 3 and 4 until the values of ¢>j converge.
STEP 6. Determine the coefficients of the mapping function from the con-
verged data.
JAMES'S METHOD. This method uses the truncated series of the form
(10.4.5)
which is applied at equispaced points on the circle. This yields
L
N-l
log
(
dz
)
= Ck ei21rjk/ N. (10.4.6)
d( j k=O
The term on the left side is related to the geometric variables by the relation
log ( dZ)
I dz 1
d( j = log d( j + t. arg
{ dz }
d( / (10.4.7)
Sj = J{OJ Id(
dZI
dB, (10.4.8)
o
arg { -dz} =7+B·--
31l" (10.4.9)
d( j J J 2'
where S j is the arc length on the contour r in the (-plane and 7j is the angle
on that contour (for the convention, see Fig. 10.4.1). Then the iterations are
performed as follows:
STEP 1. Compute the values of sand 7 at the defining points in the z-plane,
and determine the curve-fit coefficients of 7 vs s.
STEP 2. Approximate the values of I:~ Ij at equispaced points on the circle
(I ~~ I = 1.0 is usually assumed).

j
STEP 3. Integrate (10.4.8), and obtain approximate values of Sj'

STEP 4. Use the curve-fit coefficients to determine the values ofTj correspond-
ing to the approximate values of Sj, and compute arg {~~} j from (10.4.9).
STEP 5. Compute the conjugate function to determine the values of log I~~ I j
corresponding to the latest values of arg { ~~ } j' and take the exponential to
update the values of I~~ I

j.
STEP 6. Repeat steps 3 through 5 until the values of I~~ I

j
converge.
STEP 7. Compute the coefficients of the mapping function from the converged
data.
--
1=0
1=1t
Fig. 10.4.1.
Halsey (1982) found, by comparing the numerical data and convergence

properties of both of these methods, that James's method is more suitable
than Theodorsen's for a larger class of single- and multi--element boundaries.
For example, an application of Theodorsen's method to a slightly complicated
boundary, such as a cambered ellipse, where polar coordinates are used (log p
and 4», leads to functions with multiple values. This makes numerical interpo-
lation inaccurate, if not impossible. On the contrary, in James's method the use
of intrinsic coordinates (7 and s) always yields a family of interpolated func-
tions so long as the boundary has no corner singularity. Thus, James's method
is more suited for mapping complicated boundaries. The three--element looped
boundary shown in Fig. 10.4.2 failed to be mapped by Theodorsen's method,
but James's method presented no computational difficulty. Even when the use
of polar coordinates is not appropriate, James's method still works smoothly in
many cases where Theodorsen's method fails. These include elliptic boundaries
of different ratios b/a of the minor and major axes. For example, for b/a = 0.9
both methods require 10 iterations to reduce the residuals to less than 10- 5 ;
for bja = 0.4 Theodorsen's method requires 25 iterations, but James's 15, to
reduce the residuals to 10- 5 ; and for bja = 0.3 Theodorsen's method fails to
converge, whereas James's method succeeds even down to bja = 0.005 with
only a maximum of 25 iterations.
Fig. 10.4.2. Three-element loop.
Halsey (1982) has found that both methods fail to satisfy Warschawski's
sufficient conditions (8.2.5)-(8.2.6) for convergence established in §8.2.
A detailed examination of all computed data has revealed the mechanism

for the failure of iterations to converge in Theodorsen's method. In general,
equispaced points on the circle transform into more closely spaced points on
the boundary r I~~ I < 1 and into more sparsely spaced points
for values of
on the boundary for values of I~~ I > 1. Thus, the thinner a boundary gets, the
smaller some the values of I~~ I become and the more closely spaced some of
the points become. As such, computation of the conjugate harmonic function
in Theodorsen's method is an approximation to the difference in arguments
between the corresponding points in the z- and (-plane, which leads to com-
putation of arguments in the (-plane. For boundaries where the points are very
closely spaced, small errors in the computed arguments cause a breakdown in
the ordering of the points where the argument at a point number j becomes
smaller than that at the point number j - 1. Hence, subsequent computations
contain large errors, and the iterations fail to converge. Moreover, this kind of
failure of Theodorsen's method is more likely if a larger number of points is
used. For example, if the number of points is increased to 257, the iterations
fail to converge in the case of an ellipse with bj a = 0.4, whereas with only 17
iterations the iterations do converge even with bja = 0.2. This mechanism of
failure of convergence is absent in James's method.
10.4.2. von Karman-Trefftz Transformations. These transfor-

mations are useful in analyzing potential flow over multi-element airfoils. First,
we shall consider a single-element airfoil and modify Theodorsen's method by

using a von Karman transformation and FFT. There are four basis procedural
steps in the problem of conformally mapping any airfoil onto a circle: (i) re-
move the effects of slope discontinuities in the airfoil contour, and expand the
regions of rapid flow changes (e.g., the nose region) by conformally mapping
the airfoil point-by-point onto a nearly circular contour; (ii) translate the coor-
dinate system so as to place the centroid of the nearly circular region on or near
the origin; (iii) use interpolation and obtain a continuous representation of the
nearly circular contour and thus of the airfoil; and (iv) map the nearly circular
contour onto a circle.
The analytical and computational tools to perform these four steps are as
follows. For step (i) a von Karman-Trefftz transformation is used. Let Z denote
a complex coordinate in the airfoil plane (z-plane), and ( a complex coordinate
in the nearly circular contour plane «(-plane). Then the von Karman-Trefftz
transformation is given by
z - ( = ( z - z s -/'LZl)l/l<,
(10.4.10)
z+( z+Zs-/'LZl
where Zs and Zl are complex constants, /'L = 1 - T {rr, and T is the trailing edge
included angle. This transformation is singular at
Z=Zs+/'LZl=ZTl, and Z=Zs-/'LZ/=ZNl, (10.4.11)

where the firmer point is at the trailing edge and the latter at a point midway
between the nose of the airfoil and its center of curvature (Fig. 10.4.3). Since
ZTl and Z Nl are known, we can determine Zs and Zl from (10.4.11). If the angle
T is opened up to 1r (see Fig. 10.3.2), then the airfoil is mapped into a near circle
in the (-plane where the circle (dotted) is drawn for comparison, 0 is the origin
of the coordinates and C is the approximate centroid which is the origin of the
('-plane. Since the transformation (10.4.10) contains a rational exponent, the
proper branch should be chosen by 'tracking' the transformation from a point
where it is known to the point of interest. A proper choice of the branch in
(10.4.10) implies continuity of the argument of the base (the expression within
parentheses) along a path that does not cross the boundary of the airfoil. Note
that this argument approaches zero as Z ---t 00.
In step (ii) the origin of the coordinate system is translated to the centroid
C of the nearly circular region. First an approximate centroid C is determined
by connecting adjacent points of the line segment. This translation is given by
(' = ( - C. (10.4.12)
It will be used to improve the convergence of a series expression used in step

(iv).
In step (iii) a continuous representation of the airfoil image, which has so

far been defined pointwise, is obtained. To do this, a polar coordinate system
is defined in the (-plane (Fig. 10.4.3), where log p as a function of'l/J is fitted
with a periodic cubic spline (see §9.5 for more on splines). This curve fitting
technique leads to a smooth definition of the airfoil image in the (-plane with
a high degree of accuracy.
In step (iv) the Theodorsen-Garrick transformation is used to map the near

circle in the (-plane onto a circle in the ('-plane, where instead of Theodorsen's
integral equation (8.1.6) with a cotangent kernel we shall use the Fourier series
analysis to eventually use FFT. That is how Theodorsen's method is modified.
Thus, the Theodorsen-Garrick transformation can be written as

N
(j = (exp {L (A j + iB j ) (j}. (10.4.13)
j=O
Note that the near circle in the (-plane is represented by
(10.4.14)
We shall finally map the circle in the ('-plane onto the unit circle w = eic/>.
Thus, substituting these polar representations in (10.4.14), taking logarithms
on both sides, and equating real and imaginary parts, we obtain the following
set of equations:
N
logp = Ao + L (A j cosj¢> + B j sinj¢»,
j=l (10.4.15)
N
'l/J = ¢> + B o + L (B j cosj¢> - A j sinj¢».
j=l (10.4.16)
Since log p on the near circle is known as a function of 'l/J from the periodic
cubic spline fit, the problem reduces to determining the coefficients A j and B j
by FFT and an iterative scheme, as follows: Choose 2N equispaced points on
the unit circle in the w-plane, starting at the image of the trailing edge. Thus,
"'k =
If'
(k -1)1r,
N k = 1, ... ,2N. (10.4.17)
To place the trailing edge at 4> = 0 in the w-plane, set

N
B o + L:B j = 7/Jr, (10.4.18)

j=1
where 7/Jr is the value of 7/J at the trailing edge in the ('-plane. Take B N =0
to make a closed system. Then, 7/Jk is given by
N-1
7/Jk - 4>k = 7/Jr - L: (B j + A j sinj¢k - B j cosj4>k) , (10.4.19)
j=1
which can be evaluated by the Fourier technique as follows (assuming A j and
B j are given for j = 1, ... , N - 1):
N-1
Y1 = ~ (7/Jr - L: B j ),
j=1
(10.4.20)
Yj+1 = ~ (B j + i A j ), j = 1, ... , N - 1,
Y2N-j+l = fij, j = 1, ... , N,

where the bar denotes the complex conjugate. Then
./, A,
'Pk - 'Pk -
_ ~
~ Yj exp
{i1r(j - l)(k -
N
I)} ' (10.4.21 )
j=1
which is known as a discrete Fourier transform (Cooley and Tukey, 1965; Coo-
ley, Lewis and Welch, 1970) and is evaluated by FFT technique in 0 (N log2 N)
operations for k = 1, ... , N. Note that a direct evaluation ofEq (10.4.21) takes
o (N 2 ) operations.
We can apply a similar FFT technique to solve Eq (10.4.15) and obtain

A j and B j by using a trigonometric series fit through the points (log Pk) for
k = 1, . .. , 2N as follows: Define
1 ~ { i1r(j-l)(k-l)}
Yj = 2N ~ (log ph exp - N ' (10.4.22)
k=1
where (log phis the value of log p at the k-th point given by Eq (10.4.17).
Then
A j = 2~ {Yk+l}, j = 0,1, ,N,
(10.4.23)
B j = -2~{Yj+l}' j = 1, ,N-1.
This evaluation takes 0 (N log2 N) operations.
C
Main airfoil Trailing edge
~/ ~,secondaryairfOil
ZNI ------- Zn ZN2
't2
ZT2
z-plane
Main airfoil
Trailing edge
/
Secondary airfoil
~
~-plane
Main airfoil
Trailing edge
/
···.,.~i ..
Zn ::....::.-.-.::::::::~~~~Oil
-------~ ~ ZT2
w-plane
Fig. 10.4.3. Single-element airfoil mappings.
Now, we can summarize the iterative method by the following five-step

algorithm:
1. Set A j = B j = 0 for j = 1, ... ,N - 1.
2. Evaluate'ljJj from (10.4.21) by using (10.4.20).
3. Evaluate (log p) k for k = 1, ... 1 2N. using 'ljJk and the cubic spline fit
coefficients.
4. Solve for A o• AN. and A j • B j for j = 1, ... , N - 1, using (10.4.22)-
(10.4.23).
5. Using the latest computed values of A j and B j repeat the steps 2 through 4
until the values converge.
Warschawski's sufficient conditions for convergence (§8.2) for the above

algorithm imply that
JPmax -1 < €, GIog p) I < €,

Pmin I( G,I,
If/ max
where € = 0.2954976, and the maximum and minimum values are taken on
the nearly circular contour. Note that the composite accuracy of this method
depends on the accuracy at each step. Since (10.4.20) involves trigonometric
functions, the computation of A j , B j in step 4 may not produce accurate results.
However, an error analysis in step 4 must be carried out to determine the terms
that should be taken to approximate the trigonometric functions involved in
this step. It is known (Abramovici, 1973) that about 100 terms are enough to
approximate log P in terms of the trigonometric series in 1/J. Another source of
errors lies in step 3 where a periodic cubic spline is used to interpolate log P
as a function of 1/J. These errors, though small, are due to the definition of the
airfoil only at finitely many points j which are mapped onto the circle with an
accuracy limited only by the round-off error of the computer.
10.4.3. Two-Element Airfoils. We shall use Garrick's approach

(Garrick, 1936, 1949), together with the von Karman-Trefftz transformation
and FFf, to map a two-element airfoil onto the unit circle conformally. A two-
element airfoil mapping is initially identical to a single-element airfoil mapping
except that there is a secondary airfoil which is carried through the four steps
(i)-(iv) mentioned in §lOo4.2. Then the airfoil-like shape of the secondary
airfoil is mapped onto the near circle whereas the mapping of the main airfoil
remains a circle (Fig. 1004.3). The von Karrnan-Trefftz transformation for a
simultaneous conformal mapping of the two airfoils (main and secondary) is
g(() = j(z), where
g(()
( - (T (-
= ( - (N . ( -
G
(tv' j(z) = (z - ZT2 . Z - ZT** 2) 1/1<-2, (
10.4.24 )
Z-ZN2 Z-zN2
and /'\,2 = 2 - 72lrr, 72 is the trailing edge angle for the secondary airfoil; ZN2,
ZT2, (N, and (T are complex constants, and zN2' zT2 denote the symmetric
points to ZN2, ZT2 with respect to the unit circle, i.e., ZN2 = 1/ZN2, zT2 =
1/ZT2, and (N' and (T are symmetric points to (N, and (r with respect to the
circle 1(1 = R, i.e., (N = R 2f(N, and (.r = R 2f(T (see §2.2). Moreover,
r (* = ( *)
':i!.-!:i ZT2 ZN2 1/1<2
(10.4.25)
(N (r ZN2 Zr2 '
where ZT2 and ZN2 are the points at the trailing edge and at a point midway
between the nose and the center of curvature of the secondary airfoil. From
(10.4.24) we find that ~~ ----> 1 as ( ----> 00 only if
(T + (r - (N - (N = ~ (ZT2 + zh - ZN2 - zN2) . (10.4.26)

"'2
Main airfoil (circle)
Trailing edge
/
Secondary airfoil (near circle)
1;;':' plane
Main airfoil (unit circle)
Trailing edge
/
\
ingedge
Secondary airfoil (near circle)
t-plane w-plane
Fig. 10.4.4. Two~lement airfoil mappings.
dz
Also, df ----> 00 when
I I I 1
---+--- ---=0, (10.4.27)
z - ZN2 Z - ZT2 Z - zN2 z- zN2
d(
and dg - t 00 when
1 1 1 1
- - + - - - - - - - - = O. (10.4.28)
( - (N ( - (T ( - Ci" ( - Ci"
Let the (simple) roots of Eqs (10.4.27) and (10.4.28) be denoted by ZOl, Z02
and (Ob (02, respectively. A further analysis shows that arg{J} = arg{g} at
these roots which are singular points, although the magnitudes of these roots,
in general, are different. Thus, we have
ZOm - ZT2 . ZOm - ZT.· 21 1 /K.2 = I (Om - (T . (Om - (T I

-'-------"=--, m = 1,2,
IZOm - .ZN2 ZOrn - ZN2 (Om - (N2 (Om - (N
(10.4.29)
in order that Idz/d(1 or its inverse is regular at these singular points. In this
analysis all quantities except a real R and the complex variable (N can be
computed from (10.4.24) and (10.4.25). The two unknowns Rand (N can be
computed from (10.4.25) by using (T = R 2/(T'
The above mapping transforms the secondary airfoil into a near circle and
at the same time transforms the main airfoil into a circle of radius R (w-plane,
Fig. 10.4.4). This mapping is nonsingular at all points except ZN, ZT, zN'
z~r. (N, (T, (N and (T' The near circle is mapped onto the unit circle by the
Mobius transformation
(10.4.30)
where the trailing edge image (r E of the main airfoil is mapped into w = 1 if
(rE +b _ 1
(10.4.31 )
a (rE +C - .
The image of the near circle of the secondary airfoil in the t-plane should be
mapped onto Iwl = T, T < 1, such that its center lies at the origin so that a
Fourier series used later will converge rapidly. Also the point (~ in the ('-plane
is mapped into t = 0 so that
(~+ b = O. (10.4.32)
The Mobius transformation that maps the circle ('I = R onto the circle It I = 1
with their centers at the origin is given by
1
(~ +C= O. (10.4.33)
Then the complex constants a, band c are determined as follows:

1) Set (N equal to the centroid of the near circle in the (-plane.
2) Solve (10.4.31)-(10.4.33) for a, b, and c.
3) Use (10.4.30) to transform the near circle from the ('-plane to the t-plane.
4) Approximate the centroid of the near circle, denoted by (~c' by connecting
adjacent points with straight lines and calculating the centroid of the resulting
region.
5) Compute a new value for (~ from (10.4.30), using (~c' a, b, and c.
6) Use the latest value of (~, and repeat steps 2 through 5 until convergence is
achieved.
The region infinitely far from the airfoil is mapped into the point 00 in the
t-plane. The near circle in the t-plane is determined only point-by-point, and
thus a periodic cubic spline can be used to interpolate log p on the near circle
as a function of'ljJ (as in the single-element airfoils). The mapping from the
near circle to the circle Iwl = R is carried out by the function
(10.4.34)
where R < 1 is the image of the nearly circular contour of the secondary airfoil
in the w-plane. Note that since w is purely imaginary for t = ei {3 for any real
j3, the mapping (10.4.34) maps the ('-plane onto itself in the t-plane.
10.4.4. Multi-Element Airfoils. The von Karman-Trefftz transfor-

mation for simultaneous conformal mapping of n airfoils in the z-plane is given
by g(() = f(z), where
f(z) = II
n (
z-
)l/l<-j
ZTj , (10.4.35)
j=l Z - ZNj
where "'j = 2 - 7j rrr,7j is the trailing edge included angle of the j-th air-
foil; ZTj and ZN j denote the complex coordinates of the trailing edge and the
point midway between the nose and the center of curvature of the j-th air-
foil, and (Tj and (Nj are suitably chosen complex coefficients. The condition
lim Idz/d(\ = 1 is satisfied only if
(->00
(10.4.36)
The coordinate system in the (-plane is defined by (Tl + (N1 = O. Note that
dz/ df becomes unbounded when
L -",.1
n
j=1 J
(1
Z-ZT'J
- 1 ) -0
Z-ZN'J - ,
(10.4.37)
and d(/ dg becomes unbounded when
J=1
L (1
n
(_(T J
1)
- (_ (w
J
- O. (10.4.38)
In general, there exist (2n - 2) finite roots of Eq (10.4.37), all different from
ZTj and ZNj. There are also (2n - 2) finite roots of Eq (10.4.38), all different
from (Tj and (Nj. Let these roots (which are also singularities) be denoted by
ZOj and (OJ, j = 1, ... , 2n - 2. Then dz/d( and its inverse will be finite at all
these singular points only if
g((Oj)=f(zOj), j=1, ... ,2n-2. (10.4.39)
The 2n complex quantities (Tj and (N j, j = 1, ... , n, are uniquely deter-

mined from (10.4.35)-(10.4.39) by using an n-dimensional complex Newton-
Raphson iteration scheme. For n = 1 this mapping reduces to that of the
single-element airfoils defined by (10.4.10).
10.5. Problems
PROBLEM 10.5.1. Show that the conformal mapping
lZ2 + 2mz + n
w = -;::----- (10.5.1 )
pz2 + 2qz + r
degenerates to a bilinear transformation if (lr - np)2 = 4(mr -nq)(lq - mp).

If this condition is not satisfied, then the above mapping can be reduced to the
10.5. PROBLEMS 293
following forms:
w - (3 = k (Z - f-L) 2 when q2 # pr, lq # mp;

w-o z->.
w - (3 = k (z _ f-L)2 when q2 # pr, lq # mp, polO;
w-o
(10.5.2)
w-(3=k (z-f-L)2 wh enq 2=pr,lq#mp,p#O;
z->.
k
w - (3 = (z _ >.)2 when q2 = pr, lq = mp, polO;
w - (3 = k (z - f-L)2 when p = 0, q = 0,
where 0 and (3 are two unequal roots oflz 2+2mz+n-w (pZ2 + 2qz + r) = 0,
z = >. and z = f-L are the two real values that correspond to w = a and w = (3,
respectively, and>. # f-L since 0 # (3. (Piaggio and Strain, 1947.)
PROBLEM 10.5.2. Let the circle r pass through the point z = ia, a > 0,
and let the point z = -i E Int (r). Show that the mapping (10.2.4) maps the
circle r onto a Joukowski airfoil. (Pennisi et aI., 1963, p.335.)
PROBLEM 10.5. 3. Show that underthe mapping (10.2.4) one half of the
unit circle Iwl < 1 is the image of the circle Iz - il = .j2, and the other half
that of the Iz + i I = .j2. (Pennisi et aI., 1963, p.335.)
PROBLEM 10.5.4. Under the mapping (10.2.4) with a > 0 show that
(i) two distinct points Zl, Z2 are mapped into the same point iff Zz = a2/z1'
(ii) a circle passing through the point z = -a and containing the point z = a
in its interior is mapped bijectively in the w-plane;
(iii) the circle Izi = (a + b)/2, a > 0, b > 0, is mapped onto an ellipse
u = a cos ¢, v = b sin ¢ in the w-plane with foci at w = ±2a. (Pennisi et aI.,
1963, p.335.)
PROBLEM 10.5.5. Show that under the mapping w

n +
= ( _z__1)2 ,n> 0
zn -1
an integer, the image of the sector Izi < 1, z = Izi eilJ , 0 < () < 1r In, is the
upper half-plane 8'{ w} > O. [Hint: Use the chain of mappings ( = zn and
w = (( +
(-1
1)2, in that order, to transform the sector onto the semi-circle
1(1 ::; 1 and 8'{(} 2: 0 onto 8'{ w} 2: 0.] (Pennisi et a\., 1963, pp.336-337.)
PROBLEM 10.5.6. Using the formula for density distribution 1-£(8) of a

charge on an ideally conducting circular cylinder, given by 1-£( 8) = ~ Idz 1-1 ,
211' d( 1(=1
where e denotes the charge per unit length of the cylinder, show that the charge
density in a strip of width 2a in the (x, y)-plane along the segment -a < x < a
=~
is given by I-£(x)
211' va 1
2 -
nikov and Tikhonov, 1978, p.216.)
x2
. [Hint: use the mapping (10.1.9).] (Svesh-
PROBLEM 10.5.7. Apply Theodorsen's and James's methods to approx-

imately transform the unit circle conformally onto the one-element boundary
of
(a) teardrop wing profile of NACAOOlO shown in Fig. lO.5.l(a).
(b) a composite elliptic wing profile composed of upper and lower curves from
two ellipses with the same major axis but a different minor axis (Fig. 10.5.I(b)).
Fig. lO.5.1.(a), (b).
REFERENCES USED: Abramovici (1973), Andersen et al. (1962), Caratheodory

(1969), Carrier, Krook and Pearson (1966), Cooley and Tukey (1965), Cooley, Lewis
and Welch (1970), Gaier (1964), Garrick (1936, 1949), Glauert (1948), Halsey (1979,
1982), Ives (1976), James (1971), Kantorovich and Krylov (1958), Kober (1957), Pi-
aggio and Strain (1947), Pennisi et al. (1963), Phillips (1943, 1966), Theodorsen
(1931), von Karman and Trefftz (1918), Warschawski (1945).
Chapter 11
Doubly Connected Regions
Some well-known numerical methods for approximating conformal mapping

of doubly regions onto an annulus or the unit disk are presented. There is
a definite need for a simple yet accurate method for mapping a general dou-
bly connected region onto a circular annulus. According to Kantorovich and
Krylov (1958, p. 362) the problem of finding the conformal modulus is 'one
of the difficult problems of the theory of conformal transformation'. As such,
analytic solutions have been determined for a very restricted class of doubly
connected regions, like those mentioned in Muskhelishvili (1963, §48). Nu-
merical solutions are also confined to a limited class of regions where either
one boundary is circular or axisymmetric. Most common methods use inte-
gral equations, iterations, polynomial approximations, and kernels. We shall
develop Symm's integral equations and the related orthonormal polynomial
method. A dipole formulation that leads to the method of reduction of con-
nectivity shall be presented. Another useful method for multiply connected
regions, based on Mikhlin's integral equation, that also works for simply and
doubly connected regions as well will be discussed in Chapter 13.
11.1. Conformal Modulus
Let f 1 and f o denote two Jordan contours such that f 1 C Int(f o) and 0 E
Int (f 1)' Let n denote the doubly connected region
295

296 CHAPTER 11. DOUBLY CONNECTED REGIONS
The conformal mapping problem is to determine the function w = In (z) which

maps n univalentiy onto an annulus A(Pl,P2) = {w : Pl < Iwl < P2}, where
Pl,2 are real numbers such that 0 < Pl < P2. Thus, l and o are mapped r r
conformally onto two circles Iwl = Pl and Iwl = P2, respectively. The ratio
M = pd Pl > 1 is called the conformal modulus of n. Let the parametric
equation of the boundary an r r
= l U o be z = -y(s), 0 ~ s ~ L, so that l = r
{-y(s), : 0 ~ S ~ i} and r o = {-y(s) : i ~ s ~ L}. For simplicity, we shall
write I instead of In hereafter, and use the notation -y(0) = -y(0+) = -y(l-),
and -y(l) = -y( i+) = -y( L-) = -y( L). Then the boundary correspondence
function <1> for the function I is defined by
<1>(s) = arg {f (r(s))}, (11.1.1)
where
I (r(s)) = p(s) ei and p(s) ={ Pl,

o ~ s ~ i,
4>(s), (11.1.2)
P2, i ~ s ~ L.
y V
b
It
w= logz aI2
X ~
U
r2 0 logr. logr2
aI2
-It
b
z-plane w-plane
Fig. 11.1.1.
11.1.1. Conformal Invariants. If an annular region A (Tl, T2), Tl <

Izi < T2, is mapped conformally onto a parallel strip {log Tl < U < log Td
by the function w = log z, then the annulus corresponds to an enumerable
set of congruent rectangles, all with sides a and b such that b = log T2, and
Tl
a = 27r. Since every doubly connected region can be mapped conformally
onto an annulus of the type shown in Fig. 11.1.1 and since two annuli with
different ratios TdTl cannot be mapped conformally onto each other, the set of
all doubly connected regions falls into classes of conformally equivalent regions
11.1. CONFORMAL MODULUS 297
where every class is characterized by the ratio r2/rl of the radii belonging to
that class (see §1.4). Hence, the ratio r2/rl, known as the conformal invariant,
is related to the ratio a/ b of the sides of the rectangles by
b 1 r2 r2
- =- log - , or - = e2nb / a . (11.1.3)
a 271" rl rl
The linear transformation
i-Vkw 1 (-1
(= 1 + Vkw' or w = iVk (+ l' (11.1.4)
where k is defined by (2.3.11), maps the half-plane ~ {w} 2': 0 onto the circular
region 1 (I :S 1 such thatthefour points corresponding to the points w = ± 1, ± ~
are the vertices of a rectangle whose center is at ( = 0 (Fig. 11.1.2). Then the
angle 7/J between the diagonals of the rectangle is given by
7/J 2Vk
tan - = - - . (11.1.5)
2 1- k
Combining the mapping (11.1.4) and the mapping (2.3.13) of the upper half-
plane ~ { z} > 0 onto the rectangle, we obtain a conformal mapping of the
rectangle onto the disk 1(1 < 1. Thus, the angle 7/J is another conformal
invariant for doubly connetced regions. A table of complete elliptic functions
of the first kind for k from 0 to 1 and of k1, K(k), K(kt}, alb, r2/rl and 7/J
(see Fig. 11.1.2) is available in Andersen et al., 1962, pp.165-166 (also see
§2.3).
-11k o I
D A
w-plane
/z=sn(W,k) ~-plane
A3 A2
-o+i b r " - - - - - t - - - - - - - , ' o+i b
-0 o o
z-plane
Fig. 11.1.2.
11.1.2. Area Theorem. The star of a region D is defined as the

largest region contained in D that is starlike with respect to z = 0 (see §8.5).
Corresponding to the annulus A(P1, P2) in the w-plane, there exists a doubly
connected region D in the z-plane, with r 1 and r 0 as the inner and outer
boundaries such that r 1 lies outside the disk Iz I < r (1 - /5r), where br --+ 0
as r --+ O. Let us partition the part of the star of D lying outside r 1 by rays
emanating from z = 0 such that (i) the rays are mapped onto themselves by
a rotation about z = 0 through an angle 2k7r In, k = 1, ... ,n; and (ii) the
variations in Iz I on the portion of r 1 between two consecutive rays (excluding
the rays themselves) is less than a preassigned quantity E: > O. Since r 1
is an analytic curve, such a system of rays exists. There are nm such rays,
where m is an integer. We denote them in order by l1> l2, . .. , lnm, where
lmk+q is obtained from lq by a rotation through an angle 2k1f In. Let rk, k =
1,2, ... ,nm, denote the largest distance from z = 0 to the part G k of the star
lying between the rays lk and lk+l and outside r 1 for k = 1, 2, ... ,nm, where
Gnm+l = G 1 (Fig. 11.1.3). The function ( = log (~) transforms this system
of regions Gk into a system of regions Hk lying, respectively, in rectangles
C:)
nm
with sides of length log and O:k, where L O:k = 21f and O:mk+j =1= O:j'
k=1
~= log~
r
-+---------'--,,---~
o r
log~
Fig. 11.1.3
If the region H k is mapped onto a rectangle of sides ak and bk such that the
boundary segments of Hk are mapped into the side of length bk, then
ak O:k
->
- - rk'
- (11.1.6)
bk log-
r
which yields
nm m n-1 m n-1 1
""' ak > ""' ""'
L.J bk - L.J L.J
k=1 j=1 k=O
O:mk+j
rmk+j =
log - -
L L
j=1
O:j
k=O
-I-"'rm-:-k-+-j .
og - -
r r
Since the system of regions G k is the image of a system of strips contained in

the annulus PI < jwl < P2 under the mapping w = f(z), we have
2w m . n-l 1
- p i 2:
log -
I:
j=1
exj . mm
J
I: log - -
k=O
rmk+j
PI r
(11.1. 7)
2w
. n-l 1
= - . mm '""
n j L..- I rmk+j .
k=O o g - -
r
Using the inequality.!.

n
t
k=1
Cn > Ck 2: 0, twice, we find from
(11.1.7) that
n-l 1 n n2
I: Iog - - >
k=O
rmk+j -.,r=n=n=-=I=I==r=m=k=+=J·
,. og
2: ""n-l
LJk=1 1og rmk+j'
-r-
r k=1 -r-
and hence, (11.1.7) yields
n-l
n log -P2 ~ max '"" rmk+j
L..-1og - -,
PI J k=1 r
or
II
n n-l
M n = P2 :::; max rmk+j, (11.1.8)
( )
PI J k=O r
where M is the conformal modulus of the region A (PI, P2).
Antonjuk (1958) has proved the following theorem for functions that are
regular in an annulus:
THEOREM 11.1.1. Suppose that the function w = f(z) is regular

in the annulus A(I, M) = {I < Izi < M and satisfies the conditions
If(z) 2: 1 and ~
2zw c
1 f'((z)) dz 2: 1, where C is a contour in A(I, M)
f z
which is not homologous to z = 0. Let Aj denote the star of a finite
doubly connected Riemann surface AI onto which the annulus A(I, m)
is mapped by the function w = f(z) with respect to the system of rays

emanating from the point w = O. Then
(11.1.9)
where P* is the area of the star Aj and p* the area of the preimage of
Aj. The equality in (11.1. g) holds for functions of the form f(z) = cz,
lei = 1.
CASE STUDY 11.1.1. The region A(p, 1) = {p < 1(1 < I} is mapped
conformally onto the unit disk It I < 1, slit from - L to +L (Fig. 11.1.4) by the
function
2iK (
p
t=Lsn(-:;;:-log +K,k), k=L 2 , K=K(k), (11.1.10)
s-plane t-plane
w-plane z-plane
Fig. 11.1.4.
(Nehari, 1952, pp.293-295). Let
( = p/w, t ="2L (z + z -1) , Izl = 1. (11.1.11)

Then the circle ItI = 1 is mapped onto the boundary f o in the z-plane by the
mapping (11.1.11), where f o is in polar coordinates defined by
. /~
r = r(B) = Vp . /~
+ sin2 B- Vp - cos 2 B. (11.1.12)
For example, if we take k = L 2 = sin 46° , as in Gaier (1964, p.222), then the
values of r n = r (~;) for n = 0, 1,2, ... ,9 are given in the following table:
n rn n rn
0 0.554435 5 0.417632
1 0.543464 6 0.395150
2 0.515529 7 0.379357
3 0.480595 8 0.370041
4 0.446599 9 0.366967
We find that
(i) The modulus M of the region Do is given by (Nehari, 1952, p.294)
1 {1rK'(k)} 1
M = P = exp '4 K(k) = y'q(k) >::! 2.166187,
for k = sin 46° (for the function sn and q, see §7.4).

(ii) Let w = ei ¢ be the map of z = eia , 0: = o:(c/». Then
(I--d t=Lsn((1+2c/>/1r)K,k), (11.1.13)

Zl--->t t=Lcoso:, (11.1.14)
where
coso: = sn( (1 + 2c/>/1r) K, k) = sn( (1 - 2c/>/1r) K, k).
Set 0:' = 1r/2-0:. Then (1-2c/>/1r)K = F(o:',k), where Fisthehy-

pergeometric function. In particular, let c/>n = ~;, n = 1,2, ... ,8, and set
r = 90-10 n. Then F(o:', k) = rK/90, r = 80,70, ... ,10, and k = sin 46° .
This yields the values of o:~, and hence, of O:n'
(iii) Let w = p ei ¢ map the z-plane onto the w-plane (Fig. 11.1.4). We shall
determine (3 = (3(c/» - arg{z}. First, we use the mapping
2K log M
( I---> t : t = L sn (K + iK + v, k) , K= v= 2K c/>.
1r
(11.1.15)
Since t = ei'l/J maps tonto It I = 1, we have 1 = L sn (K + i K, k) for cP =I- O.

Set u = K + i 1>,. Then sn u = 1/ L, cn u = Jl - 1/£2 which is purely
imaginary, and dn u = v'1 - L2 which is real. Also, sn v, cn v, and dn v are
real. Hence, from (11.1.15), by equating real parts, we get
cos'I/J = cn(v,k)dn(v,k)2' (11.1.16)

1 - £2 [sn (v, k)]
where v = V n = nK/9, n = 1,2, ... ,8. Since, in view of (11.1.14), the

mapping z t----+ t gives
cos'I/J = ~ (r + ~) cos f3, sin'I/J = ~ (r - ~) sin f3,
we eliminate r, use (11.1.16), and obtain
(11.1.17)
If we set V n = nK/9, n = 1,2, ... , then we determine the values of f3n =

f3( cPn) from the table given above.•
CASE STUDY 11.1.2 (Dirichlet problem for the annulus). Let two real-
valued, 27r-periodic and continuous functions Ul (B) and U2 (B) be defined on the
boundary of the annulus A(rl, r2) = {rl < Izi < r2}. The Dirichlet problem
for this region deals with determining a function u(r, B) which is continuous
in the closed region A(rI, r2) U f 1 U fa = {rl ::; Izl ::; r2}, harmonic in
A(rl' r2), and takes the boundary value Ul (B) on f 1 for z = rl ei II and U2( B)
on fa for z = r2 eill . The harmonic function v(r, B), conjugate to u(r, B), is in
general not single-valued, and thus the function f(z) = u + i v will have two
summands. One is a single-valued function that can be expanded in a Laurent
series for the annulus, and the other is log z with a real coefficient A. Thus,
L
00
u +i v = 'Yn zn + A log z, 'Yn = an + i f3n, (11.1.18)

n=-oo
which, after separating into real and imaginary parts, yields
+L
00
U = ao [(anr n + a_nr- n ) cos nO - (f3n rn - f3_n r - n ) sinnOJ

n=l
+A log r,
130 + L [(f3n rn + f3_n r - n )

00
v= cos nO + (anr n - a_nr- n ) sinnOJ + O.

n=l
(11.1.19)
In the first equation in (11.1.19) we set r =
and r = r2. This gives usrl
the Fourier series expansions for the functions Ul(O) and U2(O), respectively,
where the Fourier coefficients are given by
a6 l
) = ao +A j7f Ul(O) dfJ,
-7f
log rl = -
1
21r
2 1 j7f
a6 ) = ao + A log r2 = - U2(O) de,
21r -7f
1
a~l) = an rl + a_ n r i n = -2 j7f Ul(e) cos nO de,
-7f 1r
an(2) = a n r2n +a_ n r 2-n = 21 j7f U2 (e) cosn e de ,
1r -7f
b~l) = f3-n r i n - f3n rr = -
1 j7f Ul (e) sin nO de,
21r -7f
1 j7f
b~2) = f3-n ri n - f3n r2 = - U2(e) sin ne de.
21r -7f
Hence
(2) (1)
A- ao -an
- log M '
(1) -n (2)-n
an r 2 - an r l
an = --n-=---n---n-n~'
r
rl 2 - rl 2 r
-- bn r l-n
(2)
-
b(l)-n
n r2
13n rrr 2n -
rl n r2
,
(11.1.20)
CASE STUDY 11.1.3 (Neumann problem for the annulus). Let F l (e) and
F 2 (e) denote the normal derivatives of the harmonic function u(r, e) on the
boundaries f l = {izi = rIl and f o = {izi = r2} of the annulus A(rl, r2).
Then the function f(z) = u + i v has the same representation as in (11.1.18),

where separating real and imaginary parts and satisfying the Neumann condi-
tions on the boundaries r l and r o, respectively, we get
L n[ (an r~-l
00
F1(B) = - a- n r~-l) cosnB

n=l
(11.1.21)
L n[ (an r~-l
00
Fz(B) = - - a_ n r~-l) cosnB

n=l
- (fJn rz
n-l
- fJ-n rzn -)lsmnB]
.
- -A.
rz
The coefficient A is determined in two ways which are equal:
(11.2.22)
The other coefficients in the Fourier series expansions of F1 (B) and Fz (B) are
given by
(11.1.23)
Thus, after solving (11.1.23), we find that the coefficients a±n and fJ±n in the
series (11.1.21) are given by
11.2. SOURCE DENSITY 305
11.2. Source Density
Let w = inez) map conformally a finite, doubly connected region n, bounded

by two Jordan contours f 1 and f o, f 1 C f o, onto the annulus A(p, 1), with
M = 1/ p, p < 1, where p is initially unknown and is to be determined. Let us
assume that the origin lies inside the region Int (f 1). Then the function in (z),
which is uniquely determined except for a rotation, can be represented in the
form (as in §9.1)
inez) = e10g z+g(z)+ih(z), (11.2.1)
where 9 and h are conjugate harmonic functions in n such that
for z E f l ,
lin(z)1 = { p, (11.2.2)
1, for z E fo.
Thus,
inez) = e10g Izl+g(x,y)+i [arg{z}+h(x,y)], Z = x + i y, (11.2.3)
where the boundary conditions (11.2.2) become
log p - log Iz I, z E f 1,
g(x,y) = { (11.2.4)
-log Izl, z E fo·
As in §9.1, we shall represent g(x, y) as a single-layer logarithmic potential
g(x, y) = 1
r=r,+ro
log Iz - (I !J-(() Id(l, (11.2.5)
where !J-(() is the source density on f. Also,
h(x,y) = [arg{z - (} !J-(() Id(1 + q, (11.2.6)
where q is an arbitrary constant corresponding to an arbitrary rotation in the

mapping function in (z) defined by (11.2.1). Then the boundary conditions
(11.2.4) become
r log Iz _ (I !J-(() Id(1 = { log p -log Izl, z E f l

, (11.2.7)
Jr -log Izl, z E f o,
and the condition (9.1.7) on the single-valuedness of h reduces to
(11.2.8)
as in (9.1.13). Note that Eqs (11.2.7) and (11.2.8), known as Symm's integral
equations, are coupled equations for J.L(() and p and possess a unique solution
(see lawson, 1963). Once J.L(() is determined, the functions 9 and h can be
computed from (11.2.5) and (11.2.6), respectively, and hence, the mapping
function fn(z) from (11.2.1).
The numerical computation of J.L((), g, h, and fn is carried out by the

ONP method as in §9.2, i.e., by partitioning the boundary f into N sections
GI, ... ,GN and approximating J.L(() by J.Lj which is constant over each G j ,
j = 1, ,N. Then Eq (11.2.7) is computed at each node Zk = Xk + i Yk,
j = 1, ,N, together with Eq (11.2.8). This yields a system of (N + 1)
linear equations in (N + 1) unknowns J.LI, ... ,J.LN and log p. The solution of
this system leads to the approximate value of p. Then the functions 9 and h
are approximated by finite terms corresponding to the integrals in (11.2.5) and
(11.2.6), and the mapping function fn(z) is finally obtained from (11.2.1). The
method follows the same set of steps as in §9.2.
Let fn(z) and .0 denote the numerical approximations to fn(z) and p. Then
1.0 - pi «max
z
Ifn(z) - fn(z)l, (11.2.9)
and, as Symm (1969) has noted, .0 is more accurate than fn(z). In fact, by
the maximum modulus theorem, Ifn(z) - fn(z)j takes its maximum value
somewhere on the boundary f = f l U fa, and, as in §9.2, this maximum value
I.
rarely exceeds 2 max Ifn(z) - fn(z) But since
z
' () ()/ {Ip-pl, atthenodesoff

Ifn z - fn z = 0, at the nodes offo,
l ,
then, in view of (11.2.9), we should compute Ifn(z)1 at some point z = Zj

between the nodes (which may be end points of G j ). Thus, the point Zj is
called the internodal point for G j . The error E in Ifn(z)1 then is given by
E = max {max Ilfn(z)j-

Zj ZErl
pi, max Ilfn(z)l- II}·
zEro
(11.2.10)
11.2. SOURCE DENSITY 307
If the doubly connected regions are symmetric about one or both coordinate
axes, then the total number of equations to be solved reduces from (N + 1) to
(N /2 + 1) or (N/4 + 1), respectively. We shall denote the approximate values
of u, v, and w = u + i v by il, f), and w,
respectively.
CASE STUDY 11.2.1. Consider a pair ofLima~ons
rl = {x=al cost+b l cos2t,y=al sint+b l sin2t, al >0, bl >O},

ro= {x = a2 cost + b2 cos2t,y = a2 sint + b2 sin2t, a2 > 0, b2 > O},
(see Fig. 11.2.1 with al = 5, a2 = 10, b2 = 3, and bl = b2/4) where

t = 0 (271"/N) 271" defines the distribution of nodes on each boundary and N =
2(n-1). Thevaluesofal,a2,bl,b2arechosensuchthatbI/b2 = (aI/a2)2
which ensures that the function
which maps r o onto the unit circle (see Muskhelishvili, 1963, §48, who has
determined analytic solutions for some doubly connected regions like Pascal's
lima~ons, epitrochoids, hypotrochoids, and elliptic rings ), also maps r 1 onto a
concentric circle of radius p = aI/a2, where M = 1/ p. Because of symmetry
about the x-axis, we take t = 0(71"/10) 71". The values of il and f) are given
below in Table 1.
y
10
~--";+--_-.t-- _ _""""-+_ _--:7_-+X

-7 -5 13
-10
Fig. 11.2.1.
Symm (1969) has taken al = 5, a2 = 10, b2 = 3 and bl = b2/4 and has shown
that the error E increases as b2 increases and that the boundary r 0 gradually
changes from a circle (b 2 = 0) to a cardioid (b = 5). In each case E decreases
as N increases. However, p varies very little, which indicates that even a crude
partition of r is sufficient for a good approximation of p. •
Table 1. Values of uand v.
x y U v
13.0 0.0 1.00000 -0.00006
11.9376 4.95353 0.95106 0.30902
9.01722 8.73102 0.80902 0.58778
4.95080 10.9433 0.58779 0.80902
0.66316 11.2739 0.30902 0.95106
-3.0 10.0 0.00002 1.00000
-5.51722 7.7472 -0.30902 0.95106
-6.8049 5.237 -0.58779 0.80902
-7.16312 3.02468 -0.80912 0.58779
-7.08351 1.32681 -0.95106 0.30902
-7.0 0.0 -0.99999 0.00002
5.75 0.0 0.49999 -0.00001

5.36205 1.98592 0.47552 0.15451
4.27685 3.65222 0.40452 0.29389
2.70716 4.75838 0.29389 0.40542
0.938322 5.19612 0.15451 0.47553
-0.75 5.0 0.00002 0.49999
-2.15185 4.31444 -0.15452 0.47553
-3.17069 3.33179 -0.29389 0.40452
-3.81322 2.22563 -0.36698 0.42726
-4.14852 1.10425 -0.47554 0.15452
-4.25 0.0 -0.49999 0.00003
11.3. Dipole Distribution
It is known that in a simply connected region D bounded by a Jordan contour

r: {z = I'(s)}, where s denotes the arclength along r, 0 ~ s ~ L, the dipole
11.3. DIPOLE DISTRIBUTION 309
distribution density /-l(8) satisfies the integral equation
/-l(8) = ~71" [9(8) - (L /-l(t) aa

Jo nt
(log~)
rtz
dt] (11.3.1)
where nt is the inward normal at a point ( = l'(t) E r, a ::; t ::; L, 8 "I t,

and rtz = Iz - (I, zED, and 9(8) denotes the boundary value of the potential
function u(z) on r (see Case Study 7.4.3, and Problem 7.6.4). The following
result is useful in numerical evaluation of conformal mapping: If the density
/-l(t) can be determined by solving Eq (11.3.1), then the Dirichlet problem and
hence the problem of conformal mapping is reduced to quadratures.
i;-plane
z-plane
Fig. 11.3.1.
This dipole distribution formulation can be used for a doubly connected

region 0 in the (-plane bounded by two Jordan contours r l and r o, r l C r o,
and a E Int (r1), by transforming the region 0 into a simply connected region
by the function z = log «( -
(0) which transforms 0 into an irregular strip 0'
with period 2i7l". Then the region 0' is further transformed into an irregular
circlelike region C in the w-plane by the function w = tan 71" z , where a is the
2a
meanwidth of 0'. Note that the region C may have infinitely many extrema
Chumps') in the neighborhood of the two points corresponding to ±oo (Fig.
11.3.1). Thus, the boundary problem for the region 0 reduces to a boundary
problem for the strip 0', where the boundary values are 2i7l"-periodic. Without
loss of generality, we shall consider the general case of the period ib, where
b need not be 271". Let the equations of the two boundaries G I and G 2 of the
strip 0' be G 1 : Z = "I (81) and G 2 : Z = "I (82), where 81 and 82 are the arc
lengths on G 1 and G 2 , respectively. If one period covers the arc lengths £1 and
£2 such that "I (81 + £d = "I (81) + ib, and "I (82 + £2) = "I (82) + ib, then
the kernel in Eq (11.3.1) becomes
_8 (log _1 ) = !R { _8 (log ----:--:---1--,--,-) }

8nt rtz ant "1(8) - 'Y(t)
!R{ 1 8'Y}!R{ i 8'Y}
= 'Y(s) - 'Y(t) ant = 'Y(s) - 'Y(t) 8t
= -~ {'Y(s) ~ 'Y(t) 'Y'(t)},

(11.3.2)
where n is the inward normal, z = 'Y(s) En', 'Y(t) E G 1,2, and the potential
at a point zEn' is given by
1
00 1
u(z) =
-00
~ { "I ()
s -
( ) 'Y~ (t2) } fJ,2 (t2) dt2
"12 t2
(11.3.3)
1
00 1
-
-00
~ { "I ()s - "II () t1
'Y~ (td} fJ,1 (td dt1,
where the parameters t1 and t2 run from -00 to +00 as z traverses from -ioo to
+ioo. Let us assume that the dipole densities fJ,1 (t1) and fJ,2 (t2) are periodic,
i.e., fJ,1 (t1 + £1) = fJ,1 (t 1), and fJ,2 (t2 + £2) = fJ,2 (t2). Then the integrals in
(11.3.3) can be written as sum of integrals, and we have
u(z) = L
00 lL2 8' { ()
1
( ) . b 'Y~ (t2) } dt2
iLl ~ { ()
n=-oo 0 "I S - "12 t2 - m
(11.3.4)
1
- L "I S - "II ( 1t )- ·m b 'Y~
00
o (td} dt1·
n=-oo
Note that the two integrals in (11.3.3) and the two sums in (11.3.4) are conver-
gent in the Cauchy sense. Since the series in (11.3.4) are uniformly convergent,
we can interchange the integration and summation. Using the formula (1.2.15)
we find that
00 1 1
L
n=-oo
'Y(s) - 'Y(t) - inb = "1(8) - 'Y(t)
+ 2ib
1 LOO{ 'Y(s) - 'Y(t)
II }
+ 'Y(s) - 'Y(t)
n=l - n +n
ib ib
_ !!.- ('Y(S) - 'Y(t))
- ib cot 7f ib '
and (11.3.4) becomes
u(z) = 1f Jo
fL2 ~ ib {I cot
(I'(8)-I'2(t 2) ) ,
1f ib 1'2 (t2)
}
J.12 (t2) dt2
- 1f JfL'o{l
o
:s ib cot
(,(8)-I'I(td),
ib 1f
}
II (td J.11 (tl) dtl·
(11.3.5)
Let the point z E 0' approach a point Zs on the boundary of 0'. Then, in view
of Problem 7.6.4, if g(8) = u+ (zs) is the prescribed boundary value, then
lim u(z) = U+ (8 s ) = U(zs) + 1f g(8), and the integral equation for p,(8) is
z-+zs
1
p,(8) = ;: g(8) + JfL
o
,
{I
ib ~ cot
(1'(8) - 1'1 (td)
(tl) J.11
1f dtl
ib I'~
}
(td
f {I
- Jo
L
2
~
(I'(8)-I'2(t 2) ) ,
ib cot 1f (t2)
}
(t2) dt2'
ib 1'2 P,2
(11.3.6)
This integral equation can be solved by an iterative method, e.g., the one in §7.4.
Although this method of reduction of connectivity seems especially suitable
for those doubly connected regions that can be easily transformed into parallel
strips, yet no numerical study has been done for it. In some special cases,
however, the analysis becomes simpler, and it is presented in the following two
case studies.
CASE STUDY 11.3.1 (Andersen et aI., 1962). Consider the symmetric

doubly connected region 0 which is transformed into the unit disk (or an an-
nulus) by a chain of conformal maps fI, fz, 13, 14, Is,!6 and h, as shown in
Fig. 11.3.2.
Note that in the z-plane Zo E r 0 is chosen as a point on the axis of symmetry.

The mapping goes from the z-plane through the zl-plane, z2-plane, z3-plane,
z4-plane and finally to the unit disk in the w-plane. The conformal maps are
as follows:
1fZl
fI : ZI = log (z - zo) ; fz: Z2 = tan - ;
2a
13 : Z3 = (q + 1)z2 - 1 + q;
(q - 1)Z2 + 1 + q
1 Z3 - 1 Z4 - i
14: Z 4 = - - - - - ; 15: w=--.;
i z3 Jk
1 + Z4 +t
16 : Z6 =
t
Jo
4
dt
J(l _ t 2) (1 _ k 2t 2) ; h: Z7 = exp
{21fZ4 }
--:;;;- .
d~----4C
z-plane a
e 0
d
a
w-plane
Fig. 11.3.2.
The five points marked as a, b, c, d, and e are traced through all of these
mappings. The maps II and h are used in Fig. 11.3.1; the map h carries the
points a and b into the points e±iQ and the points c and d into the points e±i/3,
respectively, on the unit disk C 1, which are mapped by h into the end points of
two diameters of the unit disk C 2 , where the angle 'I/J between these diameters
is given by (11.1.5) and k = :::~~~~~ (see §11.1.1); map f4 carries the unit
disk C2 into the upper half-plane H, and finally f5 maps the upper half-plane
H onto the unit disk in the w-plane. Note in passing that f6 maps the upper
half-plane H onto the rectangle R which is mapped onto an annulus by h.
Conversely, by using the maps fi 1 , fi 1 , and nin that order, an annulus can
be mapped onto the unit disk.
In this example the numerical computations are needed only for the mapping
h (n' f-+ Cd; other mappings are straightforward. The relation between the
mapping h (n' f-+ C 1 ) is provided by Gershgorin's integral equation (7.2.6)

and the relation between their interior points by the Poisson integral (6.4.6).
The mapping f6 (H f-+ R) is defined by the elliptic integral (2.3.13).•
CASE STUDY 11.3.2 (Andersen et aI., 1962). In the nonsymmetric case

the the horizontal lines in the z-plane need not go into straight horizontal lines
in the w-plane, as shown in Fig. 11.3.3. We shall apply the above method
of reduction of connectivity (from 2 to 1) to a periodic irregular strip n' onto
a parallel strip S of width 71'/2. Then we can use the chain mapping of Case
Study 11.3.1 to obtain the conformal mapping of a doubly connected region
onto a unit disk or an annulus. This method involves the following steps:
1. Compute Green's function for the parallel strip S. This Green's function
is not the usual Green's function with a logarithmic singularity; besides, it is
Green's function with some unspecified period iq.
2. Use this Green's function to derive the integral equation for the boundary
correspondence between the boundaries of the strip n' and S.
iq
w =f(z)
(-It/4,O) (It/4,O)
- iq
Fig. 11.3.3.
STEP 1. To compute Green's function, note that the function w = tan z,

where z = 0 goes into w = 0, maps the parallel strip S : { - ~ ::; 3? {z} ::; ~}
onto the unit circle Iwl ::; 1. The usual Green's function for the point z = 0 is
Q(z,O) = log tan z. Then the iq-periodic Green's function is given by
L
00
Q(z, q) = log tan(z + inq), (11.3.7)

n=-CX)
where the series is convergent in the Cauchy sense. This periodic Green's
function can be represented in terms of elliptic theta functions as
L log [tan(z + inq) tan(z -

00
Q(z, q) = logtanz + inq)]

n=1
IT tanh(iz -
00
= logtanz -log { nq) tanh(iz + nq)}

n=1 (11.3.8)
where we have set u = eiz and h = e- q , and 19 1 ,2 are elliptic theta functions
such that 19 1 (z+rr /2, h) = 19 2 (z, h). These functions are numerically evaluated
by the formulas (Abramowitz and Stegun, 1968)
L( _l)n-1 h(2n-1)2/ 4 sin(2n -

00
19 1 (z, h) = 2 1)z,
n=1
(11.3.9)
L
00
19 2 (z, h) = 2 h(2n-1)2/4 cos(2n - 1)z,

n=1
where the series are convergent for h < 1, i.e., q > O. If q is not too close to
unity, the series converge very rapidly, and the computation of 19 1 ,2 is straight-
forward.
STEP 2. We shall derive Gershgorin's equation for the periodic strip n'
which we assume consists of congruent regions . .. , n'--1' n~, n~, n~, ...
n
(Fig. 11.3.3). Let C denote the contour C = L n U Un U ( U (C j U Gj ) ),
j=-n
where G j consists of the boundary curves on both sides of OJ, C j is a circle

around Zo + jib and two lines of the cut connecting this circle and Zl + jib, Uj
is the upper boundary line, and L j the lower boundary line of OJ (Fig. 11.3.3).
The point Zo corresponds to w = 0, and Zl to w = 1r/ 4. The variable (traverses
C in the positive sense. If z E Int(C), then ( = z is the only simple pole of
f(z), in which case by Cauchy's formula
g(J(z), q) =~
2t1r c
1 g(J((), q)
(- z
de·
Now, let the radii of the circles C j tend to zero. Then the above integral along
these circles vanishes because g(J(() , q) ~ log (( - Zo - jib), and thus
9 (J(z), q) = ~
2t1r
1(- z
9 (J((), q) d(
t
C\{C j }
+~ o jib
r
+ g+ (J((),q) - g- (J((),q) de,
2m j=_n}Zl+jib (-z
where g+ (J((), q) - g- (J((), q) = 2i1r (for this notation, see §1.4). There-
fore,
g(J(z), q) = ~
2t1r
1 C\{Cj}
9 (J((), q) d( +
(- z
t
j=-n
log Zo
Zl
+ ~~b - Z.
+ Jtb - z
Since G j = Go + jib and 9 (J(( + jib), q) = 9 (J(() = iq, q) = 9 (J((), q),

we find that
9 (J(z), q) = 1-2~
t1r
9 (J((), q) d(
z
l I nj~n
U"uL_" ( -
1
+ 2i1r Go 9 (J((), q) (+ jib - z d( (11.3.10)
~ Zo + jib - z
+ .L...J log
Zl + Jt. 'b - Z
== It + 12 + h
J=-n
Let n ---- 00. Then 12 ---- O. Using the formula (1.2.15), we have
~ 1 =~ ~ 1 ~ cot _1r.;. (_-_z--,-)

:.(
as n ---- 00,
j~n ( + jib - z iL j~n (i~ z +j iL iL
and using formula (1.2.14), we find that, as n --+ 00,
ZO-Z)2 . Zo-Z
IT Zo + ~ib - z = Zo - z IT 1- ( ----;:;r;- 2 --+ sm -j2-'b-
,
J=-n
zl+)tb-z zl-z,
J=1 1-
(ZI-
--
Z) . ZI- Z
sm -)"-'b-
jib •
Hence, as n --+ 00 in the Cauchy sense, we find from (11.3.1) that
h+I3=~
11
21l' Co
1r
Q(j((),q)7- cot
2
1r((-z)
'L d(+log
sm--
t
jib
Z -z
. Zo - Z
sin _I_ _
jib
= 9 (j(z), q).
(11.3.11)
Since 9 (j(z), q) is purely imaginary, let 9 (j(z), q) = i it> (j(z), q), where
it> is a real-valued function that defines the boundary correspondence and, by
taking the imaginary part of (11.3.11), is defined by
it> (j(z), q) = 2~ !co it> (j((), q) ~{i~ cot 1r(~~ z) d(}

.-
sm ZO-Z}
-
jib
(11.3.12)
+ arg == J 1 + J 2 •
{ sm. ----;:;r;-
Zl - Z
Now let the point z approach a point Zs on the boundary of Go. Then
1 { 1r 1r(( - z) }
J1 --+ 21r Jc{ it> (J((), q) ~ iL cot iL d(
1
+ 2 it> (J(zs), q).
o
(11.3.13)
Hence, as z --+ zs, from (11.3.12) and (11.3.14) we obtain
it> (J(zs), q) = ~ !co it> (j((), q) ~L~ cot 1r(~~ z) d(}

.-
sm ZO-Z}
-
"b
(11.3.14)
+ 2 arg )2 ,
{ . zl - z
sm ----;:;r;-
which is Gershgorin's integral equation for the periodic strip nt .
11.4. PROBLEMS 317
Eq (11.3.14) can be simplified by taking the parametric equation of the

curve Go as ( = ((t), where t is the arc length on Go. Since Go has two
parts, G~ and G~, this parametric equation represents two equations. If we set
w(t) = f (((t)), Zs = ((s), and if.> (J(((t)), q) = ¢(t, q), then Eq (11.3.14)
becomes
¢(s) = ~
1l'
(rJc ' + JCIIr ) ¢(t,q)~{~zL
o 0
cot 1l'(((t).-((s)) ('(t)}d(
zL
sin _zo_----:,(.o.-(s--,-) }
jib {'l?I(W(S),q)}
+ 2 arg { . ZI _ ((s) = arg 'l?2 (w(s), q) .
sm "b JZ
(11.3.15)
Since w(s) = ±~ + i y(s) for ((s) E Go, we find from (11.3.9) that
'l?1 (w(s), q) ±A (y(s)) + i B (y(s))

(11.3.16)
'l?2 (w(s), q) A(y(s))=t=iB(y(s)) ,
where the upper and the lower sign is chosen according as w( s) = ± ~ + i y( s),
and
A(y) = f..j2
n=1
hn(n-l) cos (2n ~ 1)1l' cosh(2n - l)y,
~
B(Y)=~V2hnn-
In (1)
sin
(2n-l)1l'
4 sinh(2n-l)y=i A Y+2"
1 ( i1l')
.
Hence the simplified form of Eq (11.3.14) is
¢(
s) q = arg {
'l?1 (w(s), q)}
=
{2 arg{A (y(s)) + i B (y(s))} on G~,
, 'l?2(W(S),q) 1l'-2arg{A(y(s))+iB(y(s))} onG~.
(11.3.17)
This equation can be numerically computed by an iterative method.•
11.4. Problems
PROBLEM 11.4.1. Prove that a doubly connected region n can be mapped

univalently onto an annulus A(I, R), R > 1, and the mapping is unique if a
point Zo on the boundary of n is transformed into a point Wo on the boundary

of A(l, R). (Goluzin, 1969, p.208; Wen, 1992, p.97.)
PROBLEM 11.4.2. Prove that a necessary and sufficient condition for the
existence of a univalent mapping function, which maps the annulus A (PI, P2)
onto the annulus A (rl' r2), is P2/ PI = r2/rl' (Goluzin, 1969, p.208; Wen, 1992,
p.97.)
PROBLEM 11.4.3. If a doubly connected region n contains a doubly

connected region ~, ~ c n, ~ f:. n, then show that the conformal modulus
of n is greater than that of ~. (Goluzin, 1969, pp.209-21O; Wen, 1992, p.133.)
PROBLEM 11.4.4. Show that the kernel in Eq (11.3.1) can be evaluated

from
-a
an s
(I rsz1) _
og- - cos (ns,r sz )
r sz
e(s) [y -1](s)] -1]/(s) [x - ~(s)]
[x - ~(s)j2 + [y -1](s)]2
where ((s) = ~(s) +i 1](s), and z = x+i y. (See Case Study 7.4.3; also Andersen
et al., 1962, p.178.)
PROBLEM 11.4.5. Consider coaxial ellipses
which are symmetric about both coordinate axes. Take the distribution a =
o (27r/ N) 27r for the partition of the boundaries, and N = 4(n - 1). Note that
if (i) al = 5, bl = 1, a2 = 7, b2 = 5, (ii) al = 6, bl = 2, a2 = 9, b2 = 7, or
(iii) al = 7, bl = 2, a2 = 9, b2 = 6, then the ellipses f l and f o are confocal,
i.e., ay - by = a~ - b~. In these cases there is an exact mapping function
with P = al + bbl. By computing E show that E decreases as N increases.

a2 + 2
(Symm, 1969.)
11.4. PROBLEMS 319
PROBLEM 11.4.6. Show thatthe curve f 0 in Case Study 11.4.1 gradually

changes from a circle (b 2 = 0) to a cardioid for values of b2 up to 5 and the
point where f 0 crosses the negative x-axis becomes a singularity in the final
case. (Symm, 1969.)
PROBLEM 11.4.7. Consider Cassini's ovals
f l = {[ (x + bl )2 + y2][ (x - bt}2 - y2] = aD,

f o = {[ (x + b2)2 + y2] [(x - b2)2 - y2] = a~},
(see Fig. 9.2.1), which have symmetry about both coordinates axes. Partition
f by taking equispaced points on the x-axis on each boundary with the same
number of points on f l and f o, and take N = 4(n - 1). Note that if (i)
al = 2, bl = 1, a2 = ~2506 :::::; 7.07389, b2 = 7, or (ii) al = 9, bl =
6, a2 = 4
~1l1l6 :::::; 10.26803, b2
a1- b1 = bbr ' and In. this case
= 8, then ---::r--b
a2 - 2
4 2
2
the exact mapping function is given by
with p = a 2 bl . Show that E decreases as N increases. (Symm, 1969.)

al b2
PROBLEM 11.4.8. Consider a square in a circle, defined by f 1 = {-b :::;

x, y :::; b} and f o = {x 2 + y2 = a2}, where a> b> O. Because of symmetry
about both coordinate axes, partition f land r 0 with equi-spaced points in equal
number on each boundary, and take N = 4(n - 1). Show that E decreases as
N increases, and that E decreases rapidly away from Z. (Symm, 1969.)
REFERENCES USED: Abramowitz and Stegun (1968), Andersen et al. (1962),

Antonjuk (1958), Gaier (1964), Goluzin (1969), Jawson (1963), Kantorovich and
Krylov (1958), Muslhelishvili (1963), Nehari (1952), Symm (1969), Wen (1992).
Chapter 12
Singularities
As Radon (1919) and Carleman (1922) have shown, the method of integral
equations can also be used to solve the problem of potential theory when the
boundary r of a simply connected region D contains corners. In such cases
Carleman separates the kernel into two parts, one of which corresponds to
the corner singularities, whereas Radon uses the Stieltjes integral equations
to solve this problem. We shall derive the analogues of Gershgorin's inte-
gral equation and then obtain Arbenz's integral equation which uses Radon's
method to determine conformal maps for boundaries with corners and has a
unique solution. The cases of interior and exterior mapping functions f (z)
and fE(Z) are related to each other through inversion by the relations (7.3.12).
We are interested in the behavior of these univalent maps and those of dou-
bly connected regions at singularities on and near the boundary, which are
corner-type or pole-type. The nature and location of such singularities are
determined.
12.1. Arbenz's Integral Equation
Let us assume that the boundary r of a simply connected region D consists of

two Jordan curves r 1 and r 2 (r 1 U r 2 = f), and suppose that r 1 and r 2 meet
at a point Zo = 1'(80) and form a corner with interior angle o:n, 0 < 0: < 2.
Suppose that Zo is a regular point of both curves. Let f(z) E J(O(D) denote
thefunction that maps D univalentiy onto the unit disk such that f(O) = O. Let
the parametric equation of the boundary r be z = 1'(8), 0 ~ 8 ~ L, which is
320

12.1. ARBENZ'S INTEGRAL EQUATION 321
positively oriented with respect to the region D. Then f ('y(s)) = ei¢(s) and
fE ('y(s)) = ei¢E(S), where the boundary correspondence functions </>(s) =
arg {f ('y(s))} and ¢>E(S) = arg {fE ('y(s))} are continuous principal ar-
guments which playa significant role in integral equation methods. Let the
function B(s) = arg b(s)} be defined for a ~ s ~ L, such that it has at most
finitely many jump didcontinuities of magnitude less than 1r in the interval
[0, L]. This yields finitely many subintervals of [0, L], in each of which B( s) is
continuous and has bounded variations. Thus, at a corner point on r we have
IB(s+) - B(s-)I < 1r, and the boundary r is called a contour with bounded
variation. The following result is due to Radon (1919).
THEOREM 10.1.1. If r is a contour with bounded variation, then

Bs(t) = arg b(s) - -y(t)}, defined in (7.1.1), is of bounded variation for
every fixed s E [0, L] and is uniformly bounded for all s E [0, L].
The Stieltjes integral equations that arise in Radon's method have the form
1£ ¢>(t) dtBs(t) = g(s), where ¢>(t) is continuous in [0, L], and the subscript
t denotes the variable of the Stieltjes integration. In fact,
(12.1.1)
where a1r = a(s)1r, s E [0, L], denotes the interior angle at the corner point
-y (s) on the boundary r.
To derive the Stieltjes integral equation associated with Gershgorin's inte-

gral equation (7.2.6), let z = -y(s) be a corner point on r. Then, in view of
(1.2.10), the left side ofEq (7.2.2) becomes a log f(z), and then instead ofEq
(7.2.6) we obtain
a¢>(s) = -1
1r
1£
0
¢>(t) dtBs(t) - 2(3(s), (12.1.2)
where the integral takes the Cauchy p.v. at t = s. Since, by (12.1.1),

322 CHAPTER 12 ; SINGULARITIES
r£ ¢(t) dtBs(t) = .!. l~ ( Jor-

Jo 71" c 0
c
+ l + 1£ )
s c
+
s-c s+c
¢(t) dtBs(t)
= 1£ ¢(t) dtBs(t) + 0'71" ¢(8),

from (12.1.2) we find that
.!.
.71"
r£ ¢(t) dtBs(t) = 2 (3(8),
Jo
8 # O. (12.1.3)
,
Since 1 £
dtBs(t) = 0, the solution of Eq (12.1.3) is determined up to an
additive constant, and hence, it is not unique. This situation is avoided in
Arbenz's integral equation which can be derived from (12.1.3) as follows: For
8 # 0 set
(12.1.4)
Also, Bs(t) = Os(t) - Bo(t), where

Bo(O+) for t = 0,
Bo(t ) = { (12.1.5)
Bo(L) = Bo(L-) for t = L.
The angle Bs(t) is shown in Fig. 12.1.1. Then for 8 # 0 we find from (12.1.3)
that
-1 1£ ¢(t) dtBs(t) = -1 1£ ¢(t) dtBs(t) + ¢(8) - -1 1£ ¢(t) dtBo(t)
1£
71" 0 71" 0 71" 0
= 2(3(8) + ¢(8) - -1 ¢(t) dtBo(t).

71" 0
(12.1.6)
To determine the integral in (12.1.6), let 8 -+ 0+ in (12.1.3), and use Hally's
theorem which states that lim Bs (0) = Bo(0) + 71". Then the limit value of this
s---+o+
integral is given by
12.1. ARBENZ'S INTEGRAL EQUATION 323
which yields
If we require that the boundary correspondence function be <p( 0+) = - 2 {3( 0+),
then <p( s) is uniquely determined from the integral equation
~ Jr
L
<p(t) dtBs(t) = <p(s) + 2 {3(s), s :/= 0, (12.1. 7)
1f o
which is known as Arbenz's integral equation. Note that the integral in
(12.1.7) is not evaluated as a Cauchy p.v. as in (12.1.3).
Fig. 12.1.1. Fig. 12.1.2.
The discretization method should be used when the boundary r is repre-

sented geometrically rather than analytically. In order to discretize Arbenz's
equation (12.1.7), we partition the interval [0, L] into N subintervals [tk-b tk]
of equal or unequal length with Sk as an interior point, tk-l < Sk < tk>
k = 1,2, ... ,N. In practice, it is useful to take a corner point coincident
with a partition point and have more subintervals in its neighborhood. Then for
S = Si Eq (12.1.7) becomes
(12.1.8)
The solution <p( s) is continuous and can be taken as

N
<Pi = <p (Si) = L aik <Pk, (12.1.9)
k=l
where
(12.1.10)
324 CHAPTER 12 : SINGULARITIES
For i = k the integral in (12.1.10) takes the principal value of Stieltjes integra-
tion. In fact, if Si is a comer point, then the arcs at Si subtend the interior angle
a{71". Let K(Tr denote the angle between the chords (Si, t i - 1 ) and (Si, ti) (Fig.
12.1.2). Then
(12.1.11)
and
L ai k = -;1
N 1£ dtB si (t) = ai, (12.1.12)
k=l 0
where the integral takes the p.v. of Stieltjes integration.
CASE STUDY 12.1.1 (Gaier, 1964, p.57). For the mapping ofthe square
{-I < x, y < I} onto the unit disk Iwl < 1, we have discretized the boundary
of the square with N = 40 subintervals, where t = 0 and t = L at the point
to, although in Fig. 12.1.3 only quarter regions of each boundary are presented
because of the symmetry of the square with respect to the x and y axes and the
symmetry of cP( s), i.e., cP20+j = 7r + cPj and cPlO+j = 7r /2 + cPj·
57 56 V
55
/6
54
53
52
5,
/0
o""----------+~
X $ u
I 40 0
5 39
Fig. 12.1.3.
12.2. BOUNDARY CORNER 325
Thus,
1
- L Bdtk-l, tk) ¢k = 2{3 (Si)'
40
(12.1.14)
1r
k=l
The coefficient for ¢k is 0.5 for k = i when i = 5, 15,25,35, and is equal to
1 otherwise. The following results, obtained after computing Eqs (12.1.3) and
(12.1.6), are compared with the exact solutions in the following table:
Eq (12.1.3) Eg (12.1.6) Exact

1 0.256503 0.256738 0.256319
2 0.480278 0.481440 0.479890
3 0.648856 0.654261 0.648240
4 0.751901 0.789932 0.751028
5 0.785394 0.785406 0.785398
6 0.818887 0.780878 0.819768
7 0.921933 0.916549 0.922556
8 1.090511 1.089369 1.090906
9 1.314286 1.314071 1.314477
The exact boundary correspondence function is given by
cos ¢ = dn (K y), K =K ( ~) , (12.1.15)
where dn is one of the twelve Jacobian elliptic functions which is a meromorphic

function with pole at i K'(m), K(m) = K'(1 - m), and dn 2 u = 1 - sn 2 u
(see Case Study 3.3.2). The inverse mapping is given by
(1/2)
z= A ~ n w-4n+l~ (12.1.16)
L.J 4n-l
n=O
where A is a constant (cf. with Case Study 4.2.2 and 4.5.1, and for the Green's
Function and the associated mapping function, see Case Study 12.5.1).•
12.2. Boundary Corner
Symm's integral equations (9.1.6) and (9.1.12)-(9.1.13) for determining the

density function p,( s) can be written as
1£ J-L(S) log h(t) -,(s)1 ds = k(t), t i= s, (12.2.1)

where
k(t) ={ -log h'(t) I for z = ')'(8) E D = Int (r),

(12.2.2)
1 for z = 1(8) E D* = Ext (f).
This equation has a unique solution provided that the capacity cap (r) i 1 (see
§1.3, 1.4). The density function f.-t( 8) is related to the boundary correspondence
function 1>(8) by
1>'(8) = { -27r f.-t(8) for z E Int(r),

(12.2.3)
27rif.-t(8) forz E Ext (f),
where i is Robin's constant (i = -log {cap (f)}, see §1.3).
If Zo is a corner point on a portion of the boundary f with interior angle

a7r, where a = p/q (p and q are relatively prime), then we should analyze the
asymptotic behavior of the mapping function in the neighborhood of the point
Zo E f. If D is a polygonal region, we know from the Schwarz-Christoffel
transformation that in the neighborhood of Zo
L ak (z -
00
j(z) - j(zo) = zo)k/a, (12.2.4)

k=l
or, if j E KO(D) (i.e., j(O) = 0), then
L ak
00
j(z) = [tz - zo)k/a - (_zo)k/a] , al i 0, (12.2.5)

k=l
(seeCopson, 1975, p.170). Thus, ifl/a is not an integer, then j(z) has a branch
point singularity at Zo (see §2.l). This corresponds to Case 3 in Theorem 12.2.1
given below. Lichtenstein (1911) was the first to show that if a corner point is
located at the origin, then d~~) = zl/a-l h(z), where h(z) is a continuous
function such that h(O) i 0, and a is irrational. Warschawski (1932, 1955)
proved this result for all a. In the case when the two arcs at Zo are straight line
segments with a = 1, Lewy (1950) proved the stronger result that j(z) has an
asymptotic expansion in powers of z and log z. This result was generalized
by Lehman (1957) in the development of the mapping function at an analytic
corner as z ~ Zo, as follows:
12.2. BOUNDARY CORNER 327
THEOREM 12.2.1. Let Zo E r denote a corner point. Then there are

three cases of asymptotic expansions for the mapping function fez) as
z -> zo:
CASE 1. If a is rational, a = plq where p and q are relatively prime,
then as z -> Zo
fez) = f (zo) +L Bk,l,m (z - zo)k+l/a (log (z - zo))m, B O,l,O =f. 0,

k,l,m
(12.2.6)
where k, land m are integers, k 2: 0, 1 ::; l ::; p and 0::; m ::; klq. The
terms in (12.2.6) are ordered such that the term containing Bk',l',m'
always follows the term containing Bk,l,m if k' + l'la 2: k + lla and
m'<m.
CASE 2. If a is irrational, then as z -> Zo
fez) = f (zo) +L Bk,l (z - zO)k+l/a , B O,l =f. 0, (12.2.7)

k,l
where k and l are integers, k 2: °
and l 2: 1.
CASE 3. In the case when the two boundary arcs at Zo are straight line
segments, then the asymptotic expansions (12.2.6) and (12.2. 7) simplify
to
+L
00
fez) = f (zo) Bl (z - zo)l/Ot., B 1 =f. O. (12.2.8)

1=1
Cases 1 and 2 correspond to the situation when D is not a polygonal region,

whereas Case 3 applies when D is a polygonal region. In this case, to eliminate
the effect of a branch point singularity at zo, if such a singularity exists, we
augment the basis {¢>j(z)}7=1 by the functions (Papamichael and Kokkinos,
1981)
¢>(z) = .!!:- {(z - zo)k/Ot. - (_zo)k/a} - d = ~ (z - zo)k/a-1 - d,

dz a
(12.2.9)
where
~ (_zo)k/a-1 for RM,
d ={ a
o forBKM.
Note that (z - zo//a is the dominant term in each ofthe asymptotic expansions
(12.2.6)-(12.2.8). It appears that the mapping z 1-+ Zl/Ot. that transforms an
angle a1l" atzo E r into the angle 1l" atthe point Wo = f(zo) (see §2.1) will solve
the corner problem. But this does not happen because if 1/a is not an integer, a
branch singularity always occurs at the corner zo, and when a is an integer, the
existence of the logarithm in (12.2.6) makes the corner Zo a logarithmic branch
point singularity even if 1/a is an integer.
In the RM and BKM (§4.2) the minimum polynomial is constructed by

taking the basis set as that of monomials zi, j = 0, 1,2, .... Then the singular
basis function ¢( z) associated with the corner singularity at Zo Erin the above
three cases of asymptotic expansions (12.2.6)-(12.2.8) has the form
:z {(z - zo)k+l/o: (log (z - zo))m} in Case 1,
¢(z) = dd {(z - zo)k+l/O:} in Case 2, (12.2.10)

z
dd {(z - zO)l/o: } in Case 3,
z
which is used to augment the basis set in RM and BKM when determining the
mapping function f(z) from (4.2.25). It may be noted that the function f(z),
originally defined on D, can be extended by analytic continuation through
a portion of its boundary into Ext (r). This procedure is used in §12.4 to
investigate the nature and location of poles and pole-type singularities of f (z)
near the boundary r. The singularities of the Bergman kernel K(z, a) in the
region Ext (r) also affect the convergence of the polynomial series (4.2.20).
These singularities are either poles of K(z, a) that lie close to the boundary r
or branch point singularities of the boundary itself. Their effect should always
be taken into account when determining the mapping function for the exterior
regions (see §12.5).
12.3. Singularity Behavior
In the neighborhood of a corner point Zo = "/ (80) let the function ,,/(8) have a
Taylor series representation
f:
n-I
(8 - ~ot ,,/(n) (80),
n.
8 < 80,
,,/(8)=,,/(80)+
{ ~-LJ
(8 - 80)
,
n
"/
(n) ( +)
80' 8 > 80,
(12.3.1)
n=1 n.
12.3. SINGULARITY BEHAVIOR 329
where ')'(n) (s~) = lim {dnd')'~s)}. The boundary correspondence function

s
s--+st
¢(s) associated with the mapping function f is given by
A-,'( ) = . f' (')'(s)) f (')'(s)) (12.3.2)

If' s Z If (')'(s)) 12 .
Then, by (12.2.3), the density function p,( s) is related to f by
~{f'(')'(s)) f(,),(s))}
p,( s) = (12.3.3)
27r
where If (')'(s)) I = 1. Note that for the mapping function !E this relation is,
in view of (12.2.3), given by
~ {fe (')'(s)) !E (')'(s))}

p,(s) = - 2 A • (12.3.4)
7r')'
Hence, using (12.2.6)-(12.2.8), (12.3.1), and (12.3.3), a formal asymptotic

expansion for p,( s) as s -4 So is given by
I:~l aj 'l/Jj (s - so), s < so,

p,(s) = + (12.3.5)
{ I:j=l a j
00
'l/Jj (s - so) , s> so,
where the functions 'l/Jj depend on the value of (}:, 0 < (}: < 2 (see Problem
12.7.1). Then from (12.3.5) we conclude the following:
(a) If 1 < (}: < 2, i.e., if the comer Zo is re--entrant, then p,( s) becomes
unbounded at s = so.
(b) If 1 ~ q < (}: < 1, where q ~ 1 is an integer, then p,( s) becomes unbounded
at s = so.
(c) If (}: = ~, where q ~ 1 is an integer, then (12.3.5) does not involve rational
q
powers of (s - so). Since at
i- at, the function p,(q-l)(S) has a jump
discontinuity at s = so. Moreover, for some j > 1, one of the functions 'l/Jj
in (12.3.5), obtained from the expansion (12.2.6), is a function of the form
0'2q-l log 0', where 0' stands for (s - so) or (so - s). Thus, in general, the
left and right (2q - 1)-th derivatives of p,( s) at s = So become unbounded.
(d) In Case 3 of Theorem 12.2.1, without loss of generality, we take ")'(8) in the
form
( 80 - 8 ) e ia". , 8 <
_ 80,
80) + {
")' ( 8 ) =")' ( (12.3.6)
8 - 80, 8 ~ 80.
Th en, ol,.()
'f/J a -- a -Hj/a f or J. -- 1 , 2 , ... , an d a + - ( - 1)i+ 1 a j- f or
j -
j = 1,2, .... In this case conclusions (a) and (b) remain the same, but (c)
changes, viz., if a: = l/q (q ~ 1 an integer), then
(d.l) if q is odd, then 1-£( 8) has no singularity at 8 = 80; and
(d.2) if q is even, then l-£(q-1)(8), in general, has a jump discontinuity
at 8 = 80.
12.4. Pole-Type Singularities
Sometimes the function I (or IE) has poles or pole-type singularities adjacent
to the boundary r, which are located in the region D* (or D). We shall examine
the nature and location of such singularities. This analysis, based on the work
ofPapamichael et al (1986), will be confined to the function I. Thus, we shall
determine the nature and location of poles and pole-type singularities of I in
D*, which are obtained by considering an analytic continuation of I across
the boundary r into D*. The procedure expands the domain of I as much
as possible provided the function I on the extended domain agrees with the
original.
To determine the dominant poles of the mapping function w = I(z) in

Ext (r), which are actually the poles of the analytic continuation of I across
r into Ext (r) that lie close to r, we can use the Schwarz reflection principle
in the case when the boundary r consists of straight line segments and circular
arcs. Then I has simple poles at the (finite) symmetric points of the origin
with respect to the straight line segments and the circular arcs (see §IA). But
when the boundary r is more analytic than straight line segments and circular
arcs, then in many cases the dominant poles of I can be determined by using a
generalized symmetry principle as follows: Let the parametric equation of an
analytic arc f' of r be given by
z= ")'(8), 81 < 8 < 82. (12.4.1 )
Let G* be a simply connected region in the complex (-plane, ( = 8 + it, that

satisfies the following two conditions:
12.4. POLE-TYPE SINGULARITIES 331
C1. The function z = -y( s) is univalent in G* .

C2. The straight line L = {( : ( = s + it, Sl < S < S2, t = O} divides the
region G* into two partitions G l and G 2 which are symmetric to each other
with respect to Land G* = G l U L U G 2 , and the image D l of G l under the
transformation (12.4.1) is contained in D (D l ~ D).
Obviously, Gland G 2, each subsets of G*, are defined by
G 1 = {(: (E G* and t > O}, G 1 = {(: (E G* and t < O},

or
G 2 = {( : ( E G* and t < O}, G z = {( : (E G* and t > O}.
Under conditions (i) and (ii) above the function (12.4.1) maps the region G*
r
conformally onto a region D* = D l U U D z such that the regions G l , G 2
and the straight line L are mapped onto the regions D l , D z and the arc r,
respectively. Then the function
h(() =f (-y(()) , (12.4.2)
where f (z) maps the region D onto the disk IwI < R, is univalent in the region
G l UL, and w = h(() maps the straight line L onto an arc ofthe circle Iwl = R.
Thus, by the reflection principle the function
h(() (E G l U L,
H(() = { 1_, (E G 2 , (12.4.3)

h(()
is meromorphic in G z and defines an analytic continuation of h across L into

G 2 • If", denotes ~he inverse of {, then the function
f(Z)' zEDlUr,
F(z) = H (1J(z)) = { 1 zED , (12.4.4)
2
f({3(z))
where (3(z) = -y (1J(z)) is analytic in Dl and meromorphic in D 2 , defines an

r
analytic continuation of f across into D 2 , where the points z and (3(z) are
symmetric points with respect to the arc r.
Hence, we have the following result
(Papamichael and Kokkinos, 1983):
THEOREM 12.4.1. The following cases hold:

(aJ If 0 E D l , then -y(z) has exactly one zero in G l , i.e., the function
F(z) has a simple pole at a point Zo E D 2 , where Zo = l' (0) = a(O) is

the inverse point of the origin with respect to the arc 1".
(b) If 0 E aDl\1", then I'(z) has at least one zero (0 E aGl\L, and the
function I'(z) need not be one-to-one in the neighborhood of the points
(0 and (0,
(c) If 0 E D l U (aD l \1"), then F has no poles in the region D 2 U
(aD 2 \1").
To determine the behavior of F(z) at the point Zo = 1'((0) E aD 2 \1" in

part (b) ofthe above theorem, let us assume that l' is analytic at (0 and (0. Then
(12.4.5)
and
(12.4.6)
where 1'1 and 1'2 are analytic and nonzero at the points (0 and (0, respectively.
Thenthemappingfunctionisf(z) = z!l(z),!l(O)::j:. O. Hence, from (12.4.4),
for z E D 2 the function G(z) = Ftz) can be written as
(12.4.7)
where g1 is analytic at zoo Since
(12.4.8)
we find that
(12.4.9)
where a1 and 1]1 are analytic at zoo Hence, from (12.4.8) and (12.4.9) we get
and thus, from (12.4.7)
(12.4.10)
where 92 is analytic at zoo This leads to the following:

THEOREM 12.4.2. The following cases hold:

(b.l) Ifm = n = 1, then F has a simple pole at zoo
(b.2) If m = 2, n = 1, then F has a double pole at zoo
(b.3) If m = 1, n = 2, then F has a branch singularity of the form
(z - zO)-1/2 at the point zoo
G*
L
1 0 52
- q
Fig. 12.4.1.
CASE STUDY 12.4.1. Let the arc f' be defined by the ellipse E
(x - X )2 (y _ y )2
+ r
A
a2 c b2 c = 1, a > b, and let the parametric equation of

bez = "((8) = zc+aecos(s-iq), O:S S1 < 8 < S2 < 21T, where
Zc = Xc + i Yc is the center C , e = J1 -
b2 ja 2 the eccentricity of the ellipse,
coshq = lie, and 82 - 81 < 21T. Then the function z = "(((), ( = 8 + it,
is univalent in the strip {(: (+ s + it, 81 < 8 < 82, -00 < t < q}, and the
region G* is a symmetric subregion of the rectangle {( : (+ 8 + it, S1 <
8 < S2, -q < t < q} (see Fig. 12.4.1). Consider the case when G* is the
entire rectangle. Then the region D* = D 1 U f' U D 2 can be determined by
finding the images of the four sides of this rectangle under the transformation
z = "(((), ( = 8 + it. Assuming that the regions G 1 and G 2 are defined by
(12.4.1), the four typical regions D* are presented in Fig. 12.4.2(a)-(d) which
correspond to the following four cases, respectively:
1T 1T
(a) 81 = 0,0 < 82 ::; 2; (b) 81 = 0, 2< 82 :S 1Tj
31T 31T
(c) 81 =0,1T<S2:S
2; (d) 81 = 0, 2 < 82 < 21T.
In each Fig. 12.4.2, the equations of the different arcs are
r=arcPQ: {z=zc+aecos(s-iq), 0<S<S2},
r' = arcP'Q' : {z = 'Y(s - iq), 0 < s < S2},
r*=arcQ'R: {Z='Y(S2+it), -q<t<q}.
P'
(a) (b)
r' -----
,
r
Q r*
----Q'
(c) (d)
Fig. 12.4,2.
Arc r' is that of an ellipse E', and f* that of a hyperbola orthogonal to both
ellipses E and E' (the right branch of the hyperbola if cos S2 > 0 and the left
branch if cos S2 < 0). In the case when S2 = 7r /2 or 37r /2, the hyperbola
degenerates into a vertical straight line through the center C (R coincides with
C and r becomes a part of the minor axis), whereas when s = 7r, it degenerates
into the major axis, R coincides with the focus F2, and r becomes a part of
the major axis. The region D 1 is shaded in each figure, and D 2 is the region
bounded by the arcs r, r' and the subarc Q'Q of f*. Note that in figures (c)
and (d) the region D 1 includes a cut on the major axis from F 2 to R (because
in this case the mapping z = 'Y(() yields a common image Zc - a e cos s of the
points (7r ± s) + i q, s > 0),
The region D* associated with any arc r with 0 S; S1 < S < 27r can be
obtained from Fig. 12.4.2 (a)-(d). For example, the region D* associated with
r
an arc for 0 < 81 ~ 1r /2 and 1r < 82 < 31r /2 is obtained by deleting the
region of figure (a) from that of figure (c). The region D* associated with an arc
r, which includes the two vertices Zc ± a of the ellipse E can also be obtained
from these four figures. For example, if -1r/2 < 81 < 0 and 1r < 82 < 31r /2,
then the region D* is the union of the region of figure (c) with the region
obtained by reflecting the region of figure (a) on the major axis.
Now, using the results of Theorems 12.4.1 and 12.4.2, we conclude that
"y(() has exactly one zero in C 1 at the point (0 = cos- 1 (- ::) + iq, which
means that the function f has a simple pole at the point
where the square root is chosen such that 0 < arg { J a 2 - b2 - z~ } < 1r.
If 0 E aD 1 \r, then the origin lies on the major axis between the foci F 1
and F2, i.e., -a e ~ Xc ~ a e and Yc = O. Then, the following three situations
arise:
(i) If the origin lies on a cut in the region D 1 but does not coincide with either
focus of E, then there are two distinct values of cos- 1 (-xc/a e) in the interval
(81,82), and associated with these two values there are two distinct zeros of
"y(() on the side t = q of C 1 . Hence, f has two simple poles at the two points
_ -2b2xc±2iabJa2-b2-x~ rA
(12.4.12)
,
Zo- 2 b2 E.
a -
(ii) If the origin does not lie on a cut of D 1 and does not coincide with either
focus of E, then there is exactly one value of cos- 1 (-x c / a e) in the interval
(81,82), and so "y(() has a zero on the side t = q of C 1 . Thus, f has a simple
pole at Zo given by (12.4.12), where a proper sign is chosen so that Zo lies on
r'.
(iii) If the origin coincides with either focus of E, i.e., Xc = ±ae, Yc = 0, then
"y(() has a double zero at (0 = i q and (0 = 1r + i q. Hence, f has a double
pole at one of the vertices of the ellipse E', i.e., at one of the points
(12.4.13)
where the ± sign is chosen according as the origin is at F 1 or F 2 .

If the origin is not in D 1 U (r\f), then f has no poles in D 2 U f'. If the

origin lies in aD 1 \f, then f has a simple pole at the point Zo given by (12.4.11),
except when the origin coincides with one of the vertices of E', i.e., when
a 2 + b2
=± = O.
Xc
va 2 - b2
,and Yc (12.4.14)
In this case f has a singularity of the form (z - ZO)-1/2 at one of the foci of
2
. at one 0 f th ' Zo =± 2b
E , I.e., e pomts
va 2 - b2
.•
CASE STUDY 12.4.2. (a) Let the boundary r of the region D be the union
of an elliptic curve r 1 and the straight line segment r 2, defined by
r1 = {z: z=4coss-2e+ibsins, -7r/2<s<7r/2, O<b<4},
r2 = {z; z = X + iy, x = -2e, -b < y < b},
V16 - b2
where e = 2 (see Fig. 12.4.3).
y y
r -_ _ r,
~ ~
c-ic;+--_ _----;;-I--_ _---:+-t-"x --':c+- ~F';ct-j_.X
(-2e, O) 0 F (-4e, 0) 0
(a) (b)
Fig. 12.4.3.
There are two poles of f with respect to the curve rb and in view of
(12.4.12) they are at
b2 ± 4V3bi
z 1,2 -- -----c::::=:::::;=;<:-
V16 _ b2 .
There is one pole with respect to the line r 2 at Z3 = -V16 - b2 , which is the
mirror image of 0 in r 2 .
(b) If we translate the region D by 2e in the negative x direction, then the origin
ocoincides with the focus PI, and the new region D' is bounded by arcs
r3 = {z; z = 4(coss - e) + ib sins, -7r/2 < s < 7r/2},
r4 = {z: z = x + iy, x = -4e, -b < y < b}.
Then, in view of (12.4.13), the function f has a double pole with respect to the
2
curve r 3 at Z4 =
2b , and with respect to the line 4 it has a simple r
V16 - b2
pole at Z5 = -2 V16 - b2 , which is the mirror image of 0 in r 4. Note that
the boundary of the region D' is very close to the origin. In view of (6.2.3), in
such a situation the mapping function f is connected to the mapping function
h of part (a) by
f( z) = ~ h(z) - a
a l-ah(z)' a=
f
1
(ae)
2 .•
CASE STUDY 12.4.3 Let the region D be bounded by the straight line
segments
AB: {z: z = X + iy, -2 < x < 2, y = -1/3},

Be: {z: z=x+iy, x=2, -1/3<y< I},
AE: {z: z=x+iy, x=-2, -1/3<y<I},
Zo y
"' \
\ c
\
\
\
\
\
_t- -+--- +-- x

o
A L-- -I- ---' B
Fig. 12.4.4.
and the cubic arc
EDC: {z: z = ')'(8), -2 < 8 < 2},
where ')'(8) = S
(2 1 1)
+ i "3 + 4 s - 4883
~
(see Fig. 12.4.4). The arc CDE has
a point of inflection at x = O. The function ')'(() has a zero inside the boundary
of the region D at the point
(0 = -0.160784962923 - 0.626680456065 i.
Then
Zo = 'Y ((0) = -0.321569925846 + 1.25336091213 i.
Also, since c ((1) - c ((2) = ((1 - (2) R ((1, (2), where
and R ((1, (2) =1= 0 for all (1 and (2 in the rectangle G = {( : (= s+it, -2 <
s < 2, -1 < t < 1}, the function 'Y(() is one-to-one in G. Thus, there exists
a simply connected region G* that contains the points (0 and (0 and is such
that the conditions 01 and 02, mentioned in the beginning of this section, are
satisfied. Hence, in view of Theorem 12.4.l(a), the function f has a simple
pole with respect to the arc ODE at the point zoo •
L
----"c~'_-----T:-:--_"-,-_ _-t--"'Q:-- Z3
K, M ",'0
N
Fig. 12.4.5.
CASE STUDY 12.4.4. The region D bounded by the elliptic arc LMN
which is defined by z = 5 coss - 17/2 + 3i sins, -1r/5 < s < 1r/5, the
~
straight lines N P and LR, and the circular arc PQR whose center is at the
point K and radius is KQ, where Q = (7/2,0) and K is the point where the
normals to the ellipse at Land N intersect the x-axis. The coordinates of
the center 0 of the ellipse are (xc,O) = (-17/2,0), and the focus F1 is at
(-9/2,0). Thus, the origin 0 and the focus F1 are inverse points with respect
to the elliptic arc LMN (Fig. 12.4.5). Then, in view of (12.4.14), the mapping
function f has (i) a singularity of the type (z + 9/2)-1/2 at F1 , and (ii) a simple
pole at the mirror image Zl, Z2 of the origin with respect to the line segments
N P and LR, and at the geometric inverse Z3 of the origin with respect to the
~
circular arc PQ R. •
In §7.3.3 we considered the case of the function w = fE(Z) which maps

the region Ext (D) univalently onto the region U* = {Iwl > I} such that
!E( 00) = 0 and lim fE(z) > O. In this case, by using the inversion Z f--+ Z-l,
z-+oo
we reduced the problem to that ofthe mapping of interior regions by the function
f(z).
o
Ext(r)
w = g(z) R
---+---=+----1-'-'-
z-plane
z= l/~ w-plane
r*
~-plane
Fig. 12.5.1.
In this section we are concerned with the following mapping problem:

Assume that the origin lies inside a simply connected region D with the Jordan
boundary r. Let a function w = g( z) map the region Ext (r) conformally
onto the disk B(O, R) = {Iwl < R} in the w-plane. Also, the transformation
( = l/z maps the boundary f onto a Jordan contour f* so that z = 00

goes into ( = O. Let w = f( (), f(O) = 0, f' (0) = 1, map the region D*,
bounded by the contour f*, conformally onto the disk B (0, R) (Fig. 12.5.1).
Thus, w = g(z) = f(l/() maps the region Ext (f) conformally onto the
disk B(O, R) such that g( 00) = O. Hence, determining the mapping function
9 reduces to determining the interior mapping function f in such problems
of exterior regions, which correspond to Case 2 of §5.4. Also, the function
w = 1/g( z) maps the region Ext (f) conformally onto the exterior of the circle
Iwl = 1/R, and the quantity d = 1/ R is the transfinite diameter of the region
D U f (see §1.1).
Note that the function 9 is different from the function f E studied earlier.
We shall use the RM and BKM to approximate the mapping function g. Since
f E K:,l(D*), the basis in RM is taken as {cPj()}, as in §4.1, so that <P1(0) = 1
and cPj(O) = 0 for j = 2,3, ... , which leads to the complex linear system
n
L (cPj, cPi) Cj = -(cP1' cPi)' i = 2, ... ,n, (12.5.1 )
j=l
which is solved for the unknowns Cj, j = 2, ... , n. Thus, the n-th RM
approximations for the mapping function f() and the radius R are given by
(12.5.2)
where
n
IP n ( ) = cP1() + L cPj(), (12.5.3)
j=2
is the n-th approximation of f'().
In the RM and BKM (§4.2.2), since the basis set {cPj ()} is a known com-
plete set, first we approximate the Bergman kernel K (, 0) of D* by a finite
Fourier sum. Then
f'() = K(, 0) (12.5.4)
K(O,O)'
and R = (n K(O, 0))-1/2, as in (4.2.23) and (4.2.24). The details ofthe process
are the same as in the five steps given in §4.2.2.
The basis set is taken as the set of monomials {(j-1}, j = 1,2, ....
Depending on the singularities of K(, 0), the boundary singularities, and the
poles of f ((), however, this basis is augmented by the functions </>( z) defined
in (12.2.10).
CASE STUDY 12.5.1. Consider mapping the rectangle
nab = {(x, y) : Ixl < a/2, Iyl < b/2}, a 2:: b,

onto the unit disk U. In view of the Case Study 2.3.2, the mapping function
f (z), known in terms of the elliptic functions, is given by
(-a
f(z) = /3 --;;--::, (= sn (z, k), 1/31 = 1, ~{a} > O.
,,-a
Alternately, using the Green's function method, it is also known that the map-
ping function f(z) is related to Green's function Q(z, zo) of the rectangle nab
with a pole at Zo by
f(z) = exp{ -2 7l' Q(z, zo) + i 7t(z)}, (12.5.5)
- --+- -- ---- -- ~ - --- ---- +--------~ --------+--

• I I t I
,
,,
--"* ---------.- --------
t-- -~
I • I I I
-)IE- - - - - - - - - ...... - - - - - - - - -)(. - •
I , I , I
--+--------+----1- --- +--

I , I I I
-1-j-b/2 --------
I , I I
: : :--an-": :
- - ~ - - - - - - -- -t - - - - - - - -1If - - - - - - - - -t - - - - - - - - -¥ _.
I I , ,
, I I I
I , , ,
, I , ,
- - -~ - - - - - - - - -$< - - - - - - - - + -------- ~ --------+--

; I I , ,
Fig. 12.5.2.
where 7t(z) is the conjugate harmonic function of Q(z, zo). The method of
images can be used to express Green's function of nab as a double sum of
logarithm functions (see Kythe, 1996, p.81). In particular, at Zo = 0
Q(z, zo) = 2
1
7l'
f
m,n=-oo
(_I)m+n log IZ _1Zmn I' (12.5.6)
where Zmn = rna +i nb (Fig. 12.5.2). Since the conjugate harmonic function
oflog Iz - zmnl is arg{ Z - zmn}, we find from (12.5.5) and (12.5.6) that the
function f(z) that maps the rectangle nab onto U such that f(O) = 0 is given
by
00 IT (z - zmn)
f(z) = exp {m ~oo (_l)m+n log (z - zmn) = m+[ieven (z _ zmn) }.
, m+n=odd
(12.5.7)
As noted in Case Study 4.2.3, in the present case both f and the kernel function
K(z,O) have poles at all 'negative' images of the point Zo = 0 with respect
to the four sides of nab (these points are identified by an x in Fig. 12.5.2).
The poles at z = ±a and z = ±ib affect the convergence of the representation
(4.2.10) of K(z, 0), even when a = b. But their effect is more significant the
thinner the rectangle becomes, because in such cases (b « a) the distance of
the poles at ±ib from the boundary of nab gets smaller compared with the
dimensions of nab, Then the mapping function f(z) from (12.5.7) is given by
z
f(z) = (z2 _ a2) (z2 + b2) g(z), (12.5.8)
where g(z) is analytic in the region {(x, y) : Ix/al + Iy/bl < 3}. Also, since
from (4.2.25)
K(z,O) = JK(~,O) j'(z), 1'(0) = 0,
the set
where the prime denotes differentiation with respect to z, is best suited as the
basis set for both BKM and VM (see the next case study).•
CASE STUDY 12.5.2. Consider the rectangle na1 = {(x, y) Ixl <
a, Iyl ::; I} (Fig. 12.5.3).
CASE 1 (a =f. 1). Since the region has fourfold symmetry about 0, the odd
powers of z do not appear in the polynomial representation (4.2.20) ofthe kernel
function K(z, 0). Hence we take the basis set as {4>; (z)} = {z2(j-l)}N .
J=l
CASE 2 (a = 1). Since the region has eightfold symmetry about the origin
and the polynomial representation of K(z, 0) has only powers of z that are
multiples of 4, we take the basis set as {cP; (z)} = {z4(j-1)}N .
J=1
(-a, 1) (a,1)
D C
A B
(-a,-I) (a ,-1)
Fig. 12.5.3.
The augmented basis (AB) is obtained by adding to the above orthonormal

basis set the four singular functions that correspond to the four poles at z = ±2a
and z = ±2i. Because of the symmetry of the region, these four singular
functions are combined into two functions ( 2 Z 2 )' and ( - / - ) ' when
z - 4a z +4
a ::f. 1. In the case when a = 1, these singular functions simplify to a single
function ( 4 Z ) '. Hence the AB is given by
z -16
cP1(Z) = (z2 ~ 4a2 )', cP2(Z) = (Z2: 4)"

cPj+3 = z2J, j = 0,1, ... , when a ::f. 1;
cP1(Z) = (Z4~16)" cPj+2=z4

j
j=O,I, ... , when a = 1..
CASE STUDY 12.5.3. Consider the bean-shaped region D bounded by

the contour (Fig. 12.5.4)
r : {z: z = ')'(s) = ~ [0.2 cos s + 0.1 cos 2s - 0.1
+i (0.35 sins + 0.1 sin2s - 0.02 sin4s)], -71':::; s :::; 71'}.

The conformal mapping of this region D was found by Reichel (1985) who,
based on geometric considerations, predicted that the function f has a simple
pole at Z >:::: -0.61. Papamichael, Warby and Hough (1986) have shown that in
the neighborhood of the s-axis (= {( : ( = s + it, -7f ~ s ~ 7f t = O}) the
function f has (i) a simple pole at each of the points ZI = -0.650225813375
and Z2 = 1.311282520094; and (ii) a singularity ofthe form..jz - Zj, j = 3,4,
at the points Z3,4 = ±0.565672547402 =f 0.068412683544 i. Hence, for BKM
with augmented basis (AB), we take
'Ij;(z) = .!!:- { J..;z=z; - ..jZ3=Z4} ,

dz Z - ZI
and the AB consists of the functions
<Pl(Z) = ( _ Z )',
Z - ZI
. 1
<P3(Z) = ( _ Z )', <P4+j(Z) = Z1-, j = 1,2, ....
Z - Z2
0.75
Fig. 12.5.4.
12.6. DOUBLY CONNECTED REGIONS 345
12.6. Doubly Connected Regions
In Gaier's variational method we first determine the approximate function
H(z) = fn(z) _ ~ (12.6.1)

fn(z) z
by taking a finite series representation as
n
H(z) = L aj ¢j(z),
j=1
where {¢j(z)} is the basis set of functions in £2(n) which possess single-
valued indefinite integrals in n. This set is augmented by adding appropriate
singular functions to account for singularities on the boundary an = f 1 Ufo
and in Int (f 1) U Ext (f 0). In fact, since the variational problem to minimize
J
the integral
IIul1 2 = llu(zW dx dy, u E Je1(n), (12.6.2)
(see §4.2) has a unique solution Uo such that Uo is orthogonal to Je°(n), the
function H is related to Uo by
uo(Z)
H(z)=~. (12.6.3)
Let us denote
A(z) = log fn(z) - log z. (12.6.4)
Then H(z) = A'(z), and for each function ¢j E £2(n) which is continuous
on an, we have, in view of Green's formula (1.1.29),
(12.6.5)
The modulus M = r2/r1 of n is related to the function H by
log M = 2- {;. r ~ log Izi dz - IIHI1 2 }.

211' ~ Jan z
(12.6.6)
Gaier's variational method (VM) resembles the RM (§4.2) in many ways.

Letthebasisset{<pj(z)} E L 2(0) be such that <P1 E KO(O). LetK~\ m = 0,1,
denote the n-dimensional counterparts of Km(o), i.e.,
(12.6.7)
where En = span (<P1, <P2, ... ,<Pn). The setK~ is nonempty for n = 1,2, ... ,
and the n-dimensional problem corresponding to (12.6.2) is as follows:
PROBLEM I:: In the class K;(O) minimize Ilull, defined by (12.6.2),

over all U E K;(O).
As in Problem In, the following results hold for the above problem:
(i) the problem I: has a unique solution Uo;
(ii) the minimal function Un is orthogonal to K~(O); and
(iii) the sequence {un} ---+ Uo uniformly in 0, i.e., in view of (12.6.3),
Un(Z)
~ ---+ H(z), (12.6.8)
almost uniformly in 0 (i.e., there is mean convergence in every compact subset

of 0). Hence, lim Ilu n - uoll = 0. Let
n->oo
(12.6.9)
Note that hj -j 0. Also, since K~ (0) = {u E En : (H, u) = O}, the set

{h 1<Pj(z) - hj <P1 (z)}, j = 2,3, ... ,n, is the basis of KO(O). Thus, if we
take
n
un(z) = L Cj <Pj(z), (12.6.10)
j=l
then, since Un is orthogonal to KO(O) and (H, un) = 1, we obtain the linear
(n x n) system of equations
n
L {h 1(<pj, <Pi) - hi (<pj, <P1)} Cj = 0, j = 2,3, ... ,n,
j=l
n
(12.6.11)
LhjCj = 1,
j=l
which determines the coefficients Cj. Then, in view of (12.2.8), the formula
Hn(z) = ~~~~~ gives the n-th approximation of the function H(z) = A'(z)
and the n-th VM approximation of the mapping function In (z) == I (z)
(12.6.12)
Also, from (12.6.6)
M = exp {2~ (~ hn ~ log Izi dz -II Hn I1

2
) }, (12.6.13)
which gives the n-th VM approximation of the modulus M of the region n

(Mn gives an upper bound to M).
Let us assume that the mapping function In is normalized so that the region
n is mapped conformally onto the annulus A(p, 1). Then the density function
J-L(s) (see §1O.2) is related to the boundary correspondence function ¢n(s) by
(12.6.14)
where 1'1 = log M = - log p.
Let {¢;(z)} denote the orthonormal basis of L 2 (n). Then the function H
has the Fourier series expansion
00
H(z) = 2:,6j ¢j(z), (12.6.15)

j=1
where the Fourier coefficients are given by ,6n = (¢;, H ). Then the VM
follows the same five-step procedure explained in §4.2.2, which leads to the
n-th approximation
n
Hn(z) = 2:,6j¢;(z), ,6j = (¢j,H), j = 1,2, ... ,n, (12.6.16)
j=1
which, from (12.6.12), yields the n-th approximation In(z) of the mapping
function In(z).
The basis setis taken as the set {zj} ~-oo which is a complete set in L 2 (n).
But the use of this set results in the same kind of problems as in the RM and
BKM for simply connected regions. Due to the presence of singularities of the
function H in the complement of n and corner points on the boundary, this
basis set is augmented by adding singular functions related to each singular
behavior. Thus, in the neighborhood of a branch point singularity at Zj E an
the asymptotic expansion of the mapping function involves fractional powers
of (z - Zj), for which Lehman's theorem (§1O.2) is used to account for the
singularity problem and determine the augmented basis (AB). For a corner
point Zj E an the asymptotic expansion of H is given by (10.2.6) which
augments the basis by singular functions of the form
(12.6.17)
where r = k + lla., a. = plq > 0, k = 0,1,2, ... , and 1 :::; l :::; p. A

branch point singularity occurs if p i= 1. If the arcs at a corner point are
straight line segments, then the exponent k + lI a. in (12.6.17) is replaced by
lla., l = 1,2, ... (see (10.2.8)). Depending on the rational values of k + lla.,
':-00
the first few singular functions (12.6.17) are added to the basis set { zj}
to form the AB. Note that the singular functions in (12.6.17) for Zj E fa are the
same type as used for interior mappings and those in (12.6.17) for Zj E f 1 are
those used for exterior mappings of the simply connected regions discussed in
Chapter 10. The function in may involve logarithmic terms if the asymptotic
expansion (10.2.6) is valid, but these logarithmic singularities can generally be
ignored as they produce no serious computational problem.
CASE STUDY 12.6.1. Consider the doubly connected region n bounded

by a circle in a square, defined by
n: {(x,y):lxl<I,lyl<l}n{z:lzl>a, a<I}.
There are no corner singularities, so no AB is required. The region has eightfold

symmetry about the origin, and thus, the basis set is taken as z( _1)H1 (2 j +1),
j = 1,2, ....•
CASE STUDY 12.6.2. LetG a = {(x,y): Ixl < a,lyl < a} define a
square region. Consider the doubly connected region n as a square in a square
(square frame, Fig. 12.6.2) defined by
n= {G 1 n compl (Oa) , a < I}.

Let Zj denote the four corners of the inner square. Then the singular functions
associated with the branch point singularities at these corners are the functions
cPrj(z),j = 1,2,3,4, where
2l
r=k+
3, k=0,1, ... ,andl:Sl:S3. (12.6.18)
n
Since the region has eightfold symmetry about the origin, these four singular
functions cPrj(z) are combined into a single function
4
¢r(Z) = cPr1(Z) + L ei8j cPrj(Z), (12.6.19)
j=2
where the arguments OJ are chosen such that
Since OJ depend on the branches used in defining the functions cPrj (z), care
must be taken while constructing the singular functions of the form ¢r (z). The
AB is
- 2 4 5 7
Z
(_1)H1 (2j+l)
,J. -- 1 , 2 , ... ., cPr(z), r = 3' 3' 3' 3'·
y y
Zz Zt
a x
,-_-+_-+_ X 0
Z3 <4
Fig. 12.6.1 Fig. 12.6.2.
CASE STUDY 12.6.3. Let
Gab = {(x, y) : Ixl < a, Iyl < b} U {Ixl < b, Iyl < a}, (12.6.20)
and
G c = {(x,y) : Ixl < c, Iyl < c}.
Then consider the doubly connected region n which is a cross in a square (Fig.
12.6.3), defined by
n = Gc n compl ((;ab), a < c, b < c.
Let the eight comers A, B, a, D, E, F, G, H of the cross-shaped region
Gab be denoted by Zj, j = 1,2, ... ,8, respectively. The singular functions
associated with the branch point singularities at Zj, j = 1, . .. ,8, are given by
¢rj(z) = z2
1(1 l)r-l-; - ~ ,
J
as in (12.6.17), and r is defined by (12.6.18). In view of the symmetry these
eight singular functions can be combined into two functions
3
¢rj(Z) = ¢rj(Z) +L eilJ2k+j ¢r,2k+j(Z), j = 1,2, (12.6.21)
k=l
where the arguments 82k+j are chosen such that
e i7r 2
/ ¢rj (e i7r 2
/ z) = ¢rj(Z), j = 1,2.
as in Case Study 12.6.2.•
D C
E B
x
A
F
G H
Fig. 12.6.3.
12.7. PROBLEMS 351
12.7. Problems
PROBLEM 12.7.1. Let a be rational. Use (10.2.6), and show that the first
four functions "l/Jj in the formal asymptotic expansion (10.3.5) for the density
function J.L( 8) are given by:
"l/J1((j) = (j-l+1/"', 0 < a < 2,
(j1/""o<a<1,
"l/J2((j) = (j log (j, a = 1,
{ (j-1=2/"" < a < 2,
1
(jl+1/"', 0 < a < 1/2,
(j3 log (j, a = 1/2,
"l/J3((j) = (j-l+2/"' , 1/2 < a < 1,
{
(j1/"', 1::; a < 2,
(j2+l/"', 0 < a < 1/3,
(j5 log (j, a = 1/3,
(j-l+2/"', 1/3 < a ::; 1/2,
"l/J4((j) =
(jl+1/"', 1/2 < a < 1,
(j2 (log (j)2, a = 1,
(j-1+3/"', 1 < a < 2,
where, in particular, aT, j = 1,2,3, satisfy the relations
1
a--.x
1 - /"'a+1 , O<a<2, }
- -- - /\\ 2/'" a+
a2 2' 1< '" < 2,
-<-< ,
and
(80) I .
.x = ",'(sri)
a- - -.x /"'a+
3 -
2
3' 1/2 <_ a < 1 I
(Papamichael, Warby and Hough, 1986.)
PROBLEM 12.7.2. Discuss the nature of the singularity of the cardioid f o

in Case Study 11.2.1 when b2 = 5 (Fig. 11.2.1).
" 2 2
PROBLEM 12.7.3. Letfbeanarcoftheparabola(y - Yv) = 4a (x - xv),
a > 0, where Zv = Xv + i Yv denotes the vertex, and let the parametric equation
off' be
z = ,(s) = Zv + a [1 + (s + i)2], S1 < S < 82.
Let G* be taken as an appropriate subregion of the rectangle {( : ( = 8 +

it, 81 < 8 < 82, -1 < t < I}. Show that the conditions Cl and C2 of §lOA
are satisfied, and z = ,/,(() maps G* onto the region D* = D l u u D 2 . Also, r
if 0 E D l U ((WI \I') , then show that
(i) if Zv does not lie on the half-line I: {(x,y) : x < -a, y = O} and does
not coincide with the points (-a, 0) and (3 a, 0), then f has a simple pole at
the point ZQ = 2 i Yv - 4 a [1 - i (-zv/a - 1)1/2], where the square root is
chosen such that arg {(-)1/2} < 1r;
(ii) if Zv E I, then f has a simple pole at one or both of the points ZQ =
-4 a [1 ± V-xv / a-I] , depending on the values of 81 and 82 taken in the
equation Z = ,/,(8) of I';
(iii) if the focus of the parabola is at the origin, i.e., if Zv = -a, then f has a
double pole at the point ZQ = -4 a;
(iv) if Zv = 3a, then f has a singularity of the type (z - zQ)-1/2 at the point
ZQ = 4 a which in this case is the focus of the parabola. (Papamichael, Warby
and Hough, 1983.)
PROBLEM 12.7.4. Let I' be the right branch of the hyperbola
r
and let the parametric equation of be Z = ,/,(8) = Zv + a e cosh(8 + i q),
where Zv = Xv + i Yv is the center of the hyperbola, e = VI
+ b2 /a 2 , and
cos q = 1/e. Show that the conditions Cl and C2 of §lOA are satisfied by
taking G* as a symmetric subregion of the rectangle in Fig. 12.4.1; also, show
that
(i) if Zv does not lie on the half-line 1= {(x,y) . x < -ae y - O} and
does not coincide with Ihe poin~ (-0 e, 0) and ( ':: ~ ::, 0)', Ih: f has a
simple pole at the point
(ii) if Zv E I, then f has a simple pole at one or both of the points ZQ
v 2 b v
V
2 b2 x ± 2 i a b x 2 - a 2 - b2
, depending on the values of 81 and 82 taken
a + 2
in the equation Z = ,/,(8) of I';
12.7. PROBLEMS 353
(iii) if the focus of the hyperbola under consideration is at the origin, i.e., if
2b 2
Zv = -a e, then f has a double pole at the point Zo = - -r:::;<=::::::::;::;;:
ya 2 + b2
2 2
(iv) If Zv = b - a ,then f has a singularity of the type (z - ZO)-1/2 at
va 2
+ b2 2
the point Zo
va 2b
2
+ b2
which in this case is the focus of the hyperbola.
Fig. 12.7.1.
PROBLEM 12.7.5. Consider two parabolic arcs f 1 and f 2, defined by
arc ABC: {z: z = 0.6 + 0.4 a 2 - 8

2
+ 2 i a 8, -1 < 8 < I},
arcCDA: {z: z=-0.4-0.6a 2 +8 2 -2i8, -a<8<a, a>l},
respectively, which intersect orthogonally at the points A and C (Fig. 12.7.1).
Show that the mapping function f has (i) two simple poles with respect to
the arc CDA at the points Z1,2 = -4 ± 4iYO.6a 2 - 0.6, and Oi) a simple
pole with respect to the arc ABC at the point Z3 = 4 a (a - VO.6 a 2 - 0.6).
PROBLEM 12.7.6. Consider the region D bounded by two hyperbolic arcs
f 1 = arc ABC: {z : z = Xo - 2 cosh 8 +i (YO - sinh 8), 81 < 8 < 82},

f 2 = arc CDA: {z : z = -Xo + 2 cosh 8 + i (-Yo + sinh 8), 81 < 8 < 82},
where Xo = cosh 81 + cosh 82, Yo = (sinh 81 + sinh 82) /2 are the coordinates
of the center Zo of the arc ABC (see Fig. 12.7.2). Take 82 = 1, and choose
81 such that 4 tanh 81 . tanh 82 +1 = O. Show that f has simple poles at

the points Z1,2 = ±zo =F ~ (320 + 4 i J 5).
25 - (Papamichael, Warby and
Hough, 1983.)
(O,2),t2
G,-_-:-,:F_~E''''
2_/3 _ ,,"
-0.6
..':-:,-,,-;,_,;-,--:-:B . - - - - - - - " C
A (O,-2).l.4
Fig. 12.7.2. Fig. 12.7.3.
PROBLEM 12.7.7. Consider the quadrilateral in Fig. 12.7.3. Show that
the AB is <Pj = (_Z_)', j = 1,2,3,4; <Ps = 1; <P6 = y'z - ZE; <P7 = z;

Z - Zj
<Ps = z2; <P9 = z3; <P1O = (z - zEf/2; and <P10+j = z3+j, j = 1,2, ... ,
where ZE = 2 + i. Take the parametric equations of the sides as
AC: Z=(2+ ~)t+ZA' O:::;t:::;2,
CE: Z = - ~ (t - 3)2 e2i7r / 3 + ZE, 2:::; t :::; 3,
EG: Z = -(t - 3)2 + ZE, 3:::; t :::; 5,

GA: z=-2(t-6)i+ZA' 5:::;t:::;6,
and evaluate K(z, 0). (Levin, Papamichael and Sideridis, 1978; Papamichael and
Kokkinos, 1981.)
PROBLEM 12.7.8. Consider the octagonal region in Fig. 12.7.4. Since

this region has fourfold symmetry about the origin, show that the orthonormal
basis set is {z2(j-1)}N ,and the AB is given by
J=1
12.7. PROBLEMS 355
Take the parametric equations of the sides as
LA: z=2it 3 +Z1, -l:::;t:::;O,

3
AB: z=2t +z1, O:::;t:::;l,
BD: z=t 3zD+(1-t)3 zB , O:::;t:::;l,
DF: z = t 3zF + (1 - t)3 ZD , 0:::; t :::; 1,
FC: z = -2it 3 + Z2, -1:::; t :::; 0,
CH: z = -2t + Z2, 0:::; t :::; 1,
3
3
HK: z=t zK+(1-t)3 zH , O:::;t:::;l,
KL: z = t 3zL + (1-t)3 zK , 0:::; t:::; 1,
and determine K(z, 0). (Levin, Papamichael and Sideridis, 1978.)
G
H "
t>- , ~ .... . _. _.L .. --(I

(-10,0) 0: (10,0)
0'· •• A B
"A B(s,-)
K
(;t.: -2~,)
.••. (z},' -z},J
~~(O,-lO)
D It2
Fig. 12.7.4. Fig. 12.7.5.
PROBLEM 12.7.9. Consider the circular sector of radius 1 and angle 311'/2
(Fig. 12.7.5). Show that the AB is given by
¢5 -_ (z - zA )1/3., ¢6 = Zj ¢7 = (Z - ZA )
5/3
;
Compute K(z,O) by taking the parametric equation of the arc BCD as z =

ZA + e , 0 :::; t :::; 31r/2, and of the sides AB and DA as z = t +
it 3
(1 - 3t + 3t ) ZA and Z = (1 - 3t + 3t ) ZA - i (1 - t)3, 0 :::; t :::; 1,

2 2
respectively. (Levin, Papamichael and Sideridis, 1978.)
PROBLEM 12.7.10. Let Gab = {(x,y) : Ixl < a < 1,lyl < b < I}
denote a rectangular region. Consider the doubly connected region n which is
a rectangle in a circle (Fig. 12.7.6) and defined by
n = {z: Izi < I} ncompl (Gab).

If a i= b, the region n has fourfold symmetry. Show that the four singu-
lar functions cPrj (z) can be combined into two functions ¢rj (z) = cPrj (z) +
ei8j , j = 1,2, where r is defined in (12.6.18), and OJ are chosen such that
ei1r ¢rj (e i1r z) = ¢rj(z). If a = b, the region n has eightfold symmetry.
Show that for each value of r the four functions cPrj(z), j = 1,2,3,4, can
be combined into a single function of the same form as (12.6.19), and in each
case the monomial basis set can be taken as {z, z±(2 j +l)}, j = 1,2, ....
(Papamichael and Kokkinos, 1984.)
y y
B A
'--_-+--_-+_x
( 1\ x
C
"'-V D
Fig. 12.7.6 Fig. 12.7.7.
PROBLEM 12.7.11. Let Gab be the cross-shaped region defined by

(12.6.20). We shall consider the circle-in-a-cross region n (Fig. 12.7.7)
defined by
12.7. PROBLEMS 357
Letzj,j = 1, 2, 3, 4, denote the four comers A, B, C, Doftheouterboundary.

Show that the singular functions associated with the branch point singularities
at these points Zj are 1Jrj(z) = (z - Zjr-\ where r is defined in (12.6.18).
Use the symmetry to show that these four functions can be combined into a
single function ¢rj(Z) of the form (12.6.21), and that the AB is formed by
the monomial basis set z(-l)Hl (2j+1) plus the functions ¢rj(Z) with r =
2/3, 4/3, 8/3, 10/3. (Papamichael and Kokkinos, 1984.)
REFERENCES USED: Carleman (1916), Copson (1975), Gaier (1964), Lehman

(1957), Hough and Papamichael (1981, 1983), Levin, Papamichael and Sideridis
(1978), Lewy (1950), Lichtenstein (1911), Papamichael and Kokkinos (1981, 1982,
1984), Papamichael, Warby and Hough (1983,1986), Papamichael and Warby (1984),
Radon (1919), Reichel (1985), Warschawski (1932, 1955).
Chapter 13
Multiply Connected Regions
We shall discuss some existence and uniqueness theorems for the conformal
mappings of multiply connected regions onto canonical regions. The numeri-
cal method presented here is based on Mikhlin's integral equation formulation
on the boundary, which is a Fredholm integral equation of the second kind
and has a unique periodic solution.Then a numerical method, called Mayo's
method, that uses a fast Poisson solver for the Laplacian (Mayo 1984) is em-
ployed to determine the mapping function in the interior of the region which
can be simply, doubly, or multiply connected, with accuracy even near the
boundary. This method, in fact, computes the derivatives of the mapping
function in the first application and the mapping function itself if applied
twice.
As we have seen, most of the methods for conformal mapping compute

the boundary correspondence function only. Thus, the problem of finding the
mapping function in the interior (or exterior) is reduced to that of evaluating
certain integrals in the interior (or exterior) of the region. But these integrals
are often difficult to compute directly because of (i) the time consumed in eval-
uating the kernels at many interior (or exterior) points, and (ii) the difficulty
in evaluating the mapping function accurately near the boundary where it is
mostly needed since the kernels become unbounded at points near the bound-
ary due to the presence of pole-type singularities (see §lOA, 10.6). Mayo's
method presented here overcomes these two difficulties. In fact, this method
embeds the multiply connected region to be mapped into a rectangular region,
computes the integrals on the boundary of this rectangle, and finally evaluates
the mapping function at all points of the region to second order accuracy.
The choice of Mikhlin's integral equation reduces the conformal mapping

problem to that of a Dirichlet problem for simply connected regions. For
358

13.1. EXISTENCE AND UNIQUENESS 359
regions of higher connectivity it reduces the problem to a modified Dirichlet

problem which is an elliptic boundary value problem for which the boundary
values are prescribed only up to additive constants on all but one of the boundary
contours, and the constants are determined by the condition that the problem
must be single-valued in the region.
13.1. Existence and Uniqueness
Let n denote a multiply connected region of connectivity n + 1, n = 0,1, ... ,

such that a Jordan contour f o contains n Jordan contours f j , j = 1,2, ... , n,
in its interior and the origin is an interior point of f 1 (Fig. 13.1.1). The
connectivity is taken as n + 1 simply because the value of n tells the number
of 'holes' inside fo. The boundary of the multiply connected region shall be
n
denoted by f = on ( = Uf j) .
j=O
Fig. 13.1.1. An (n + 1)--eonnected region.
We shall define some canonical regions, besides the unit disk and the an-
nulus. A region with parallel cuts (slits) is understood to be a region obtained
from the extended complex plane <Coo by removing several mutually parallel
line segments inclined at an angle B to the positive real axis (we call them par-
allel finite cuts of inclination B). By a region with spiral cuts we mean a region
which is obtained by removing several logarithmic spirals from <Coo. Let a and
360 CHAPTER 13. MULTIPLY CONNECTED REGIONS
c be real constants. The equation
~{e-i<> logw} = c (13.1.1)
represents a logarithmic spiral in the w-plane with the origin as its asymptotic
point, where a is the angle between the logarithmic spiral and a fixed ray
emanating from the origin (a is known as the oblique angle of the spiral cuts).
For a = 0 the logarithmic spiral reduces to a ray arg {w} = c emanating
from the origin, and for a = 7r /2 it becomes a circle Iwl = e C (unit circle for
c = 0). The following theorem establishes the existence and uniqueness of
the conformal mapping from a multiply connected region onto a region with
parallel or spiral cuts. A result due to Hilbert is as follows:
THEOREM 13.1.1. Let n be a multiply connected region in the ex-

tended z-plane Coo and () a real number. Then there exists a univalent
meromorphic function w = fe(z) in n such that (i) it maps n con-
formally onto a region with parallel finite cuts of inclination () in the
extended w-plane; and (ii) it maps a given point z = a into w = 00, and
in a neighborhood of z = a the function fe(z) may be represented by a
series of the form
1
fe(z) = - -
z-a
+ al (z - a) + ... , (13.1.2)
or
fe(z) = z + -al + ... , (13.1.3)
z
according as a is finite or not. Each of these functions is unique for the
region n.
A similar theorem holds if the image of n has spiral cuts of oblique angle
a in the w-plane. The mapping functions are the same as (13.1.2) and (13.1.3)
which in this case are denoted by <> instead of fe. A proof of this theorem
can be found in Goluzin (1969, p.213) or Wen (1992, p.l18). Two results on
the existence and uniqueness of conformal mapping of a multiply connected
region n contained inside the unit disk Izi < 1 onto a region inside the unit disk
Iwl < 1 with concentric finite circular cuts and inside an annulus r < Izl < 1
are as follows:
THEOREM 13.1.2. Let n be a multiply connected region of connectiv-

ity (n + 1) inside the unit disk Izi < 1 where r = Izi = 1 is the boundary
component of nand 0 E n. Then there exists a unique, univalent ana-

lytic function w = f(z) in n such that (i) it maps n conformally onto a
region G inside the unit disk Iwl < 1 which has n circular cuts centered
at w = 0 and (ii) it maps the unit circle Izi = 1 conformally onto the
unit circle Iwl = 1 with f(O) = 0 and f(l) = 1.
THEOREM 13.1.3. Let n be a multiply connected region of connec-

tivity n + 1 inside the annulus r < jzl < 1 where r o = {Izl = I} and
r 1 = {Izl = r} are the two boundary components ofn. Then there exists
a unique univalent analytic function w = f(z) in n such that (i) it maps
n conformally onto a region G in the w-plane formed by removing n con-
centric circular arcs centered at w = 0 from the annulus p < jwl < 1,
where 0 < p < 1, and (ii) it maps the unit circle r o conformally onto
the unit circle Iwl = 1, and the circle r 1 onto the circle Iwl < p, with
f(l) = 1.
A region whose boundary consists of a finite union of circles is known as

a circular region. We shall consider the conformal mapping of a multiply
connected region n onto a circular region G. This can be accomplished by a
chain of two conformal mappings: (i) that of n onto a region G with parallel
cuts and (ii) that of G onto a circular region Ll. The former mapping is already
established in Theorem 13.1.1. We need to discuss only the latter mapping.
We shall state the uniqueness theorem for conformal mappings onto circular
regions; the existence of this mapping can be proved by the continuity method
(see Wen, 1992, p.118).
LEMMA 13.1.1. Let D be an (n + I)-connected circular region ob-

tained by removing n disks from the unit disk Izi < 1, with 0 E D, and let
Ll be an (n + 1) -connected circular region obtained by removing n disks
from the unit disk Iwl < 1, with 0 ELl. Ifw = f(z) maps D conformally
onto Ll such that (i) f(O) = 0, f(l) = 1, and (ii) f((j) = (j, j = 1,2,3,
where (j are three distinct points on Izl = 1, then f(z) = z.
This lemma establishes the identity mapping; its proof is available in Wen
(1992, p.1l8).
THEOREM 13.1.4. Let n be a multiply connected region in the ex-

tended z-plane. Then there exists at most one univalent meromorphic
function w = fez) in n, which maps n conformally onto a circular re-

gion G in the w-plane, such that the point z = 00 goes into w = 00, and
in a neighborhood of z = 00 the function fez) has the series expansion
(13.1.2).
PROOF. Suppose that w = hz) and w = h(z) are two univalent mero-
morphic functions in n, each of which satisfies the hypothesis of the theorem.
Then the function h 1
Ul
(w)), where z = f 11 (w) denotes the inverse of
w = h(z), is univalent and meromorphic in the region G, maps G onto an-
other circular region G' in the (-plane, maps the point w = 00 into the point
( = 00, and in the neighborhood of the point w = 00 has the series expansion
(13.1.4)
Now we must show that if ( = F( w) is the function that maps the region G
conformally onto G', such that F( 00) = 00, and has the series representation
(13.1.4) in the neighborhood of the point w = 00, then F(w) = w. In fact,
by using linear transformations of wand (, we can map the regions G and G',
respectively, onto circular regions D and f:j. of Lemma 13.1.1. The univalent
function obtained from these linear transformations satisfies conditions (i) and
(ii) of this lemma and thus represents an identity mapping. Hence, ( = F( w)
is a linear transformation with F( 00) = 00, and therefore, F( w) = a w + b.
But since F(w) has an expansion of the form (13.1.4) in the neighborhood of
w = 00, we require that a = 1 and b = 0, i.e., F(w) = w. It proves that
h Ul 1
(w)) = w, and hence, h(z) == h(z) .•
The function fo(z), defined by (13.1.2) or (13.1.3), can be evaluated for

arbitrary 0 from the equation
fo(z,a) = eiO [cosO fo(z,a) - i sinO f1r/2(z,a)]. (13.1.5)
The difference d(z) between the two sides ofEq (13.1.5) is regular in the region
n, and d( a) = O. Also, all values taken by d( z) on any contour r j lie on a circle
(in the extended sense) ~ {e- iO w} = c. The above equation also enables us to
compute the function fo(z, a) for arbitrary 0 if we know the functions fo(z, a)
and f1r/2(Z, a). To get these relations in a symmetric form, we set
"21
1
P(z, a) = [J1r/2(Z, a) - fo(z, a)] , Q(z,a) ="2 [f1r/2(z,a) + fo(z,a)] .
(13.1.6)
Since ~ Uo(z, an
= const, and ~ U1r/2(Z, an = constonrj,j = 0,1, ... ,n,
for fixed a E n, i.e., since
fo(z, a) = fo(z, a) + const, f1r/2(Z, a) = - f1r/2(Z, a) + const,

we have on r j, in view of (13.1.6),
P(z, a) = - Q(z, a) + qj(a), (13.1.7)
where qj (a) are independent of z E r j' Now, for fixed a, bEn, we set
1
P(z,a,b) = 2 [log f1r/2(Z, a, b) -log fo(z, a, b)] ,
(13.1.8)
1
Q(z, a, b) = 2 [log f1r/2(Z, a, b) + log fo(z, a, b)] .
Then
d ----
P;(z, a, b) = dz P(z, a, b) = P(b, z) - P(a, z),
(13.1.9)
Q~(z,a,b) = ddzp(z, a, b) = Q(b,z) - Q(a,z).
Since ~ {log fo(z, a, b)} = const, and ~ {log f1r/2(Z, a, b)} = const on r j ,
from (13.1.8) we find that
P(z,a,b) = -Q(z,a,b) + qj(a, b), (13.1.10)
where qj (a, b) are independent of z E r j' We also consider an integral on the

entire boundary r of the region n in the positive direction:
It = ~
Jrr P(t,z) P;(t,a,b) dt
i
2211"
= l: -.
n 1
2m r.3
P(t,z) dP(t, a, b)
t -2~ Jrr
j=O
= [-Q(t,z)+qj(z)] d(-Q(t,a,b) + qj(a, b))

211"
j=O 3
=t-2~ Jrr 211"

Q(t,Z)d(Q(t,a,b))
j=O 3
= -~
2211" Jrr Q(t,z)Q~(t,a,b)dt,
because the function Q(t, a, b) is single-valued on each contour f j . Since

Q(t, z) and Q(t, a, b) have simple poles in n at the point z and the points a
and b, respectively, by using the residue theorem, we get II = - Q'z (z, a, b) -
Q(a, z) + Q(b, z). But since II = 0, we obtain the formula
Q~(z, a, b) = Q(b, z) - Q(a, z). (13.1.11)
Similarly, if we consider the integral 12 = ~

2Z7r
1
r
Q(t, z) P!(t, a, b) dt and
follow the above technique, we obtain the formula
P~(z, a, b) = P(b, z) - P(a, z). (13.1.12)
Note that the functions P(z, a) and Q(z, a) are themselves not necessarily
analytic functions of a, as shown by taking the region n as Izi > 1. Then, in
this case
1 z-a 1 1- liz
P(z,a) = 1-la12 1- az' Q(z,a) = 1 - Ia 12 - -
z-a
In the next section we shall use the above formulas to solve the Dirichlet problem
and construct Green's function for the multiply connected region n.
13.2. Dirichlet Problem
Let n be an (n+l)-eonnected region in the finite z-plane, bounded by (n + 1)

Jordan contours f j , j = 0,1, ... ,n (Fig. 13.1.1). Let Uj(z), j = 0,1, ... ,n,
be a set of harmonic functions that have the boundary values Uj = fj j k on f j,
k = 0,1, ... ,n (i.e., the boundary value is 1 on f j and zero on fk, k =I- j),
n
and satisfy the relation L Uj(z) = 1 because the sum on the right side is
j=O
a harmonic function and it must be 1 everywhere on the boundary f = an.
The conjugate harmonic function Vj(z), j = 0,1, ... ,n, in general, is not
single-valued. Let [~Vj(z)]rk denote the increment in the function Vj(z) as
Z traverses the contour fk in the positive sense (marked by arrowheads in Fig.
13.1.1), and suppose that
(3.2.1)
where Pk,j are constants. Then the set of analytic functions Wj(z) = Uj(z) +
i Vj(z), j = 0,1, ... ,n, satisfies the following conditions:
(i)Wj(z) = - Wj(z) + Ck,j on rk> where Ck,j is a constant;

(13.2.2)
(ii) [~Wj(z)]rk = 2i7rPk,j;
(iii) wj (z) is regular in n.
Now consider the integrals
h = ~
2t7r Jrr P(t, u) wj(t) dt, 12 = -2~
t7r JrrP(t, u, v) wj(t) dt. (13.2.3)
In view of Cauchy's theorem, each integral is equal to zero. We shall use the
formulas (13.2.2), (13.1.7), and (l3.l.l0) to obtain
n
= -wj(u) + Lqj(U)Pk,j,
j=O
which gives
n
wj(u) = Lqj(U)Pk,j, k = 0,1, ... ,no (13.2.4)
j=O
Similarly, for the integral 12 by using integration by parts we get
1
h = - -.-
2t7r
1r
Q(t,u,v)wj(t)dt + Lqj(U,V)Pk,j
n
j=O
n
= - Wj(U) - Wj(V) +L qj(U, v) Pk,j,
j=O
which yields
n
Wj(V) - Wj(U) = L qj(u, v) Pk,j, k = 0,1, ... ,n. (13.2.5)
j=O
Formulas (13.2.4) and (13.2.5) express the solution of the Dirichlet problem for
a multiply connected region in terms of the functions Qj(u) and qj(u, v) which
define univalent mappings.
To find Green's function 9(z, zo) for the region n, let the corresponding
analyticfunction be denoted by F(z, zo) so that 9(z, zo) = ~ {F(z, zon. The
function F(z, zo) has a logarithmic singularity at the point Z = zo E n and is
not single-valued in n. The function F(z, zo) has the following properties:
(a) F(z, zo) = - F(z, zo)+const on f j ;
(b) [~F(z, zO)]r = -2i1T Wj(zo), where Wj is defined in (13.2.2);
J
(c) F' (z, zo) is regular in n except at a simple pole z = Zo with residue -1.
Now, consider the integrals
h =~
2t1T JrrP(t, u) F~(t, zo) dt,
and
14 = ~ r P(t,u,v)F~(t,zo)dt,
2t1T Jr
where u, v E n. Evaluating these integrals first by using the residue theorem
and then using (13.1.7), (13.1.10) and the above mentioned property (b), we
obtain
n
F~(u, zo) = Q(zo, u) + P(zo, u) - L qj(u) Wj(zo),
j=O
n
F(v,zo) - F(u,zo) = Q(zo,u,v) + P(zo,u,v) - Lqj(u,v)Wj(zo).
j=o
(13.2.6)
After separating the real parts in (13.2.6), we find that
n
9(v,zo) -9(u,zo) = log If"/2(ZO,U,v)j- Llogpj(u,v) 'Wj(zo),
j=O
(13.2.7)
where Pj (u, v) are the radii of the circles on which lie the images of the contours
f j for j = 0,1, ... ,n under the mapping W = f"/2(Z, U, v). Using the
13.3. MIKHLIN'S INTEGRAL EQUATION 367
symmetry property of Green's functions and replacing Zo by z in (13.2.7), we

also get
n
Q(v, zo)-Q(u, zo) = R {f7r/z(zo, u, v)}- L logpj(u, v)·Wj(zo). (13.2.8)
j=O
Together the two formulas (13.2.7) and (13.2.8) yield

n
F(z, v) - F(z, u) = log f7r/z(zo, u, v) - L log Pj(u, v)· Wj(zo), (13.2.9)
j=O
which is the analytic representation of functions that are meromorphic in the

region n and real on its boundary f. These functions are known as Schottky
functions.
13.3. Mikhlin's Integral Equation
As we have seen in §9.1, the function W = f(z) that maps a simply connected
region D with Jordan boundary f conformally onto the unit disk Iwi < 1, such
that a point Zo E D goes into the point W = 0, can be written in the form
f(z) = (z - zo) g(z), where g(z) is analytic and nonzero in D. The function
F(z) = logg(z) is also analytic in D. If ( E f, then If( ()I = I(-zollg«()1 =
1, and hence, R {F«()} = log Ig«()1 = -log I( - zol. The problem of finding
the function F(z) reduces to solving the Laplace equation with the Dirichlet
boundary value -log I( - zol, which determines R {F(z)}. This, followed
by finding the conjugate harmonic function, leads to determining g(z) = eF(z)
which finally yields the mapping function f(z). This formulation is the same
as in (9.1.1) in Symm's method.
In the case of a doubly connected region n

in the z-plane bounded by
the Jordan contours f o and f l , f l C f o, such that 0 E f l , let the function
W = fn(z) map n conformally onto the annulus A(p, 1). Since the function
fn(z) is bounded and nonzero in n, the function logfn(z) is nonsingular in
n. Let
fn(z)
Fn(z) = log - - = log fn(z) -logz, (13.3.1)
z
where the function Fn(z) is single-valued and regular in n. If z traverses f o
and f l in the positive direction, both arg {fn(z)} and arg{ z} increase by 27l".
Let g(z) = ~ {Fn(z)}. Since the contours f o and f 1 are mapped onto circles,
it is easy to find the boundary values of g(z). Since I/n(()1 = (for ( E f 1
and I/n(()1 = 1 for (E f o, we find that
g(z) = {IOgp- log 1(1. (E f b

(13.3.2)
- log (, (E f 0,
which is the same as (I 1.2.4). Thus, the value of p (conformal modulus) is

determined under the condition that the function h(z), conjugate to g(z), must
be single-valued, i.e., it must satisfy the condition (Il.2.8). The mapping
function In is then determined from (11.2.1).
We shall consider the mapping of a multiply connected region n onto the

slit unit disk by the function w = w(z). Let the region n be (n + 1)-eonnected,
n 2:: 2, and bounded by Jordan contours f j , j = 0, 1, ... ,n. We shall assume
that a point Zo E n is mapped into the point w = O. As in the case of a simply
connected region, the mapping function can be written as w(z) = (z- zo) g(z),
where g(z) is analytic and nonzero in n. Thus, if W(z) = log g(z), then
u(z) = ~ {W(z)} = -log I( - zol, ( E fo. Now, suppose that the contours
fk, k = 1, ... ,n, are mapped onto the circles jwl = Pk, i.e., Iw(()1 = Pk for
(E f k . Then
-logl(-zol, (Efo,
u(z) = { (13.3.3)
logpk-Iog!(-zol, (Efk,k=l, ... ,no
The problem of determining u(z) reduces to solving a Dirichlet problem, and

the mapping function w(z) can be determined from w(z) = ell(z)+iV(z), where
v(z) = ~ {W(z)} must be single-valued.
As in §11.3, we shall assume that the mapping problem can be solved in

terms of the integral of a dipole density function f.L( s) on the boundary f == an,
which is the real part of the Cauchy integral with density f.L, i.e.,
u(t) = ~ {~ r:(z) dz}

Jr ." - z
1
2z1l'
(13.3.4)
1
= -.- a logrst ds,
f.L(s) -;::;-
2z1l' runs
where z = x + iy = I(t), rst = /( - zl, and (= I(S), 0 ~ s,t ~ L (as in

(I 1.3.1». Note that ifn is simply connected, then we have an ordinary Dirichlet
13.3. MIKHLIN'S INTEGRAL EQUATION 369
problem with boundary data J.L(() = -log Zo I(- I, (

E rand Zo E n, and the
density function J.L( s) is obtained by solving the integral equation
J.L(t) + -1
7r
1 J.L(S) ~
runs
f) logrst ds = -2 log I( - zol· (13.3.5)
But in the case of a multiply connected region a modified Dirichlet problem

is solved, where we must determine the density function J.L(s) as well as the
radii Pk which appear in (13.3.3). Let J.L(t) denote the solution of the integral
equation (Mikhlin, 1957)
J.L(t) +~ l [J.L(S) f)~s logrst - X(s, t)] ds = -2 log I( - zol, (13.3.6)
where
I if s, t lie on the same contour,
,
X(s t ) = {
° otherwise.
The radii (conformal moduli) Pk are given by
Pk =~
7r
rJ.L(s)ds.
Jr
(13.3.7)
Eq (13.3.6) is known as Mikhlin's integral equation. It is a Fredholm integral

equation of the second kind whose kernel is given by
f) 2
M(s,t) = ~logrst E L [O,L],
un s
which is bounded, because we have M(s, t) = 0.5 K:(s) for s, t E r, where

K:(s) is the curvature of the contour at s.
The numerical solution of Eq (13.3.6) is obtained by using a Nystrom

method (quadrature) with the trapezoidal rule (see, Atkinson, 1976; Delves
and Mohamed, 1985; Mayo, 1986)
where h is the mesh size, d (ti) = -2 log Iti - sol,

and So = ')'(zo). The mesh
points are taken as equispaced points with respect to some boundary parameter,
but the points used as nodes to compute the above quadrature are independent of
the mesh points which are used for a fast Poisson solver (see next two sections).
Since the trapezoidal rule is highly accurate on periodic regions, the accuracy of
the solution of Mikhlin's equation is the same as that ofthe quadrature formula
(13.3.8). This equation can also be solved by the methods developed in chapters
9 and 11, and although the equation, in general, is not symmetric, it has positive
real eigenvalues (Kellogg, 1929).
13.4. Mayo's Method
After the density function j.L( s) is computed by the quadrature formula (13.3.8),
we must still compute the Cauchy integral
W(t) =~
2n
J j.L(t) dz
z- t
at points in the interior of the region n. For this purpose a fast Poisson solver,
developed by Mayo (1984), is used (see details in the next section). This is
accomplished in the following two steps:
STEP 1. Compute
~{W(t)} = 2~ J M(s,t)j.L(t)dt. (13.4.1 )
STEP 2. Compute 8' {W(t)}.

n
The details of step 1 are as follows: Embed the region in a larger region R
which is a rectangle with uniform mesh in both x and y directions. There exists
a fast Poisson solver for the Laplacian in a rectangle (§ 13.5). Then the function
~ {W(t)} defines another harmonic function it at points in R\n. The function
it is a discontinuous extension of u from n into R\n. Define a function
U(t) = { u(t), tEn,

it(t), t E R\n.
Then we use the fast Poisson solver to compute an approximate solution of
the discrete Laplace equation '\l2U = 0 at all mesh points of R. Since both u
and it are harmonic, we set the Laplacian zero at those mesh points that have
all four of their adjacent mesh points on the same side of the boundary. But
to approximate the Laplacian at all remaining (irregular) mesh points, we take
the following approach: Since both u and it are continuous along the normal
13.4. MAYO'S METHOD 371
direction but have a jump equal in magnitude to the density p" we evaluate the
u
jumps in the derivatives of u and along the coordinate directions:
x'(s) y'(s)
Ux = P, S x'(s)2 + y'(s)2' u y = p, S x'(s)2 + Y'(S)2'
A '( ) A '( )
Ux - uy -
(13.4.2)
where the suffix indicates the variable of partial differentiation, and the prime
denotes the derivative with respect to s. Note that these derivatives contain
derivatives of p, and those of the boundary contours. Therefore, these jumps
are used to approximate the discrete difference operators at irregular mesh
points (see §13.5 for details).
-------- ~ R\n
'"
PN
n
'\2
'\p*
Pw P hI PE
Ps
/ h
Fig. 13.4.1.
As an example, suppose that a point P is inside n, but its adjacent neighbor

to the right, PE, is not inside n. Let p* denote the point where the grid line
between P and PE cuts the boundary, and let h2 = Ip* -
PEl (Fig. 13.4.1).
Using the Taylor series at P and PE, we find that
u (PE) - u(p) = [u(p*) - u(p*)] + h2 [ux(P*) - ux(P*)]

+ ~~ [uxx(P*) - uxx(P*)] + hux(p) + ~2 uxx(p) + O(h3 )
h2
= {I)3)} + hux(p) +"2 uxx(p) + O(h 3 ),
(13.4.3)
where { E (3) } denotes the sum of the first three terms on the right side which
are known quantities and can be expressed in terms of the solution of the integral
equation and the boundary data, whereas the remaining terms account for the
usual Taylor series. We also obtain the same kind of expression as (13.4.3)
for the difference U (p) - U (pE) except that there may be no boundary term.
Thus, an approximate solution of the discrete Laplacian of U can be computed
as the sum of the four difference operators at all mesh points (details in the next
section). The boundary data of U are obtained by approximating the values of
the integral (13.4.1) at mesh points that lie at the edge of R.
In step 2, the values of ~ {W(t)} = v(t) are easy to compute because, in

view of the Cauchy-Riemann equations, we can express the discontinuities of
v in terms of the discontinuities in u. Thus, the discrete Laplacian of v can be
computed easily.
Once the discrete Laplacians of u and v are computed, we apply the fast
Poisson solver twice to obtain the values of u and v at the mesh points. Thus,
the solution of (13.3.8) has second-order accuracy in h.
CASE STUDY 13.4.1. For Cassini's oval (Fig. 9.2.1) where two different
paths of the boundary are close to each other, the method described above in
step 1 may fail to give very accurate results because the kernel becomes very
large there. In this case (and others like it) the kernel is integrated exactly, i.e.,
where ¢ = arg{W(t)}. Then Eq (13.3.8) yields the system of equations
(13.4.4)
" t·
where ¢ = tan - 1t+1-
' J is the angle between the lines joining the points
ti-1 - t j
ti+1 and ti-1 to the point t j .
Mayo (1986) has considered the case of Cassini's oval
whose polar equation is r(8) = Jc2 cos 28 + a4

r( 8) cos( 8), y( 8) = r( 8) sin( 8), with c = 0.1 and a
J -c2 sin 2 28, x(8)
= 0.43 (Fig. 13.4.2).
13.5. FAST POISSON SOLVER 373
Note that all Cassini's ovals in Fig. 9.2.1 (Case Study 9.2.1) have c = 1.
.4
0.2
-0.2 0.2 x
-0.2
- .4
Fig. 13.4.2. Cassini's oval. Fig. 13.4.3.
The method described above is used for this contour with 90 mesh points
equispaced with respect to e. The exact mapping function is known (Symm,
1966, p.256). The simply connected region Int (r) is embedded in the unit
square. The maximum error found was 0.54 x 10- 2 with h = 1/32, and
0.24 x 10- 2 with h = 1/64.•
CASE STUDY 13.4.2. Consider the triply connected region bounded by the
three circles f o = {Izl = 0.35}, f 1 = {Iz -0.141 = 0.08, f 2 = {Iz +0.141 =
0.08} (Fig. 13.4.3). A total of 180 mesh points on the boundary f with mesh
size h = 1/128 were taken to solve Eq (13.3.8). The region n and its image
with the images of the grid lines, together with the unit circle and the slits, are
given in Mayo (1986).•
13.5. Fast Poisson Solver
Now we shall discuss the details of the fast Poisson solver for the Laplacian.
Consider the integral equation (13.3.6) which we write as
f.L(t) + ~
7r
r M(s,t) f.L(s) ds = 2g(t),
Jr
t E f, (13.5.1)
where the kernel M(s, t) is bounded and represents the normal derivative of
Green's function for the Laplacian in the plane. In the region R\n we define a
harmonic function U, by using the same formula as (13.3.4) in the form
u(t) = 2~ l M(s, t), fJ-(s) ds. (13.5.2)
The function u is a discontinuous extension of Uin the region R\n. Let (Xi, Yj)
denote the mesh points of the rectangle R, and let U be defined on R by
if (Xi, Yj) En,

(13.5.3)
if (Xi, Yj) E R\n.
Since u and uare both harmonic, a five-point discrete Laplacian defined by
will be zero (up to terms of second order) at those mesh points of R whose four
adjacent neighbors are on the same side of the boundary. Let S denote the set
of irregular mesh points. At these points although the analytic Laplacian of u
u
and is each zero, the discrete Laplacian \7~ U is not zero. The central idea
for solving the conformal mapping problem is to compute \7~ U at the points
u
in S, evaluate values of on oR, and then apply the fast Poisson solver on R.
Now, it is possible to approximate this discrete Laplacian without explicitly

u
solving for u or anywhere because we need only compute the jump disconti-
u
nuities between u and and those in their derivatives at the boundary in terms
of the density fJ-. The procedure to compute these jump discontinuities is as
follows: Since the discontinuity between the tangential derivatives of u and at u
a point on the boundary is equal to the value of the density at that point, we have
Us - Us = fJ-'(s}. Also, there is no discontinuity between their normal deriva-
tives, i.e., ~u = ~u . These two results and the knowledge of the direction
un s un s
of the contour lead to the formula (13.4.2) which computes the discontinuities
between U x and Ux and between u y and Uy. Higher order derivatives of u and
u can be obtained by differentiating (13.4.2).
Since the function U is the real part of the Cauchy integral with the same
density function fJ-, we shall consider
Wet) = ~
2211"
r fJ-«() de.
Jr (- z (13:5.4)
13.5. FAST POISSON SOLVER 375
Since the kernel in Eq (13.5.1) is the real part of the kernel in (13.5.4), we have
~ {_I d(/dS} = _1 y'(s) [x(s) - x(t)] - x'(s) [y(s) - y(t)] ds

2i7r ( - z 2pi [x(s) - x(t)J2 + [y(s) - y(t)J2
1
= - M(s, t) ds,
27r
(13.5.5)
where ( = ,(s) = x(s) + iy(s) and z = ,(t) = x(t) + iy(t).
Thus,
u(z) = ~ {W(t)} for zED, and u(z) = ~ {W(z)} for z E R\D. Then
u
the jump between u and can be computed from the jump discontinuities of
Cauchy integrals across the boundary contour (recall that Cauchy integrals are
analytic functions). Thus, for example,
J.L(z) + ~ J (J.L(() de, if z ---> f _,

~7r r - z
2 W(z) = 1 J.L(() (13.5.6)
{ -J.L(z) + -:- J- de, if z ---> f +,
~7r r ( - z
where z ---> f ± stands for whether z approaches from inside D or from outside
D. Hence, there exists a discontinuity of magnitude J.L(z) in W as z crosses f.
Since j..£(z) is a real function, we find that u(z) - u(z) = J.L(z) if z E f.
The discontinuities in the first and second derivatives of W(z) can be com-
puted as follows: Since by integration
~ W(z) = _1 (~J.L(() d( = _1 (J.L'(() d(

dz 2i7r Jr dz ( - Z 2i7r Jr ( - z '
we find that the derivative of a Cauchy integral with density J.L is another Cauchy
integral with density J.L', and thus, W'(z) has discontinuity of magnitude J.L'(z)
as z crosses f. Also, since W(z) is analytic and ux(z) = ~ {W'(z)}, we have
{dJ.L/dS} J.L'(s) x'(s)

ux(z) - ux(z) = ~ {J.L (z)} = ~ dz/ds = x'(s)2 + y'(s)2'
A ,
(13.5.7)
J.L'(s) y'(s)
x'(s)2 + y'(s)2'
A ,
Uy(Z) - Uy(Z) = -~ {J.L (z)} =
and since ~ {W"(z)} = uxx(z), we have
uxx(z) - uxx(z) = ~ {J.L"(Z)}, Uyy(z) - Uyy(z) = -~ {J.L"(Z)}.
These discontinuities can be used to approximate the discrete Laplacian at mesh

points near the boundary.
An approximation of the discrete Laplacian of U at points of the set Scan

be computed as follows: If we consider a point PE to the right of a point pEn
(Fig. 13.4.1), we find that the difference U (PE) - u(p) is given by (13.4.3), of
which the first three terms { 2::(3)} can be computed in terms of the density
function J-L and the distances of the irregular mesh points from the boundary.
Now, if PW is the mesh point to the left of pEn, then
h22
-hux(p) + 2 uxx(p) + 0 (h 3 ), ifpw En,
U (pw)-U(p) = h2
{
{2::(3)} - hux(p) + 2
2 uxx(p) + 0 (h 3 ) , rt
if PW n.
(13.5.8)
Thus, in either case, from (13.4.3) and (13.5.8) we find that
Similarly, ifpN andps are points above and below p, respectively (Fig. 13.4.1),
then
Hence, \7 2 u(p) = uxx(p) + Uyy(p) = 0 yields

(13.5.9)
By an analogous argument, if p is in R\n, then \7 2 u(p) = 0 will also lead to

formula (13.5.9). This formula gives second-order accuracy in approximating
the discrete Laplacian of U at points of the set S. If we want to reach fourth-
order accuracy in this approximation at points of S, we must use the fourth
order Taylor series expansion. Then, for example, at the point p E we will have
U (PE) - u(p) = [u(p*) - u(p*)] + h2 [ux(P*) - ux(P*)]

+ ~~ [uxx(P*) - uxx(P*)] + ~~ [uxxx(P*) - uxxx(P*)]
h2 h3
+ hux(p) + 2 Uxx(p) + (3 Uxxx(p) + 0 (h 4 )
h2 h3
= {~)4)} + hUx(p) + 2 uxx(p) + (3 uxxx(p) + 0 (h 4 ) ,
where { 2::(4)} is the sum of the first four terms on the right side in the above
expression. Then
(13.5.10)
13.6. PROBLEMS 377
This shows that if the solution of the integral equation is known almost accu-
rately, we can can compute an approximate solution with second order accuracy.
Mayo's method solves Mikhlin's integral equation with machine accuracy be-
cause of the use of trapezoidal rule in the quadrature formula with smooth
boundary data. However, splines can be used to compute more accurate values
for the derivatives of the density function. It has been found that in practice
second-order accuracy is sufficient to obtain an accurate solution.
The computational algorithm consists of the following steps:

STEP 1. Embed the region n in a rectangle R. This rectangle is chosen at
least 3h distance away from r.
STEP 2. Find all irregular mesh points and their distances to the boundary in
the x and y directions.
STEP 3. Solve the integral equation by using the quadrature formula (13.3.8),
which replaces the integral equation by a sum at a set of boundary points. This
yields a dense linear system of equations
J.1, (t i ) +L Wi K(i,j) P, (t j ) = 29 (t i ), i = 1, ... , n, (13.5.11)
where the points used as nodes are different from the mesh points. These nodes
are chosen as equispaced points with respect to the parameter used on the
boundary, and the trapezoidal rule is used for quadrature. (in cases where the
boundary data is not smooth, or for points near those boundary portions where
the curvature is large, a Galerkin method with augmented bases containing
singular points is needed (see §12.5). System (13.5.11) is solved by the Gaussian
elimination method.
STEP 4. Interpolate the values of the density with a quintic spline which yields
sixth order accuracy for values of the density at intermediate points.
STEP 5. Compute the discrete Laplacian at irregular points in the set S by
using (13.4.3).
STEP 6. Compute the values of U at the edge of the grid.
STEP 7. Apply the fast Poisson solver.
STEP 8. Compute the derivatives U x and u y by (13.5.7).
STEP 9. Compute the conjugate function v(z).
13.6. Problems
PROBLEM 13.6.1. Let A denote a family of univalent meromorphic func-

tions of the form f(z) = z + al + ... on the exterior region Izi > R,
z
o< R < 00. Let U(J) = ~ {e- 2i8 ad, f E A. Show that there exists a
r 2 e- 2i8
unique function fo(z) = z + E A such that U(J) attains the max-
z
imum Uo(J) = R , and w = fo(z) maps the region Izl > R conformally
2
onto the slit w-plane obtained by removing the line segment joining the points
±2Rei8 . (Wen, 1992, p.102.)
PROBLEM 13.6.2. Suppose that the function f(z) defined by (13.1.3) is

univalent and meromorphic in the exterior region E = {izi > R, 0 < R <
oo}. Show that If(z) - aol :::; 21z1 on E. (Wen, 1992, p.l03.)
PROBLEM 13.6.3. Let A denote the family of univalent meromorphic

functions of the form f(z) = z + al + ... on the exterior region Izl > R,
z
o < R < 00. Let U(J) = ~ {e- 2i1r8 ad, f E A. Show that there exists a
R2 e- 2i1r8
uniquefunctionfo(z) = z+ E A such that U(J) attains a maximum
z
Uo(J) = R 2 and w = fo(z) maps Izi > R conformally onto a region obtained
from the extended w-plane with a cut along the line segment joining the two
points ±2Rei8 . (Wen, 1992, p.l02.)
PROBLEM 13.6.4. Show that the analytic function F(z, zo) defined on a
multiply connected region n such that ~{F(z, zo)} is Green's function for n
is not single-valued in n. (Goluzin, 1969, p.286.)
PROBLEM 13.6.5. Derive (13.4.3) by writing the Taylor series expansions

of u about p evaluated at p*. (Mayo, 1984, pp.289-290.)
PROBLEM 13.6.6. Consider the triply connected region bounded by n

fo = {Izl =
0.35}, which is a circle, and f 1 and f 2 which are ellipses
x 2 /a 2 + y2/b 2 = 1 with a = 0.06, b = 0.13 and centered at (0.17,0) and
(-0.17,0), respectively. By embedding the region in a unit square and using
the mesh size h = 1/128 and 180 mesh points on the boundary, plot the images
of the grid lines and the image region with the slits. (Mayo, 1986, p.152.)
REFERENCES USED: Atkinson (1976), Goluzin (1969), Kellogg (1929), Mayo

(1984, 1986), Mikhlin (1957), Symm (1966), Wen(1992).
Chapter 14
Grid Generation
Exact solutions of boundary value problems for simple regions, such as a circle,
square or annulus, can be determined with relative ease even in cases where
the boundary conditions are rather complicated. Although Greens functions
for such simple regions are known, the solution of a boundary value problem
for regions with complex structures often becomes more difficult, even for
a simple problem, such as the Dirichlet problem. One approach to solving
these difficult problems is to conformally transform a given region into the
simplest form. This will, however, result in change not only in the region
and the associated boundary conditions but also in the governing differential
equation. Grid generation methods using conformal mappings are presented
for problems dealing with a cascade of blades, and inlet flow configurations.
14.1. Computational Region
Conformal mapping has been used to generate orthogonal boundary-fitted

coordinates in solving various boundary value problems in simply connected
regions. A useful work in this area is the book by Thompson, Warsi and
Mastin (1985). A grid is an integral part of finite difference or finite element
methods. A discrete model becomes more efficient when it is constructed
by using natural coordinate systems and maintaining a uniform connectivity
pattern between grid nodes. These two requirements are met when the grid is
obtained by coordinate transformations using conformal mapping methods so
that the boundary of the physical region is represented by constant coordinate
379

380 CHAPTER 14. GRID GENERATION
lines. Besides this adaptive feature, the conformal maps can be made to adapt
to certain salient features, such as singularities. A grid generation methodology
must be able to control the grid spacing effectively, especially near the boundary
(see Tamamidis and Assanis, 1991).
Computational methods, like the finite differences or the finite elements,

for solving boundary value problems are usually simple if the physical region
has regular geometry over which a uniformly distributed grid can be imposed.
However, if the region has arbitrary irregular geometry, such a region is first
transformed into an associated computational region with regular geometry,
like a rectangle or circle. In such cases the difficulty arises not only from
the transformation of the governing equation(s) but also from the boundary
conditions. The coordinate transformation and conformal boundary maps are
generally used to transform an irregular physical region into the corresponding
computational region. But such transformations and conformal mappings, in
general, are very difficult to construct except in relatively simpler cases.
First, we shall gather some transformation formulas from the physical

(x, y)-region into the computational (~, 1] )-region.
14.1.1. Coordinate Transformations. To find the transformation

from independent variables x, y of the physical plane into a set of independent
variables ~, 1] of the computational plane, let us assume that
~ = ~(x, y), 1] = 7](x, y), (14.1.1)
or inversely,
x = x(~, 7]), y = y(~, 7]). (14.1.2)
The Jacobian J of the transformation is given by
(14.1.3)
.where the subscripts denote partial differentiation with respect to the indicated
variable. If u = u(x, y), then
au
ay = J1 (
-x'TJ
au au)
a~ + x{ a1] . (14.1.4)
For the gradient, the transformation formulas for the conservative form are
14.1. COMPUTATIONAL REGION 381
and for the nonconservative form are
(14.1.6)
Let U = Ul i + U2 j. Then for divergence the transformation formulas for the

conservative form are
(14.1.7)
and for the nonconservative form are
CASE STUDY 14.1.1. Using formulas (14.1.4), the equation of continuity
8u 8v_
8x+8y- O, (14.1.9)
when transformed from (x, y)-coordinates of the physical plane into (~, "7)-
coordinates of the computational plane, becomes
8u 8u 8v 8v
y - - y{ - - x - + x{ - = O.• (14.1.10)
TJ 8~ 8"7 TJ 8~ 8"7
82 82
CASE STUDY 14.1.2. The Laplacian \72 == 8x 2 + 8 y2 is transformed
into the following forms:
Conservative form:
J\72U={~YTJ [(YTJU){-~ (y{U)TJ] -xTJ [-(XTJU){+(X{U)TJ]L
+ {-~ y{ [(YTJ u){ + (y{ u)1j] + ~ xTJ [- (XTJ u){ + (x{ U)TJ] L·
Nonconservative form:
\72 u = ;2 [(x~ + y~) U{{ - 2 (X{ xTJ + y{ YTJ) u{TJ + (x~ + yD UTJTJ ]
+ [(\72~) u{ + (\7 2"7) U TJ ] .•
Next, to present the basic concept of numerical grid generation for different
boundary value problems, we shall consider a very simple one-dimensional
transformation and show how the computational region with uniformly dis-
tributed grids is obtained and how the governing equations are changed.
CASE STUDY 14.1.3. Consider a plane steady state boundary layer flow
over a flat rectangular plate {O ~ x ~ a, 0 ~ y ~ b}. If this problem is
solved by the finite difference or finite element method, a rectangular grid is
constructed over the physical region with more nodes concentrated near the
wall (x-axis) where the gradients are assumed to be larger than elsewhere (see
Fig. 14.1.1). This grid is uniform along the x-axis but nonuniform along the
y-axis.
(a) (b)
Fig. 14.1.1. (a) Physical region, (b) Computational region.
In order to transform the grid in Fig. 14.1.1(a) into the uniform grid of Fig.
14.1.1 (b), we use the coordinate transformation
_ 1 _ In ¢(y)
~ =x, 'TJ - In A '
(14.1.11)
where
A= a+l, 1 < a < 00. (14.1.12)

a-I
The inverse transformation is given by
(a+l)-(a-l)A l -7)
x =~, y=b 1 + A!-7) . (14.1.13)
This transformation (Roberts, 1971) makes the grid spacing uniform along the
y-axis. The parameter a is known as the stretching parameter. The grid
14.1. COMPUTATIONAL REGION 383
concentration as a --> 1 is presented in Fig. 14.1.2 where values of y are

plotted for different values of a and fI with b = 1.
0.8
11 = 0.8
0.6 ( 11 =0.6
~
0.4
11 = 0.4
0.2
11 = 0.2
~ a
0.2 0.4 0.6 0.8 1.2
Fig. 14.1.2. Grid concentration as a --> 1.
Next, we shall transform the differential equation from the physical region
into the computational region by the transformation (14.1.11). For brevity, let
us consider the equation of continuity (14.1.9). Since
~x = 1, ~y = 0, fix = 0, fly =
2 [
b I: A a
1
- ( 1- t) 2]-1
(14.1.14)
for the geometry of the plate, the continuity equation is transformed into
au av (14.1.15)
8~ +'I7y 8'17 = 0,
where fly is defined in (14.1.14). Note that the effect of uniformizing the grid
makes the governing equation rather complicated compared to the original form.
Moreover, after the problem is solved in the computational region, the solution
is transformed back to the physical region by using (14.1.11).•
14.1.2. Orthogonal Method. The use of conformal mappings to

generate grids has some important limitations: (i) they are applicable to plane
problems, (ii) they have no control over the interior grids, (iii) multiple-valued
mapping functions are difficult to implement, (iv) orthogonality is lost in arbi-
trary distribution of boundary points, (v) a very small change in the shape of the
original boundary results in changes in the location of image boundary points,
and (vi) finding a boundary map is in itself a difficult task.
y 11
(a, b)
1-----------',
w =f(z)
D
-I-----------L-x
o o
(a) (b)
Fig. 14.1.3.
A consequence of the Riemann mapping theorem is that a rectangle R =

{O ~ x ~ 1, 0 ~ y ~ b}cannotbemappedunivalentiyontoanarbitraryregion
D with four sides, as shown in Fig. 14.1.3, unless the ratio alb is restricted to
a particular constant m, known as the conformal module. Then the mapping
W = f(z) can be constructed by solving the partial differential equation
m 2 w{{ +wT/T/ -0
- . (14.1.16)
This yields a quasiconformal map (see Lehto and Virtaanen (1973). Since m
is domain-dependent and not known a priori, this approach is not feasible for
grid generation. On the other hand, orthogonal transformations in the plane
can be regarded as quasi-eonformal mapping with a real dilation (Knupp and
Steinberg, 1993).
14.2. Inlet Configurations
We shall develop the orthogonal grid generation method by using conformal

mapping. In this method the physical region is mapped onto the computational
method by one- or two-step conformal maps. First, we shall consider the grid
generation for a case study involving inlet configuration in flow problems and
determine the basic mappings by using both one-step and two-step methods.
CASE STUDY 14.2.1 (One-step method). Consider the physical region

in the z-plane in Fig. 14.2.1(a). The boundary of this region, defined by
14.2. INLET CONFIGURATIONS 385
Y = Yl(X) and Y = Y2(X), is given by two sets of data:

zi n) = x~n) + iy~n), 2:::; n:::; nl -1,
(14.2.1)
z~n) = x~n) +iy~n), 2:::; n:::; n2 -1,
where X2 takes the minimum value X~3 at n = n3, i.e., at the point E. This
region is mapped onto the computational region in the ( = (~, 1])-plane, which
is a rectangular strip 0 :::; 1] :::; 1 (Fig. 14.2.1).
11
A
.. _ .. ,~
~
C=--D F
~C
B
--
z=!(U
o
(a)
(b)
Fig. 14.2.1. (a) Physical plane (z-plane);

(b) Computational plane ((-plane).
The mapping function is given by
z = h [( - ~ (1 + e- 7r
()] + A1 + i B1
~ [A . (j -1)11"(( - ~o) + . B (j - 1)11"(( - ~o)]
+~ j sm L 2 j cos L ' (14.2.2)
j=2
where h, A j , B j (j = 1,2, ... ,11:), ~o and L are yet unknown real constants
to be determined. A particular case from (14.2.2) for A j = B j = 0, j =
1,2, ... ,II:, is given by
(14.2.3)
which maps the upper half-plane Y ~ 0 with a cut at y = h, x ~ 0, onto the

rectilinear strip 0 :::; 1] :::; i in the (-plane (Fig. 14.2.1). If we rewrite (14.2.2)
as
x + iy = h [(~ + i1]) - ~ (1 + e- ({+iT/»)] + A

7r
1 + iB1
~[A . (j-1)1I"(~+i1]-~O)'B (j-1)1I"(~+i1]-~O)]
+~ j sm L + 2 j cos L '
j=2
(14.2.4)
set TJ = 0, and equate real and imaginary parts, we get
(14.2.5)
and
Y = Yl = B 1 + ~B
~ j cos
(j-1)7r(~+iTJ-~o)
L . (14.2.6)
j=2
y Tj
hl=======
-'---'-":Io~---'---'---X
z-plane ~-plane
Fig. 14.2.2. The map z = h [( - ~ (1 + e- 7r

()].
Again, if we set TJ = 1 in (14.2.4) and equate real and imaginary parts, we

obtain
and
Y = Y2 = h+ B 1
+ t [A j
sinh (j - 1)7r + B cosh (j - 1)7r] cos (j -
j
1)7r(~ - ~o) .
j=2 L L L (14.2.8)
Note that (14.2.6) and (14.2.8) imply that Yl and Y2 are even, 2L-periodic
functions of ~ - ~o == t. The unknown constants are determined by the iterative
method as follows: Assume that the n-th approximations for h, A j , B j , ~o, and
L are known. Let us denote them by h(n), A)n), BJn) , ~~n), and L(n). Then
proceed as follows:
1. Substitute h(n), A)n), BJ n ), ~~n), and L(n) into (14.2.5) and (14.2.7) and
obtain the (n + 1)-st approximations for Xl and X2.
2. Find ~ = ~n3' which makes X2(~) minimum, and determine A 1 so that
C) (n3)
X2 (<,n3 = x2 .
3. Obtain the solutions of Xl (~) = X~2) and X2(~) = X~2) and take the smaller
value as 6.
4. Obtain the solutions of Xl (~) = x~nl-1) and X2(~) = x~n2-1), and take the
smaller value as 6.
5. Determine Zl (6) = z~l), z2(6) = z~l), Zl (6) = z~nl) and z2(6) = z~n2)
by extrapolation.
Steps 1 through 5 determine Xl (~) and X2 (~) for the interval
6:::;~:::;6 =6+L. (14.2.9)
6. Set L = 6 - 6 and ~o = 6·
7. Now the left sides of (14.2.6) and (14.2.8) for Y1 and Y2 are determined as
functions of ~ on the interval (14.2.9) through the functions Xl (~) and X2(~)'
8. The constants h, A j (for j = 2,3, ... ,11:) and B j (for j = 1,2, ... ,II:) are
determined by Fourier analysis. Thus,
±1Y1(t)dt,
L
B1 =
2 fL (j - 1)71't
Bj =LJ Y1(t) cos L dt, 1:::; j :::; 11:,
o
h= ±1 L
Y2(t)dt - B 1,
A j = csch
(j-1)71'
L
[2L JofL Y2(t) cos
(j-1)71't
L dt - B j cosh
(j-1)71']
L '
(14.2.10)
for 2:::; j :::; 11:, where t = ~ - ~o.
9. All (n+ 1)-st approximations, denoted generically by ¢(n+l) , thus obtained
are replaced by (1-r) ¢(n) +r ¢(n+l), wherer is a relaxation constant, usually
0.5. This assures the convergence of the successive approximations.
10. Repeat steps 1 through 9 until the desired convergence is achieved.
Note that the first approximation is given by h = y~n3), A j = B j = 0

(1 ~ j ~ x:), 6 = 0, and L = x~nl-l) - x~2). Since the hyperbolic functions

are involved, double precision should be used in all computations.•
CASE STUDY 14.2.2 (Two-step method). Consider the same flow prob-
lem as in Case Study 14.2.1. The mapping is carried out as follows: First, map
the region ABC'DEF in the z-plane onto the rectilinear strip 0 ~ TJ ~ 1 in
the (-plane (see Fig. 14.2.3) by
z = h [( - ~ (1 + e- 1Pi () ] + Al + {; KI A k sin (k - 1)~:( - ~o) ,

(14.2.11)
where the constants h, Ak, ~o and L 1 are computed by an iterative method
similar to that in Case Study 14.2.1. Note that the mapping (14.2.11), which is
the mapping function for the inlet without center bodies, is obtained by taking
Bk = 0 (k 2: 1) in (14.2.2). The image of the arc BC in the (-plane is denoted
by
(14.2.12)
Next, map the region ABCDEF in the (-plane onto the semi-infinite strip
+ iv) by
u ~ 0, 0 ~ v ~ 1, in the w-plane (w = u
2 1rW ~ (2k-1)1r(w-i)
(=-
1r
log cosh -
2
+ Bo + ~ B k cos L
2 2
' (14.2.13)
k=1
F F D
E~:
e:-::
A ---------<-n--- c'
z =f(~)
A c::==~,
n, ~3
z-plane ~-plane
j ~=F(w)
w-plane
Fig. 14.2.3.
such that the points B (( = 6) and C (( = ~4 + i TJ4) go into the points w = 0

and w = L 2 , respectively. The unknown constants Bk (0 ~ k ~ K2) and L 2
are determined as in §14.2.1. Thus,

n(u+iv)
E+ t. "I = 2
- log cosh ---'----'-
n 2
~B (2k-1)(u-i(1-v))
+ B 0L..- kCOS 2L
k=1 2
2 hn(u+iv)
= -; l 09 cos 2 (14.2.14)
~B [ (2k-1)nu h(2k-1)n(1-v)
+ B0 L..- k cos L cos L
k=1 2 2 2 2
. . (2k - 1) nu . h (2k - 1)n(1- v)]
+t sm L
2
sm
2
L '
2 2
which for v = 0 gives, after separating the real and imaginary parts,
."C = -1 I
og cos
h2 nu
- + B0 + ~ B
L..- k cos
h (2k - 1) n
L cos
(2k - 1) nu
,
n 2 k=1 2 2 2 ff4.2.15)
C)
( = L..- ~B . h (2k - 1) n . (2k - 1) nu ( )
"I = "11." k sm L sm L· 14.2.16
k=1 2 2 2 2
l
Hence,
2 L2
_ h(2k - 1)n . (2k - 1)nu d
Bk - -L csc L "II sm L u, 1 ::; k ::; 1',2.
2 2 2 0 2 2
(14.2.17)
Also, at the point B (w = 0) we have from (14.2.13)
(2k-1)n
6 = B o + 2: Bk
1<2
cosh 2L .
k=1 2
Thus,
(2k - 1) n
Bo = 6 - LB 1<2
k cosh 2 L· (14.2.18)
k=1 2
It is obvious from (14.2.16) that "Il(U) is an odd, 4L 2-periodic function sym-

metric about the line u = L 2 .
The semi-infinite strip CBAFED in the w-plane is mapped onto the

infinite strip 0 :::; "12 :::; 1 on the (2-plane ((2 = 6 + i "12) by
w =~ cosh-
1
(e1r6 / 2) . (14.2.19)
The lines 6 = canst or "12 = canst produce the grid lines on the z-plane. •
14.3. Cascade Configurations
The Ives-Liutermoza method (Ives and Liutermoza, 1977) deals with the map-
ping problem that transforms the region exterior to a cascade of blades first
onto a region exterior to a near circle. Then the second mapping transforms
the interior of the near circle onto the unit disk. The success of the first map-
ping depends on the solidity of cascades; if it is low, the mapping yields an
acceptable near circle, but if it increases, the image boundary degenerates into
a peanut-shaped contour, in which case the second mapping will not work at
all, and the Ives-Liutermoza method fails. To overcome this difficulty for the
mapping problem of the cascade of blades, we shall first determine the function
that maps the region directly onto a rectangle for two periods of the cascade
of blades, which can then be mapped onto the unit disk (see Fig. 11.1.2 and
11.3.2). We shall consider three cases described in the following case studies.
CASE STUDY 14.3.1. First we shall study the ordinary type of grids. Let
two periods of the cascade of blades make a row in they direction in the z-plane
(physical plane z = x + i y, Fig. 14.3.1(a)). The contour of one of the blades
is defined by a set of data Zn = X n + i Yn, n = 1, ... ,N, ZN = Zl, which are
ordered clockwise, with IXn I :::; 1. Let h denote the pitch of the blades in the
y-direction. The function that maps them onto an infinite strip in the (-plane
(computational plane, Fig. 14.3.1(b)) has the form
Z
h
= Ao [:;;: logsn((,k) -1 + '~MCj
"
cos
(j -1)7r(]
K' ' (14.3.1)
J=l
where A o is a real parameter that finally approaches unity in an iterative scheme

(see Step 3 of the algorithm given below), Cj = A j + i B j and k, 0 < k < 1,
are constants to be determined, K and K' are complete elliptic integrals of the
first kind with moduli k and k' = Vf=k2, respectively, and sn is one of the
Jacobian elliptic functions (see §2.3). In particular when A o = 1, k = e- 21r / h ,
A j = 0 = B j for j ~ 1, the function (14.3.1) represents the mapping of a
cascade of flat plates of chord 2, pitch h, and zero stagger. If we set ( = K + i TJ
in (14.3.1) and separate the real and imaginary parts, we get
h , ~{(j-1)7rK (j-1)7rTJ
x=A o [:;;: logdn(TJ,k)-1+A 1 + ~ A j cosh K' cos K'
j=2
. h (j -l)7rK . (j -l)7rTJ }]
- B j sm K' sm K' ' (14.3.2)
14.3. CASCADE CONFIGURATIONS 391
~{ (j-l)1l'K (j-l)1l'1]
y = A o [ B 1 + ~ A j sinh K' sin K'
j=2
(j-l)1l'K (j-l)1l'1]}]
+ Bj cos h K' cos K' ' (14.3.3)
where dn is another Jacobian elliptic function.
F E D
-iK
G C
K 0 K
iK
H A B
z-plane ~-plane
The following algorithm for the iterative method which starts with the data
for the initial guess as that of the flat cascade assumes that the n-th approxi-
mations A;n), B;n) and ken) are known. Then proceed as follows:
STEP 1. Substitute the known values of A;n) , Bjn) and ken) into (14.3.2) and
compute the values of A~n+l) such that X max + Xmin = O. Then compute the
constant A~n+1) such that X max - Xmin = 2. use this data on the left side of
(14.3.2) to yield the relation
(14.3.4)
STEP 2. Choose a relaxation constant co (= 0.5), and replace A~n+1) by

(1 - co) A~n) + co A~n+1), to yield a new value of A~n+1).
STEP 3. Use this value of A~n+1) to obtain k(n+1) from
(14.3.5)
In this process A o -; 1 as n increases.

STEP 4. Use (14.3.4) to write the profile of the blade intheform y = y(n+1) (1]).
Substitute this y on the left side of (14.3.3), and use Fourier series analysis on
it to obtain
(n+l) =
AJ
[K' A + a .
sm
h (j - 1)7rK] -1
K'
fK' Y sm. (j -K'1)7r1]
-K'
d
1],
.
J:::: ,
2
Bj
(n+l) _
-
[K' A + a .
sm
h (j - 1)7rK] -1
K'
fK' Y cos (j -K'1)7r1]
-K'
d
1],
.
J -
>1
,
(14.3.6)
STEP 5. Chooses a relaxation constant €1 (= 0.1), and replace Ajn+l) and
B(n+l)
J by (1-€ 1 ) A(n)+€ 1 A(n+l) and (1-€ 1 ) B(n)+€ 1 B(n+l) , respectively •
J J J J
STEP 6. Repeat steps 1 through 5 until the required convergence is achieved.
Note that Steps 3 and 5 guarantee the convergence of successive approximations
in this scheme. The grids drawn with solidity 1.58, N = 42, M = 20,
k = 4.1 X 10- 5 , K = 1.57, and K' = 11.5 can be found in Inoue (1983),
where the lines ~ =const surround the blades, and the lines 1] =const continue
across the periodic boundaries 0 A and 0 E. The constant k and the ratio K / K'
both decrease as the solidity of the blade increases. It has been observed that
this method remains successful for solidity up to 0.29.•
Fr----'''3~--___, D
Y=7t
y=o
--------~ G c
-K K
z-plane ~-plane
CASE STUDY 14.3.2. Another kind of grid is generated when the stream-
lines flow parallel to the x-axis in a special situation induced by the presence
of sources and sinks in the physical plane. We shall consider the problem of
a cascade of two periodic blades. Let the streamlines of the flow through the
cascade from left to right, as shown in Fig. 14.3.2(a), represent a family of grid
lines in the z-plane. Let Z = X + i Y denote the complex velocity potential
of the flow induced by the following distribution of sources and sinks in the
(-plane (Fig. 14.3.2(b»:
unit sinks at ( = 2mK + 2niK' ; and

unit sources at ( = 2mK + (2n - l)iK', where m, n are integers.
(14.3.7)
Since the through-flow grid is based on (X, Y) coordinates, the flow induced
by sources and sinks(4.106) is defined by (see (2.3.13»
Z = logsn((, k), (14.3.8)
i
or
(=
eZ
[(1 - t 2 )(1 - k2 t 2 )r 1 2
/ dt. (14.3.9)
Also, we have Z(K) = 0 and Z(K + i K') = -log k, where ( = K and

( = K + i K' are the stagnation points of the flow. The through-flow grids for
the cascade of Fig. 14.3.2 can be drawn with the same data as above.
D
E
•
Y=1t12
:
0
~
~~~~'
0
II
'"
-K K
~:
0
Y-1t12 A
-iK' A B
z-plane ~-plane
CASE STUDY 14.3.3. A different type of grid arises in the case when the
streamlines of the flow that starts from infinity on the right encircles the cascade
and returns to infinity to the right (Fig. 14.3.3(a». These streamlines represent
the grid as a family of coordinate lines. The flow is induced by the distribution
of sources and sinks in the (-plane which are as follows (Fig. 14.3.3(b»:
unit sinks at ( = 2mK + (4n + 1 )iK'; and

unit sources at ( = 2mK + (4n - 1 )iK', where m, n are integers.
(14.3.10)
The complex velocity potential of this flow is defined by
Z =X + i Y = log sn ( ~ ( + i K'), k 1 ) , (14.3.11)
where K 1 and K 1are the complete elliptic functions of the first kind with moduli
K' 2K' 1- k'
k 1 and k~ = }1 - kr, respectively, such that K~ = K and k 1 = 1 + k'
(see Landen transformation, §2.3). The inverse mapping of (14.3.11) is given
by
_ K
( - K'
r
io
eZ
dt
}(1 - t 2)(1 _ kr t2 )
_ ,; K'
• .
()
14.3.12
Note that Z(K - i K') = 0 and Z(K + i K') = -log k 1 , where K ± i K' are
the stagnation points of the flow. The grid lines generated by this method for
the above cascade can be drawn for the same data as in Case Study 14.4.1.
CASE STUDY 14.3.4. The design of an airfoil and wing becomes sig-
nificant in the transonic flow problem. The basic equations of fully developed
flow potential ¢ around the configuration of an axisymmetric inlet of arbitrary
geometry, shown in Fig. 14.3.4, is given by
2
(a 2 - ¢;) ¢xx - 2 ¢x¢r¢xr + (a 2 - ¢;) ¢rr + ~ ¢r = 0, (14.3.13)
r
where (x, r) denote the coordinates along and normal to the centerline, respec-
tively, and a is the velocity of sound. We shall use the subscripts int and ext to
denote the interior and exterior inlets, respectively.
The mapping of the physical region (z-plane) is carried out by functions

with scale factors that depend only on the mapping modulus. The basic idea
in the construction of a composite conformal map is to transform the physical
boundary (inlet contour) into a Jordan contour and then into the unit circle using
Fourier series. Finally the circle is mapped onto a rectangle, as in §11.1 and
11.3, supplemented by a coordinate stretching of type (14.1.11) to obtain the
computational plane. The chain of conformal mappings h, ... ,fs, presented
in Fig. 14.3.5, is described as follows:
MAPPING h from the z-plane onto the zl-plane is given by
Zl = -2r*
-,
z* - z
where z* = x* + i r * is the inversion point of the stagnation point z = x + i r.
MAPPING h from the zl-plane onto the z2-plane is
This separates the interior and exterior points at infinity and thus opens up the
closed centerline in the zl-plane. The square root of Zl + i is used with a cut
starting at the branch point Zl = -i.
MAPPING h from the z2-plane onto the z3-plane is given by
i Z2
Z3= - -.
Z2 - 2
This bilinear transformation takes the point Z2 = 2 into 00 and the centerline
into the positive real axis in the z3-plane. While approaching the interior
infinity the inside inlet in the z3-plane tends to a line with a constant imaginary
part for increasing positive values of the real part. This leads to a situation
where the flow field in the z3-plane 'opens up' as in the z4-plane.
Inversion point I Straight

z*=x*+ir* I circular tube
I
Stagnation
streamline
---
--------'------------=--------+x
o Centerline
z- plane (physical plane)
Fig. 14.3.4.
MAPPING 14 from the z3-plane to the z4-plane is given by
where the constant C3 is chosen such that ~{ lim C3 Z3} = 11" on the inlet
%3-+ 00
1I"T
interior side of the contour; thus C3 = --*, where Tint denotes the interior
2 Tint
radius for downstream in the inlet (see Fig 14.3.5 where the flow field contour
is transformed into a Jordan contour without corners).
MAPPING 15 from the z4-plane onto the zs-plane is given by
where the constant b4 takes some suitable value between 0.1 and 1. Although
the contour in the zs-plane has a continuously varying tangent, it has curvature
singularities at the two points at infinity.
Inlet
extenor
Interior and
...........exterior infinity
Cenlerline
Circle around - i Nose
.lnlel
tnlenor
zrplane zrplane
Inlet
interior Centerline
,,
os<.
,
1+!b4 Exterior }qfinily
\
Inle! Centerline
extenor
------+--..,..- - - - ~
Exterior infinity
Zqplane zs-plane
1t
nl,
e e or IN, ~" 1 te or
Inlet
exterior
e 1'in.. ",in ty
I~, In~eribr
if tn ty
e
1
Centerline
- - e q .~~ - - -+
o ~ I
IV-plane Computational plane
Fig. 14.3.5.
At this point we would like to use the Fourier series to transform the bound-
ary into the unit circle. Since the exponential function is used in the mapping
14, the far downstream region in the inlet interior is very dense around the inte-
rior infinity. Thus, the Fourier series mapping of the region in the zs-plane will
not be highly accurate in the neighborhood of the interior infinity which is at
Zs = -1. To avoid this, we use a Taylor series expansion of the mapping func-
tion Is about this point. Since the leading terms of this series do not contain the
curvature at the interior infinity, the curvature singularity becomes negligible
and can be neglected. We shall further discuss the case of interior infinity at
Zs = -1 hereafter. At the exterior infinity, which is mapped into Zs = 1 the
curvature has a finite discontinuity, where, to compute the behavior of z near the
exterior infinity, we use the transformation h, defined below, which removes
the curvature singularity at Zs = 1 before the boundary is mapped onto the unit
circle.
A transformation of the type
where
and Tint and Text denote the interior and exterior radius of the downstream
constant geometric part of the inlet. Thus,
MAPPING 16 from the zs-plane onto the z6-plane is given by
where the constant a6 has the value of about 5. This mapping is single-valued.
MAPPING h from the z6-plane onto the z7-plane is given by
M
Z6 - z~ = Z7 exp [ L (an + i .8n) z~],
n=O
where the point z6 is located near or at Z6 = O. This function maps the boundary
in the z6-plane onto the unit circle in the z7-plane such that the exterior infinity
goes into the point Z7 = 1 and the interior infinity into some point Z7 = eilh .
MAPPING Is from z7-plane to the w-plane is given by

_ _ i87/2 Z7 - 1
w- e .8 .
z7 - e' 7
This bilinear transformation maps the flow field from the unit disk onto the
upper half-plane ~{w} > O.
The problem at the interior infinity at Zs = -1 can be remedied by taking,

instead of 14, the mapping I~, defined by
where a3 and b3 are properly chosen. Then the new zs~ontour will be much
'fuller' below the point Zs = -1 instead of Zs = -1. This will affect the
subsequent mappings Is and 16, which will then be defined by
with
C{; = - 71"2 r; a4 (1 + a3/b~) ,

where, in practice, we take 1 < a3 < 2.5, and b3 ~ 2.
The Fourier coefficients in the mapping 16 are computed from
This increases the numerical accuracy over that obtained from differentiating
the function 16.
As a result of this chain of mappings the governing equation (14.3.13) is

transformed into
(a 2 - ql2) <PRR - 2ql q2 R1 <PR8 + (2

a - q22) R2
1 <P88
+a2 B ( ql -rR R -r8 + B

+ q2 (. a 2 + q22) ql
R + (2
ql + q22) (B R B
ql R + q2 8
~) = 0,
r r ( 4.3.14)
14.4. PROBLEMS 399
where ql and q2 are the velocity components in the R and () direction, respec-
tively, i.e.,
1 1
ql =B cPR, q2 = RB cPo.
Arlinger (1975) solved this problem by the finite difference method on the
rectangle shown in Fig. 14.3.5 (the computational plane).•
A Fortran code (TOMCAT) for a method of automatic numerical generation

of curvilinear coordinate system with grid lines coinciding with all boundaries
of a multiply connected region is available (Thompson, Thames and Mastin,
1977). The computer code is independent of the boundary shapes and numbers,
which are input data. The program has the following features: (i) automatic
convergence controls activated by input parameters, if needed; (ii) a choice of
several different types of initial guesses for the iterative process; (iii) gradual
addition of coordinate system control; and (iv) general movement of the outer
boundary out to its final position. Another program is available in Thompson et
al (1976) for computing the scale factors from the coordinates for use in solving
partial differential equations on a coordinate system.
An adaptive grid scheme to solve the Poisson grid generation equations by

methods related to Green's function, where the source terms are only position-
dependent, uses the boundary element method and is given by Munipalli and
Andersen (1996). All of these schemes and programs solve the types of prob-
lems discussed in the above case studies.
14.4. Problems
PROBLEM 14.4.1. Show that under the mapping of a region D in the

z-plane onto a region G in the w-plane by the function w = f (z) = u(x, y) +
i v(x, y), the Laplace equation for the function u(x, y) is transformed into
the Laplace equation for the function U(~, 7]) = u (x(~, 7]), y(~, 7])), i.e., the
Laplacian '\72 satisfies the relation
where z = F( w) denotes the inverse mapping function. (Sveshnikov and

Tikhonov, 1978, p.194.)
PROBLEM 14.4.2. Derive the condition (14.3.5) by using the conditions

that the pitch of the cascade remains h if A o = 1 and A o log dn (1], k,(n l ) =
logdn(1],k,(n+l l ). [Hint: Use sn(1],k') ~ tanh 1], sn(1],k') ~ 1, and

dn (1], k') = VI - k'2 sn 2 (1], k')] (Inoue, 1983, p.133.)
PROBLEM 14.4.3. Generate graphical representations of the grid lines in
the Case Studies in §14.2 and §14.3.
REFERENCES USED: Arlinger (1975), Inoue (1983,1985), Ives and Liutermoza

(1977), Knupp and Steinberg (1993), Lehto and Virtaanen (1973), Munipalli and
Andersen (1996), OZisik (1994), Roberts (1971), Sveshnikov and Tikhonov (1978),
Thompson et al. (1977, 1985).
Appendix A
Cauchy's P. V. Integrals
A.I. Numerical Evaluation
We shall derive approximate formulas to compute Cauchy principal value in-

tegrals. In practical problems it is often necessary to evaluate Cauchy p. v.
integrals which arise in boundary value problems or in singular integral equa-
tions. We shall consider the Cauchy p.v. integral
F(to) = ~ ( f(t) dt, to E r, (A.I.l)

Z1T Jr t - to
where r is a smooth, open arc or a Jordan contour and f E H1(r), i.e., f

satisfies the Holder condition on r with (): = 1 (see §1.1). We can rewrite
(A. 1.1) as
F(to) = ~ (f(t) - f(to) dt + f~to) ( ~

mJr t - to Z1T Jr t - to (A.I.2)
= F1(to) + F2 (to)·
The integral F1(to) is an improper integral whereas F2 (to) may be evaluated

directly. Therefore, the problem reduces to evaluating the improper integral
(A.I.3)
401
402 APPENDIX A. CAUCHY'S P. V. INTEGRALS
where
lI(t) = f(t) - f(t o).
t - to
A.I.I. Method of Separation of Singularities. If the indepen-

dent variable t in (A.l.3) is changed to the arc length parameter of r or a real
parameter in the parametric equation of r, then certain classical results can be
applied to the improper integral (A.l.3). In fact, if f(t) E HI, then II (t) is
bounded, and (A.l.3) becomes a proper integral. In particular, if r is a straight
line segment on the real axis, then the variable t is real. If r is a circle or a
circular arc with a fixed radius r, then the substitution t = r eiO also trans-
forms (3.2.3) into an integral of the real variable e. The method of separation
of singularities is very easy to use when r is a straight line segment[a, b] of
the real axis. The following result can be proved by induction.
THEOREM A.1.1. Let

g(t) - g(to) zft -t. t
G(t) = t - to' r 0,
{
g' (to) , if t = to,
where g(t) E en+! [a, b], n 2: 0, and a ::; to ::; b. Then
ift # to,
(A.1.4)
ift = to,
where to < Tk < t.
In the case when
1
F(to) = -:-
Z7l'
1 -f(t)
- w(t) dt,
r t - to
to E r, (A.1.5)
where w(t) 2: 0 is a weight function which may have integrable singularities

at certain points, e.g., at the end points, then the above theorem can be used,
provided (A. 1.5) is written as
F(to) =.;.- Jrr II (t) w(t) dt+ f~to) Jrr tw(t)

Z7l' Z7l' - to
w(t) dt, to E r, (A.1.6)
in which the second integral may be evaluated exactly. The above results are
due to Ivanov (1968) and Pukhteev (1980).
A.I. NUMERICAL EVALUATION 403
A.1.2. Gauss-Chebyshev Type Quadrature. As an application

of above result we shall find an approximate value of the Cauchy p.v. integral
I(x) = .!.
rr
1-1
1
g(t)
t- x
dt
Jf=t2
, -1 < x < 1, (A.1.7)
with weight function w(t) =

1 - t2
b'
where g(x) is sufficiently smooth.
First, note that if g(t) == 1, the value of the integral (A.1.7) is zero. In fact, it
can be proved by the contour integration method that
-1
rr
/1 -1
dt - 0
(t - x) Jf=t2 - ,
-1 < x < 1, (A.1.8)
by taking a closed contour r- containing the line segment -1 $ x $ 1, and

using the residue theorem to obtain
1 rd( 1
2irr Jr- (- z) J1=(2 = Vf="Z2'
where r- is clockwise and J1=(2 is the analytic continuation of the positive
real-valued function JI=X2 along the upper side of -1 < x < 1. As r
shrinks to -1 $ x $ 1, we get
-irr1 1 1
-1
dt
Jf=t2
(t - z) 1 - t 2
=
1
Vf="Z2'
1 - z2
z t/. [-1,1].
Then (A. 1.7) follows as z -4 x from the upper (or lower side) and using
Plemelji fonnula (1.2.13). Thus, we can rewrite (A.l.7) as
I(x) = .!.
rr
/1 -1
g(t) - g(x)
t - x Jf=t2
dt , -1 < x < 1. (A.1.9)
Now, if f E C2n [_I, 1], then we have the following Gauss-Chebyshev quad-
rature fonnula (see Abramowitz and Stegun, 1964)
1== I[f] =-
1 /1 Jf=t2 f(t)
dt ~ -
2
1
2:
n
f (tk) , (A.1.10)
rr -1 1- t n k=1
wheretk, k = 1,2, ... ,n, are the zeros ofthe Chebyshev polynomial Tn(x) =
cos (n cos- 1 x) of the first kind and degree n,
i.e., tk = cos (2k - 1 )rr ,
2n
k = 1,2, ... ,n, with the remainder Rn after n tenns
Rn = (2n)!~2n-l f2n(r), -1 < r < 1. (A.1.11)

If we set f(t) = g(t) - g(x) in (A. 1.10), we obtain the quadrature formula
t-x
I(x) .-",",~ ~
L.J
g(tk) +g(x) ~_1_,
L.J -1<x<l, xf:.tk.
n k=1 tk - x n k=1 x - tk
(A.l.12)
.
However, Since ~
L.J -1- T~ (x) and T'n ()
= -(-) U ()
X = n n-1 X , where
k=1 x - tk Tn X
_ sin (n cos -1 x)
Un-1 (X ) - r:1---::?
yl- x 2
is the Chebyshev polynomial of the second kind and degree n - 1, we find

that
and formula (A. 1.12) becomes
-1 <x< 1, x f:. tk,
(A.l.13)
which is a Gauss-Chebyshev type quadrature formula. In particular, if x = x j
is taken as a zero of Un - 1 (x), i.e., if Xj = cos j1r , j = 1,2, ... ,n - 1, then
n
formula (A. 1.13) simplifies to
1 ~ 9 (tk) .
I(xj)~- L.J---' J=I,2, ... ,n-l. (A.l.14)
n k=1 tk - Xj
This quadrature formula is useful in solving the following integral equation

of the first kind:
~
1r
1 1
-1
¢>(t) dt +
t- X
11
-1
k(x, t) ¢>(t) dt = f(x), -1 < x < 1, (A.l.15)
where k(x) E HO ([-1, 1] x [-1,1]).
A.lo3. Approximation by arcwise linear functions. The arc

length parameter s can be used for approximating (A. I.! ) in the case of simple
contours (of integration). If the contour is complicated, we shall first approx-
imate the function f(t) by an arcwise linear function and estimate error.
A.l. NUMERICAL EVALUATION 405
Let AB denote an open, smooth are, and let f(t) E HO:, 0 < a ::; 1
be defined on AB. We partition AB by A = t l , ... , t n = B and let r j
denote a partition (tj, tj+d. Let Yj = f (t j ), I:1Yj = Yj+l - Yj, and D j =
I:1Yj/l:1tj. We shall define arcwise linear functions Lj(t) on AB such that
Lj(t) = Yj + D j (t - tj) for t E r j , j = 1, ... ,N. Then, if the error is
denoted by ej (t), we have
lej(t)1 = If(t) - Lj(t)1 ::; If(t) - Yjl + IDjilt - tjl

(A.LI6)
< Cit -
-
t ·10:
J
+ 11:1 (tj)c 11 -0: It - t·1
J
< C 80:
- ,
where C is a constant and 8 = I

max I:1Yj I. This inequality implies that the
J
error ej (t) becomes very small if {tj} is very dense, i.e., if 8 is very small.
For example, let us consider the error when the integral in (A.l.l) is
replaced by ( Lj(t) dt. Then
Jr t - T
and, in view of (A.l.16),
(A.LI7)
Now, ift and T are in the same r j , we have
(A.LI8)
where c, 0 < c < a, is very small. If T E r j - l and t E rk, j ::; k, then,

since ej (Tj) = ej (Tk) = 0, we find that
lej(t) - ej(T)1 ::; !ei(t)1 + !ei(T) I ::; lej(t) - ej (Tk) 1+ lej(T) - ej h) I

::; C (It - TklO: + It - Tjn
::; C 80:-<: (It - Tk 1<: + It - TX)
::; C 80:-<: (ITk tl<: + h tn
::; C 80:-<: (ITin ::; C 80:-<: (It - Tn .
Thus, (A.U8) always holds for some constant C, and
Ih(t)1 ::; C 80:-<: i It - TI-1+<: Idtl ::; C 80:-<: i 8-1+<: d8 ::; C<: 80:-<:,
(A.LI9)
where C" is a constant that depends on c. In general, C" --+ +00 as c --+ O.
Hence, the inequalities (A.U8) and (A.U9) yield
(A.1.20)
Since c is arbitrary, we find that for very small 8 the left side of (A.I.20)
N
becomes arbitrarily small. Thus, L Lj(t) approximates the kernel f(t) in
j=1
Cauchy's p.v. integral (A.U), and
r f(t) dt ~"
N
r Lj(t) dt, (A.1.21)
Jr t - T LJ
J=1
Jr 1
t - T
which can be easily computed. For more on this topic, see the references cited
below.
REFERENCES USED: Ivanov (1968), Lu (1984, 1994), Pukhteev (1980).

Appendix B
Green's Identities
B.t. Green's Identities
n
Let be a finite domain in Rn bounded by a piecewise smooth, orientable
surface an,
and let w and F be scalar functions and G a vector function in
the class CO(n). Then
Gradient theorem: {V' F dn = 1 nF dB,

in Jan
Divergence theorem: {\l . G dn = 1 n· G dB,
in Jan
Stokes's theorem: {V' x G dn = 1 G· t dB,
in Jan
an, t
f
where n is the outward normal to the surface is the tangent vector at a
point on an, denotes the surface or line integral, and dB (or ds) denotes the
surface (or line) element depending on the dimension of n. The divergence
theorem in the above form is also known as the Gauss theorem. Stokes's
theorem in JR2 is a generalization of Green's theorem which states that if
G = (G 1 , G2 ) is a continuously differentiable vector field defined on a region
containing u n an c JR2 such that an
is a Jordan contour, then
(B.1)
407
408 APPENDIX B. GREEN'S IDENTITIES
Let the functions M(x, y) and N(x, y), where (x, y) ED, be the components
of the vector G. Then, by the divergence theorem
in (~~ + a;:) dxdy= i[McOs(n,x) + Ncos(n,y)] ds,
= i Mdx+Ndy,
(B.2)
with the direction cosines cos( n, x) and cos( n, y), where f = aD. If we take
M = f gx and N = f gy, then (8.2) yields
r (81ax axag + afay aayg ) dxdy = Jrrfagan ds _ Inrf\12 g dxdy,

In
(B.3)
which is known as Green's first identity. Moreover, if we interchange f and

gin (B.2), we get
r (afa + afa ) dxdy= rgaf ds- rg\1 2fdxdy.

g
In ax ax ayay
g
Jr an In
(B.4)
If we subtract (B.3) from (BA), we obtain Green's second identity:
in 2 2
(I\1 g - g\1 f) dx dy = l (f ;~ - 9 ~~) ds, (B.5)
which is also known as Green's reciprocity theorem. Note that Green's
identities are valid even if the domain D is bounded by finitely many closed
curves. In that case, however, the line integrals must be evaluated over all paths
that make the boundary of D. If f and 9 are real and harmonic in D c JR, then
from (8.5)
r1( f ag
- - gaf)
an
- ds=O.
an
Let D be a simply connected region in the complex plane <C with bound-
(B.6)
ary f. Let Zo be any point inside D, and let D be the region obtained by
indenting a disk B(zo, e) from D, where e > 0 is small (Fig. B.l (a». Then
aD consists of the contour f together with the contour aB(zo,e) = fE:'
~z
~ r
(a) (b)
Fig. 8.1.
B.I. GREEN'S IDENTITIES 409
If we set f = u and 9 = logr in (B.6), where zED and r = Iz - zol, then,

. a
SInce an = - ara on r 0' we get
r(u ~an
Jr
(logr) - (logr) au)
an
ds - r (~_
Jr. r
(logr) au)
ar
ds = o.
(B.7)
Now, let e -4 0 in (B.7). Then, since
lim r ~r ds = lim Jor

Jr.
27f
u(zo + eB) ~e edB = 0,
1
0--+0 0--+0
1 au ds = lim au e dB = 0,
27f
lim
0--+0 r.
log r -a
r 0--+0 0
log e -a
e
we obtain
27r u(zo) = l [u :n (log r) - (log r) :~] ds, (B.8)
which is known as Green '8 third identity. Note that Eq (B.8) gives the value
of a harmonic function u at an interior point in terms of the boundary values
of u and :~. If the contour r has no corners and if the point Zo is on the
boundary r, then instead of the whole disk B(zo,
c) we consider a half disk at
the point Zo deleted from D (Fig. B.l(b)), and Green's third identity becomes
7ru(zo) = p.v·l [u :n (logr) - (logr) :~] ds, (B.9)
where p.v. denotes the principal value of the integral, i.e., it is the limit, as
r-4 0, of the integral over the contour r
obtained by deleting that part of r
which lies within the circle of radius e and center zoo
REFERENCES USED: Carrier, Krook and Pearson (1966), Kythe (1996),

Kythe, Puri and Schiiferkotter (1997).
Appendix C
Riemann-Hilbert Problem
We shall also consider two boundary value problems of the theory of analytic
functions. They are the Hilbert and the Riemann problems. A close relation-
ship exists between the Hilbert problem and the theory of singular integral
equations. Although the latter may be developed to a large extent without
the former, it is this relationship that makes the latter simple and clear. For
example, the iterative method for solving Theodorsen's integral equation, as
outlined in Chapter 8, is based on a certain Riemann-Hilbert problem. In fact,
Theodorsen's problem is a linearized version of a singular integral equation
of the second kind.
C.l. Homogeneous Hilbert Problem
Consider the n-tuply connected region of Fig. C.U, with the boundary
f = U~=l f k • The homogeneous Hilbert problem states: Find a sectionally
analytic function <l>(z) of finite degree at infinity such that
(C.l.I)
where <l>+(t) and <l>-(t) are limiting values from the right and the left at a
point t E r (ift is an end point, then <l>+(t) = <l>-(t) = <l>(t)), and G(t),
defined on r, satisfies the HOlder condition and G(t) i= 0 at the point t E r.
If we take the logarithm on both sides of (C.l.l), we get
[log <l>(t)]+ - [log <l>(t)r = logG(t). (C.l.2)
410
C.l. HOMOGENEOUS HILBERT PROBLEM 411
As t moves along the contours r k in the positive sense, i.e., counterclockwise

for k = 0 and clockwise for k = 1, ... ,n (Fig. C.U), log G (t) increases by
integral multiples of 2i1r. Thus,
1 1
-2' [log(t)lr
Z1r k
= -2
1r
[arg{G(t)}lr = Ak,
k
k = 0,1, ... ,n, (C.1.3)
where Ak are integers (positive, negative, or zero), and [arg{ G(t)}]rk denotes
the increment of arg{ G(t)} as it goes around the contour r k. The sum
n I l
/'i, = 2: Ak = 2i1r
[log (t)lr = 21r [arg{G(t)}lr = Ak (C.1.4)
k=O
is known as the index of the Hilbert problem and also as the index of the
function G(t) given on r. The index /'i, is an integer.
Fig. c.u.
Let all ... ,an be arbitrary fixed points in the regions D 1 ,... , D:;;, , and
let the origin of the coordinates be in the region D+. Let
(z - al)-a 1 ..• (z - an)-a" iff = U~=lfk,

(C.1.5)
p(z) ={ 1 iff = fo.
Let
Go(t) = t- K pet) G(t). (C.1.6)
412 APPENDIX C: RIEMANN-HILBERT PROBLEM
Then, after traversing the contours f o , f 1 , ... ,f n, the function arg {Go(t)}
returns to its initial value, and therefore log Go(t) is a well-defined single-
valued and continuous function on f and satisfies the Holder condition there
with an arbitrarily fixed branch on each contour f k .
To determine the solution <l> (z) of the homogeneous Hilbert problem (C. I. I ),
we introduce a new unknown function
'l1(z) = { p(z) <l>(Z). in D+

(C.lo7)
zk <l>(z) In D-,
which is regular except possibly at z = 00. Then the condition (CI.I) can be
written as
(Colo8)
First we shall find the fundamental solution for the homogeneous Hilbert
problem. By taking the logarithm on both sides of (C1.8), we formally get
(C. log)
If we assume that log 'l1 (z) is single-valued, sectionally regular on r, and zero
at z = 00, then
log 'l1(z) = ~ r
log Go(t) dt,
2m Jr t - z
thus, by using the Plemelj formulas (1.2013) we find that
'l1(z) = e9 (z), (Col.lO)

where
g(z) =~
2m
r
Jr
log Go(t) dt.
t - z
(Col.ll)
Obviously, 'l1(z) is sectionally regular on r, 'l1(z) =I- 0 for all finite z, and
'l1 (00) = l. It is also a particular solution of problem (Cl08), since, in view of
(CUI), g+ (to) - g- (to) = log Go(to) for an arbitrary point to E f. Thus,
'l1+(to) = elogGo(to) = G (t )
'l1-(to) 0 0,
which is the same as (C1.9). Hence, from this particular solution we obtain
a particular solution of the homogeneous Hilbert problem (C.U) which we
represent as
_1_ for z E D+
Z(z) = p(z) (C.lo12)
{
z-I< e9 (z) for z E D-,
C.l. HOMOGENEOUS HILBERT PROBLEM 413
where g(z) is defined by (C.UO). This particular solution Z(z) is called a

fundamental solution of the homogeneous Hilbert problem (C.1.I) because
it vanishes nowhere in any finite part of the z-plane if K, > 0 and Z (00) = O.
This also holds for the boundary values Z+ (t) and Z- (t). Note that Z(z) is
of degree (-K,) at z = 00. By using the Plemelj formulas (1.2.13), we get
1
g+ (to) = 2 log Go (to) + g(to),
1
g-(to) = -2 log Go(to) + g(t o),
and then, in view of (C.U2), the boundary values Z+(t) and Z- (t) are given
by
e 9 (to) JGo(to) e 9 (to) JG(to)
Z+(to) = = ,
p(to) t~/2 Jp(to)
(C. 1.13)
e9(to) e9 (to)
Z-(to) = = / '
ta JGo(to) t~ 2 Jp(to)Go(to)
where we have used (C.1.6) which also gives
J(Wo)
(C.1.14)
p(t) .
THEOREM C .1.1. All solutions of the homogeneous Hilbert problem

(C.l.l) which have finite degree at infinity are given by
~(z) = Z(z) P(z), (C.1.15)
where P(z) is an arbitrary polynomial.
PROOF. Let ~(z) be any solution. Then by (C.U)
~+(t) = G(t) ~- (t), and Z+(t) = G(t) Z- (t).

Since Z+(t) :j:. 0 and Z- (t) :j:. 0, we find that
P+(t) P-(t)
Z+(t) = Z-(t) = G(t),
which implies that the function ~i:~ is regular in the entire z-plane, and has
finite degree at z = 00. Thus, it is a polynomial, which proves the theorem.•
Some of the consequences of this theorem are as follows:
(i) The limiting values + (t) and - (t) of any solution ( z) of the homoge-
neous Hilbert problem satisfy the HOlder condition because the function G(t)
does so.
(ii) If the polynomial P( z) is of degree n, then the degree of the solution ( z),
given by (C. Ll5), at z = 00 is (n - K;), i.e., the degree of this solution is not
less than the degree (-K;) of Z(z). The degree of (z) and Z(z) at z = 00 are
the same only if n = 0, i.e., if P(z) == const =I- 0. Hence, the index (-K;) is
the lowest possible degree of a solution of the homogeneous Hilbert problem.
(iii) The fundamental solution Z(z) has the following properties:

(a) Z (z) does not vanish in any finite part of the z-plane;
(b) Z(z) has the lowest possible degree (-K;) at z = 00; and
(c) Z(z) is a factor of every solution of the problem (C.Ll).
(iv) An application of the solution (C.l.l) produces the following result: If

0, the homogeneous Hilbert problem has no solution vanishing at z = 00,
K; :::;
except the trivial solution (z) == 0. If K; > 0, it has exactly K; linearly

independent solutions (z) , z(z) , ... ,zt<-I(z), each of which vanishes at
Z = 00. In fact, since the polynomial P(z) in (C.Ll5) is of degree at most
(K; - 1), all solutions of (C.Ll) that vanish at z = 00 must have the form
(z) = Z(Z)P"'_I(Z), where
and Co, CI, ... ,Ct<-l are arbitrary constants.
(v) If the contour r o is at infinity, then the solution (C.Ll5) holds only for
Ak = 0, i.e., for K; = 0.
(vi) The homogeneous Hilbert problems that correspond to the conditions
(C.1.16)
are known as adjoint problems of one another. Hence, if the former problem
has index K; and fundamental solution Z(z), then the latter has index (-K;) and
fundamental solution [Z (z) I . r
C.2. NONHOMOGENEOUS HILBERT PROBLEM 415
C.2. Nonhomogeneous Hilbert Problem
A generalization of the homogeneous Hilbert problem is the nonhomogeneous

problem which states: Find a sectionally analytic function <1>( z) of finite degree
at infinity such that
<1>+(t) = G(t) <1>-(t) + g(t) on r, (C.2.I)
r (t E 0, satisfy the HOlder condition on r

°
where G(t) and g(t), defined on
and G(t) =I on r.
The nonhomogeneous Hilbert problem can easily be solved by using the

results of the previous section. Let Z(z) be the fundamental solution for the
homogeneous problem, defined by (C.1.12). Then this is also the fundamental
solution for (C.2.!) with g(t) == 0, in which case Eq (C.2.1) yields
Z+(t)
G(t) = Z-(t)' (C.2.2)
which when substituted in (C.2.1) gives
cI>+(t) _ cI>-(t) g(t)

(C.2.3)
Z+(t) Z-(t) Z+(t) .
Since the function ~~:~ has finite degree at infinity, we get
<1>(z) = _1
Z(z)
r
~~
2irr Jr Z+(t) t - z
+P z
( ),
(C.2.4)
where P(z) is an arbitrary polynomial. Hence, the general solution of the

nonhomogeneous Hilbert problem (C.2.1) is given by
<1>(z) = Z(z) r~ ~ + Z(z) P(z).

2irr Jr Z+(t) t - z
(C.2.5)
The function Z(z) which is the fundamental solution for the corresponding
homogeneous problem is called the fundamental function for the nonhomo-
geneous Hilbert problem. The index of this problem is also K.
We shall examine the solution (C.2.5) when cI>(00) = O. Then the degree
of Z(z) at infinity is (-/1;), and the solution (C.2.5) will vanish at infinity for
/I; :::: 0 iff the degree of P(z) is S (/I; - 1). Hence, for /I; = 0 it suffices
to take P(z) == O. For /I; < 0, obviously P(z) == 0, and the coefficients of
z-l, z-2, ... ,z-t< must be zero in the expansion
-
1 1 get)
2i1T r Z+(t)(t - z)
dt= -
~
00 z-(j+l)
2i1T
1 t j get)
--dt
r Z+(t) ,
i.e.,
1
2i1T
rt get) j
Jr Z+(t) dt = 0, for j = 0, 1, ... ,-/I; - 1, (C.2.6)
which is a necessary and sufficient condition for the solution to vanish at infinity
for /I; < O. Hence, we have proved the following:
THEOREM C.2.I. If 0, the general solution of the nonhomo-

/I; ::::
geneous Hilbert problem (C. 2.1) that vanishes at infinity is given by
Z(z)
cI>(z) = 2i1T
r Z+(t)
Jr
get) dt
t _ Z + Z(z) P"'-l(Z), (C.2.7)
where Pt<-l (z) are arbitrary polynomials of degree S (/1;-1) and Pt<-l (z) =
o for/I; = O. If /I; < 0, then the solution is given by
cI>(z) = Z(z)
2i1T
rJ!J!L
JrZ+(t)
~
z' t -
(C.2.8)
provided conditions (C. 2. 7) are satisfied.
Note that for /I; = 0 there is a unique solution that vanishes at infinity. For
< 0 there is a unique solution vanishing at infinity, if such a solution exists,
/I;
but for /I; > 0 there is an unlimited number of solutions, and the general solution
(C.2.5) contains /I; arbitrary constants.
In view of the Schwarz reflection and symmetry principles the Hilbert prob-
lems can be applied to the upper half-plane D+ or the unit disk by using the
results obtained in §1.4 and §2.2. The general theory of Riemann-Hilbert prob-
lem is presented in the next section. The Riemann-Hilbert problem studied in
§8.8 is a linearized form of a singular integral equation of the second kind.
C.3. RIEMANN-HILBERT PROBLEM 417
C.3. Riemann-Hilbert Problem
A generalization of the Hilbert boundary problem is the Riemann-Hilbert prob-

lem. This problem deals with determining a function q>(z) = u + iv which is
regular in D+, continuous on D+ U r, and satisfies the boundary condition
!R{a+ib}q>+=au-bv=c onr, (C.3.1)
where a(t), b(t), c(t) are real continuous functions defined for t E r, satisfy
the Holder condition, and are such that a 2 + b2 =/:- 0 everywhere on r. Before
we solve this problem, note that if
(C.3.2)
is any particular solution of the homogeneous problem
au - b v =0 on r, (C.3.3)
then any linear combination
n
q>(z) = L Ok q>k(Z) (C.3.4)
k=l
is also a solution of the homogeneous problem, where Ok> k = 1, ... , n, are

real constants, and the functions q>k (z) are linearly independent. Now we shall
determine the solution of problem (C.3.1) for the circle.
RIEMANN-HILBERT PROBLEM FOR THE UNIT DISK: Let U+ denote

the unit disk with boundary r (izi = 1). Then boundary condition (C.3.1)
becomes
2!R{a+ib}q>+(t) = (a+ib)q>+(t)+(a-ib)q>+(t) = 2c onr. (C.3.5)
Let the solution q>(z) be a sectionally analytic function on r such that it can be
extended in U+ by the function q>*(z), defined by (2.2.12), i.e.,
q>*(z) = ~ (~) = q>(z) for Izi =/:-1. (C.3.6)

The function * (z) is bounded at z = 00. Boundary condition (C3.5) becomes
(a + ib) +(t) + (a - ib) - (t) = 2c, (C.3.7)
or,
+ (t) = G(t) - (t) + g(t), (C.3.8)
where
G(t) = _ a - ~b , ( ) 2c (C.3.9)
a + tb 9 t = a + ib'
Thus, the Riemann-Hilbert problem (C3.5) reduces to a solution of the Hilbert
problem (C.3.7). However, if (z) is any solution of the Hilbert problem
(C.3.7), it may not be the solution of the original Riemann-Hilbert problem
(C.3.1) because it may fail to satisfy condition (C3.6). But we can always
construct a solution of problem (C.3.5) by using the function (z). Note that if
(z) satisfies (C3.5), then taking conjugates in (C.3.5) and using the fact that
-(t) = ~-(1/t), (see (2.2.12», we get
which shows that (z) is also the solution ofthe Hilbert problem (C.3.7). Let
1
O(z) = "2 [(z) + *(z)]. (C.3.10)
Then O(z) is the solution of the Riemann-Hilbert problem (C3.5), because

1
(z) = *(z) = "2 [(z) + * (z)].
We shall determine the complete set of solutions of the Riemann-Hilbert

problem (C.3.5). First, we shall consider homogeneous problem (C.3.5) for
c([(z) + *(z)]) == O. Let K: be the index ofthefunction G([(z) + *(z)]),
i.e.,
1 1 [ a - ib]
K: = -2' [logG(t)]r = -2' log --'b
t1l" t1l" a +t r
= 2.-
211"
[arg{a - ib} - arg{a + ib}]r (C.3.11)
= .!:.11" [arg{a - ib}]r'
Thus, K: is an even integer since a(t) and b(t) are continuous functions. The
number K: is the index of the Riemann-Hilbert problem (C.3.5).
Let Z (z) be the fundamental solution of the homogeneous Hilbert problem

(C.3.8), i.e.,
C e9CZ ) for Izl < 1,
Z(z) = { C) (C.3.12)
C z-'" e9 Z for Izi > 1,
where C :I 0 is an arbitrary constant and
g(z) = ~
2m Jr
r log le'" G(t)] dt =
t - z
2-
21r
r h(t) dt,
Jr t - z (C.3.13)
where
h(t) = arg { - r'" : ~ ~:} (C.3.14)
is a real-valued continuous function defined on f. Using (2.2.12) we get
9 *()
z = -1
21r
1 h(t)
- -d
r t- z
t - 'U~ = 9 (z )- .ta, (C.3.15)
where a is a real constant defined by
(C.3.16)
Hence,
* _ { 0 e9 *CZ) = 0 e 9CZ )-iQ for Izi > 1,
Z (z) - _ C)'
C z'" e 9 z -tQ for jzl < 1,
or, for all z ¢ f,
o .
Z*(z) = C z'" e- tQ Z(z),
or, taking O/C = e iQ , we get
Z*(z) = z'" Z(z). (C.3.17)
Now we shall discuss two cases:
CASE 1. K, ~ 0: The homogeneous Hilbert problem (C.3.7) for c(t) = 0 has

a nonzero solution bounded at infinity and is given by
(z) = P(z) Z(z), (C.3.18)
where
P(z) = Co z'" + C1 Z",-1 +... + C'" (C.3.19)
is an arbitrary polynomial of degree ~ '" - 1. Thus (C.3.18) is a solution of

the Riemann-Hilbert problem (C.3.5) iff !P*(z) = !P(z), i.e., P*(z)Z*(z) =
P(z)Z(z), whereP*(z) = P(I/z) and Z*(z) = zl< Z(z). Since
zl< P (~) = Co + C1 Z + ... + CI< zl<

(C.3.20)
= Cozl< + C 1 ZI<-1 + ... + CI< = P(z),
i.e., CI< = CI<-n, n-O, 1, ... , "', wecansetCn = An+iBn , n = 0, 1, ... , ",/2,
where An' B n are real numbers with BI</2 = O. Then C m = AI<-m - iBI<_m
for m = ",/2 + 1, ... , "'. Thus, there are in all ('" + 1) arbitrary real constants.
We shall denote them by Do, Db' .. ,DI<' Then the formal general solution
of the homogeneous Riemann-Hilbert problem (C.3.5) is given by
I<
!P(z) = L D k !Pk(Z), (C.3.21)
k=O
where !P k (k = 0,1, ... , "') are linearly independent solutions of the same
problem.
CASE 2. '" ~ -2: The homogeneous Hilbert problem (C.3.7) has no nonzero
solution bounded at infinity. Hence, there is only the trivial (zero) solution of
the homogeneous Riemann-Hilbert problem (C.3.7).
THEOREM C.3.1. For '" ~ 0 the homogeneous Riemann-Hilbert prob-

lem (c. 3.7) has exactly ("'+ 1) linearly independent solutions. Its general
solution is given by (C. 3. 18), where Z(z) is the fundamental solution of
the Hilbert problem (C. 3. 7) subject to condition (C. 3. 17). For '" ~ -2,
the homogeneous Riemann-Hilbert problem (C.3.S) has only the trivial
solution !P(z) = o.
Now we shall consider the nonhomogeneous Riemann-Hilbert problem

(C.3.5). We can construct its general solution, provided that we find only one
particular solution. Then the general solution is the sum of this particular
solution and the general solution of the homogeneous problem. However, the
problem offinding a particular solution of the Riemann-Hilbert problem (C.3.7)
is equivalent to finding any particular solution of the Hilbert problem (C.3.7)
that is bounded at z = 00, because, in view of (C.3.1 0), it shall provide us with
a particular solution of the Riemann-Hilbert problem (C.3.5). Hence,
THEOREM C.3.2. For Ii ~ °

there always exists a solution for the
nonhomogeneous Riemann-Hilbert problem (C. 3. 5). For Ii "S -2 this
problem has a solution iff the following conditions are satisfied:
r
21f
Jo ei (n+l</2 D(B) c(B) dB = 0, n = 1,2, ... ,-Ii - 1, (C.3.22)
where
D(B) = 1
Va2(B) + b2(B)
exp { - -
1
41l"
1 0
21f
"I/J --
h("I/J) cot -
2
B d"I/J } . (C.3.23)
Note that conditions (C.3.22) are equivalent to the (-Ii - 1) conditions
r~ Ii
Jo D(B) c(B) cosnBdB = 0, n = 0,1, '-"2 -1,
1 21f (C.3.24)
D(B) c(B) sin nB dB = 0, n = 1, ,-~ - 1.
The solutions of the Hilbert problem (C.3.7) do not vanish at infinity, although
they are bounded there. Thus, conditions (C.3.22) become
r t n g(t)
Jr Z+(t) dt = 0, n = 0,1, ... ,-Ii - 1,
or
r
J [a(t)
r
+tnc(t)
ib(t)] Z+(t)
dt = °, n = 0,1, ... , - I i - 2. (C.3.25)
Since, in view of (C.3.13),
g+(to) = -i h(to)
2
+ -1
21l"
1- - h(t) dt,
r t - to
we set t = ei8 , to = ei80 . Then
g+(to) =
i 1 r21f
B- Bo r21f
"2 h(to) + 41l" Jo h(t) cot - 2 - dB + 41l" Jo h(t) dB.
i
In view of (C.3.20) the last term in the above expression is equal to ia/2.
Hence, Z(z) = e9 (z)-ia/2 for Izj < 1, and
eih(t 0 ) =-t K a(to) - ib(to)

o a(to) + ib(to) ,
which yields
a(to) - ib(to) exp { - 1

----:...-~-.~:.;,.
a(to) + tb(to) 41T
1 0
2
" h(t ) cot -
0--
00 dO} .
2
This proves the theorem. •
Some particular cases of Theorem C.3.2 are as follows:
1. If Ii ~ -2 and (C.3.25) is satisfied, then the Hilbert problem (C.3.7) has a

unique solution which, in view of (C.2.8) and (C.3.9), is given by
~(z) = Z(z)
i1T
r c
Jr (a + ib) Z+(t)(t - z)
dt. (C.3.26)
Since this solution is unique, it is also the solution of problem (C.3.5).
2. For Ii ~ 0, the formula
Z(z)
Q(z) = ~ Jr (a
r c
+ ib) Z+(t) (t _ z) dt, (C.3.27)
gives a particular solution of the problem (C.3.7). Then a particular solution of

problem (C.3.5), as in (C.3.1O), is given by
1
~(z) = 2" [Q+(z) + Q*(z)] , (C.3.28)
where the function Q*(z) is defined as follows: Since, in view of (38.18),

Z*(z) = z'" Z(z), Z+(t) = Z*-(t) = t'" Z-(t), and (a - ib)Z-(t)
-(a + ib) Z-(t), we have
Q*(z) = z*(z){-~
t1T
r c
Jr (a + ib) Z+(t) (t - z)
dt
+ -:-
1 1 c
t1T r (a-ib)Z+(t)
dt }
=z
'" Z( ){ 1
i1T
r ct'"
Jr (a + ib) Z+(t)(t _ z)
d
t
1
Z
- -1 ct'" dt } .
i1T r (a + ib)tZ+(t)
Substituting this in (C.3.28), we get a particular solution for the Hilbert problem
(C.3.7) for K, 2: 0 as
~ Z (z) { [ edt '" [ e r '" dt }

(z) = 2in Jr (a + ib) Z+(t) (t - z) + z Jr (a + ib) Z+(t) (t - z)
z'" Z(z) [ et-'" dt
- 2in Jr(a+ib)tZ+(t)"
(C.3.29)
3. For K, = 0, solution (C.3.29) simplifies to
~ z = Z(z) [ edt __ Z(z) [ edt

() in Jr (a + ib) Z+(t) (t - z) 2in Jr (a + ib) t Z+(t)'
(C.3.30)
CASE STUDY C.3.1: DIRICHLET PROBLEM FOR THE UNIT DISK. We

shall discuss this problem for the unit disk U+, i.e., we shall find a function u
which is harmonic in U+, continuous on U+ u r, and satisfies the boundary
condition
u = f(t), t E r, (C.3.31)
where f(t) is a real-valued continuous function given on r. This problem is

a special case of the Riemann-Hilbert problem with a = 1, b = 0, e = f(t).
Then the corresponding Hilbert problem (C.3.7) for e == 0 reduces to
(C.3.32)
This corresponds to case 1 of §C.3.1 with index K, = O. Then the fundamental

solution for the problem (C.3.32) is given by
A for z E U+,
Z(z) = { (C.3.33)
-A for z E U-,
where A is an arbitrary constant. To satisfy the condition (C.3.17) we require

that Z*(z) =' Z(z), and thus it suffices to take A = i, which yields
Z(z) = {i .-2
for z E u+,_
for z E U ,
(C.3.34)
Hence, from Theorem C.3.1 the general solution ofthe corresponding Riemann-
Hilbert problem is given by ~(z) = C i, where C is an arbitrary real constant.
Then from (C.3.30) the general solution of the nonhomogeneous Cauchy prob-
lem is given by
(z) =;..m Jrr tf(t)- z dt _ ~ r f(t) dt - iC

2m J t r
= _1_
2i1f
r f(t)
Jr t
t +z
t - z
dt + i C.
(C.3.35)
This solution is known as the Schwarz formula (see (6.4.12». Note that this
solution has been obtained by applying the results of §C.3.1, which assume
that the function f(t) satisfies the Holder condition. But the solution (C.3.35)
is also valid if instead of this condition only continuity of the function f (t) is
assumed.•
The Riemann-Hilbert problem for the half-plane can be reduced to that

for the unit disk by using the results and the notation of §1.4 and the fact that
conformal mapping of a half-plane onto the unit disk leads to an inversion.
REFERENCES USED: Kantorovich and Krylov (1958), Kythe (1996), Kythe,

Puri and SchiiJerkotter (1997), Muskhelishvili (1992), Sveshnikov and Tikhonov (1974),
Wegmann (1986).
Appendix D
Successive Approximations
D.l. Tables
The successive approximations computed for some Case Studies in Chapter 5

are presented in tabular form. The desired level of approximation is obtained
by first choosing an initial guess (in all of these tables it is taken as zero for all
cofficients). Then substituting these values into the right side of the respective
equations, the first approximation is determined. Then the values of the first
approximation are substituted into the right side of the same equations, and
the second approximation is obtained. This process is continued successively
until the desired approximation is attained.
TABLE 1
Coefficient a3 as a7 ag all
Initial Guess 0 0 0 0 0
1st Approx. _,),2 0 0 0 0
2nd Approx. _,),2 ,),4 0 0 0
3rd Approx. _,),2 _ ,),6 ,),4 _,),6 0 0
4th Approx. _,),2 _ ,),6 ,),4 + 3,),8 _,),6 ,),8 0
5th Approx. _,),2 _,),6 _ 4,),10 ,),4 + 3,),8 _,),6 _ 5,),10 ,),8 ,),10
425
TABLE 2 0I>-
l-:>
0>
Coefficient Al A2 A3 A4 As
Initial Guess 0 0 0 0 0
A
1st Approx. -- 0 0 0 0
2 >
"tl
A A2 "tl
2nd Approx. -- - 0 0 0 t':l
2 2 z
tl
A A3 A2 5A 3
3rd Approx. --+- - -- 0 0 ><
2 4 2 8 t:l
A A3 A2 9A 4 5A 3 7,X4
4th Approx. --+- --- -- - 0 U)
2 4 2 16 8 8 c::
(")
A A3 3A S A2 9A 4 5A 3 9AS 7A 4 21A S (")
Sth Approx. --+--- --- --+- - t':l
2 4 32 2 16 8 8 8 16 tI.l
tI.l
<:t':l
TABLE 3
>
"tl
Coefficient al as ag aI3 "tl
:;.l
0
><
Initial Guess 1 0 0 0 E2
k k >
'"'l
1st Approx. 1+- 0 0
16 16 <5
k 3k 2 k 7k 2 k2 Z
tI.l
2nd Approx. ---- - 0
1 + 16 + 256 16 256 64
k 3k 2 3k 3 k 7k 2 llk 3 k2 27k 3 llk 3
3rd Approx. ------- ---
1 + 16 + 256 + 1024 16 256 1024 64 + 2048 2048
0.1. TABLES 427
TABLE 4
Coefficient a3 as a7 ag
Initial Guess 0 0 0 0
1st Approx. ->. 0 0 0
2nd Approx. ->. >.2 0 0
3rd Approx. ->. + 5>.3 >.2 _>.3 0
4th Approx. ->. + 5>.3 >.2 _ 11>.3 _>.3 >.4
TABLE 5
Coefficient al as ag
Initial Guess 1 0 0
>. ->.
1st Approx. 1+- 0
16 16
>. 7>.2 >. 7>.2
2nd Approx.
1 + 16 + 256
---- 0
16 256
>. 7>.2 9>.3 >. 7>.2 19>.3 >.3
3rd Approx.
1 + 16 + 256 + 1024
-------
16 256 2048 2048
Appendix E
Catalog of Conformal Mappings
ie z -Zo
Z 1------+ e --
z; 1-zo z
A'
Z-t
Z 1------+ --:j:'""""
z t B'
.1 + z
t- +-------l Z
1-z
A B C
C'
A'
A z-1
z 1------+ Z+1
B'
B 1+z
- ----..4 Z
1-z
C
C'
428
APPENDIX E. CATALOG OF CONFORMAL MAPPINGS 429
zl----+.l
z
Z1 ----., z
A' B' C'

-2 2
A c B' A' C'

430 APPENDIX E. CATALOG OF CONFORMAL MAPPINGS
ii 1t
logz ----f Z
o
o
z ~ sin z
sin-1z ----f Z
A B C A' B' C'

-1t/2 1t/2 -I
-K+iK' K+iK'
(2.3.13)
I ..
sn (z, k) ----f Z
-Ilk -k 0 k Ilk
-K o K
Bibliography
F. ABRAMOVICI, The accurate calculation of Fourier integrals by the fast Fourier

transform technique, J. Compo Phys. 11 (1973),28-37.
M. ABRAMOWITZ and I. A. STEGUN (eds.), Handbook of Mathematical Functions,
Dover, New York, 1968.
E. N. AHLBERG and J. WALSH, The Theory of Splines and Their Applications,
Academic Press, New York, 1967.
L. V. AHLFORS, Untersuchungen zur Theorie der konformen Abbildung und der
granzen Funktionen, Act Soc. Sci. Fenn. A 1 (1930), 1-40.
Remarks on the Neumann-Poincare integral equation, Pacific J. Math. 2
(1952), 271-280.
Complex Analysis, McGraw-Hili, New York, 1966.
Ju. E. ALENICYN (10. E. AneHHcLlH), Conformal mapping of a multiply connected
domain onto many-sheeted canonical surfaces, Izv. Akad. Nauk SSSR Ser. Mat.
28 (1964), 607-644. (Russian)
Conformal mapping of multiply connected domains onto surfaces of several
sheets with rectilinear cuts, Izv. Akad. Nauk SSSR Ser. Mat. 29 (1965), 887-
902. (Russian)
CHR. ANDERSEN, S. E. CHRISTIANSEN, O. M0LLER and H. TORNEHAVE, Conformal
Mapping, Chap. 3 in Selected Numerical Methods (C. Gram, 00.), Regnecen-
tralen, Kopenhagen, 1962.
G. K. ANTONJUK (r. K. AHTOHIOK), On the covering of areas for functions regular
in an annulus, Vestnik Leningrad Univ. 13,45-65. (Russian)
K. ARBENZ, Integralgleichungenfi.i.r einige Randwertproblemefi.i.r Gebiete mit Ecken,
Dissertation, ETH Zurich (1958).
B. G. ARLINGER, Calculation of transonic flow around axisymmetric inlets, AIAA
J. 13 (1975), 1614-1621.
K. E. ATKINSON, A Survey of Numerical Methods for the Solution of Fredholm
Integral Equations of the Second Kind, SIAM, Philadelphia, 1976.
BALTRUKONIS, M. CHI and P. A. A. LAURA, Axial shear vibrations of star shaped
bars - Kohn-Kato bounds, Eighth Midwestern Mech. Conf., Developments in
Mech. 4 (1965), 449-467.
A. M. BANIN (A. M. BaHHH), Approximate conformal transformation applied to a
plane parallel flow past an arbitrary shape, PMM (IIpHKna~aR MaTXeMaTHKa
H MeqaHHKa,AKa,!{. HaYK CCCP, OT,!{eneHHe Teq. HaYK, I1HCT. MeqaHHKH 7
(1943), 131-140. (Russian)
431
432 BIBLIOGRAPHY
R. W. BARNARD and K. PEARCE, Rounding comers of gearlike domains and the

omitted area problem, J. Comput. Appl. Math. 14 (1986), 217-226; also in
Numerical Conformal Mapping (L. N. Trefethen, ed.), North-Holland, Ams-
terdam, 1986, pp. 217-226.
1. S. BEREZIN and N. P. ZHIDKOV (VI. C. Bepe3HH H H. II. lKH,lU<OB, Computing
Methods, vol. 2, Addison-Wesley, Reading, MA, and Pergamon Press, Oxford,
UK, 1965 ; translation of MeTO,D;bI BbI'IHCJIeHHH, cI>H3MaTrH3, MocKBa (Russian)
S. BERGMAN, Uber die Entwicklung der harmonischen Funktionen der Ebene und
des Raumes nach Orthogonalfunktionen, Math. Annalen 86 (1922), 237-271;;
Thesis, Berlin 1921.
Uber Bestimmung der Verzweigungspunkte eines hyperelliptischen Inte-
grals aus seinen Periodizitiitsmoduln mit Anwendungen auf die Theorie des
Transformators, Math. Zeit. 19 (1923-24), 8-25.
Uber die Berechnung des magnetischen Feldes in einem Einphasen-Trans-
formator, ZAMM 5 (1925), 319-33l.
Punch-card machine methods applied to the solution of the torsion prob-
lem, Quart. Appl. Math. 5 (1947), 69-8l.
The kernel function and conformal mapping, AMS Math. Surveys 5 (1970).
and M. SCHIFFER, Kernel functions in the theory of partial differential equa-
tions of elliptic type, Duke Math. J. 15 (1948), 535-566.
and M. SCHIFFER, Kernel functions and conformal mapping, I, II, Bull. Am.
Math. Soc. 55 (1949), 515-.
and M. SCHIFFER, Kernel functions and conformal mapping, Compositio
Math. 8 (1951), 205-249.
H. BERGSTROM, An approximation of the analytic function mapping a given domain
inside or outside the unit circle, Mem. Publ. Soc. Sci. Arts Lettr. Hainaut,
Volume hors Serie (1958), 193-198.
J .-P. BERRUT, Numerische Lasung der Symmschen Integralgleichung durch Fourier-
Methoden, Master's Thesis, ETH, Zurich, 1976.
Uber Integmlgleichungen und Fourier-Methoden zur numerischen konfor-
men Abbildung, Doctoral Thesis, ETH, Zurich, 1985.
A Fredholm integral equation of the second kind for conformal mapping, J.
Comput. Appl. Math. 14 (1986), 99-110; also in Numerical Conformal Mapping
(L. N. Trefethen, ed.), North-Holland, Amsterdam, 1986, pp. 99-110.
A. BETZ, Konforme Abbildung, Springer-Verlag, Berlin, 1964.
W. C. BICKLEY, Two-dimensional potential problem for the space outside a rec-
tangle., Proc. Lond. Math. Soc., Ser. 2, 37 (1932), 82-105.
L. BIEBERBACH, Zur Theorie und Praxis der konformen Abbildung, Rend. del Cir-
colo mat. Palermo 38 (1914), 98-112.
Uber die konforme Kreisabbildung nahezu kreisjOrmig Bereiche, Sitzungs-
bereichte der Preuss. Akad. Wissen. (1924), 181-188.
G. BIRKOFF and D. YOUNG, Numerical quadrature of analytic and harmonic func-
tions, J. Math. Phys. 29 (1950), 217-22l.
D. YOUNG and E. H. ZARANTONELLO, Numerical methods in conformal
mapping, Am. Math. Soc., Proc. Fourth Symposium Appl. Math. (1951), McGraw·
Hill, New York, 1953, 117-140.
and E. H. ZARANTONELLO, Jets, Wakes and Cavities, Academic Press, New
York,1957.
.D. M. YOUNG and E. H. ZARANTONELLO, Effective conformal transforma-
tton of smooth simply connected domains, Proc. Nat. Acad. Sci. USA 31 (1951),
411-414.
BIBLIOGRAPHY 433
D. R. BLASKETT and H. SCHWERDTFEGER, A formula for the solution of an arbi-

trary analytic function, Quart. Appl. Math. 3 (1945-46), 266-268.
R. P. BOAs, Invitation to Complex Analysis, Random House. New York, 1987.
W. BURNSIDE, On functions determined from their discontinuities, and a certain
form of boundary condition, Proc. Royal Soc. London, Ser. A 22 (1891), 346-
358.
C. CARATHEODORY, Elementarer Beweis fur den Fundamentalsatz der konformen
Abbildung, Schwarz-Festschrift, J. Springer, Berlin (1914), 19-4l.
Uber das Neumann-Poincaresche Problem fur ein Gebiet mit Ecken, Dis-
sertation, Uppsala, 1916.
Conformal Representation, Cambridge Tracts in Math. and Math. Phys.,
No. 28, London; Cambridge University Press, Cambridge, 1969.
T. CARLEMAN, Sur la resolution de certaines equations integrales, Archiv for math-
ematik astronomi och fysik 16 (1922).
G. F. CARRIER, On a conformal mapping technique, Quart. Appl. Math. 5 (1947),
101-104.
M. KROOK and C. E. PEARSON, Functions of a Complex Variable: Theory
and Technique, McGraw-Hill, New York, 1966.
M. J. CASARELLA, P. A. A. LAURA and M. CHI, On the approximate solution of flow
and heat transfer through non-circular conduits with uniform wall temperature,
Brit. J. Appl. Phys. 18 (1966), 1327-1335.
and N. FERRAGUT, On the approximate solution of flow and heat transfer
through non-circular conduits with uniform wall temperature and heat genera-
tion, Nucl. Engg. Design 16 (1971), 387-398.
S. CHAKRAVARTHY and D. ANDERSON, Numerical conformal mapping, Math. Camp.
33 (1979), 953-969.
N. V. CHALLIS and D. M. BURLEY, A numerical method for conformal mapping,
IMA J. Numer. Anal. 2 (1982), 169-18l.
S. A. CHAPLYGIN (C. A. l:.J:anJIblrJ'IH), A theory of grid wing (1914), in Collected
Papers, vol. 2, Gostekhizdat, Moscow, 1933. (Russian)
E. B. CHRISTOFFEL, Sui problema della temperature stazionaire e la reppresen-
tazione di una data superficie, Annali di Matematica (2) 1 (1867-1868), 89-
104.
J. W. COOLEY and J. W. TUKEY, An algorithm for the machine calculation of
complex Fourier series, Math. Camp. 19 (1965), 297-30l.
P. A. W. LEWIS and P. D. WELCH, The fast Fourier transform algorithm:
Programming considerations in the calculation of sine, cosine, and Laplace
transforms, J. Sound Vibration 12 (1970),315-337.
E. T. COPSON, Partial Differential Equations, Cambridge University Press, London,
1975.
L. DE BRANGES, A proof of the Bieberbach conjecture, Acta Math. 154 (1985),
137-152.
L. M. DELVES, A fast method for the solution of Fredholm integral equation, J. Inst.
Math. Appl. 20 (1977), 173-182.
and J. WALSH (eds.), Numerical Solution of Integral Equations, Clarendon
Press, Oxford, 1974.
and J. L. MOHAMED, Computational Methods for Integral Equations, Cam-
bridge University Press, Cambridge, 1985.
H. DERESIEWICZ, Thermal stress in a plate due to disturbance of uniform heat flow
in a hole of general shape, J. Appl. Mech., Trans. ASME 28 (1961), 147-149.
J. DOUGLAS, Solution of the problem of Plateau, Trans. Am. Math. Soc. 33 (1931),
434 BIBLIOGRAPHY
263-321.
A. R. ELCRAT, Separated flow past a plate with spoiler, SIAM J. Math. Anal. 13
(1982), 632-639.
and L. N. TREFETHEN, Classical free-streamline flow over a polygonal obsta-
cle, J. Comput. Appl. Math. 14 (1986), 251-265; also in Numerical Conformal
Mapping (L. N. Trefethen, ed.), North-Holland, Amsterdam, 1986, pp. 251-265.
B. EpSTEIN, A method for the solution of the Dirichlet problem for certain types of
domains, Quart. Appl. Math. 6 (1948), 301-317.
A. L. FLORENCE and J. N. GOODIER, Thermal stress due to disturbance of uniform
heat flow by an insulated ovaloid hole, J. Appl. Mech., Trans. ASME 27 (1960),
635-639.
J. M. FLORYAN, Conformal-mapping-based coordinate generation method for chan-
nel flows, J. Comput. Phys. 58 (1985), 229-245.
and C. ZEMACH, Schwarz-Christoffel mappings: A general approach, J.
Comput. Phys. 72 (1987), 347-371.
V. FOCK, Uber die konforme Abbildung eines Kreisvierecks mit verschwindenden
Winkeln, J. Reine angew. Math. 161 (1929), 137-151.
B. FORNBERG, A numerical method for conformal mapping, SIAM J. Sci. Comput.
1 (1980), 386-400.
A numerical method for conformal mapping of doubly connected regions,
SIAM J. Sci. Stat. Compo 5 (1984), 771-783.
P. FRANKLIN, Methods of Advanced Calculus, McGraw-Hill, New York, 1944.
M. S. FREIBERG, A new method for the effective determination of conformal maps,
Ph. D. Thesis, Univ. of Minnesota, 1951.
D. GAIER, Konstruktive Methoden der konformen Abbildung, Springer Tracts in
Natural Philosophy, vol. 3, Springer-Verlag, Berlin, 1964.
Integralgleichung erster Art und konforme Abbildung, Math. Zeit. 147
(1976), 113-129.
Numerical aspects in conformal mapping, in Computational Aspects of
Complex Analysis, Reidel, Dordrecht, Boston, 1983, pp. 51-78.
. Vorlesungen iiber Approximation in Komplexen, Birkhauser Verlag, Basel,
1980.
and O. HUBNER, Schnelle Auswertung von Ax bei Matrizen A zyklischer
Bauart, Toeplitz- und Hankel-Matrizen, Mitt. Math. Sem. Giessen 121 (1976),
27-38.
F. D. GAKHOV (cp. 11. raxoB), On the Riemann boundary problem, Matern. Sbornik
2(44) (1937), 165-170.
R. W. G. GANDY and R. V. SOUTHWELL, Relaxation methods applied to engineering
problems. V. Conformal transformation of a region in plane space, Phil. Trans.
Royal Soc. (London), Ser. A 238 (1940), 453-475.
I. E. GARRICK, Potential flow about arbitrary biplane wing sections, Report 542,
NACA (1936.).
Conformal mapping in aerodynamics, with emphasis on the method of suc-
cessive conjugates, Symposium on the Construction and Applications of Confor-
mal Maps, National Bureau of Standards, Appl. Math. Ser. 18 (1949), 137-147.
C. F. GAUSS, Allgemeine Auflosung der Aufgabe die Thiele einer gegebener Fliiche
so abzubilden, dapdie Abbildung dem abgebildeten in den kleinsten Thielen
iihnlich wird, Werke IV (1822), 189-216.
W. GAUTSCHI, The condition of orthogonal polynomials, Math. Compo 26 (1977),
923-924.
Questions of numerical condition related to polynomials, Recent Advances
BIBLIOGRAPHY 435
in Numerical Analysis (C. de Boor and G. Golub, eds.), Academic Press, New
York, 1978.
Condition of polynomials in power form, Math. Compo 33 (1979), 343-352.
P. R. GERABEDIAN and M. M. SCHIFFER, Identities in the theory of conformal
mapping, Trans. Am. Math. Soc. 65 (1949), 187-238.
S. 1. GERSHGORIN (C. 11. repmropHH), On the conformal mapping of a simply
connected region onto a disc, Matern. Sbornik 40 (1933), 48-58. (Russian)
D. GILBARG, A generolization of the Schwarz-Christoffel tronsformation, Proc.
Nat. Acad. Sci. 35 (1949), 609-611.
Jets and Cavities, in Handbuch der Physik, 9, Springer-Verlag, Berlin,
1960, pp. 311-445.
H. GLAUERT, The Elements of Aerofoil and Airscrew Theory, Cambridge University
Press, London, 1948.
G. M. GOLUZIN (r. M. rOJlY3HH), The solution of fundamental plane problems of
mathematical physics in the case of Laplace's equation and multiply connected
regions bounded by circles, Matern. Sbornik 41 (1934), 246-276. (Russian)
Iterationsprozesse fUr konforme Abbildungen mehrfach zusammenhiingender
Bereiche, Matern. Sbornik, New Series 6 (41) (1939),377-382. (Russian)
On conformal mapping of doubly connected regions bounded by rectilinear
and circular polygons, in Conformal Mapping of Simply and Doubly Connected
Regions, ONTI, Moscow, 1937, pp. 90-97. (Russian)
Conformal mapping of multiply connected regions on a slit plane by a
method of functional equations, in Conformal Mapping of Simply and Doubly
Connected Regions, ONTI, Moscow, 1937, pp. 98-110. (Russian)
Geometrische Funktionentheorie, Deutscher Verlag der Wissenschaften,
Berlin, 1957 ; translation of reOMeTpHqeCKaJI TeopHJI <l>yHKUHH KOMITJleKCHOro
IIepeMeHHoro, MocKBa-JIeHHHrpa.n, 1952. (Russian)
Geometric Theory of Functions of a Complex Variable, Trans!. Math.
Monographs vo!. 26, American Mathematical Society, Providence, RI, 1969; trans-
lation ofreoMeTpHqecKaJl TeopHJI <l>yHKUHtt KOMnJleKCHOro IIepeMeHHoro, HayKa,
MocKBa, 1966. (Russian)
C. Gram (ed.), Selected Numerical Methods, Regnecentralen, Kopenhagen, 1962.
T. H. GRONWALL, Some remarks on conformal representation, Ann. Math. 16
(1914/ 15), 72-76.
H. GROTZSCH, Zur konformen Abbildung mehrfach zusammenhiingender, schlichter
Bereiche (Iterationsverfahren), Berichte Verhand. Sii.chsichen Akad. Wissen. zu
Leipzig, Math.-Phys. Klasse 83 (1931), 67-76.
H. GRUNSKY, Neue Abschiitzungen zur konformen Abbild ein- und mehrfach zusam-
menhiingender Berichte, Schr. Math. Seminars Inst. Angew. Math. Univ. Berlin
1 (1932),93-140.
M. 1. GUREVICH (M. 11. rypeBHq), Theory of Jets in Ideal Fluids, Academic Press,
New York, 1965; translation of TeopHJI CTpHe 11.l\aJl'HOe >KH,lU<OCTH, rOCH-
.napCTBeHHoe 113.L\aJleJl'CTBO <l>H3.-MaT. JIHT., MocKBa, 1961 (Russian)
M. H. GUTKNECHT, Existence of a solution to the discrete Theodorsen equation for
conformal mapping, Math. Comp. 31 (1977),478-480.
Fast algorithms for the conjugate periodic function, Computing 22 (1979),
79-91.
Solving Theodorsen's integral equation for conformal maps with the fast
Fourier tronsform and various nonlinear iterotive methods, Num. Math. 36
(1981), 405-429.
On the computation of the conjugate trigonometric rational function and
436 BIBLIOGRAPHY
on a related splitting problem, SIAM J. Numer. Anal. 20 (1983), 1198-1205.

Numerical experiments on solving Theodorsen's integral equation for con-
formal maps with the fast Fourier transform and various nonlinear iterative
methods, SIAM J. Sci. Stat. Comput. 4 (1983), 1-30.
Numerical conformal mapping methods based on function conjugation, J.
Comput. Appl. Math. 14 (1986), 31-77; also in Numerical Conformal Mapping
(L. N. Trefethen, ed.), North-Holland, Amsterdam, 1986, pp. 31-77.
J. HADAMARD, Memoire sur Ie probleme d'analyse relatif Ii l'equalibre de plaques
elastiques encastrees, Memoires Acad. Sci. 33, No.4 (1908).
N. D. HALSEY, Potential flow analysis of multielement airfoils using conformal
mapping, AIAA J. 17 (1979), 1281-1288.
Comparison of the convergence characteristics of two conformal mapping
methods, AIAA J. 20 (1982), 724-726.
M. HARTMANN and G. OPFER, Uniform approximation as a numerical tool for con-
structing conformal maps, J. Comput. Appl. Math. 14 (1986), 193-206; also
in Numerical Conformal Mapping (L. N. Trefethen, ed.), North-Holland, Am-
sterdam, 1986, pp. 193-206.
J. K. HAYES, D. K. KAHANER and R. G. KELLNER, An improved method for nu-
merical conformal mapping, Math. Comput. 26 (1972), 327-334.
A numerical comparison of integral equations of the first and second kind
for conformal mapping, Math. Compo 29 (1975), 512-521.
J. HEINHOLD, Zur Praxis der konformen Abbildung, Berichte Math. Tagung, Tiibigen
(1947), 75-77.
P. HENRICI, Einige Anwendungen der schnellen Fouriertransformation, in Moderne
Methoden der Numerischen Mathematik (J. Albrecht and L. Collatz, eds.),
Birkhauser, Basel, 1976, pp. 111-124.
Fast Fourier methods in computational complex analysis, SIAM Rev. 21
(1979), 481-527.
Barycentric formulas for trigonometric interpolation, Numer. Math. 33
(1979), 225-234.
Applied and Computational Complex Analysis, vol. III, Wiley, New York,
1986.
D. HILBERT, Grundzuge einer allgemeinen Theorie der linearen Integralgleichun-
gen, 2nd ed., Leipzig-Berlin, 1924, pp. 83-94.
Das Dirichletsche Prinzip (1901), in Gesammelte Abhandlungen, Bd. 3,
Springer-Verlag, Berlin, 1935.
Zur Theorie der konformen Abbildung (1909), in Gesammelte Abhandlun-
gen, Bd. 3, Springer-Verlag, Berlin, 1935.
J. HODGKINSON, Conformal representation by means of Lame functions, J. London
Math. Soc. 5 (1930), 296-306.
and E. POOLE, The conformal representation of the area of a plane bounded
by two straight or circular slits, Proc. London Math. Soc. 23 (1924), 396-422.
F. HOHNDORF, Verfahren zur Berechnung des Auftriebes gegebener Tragfliichen-
Profile, ZAMM 6 (1926), 265-283.
R. HOCKNEY, The potential calculation and some applications, Methods in Comput.
Phys. 9 (1970), 135-211.
H.-P. HOIDN, Osculation methods for the conformal mapping of doubly connected
regions, ZAMP 33 (1982), 640-652 .
. A reparameterisation method to determine conformal maps, J. Comput.
Appl. Math. 14 (1986), 155-161; also in Numerical Conformal Mapping (L. N.
Trefethen, ed.), North-Holland, Amsterdam, 1986, pp. 155-161.
BIBLIOGRAPHY 437
D. M. HOUGH and N. PAPAMICHAEL, The use of splines and singular functions in

an integral equation method for conformal mapping, Numer. Math. 37 (1981),
133-147.
An integral equation method for the numerical conformal mapping of in-
terior, exterior and doubly-connected domains, Numer. Math. 41 (1983), 287-
307.
G. C. HSIAO, P. Kopp and W. L. WENDLAND, A Galerkin collocation method for
some integral equations of the first kind, Computing 25 (1980), 89-130.
O. HUBNER, Zur Numerik der Theodorsenschen Integralgleichung in der konformen
Abbildung, Mitt. Math. Sem. Giessen 140 (1979), 1-32.
Uber die Anzahl der Losungen der diskreten Theodorsen-Gleichung, Nu-
mer. Math. 39 (1982), 195-204.
The Newton method for solving the Theodorsen integral equation, J. Com-
put. Appl. Math. 14 (1986), 19-30; also in Numerical Conformal Mapping (L.
N. Trefethen, ed.), North-Holland, Amsterdam, 1986, pp. 19-30.
K. INOUE, Grid generation for cascades using conformal mapping, J. Comput. Phys.
52 (1983), 130-140.
Grid generation for inlet configurations using conformal mapping, J. Com-
put. Phys. 58 (1985), 146-154.
V. I. IVANOV and M. K. TRUBETSKOV, Handbook of Conformal Mapping with
Computer-Aided Visualization, CRC Press, Boca Raton, 1995.
V. V. Ivanov (B. B. MBaHoB), Theory of Approximate Methods and Applications to
Numerical Solution of Singular Integral Equations, Kiev, 1968.
D. C. IVES, A modern look at conformal mapping including multiply connected
regions, AIAA J. 14 (1976), 1006-1011.
and J. F. LIUTERMOZA, Analysis of transonic cascade flow conformal map-
ping and relaxation techniques, AIAA J. 15 (1977),647-652.
R. M. JAMES, A new look at two-dimensional incompressible airfoil theory, Report
J0918/01, Douglas Aircraft Co., Long Beach, CA (May 1971).
M. A. JAWSON, Integral equation methods in potential theory. 1., Proc. Roy. Soc. A
275 (1963), 23-32.
and G. T. SYMM, Integral Equation Methods in Potential Theory and Elas-
tostatics, Academic Press, London, 1977.
A. JEFFREY, Complex Analysis and Applications, CRC Press, Boca Raton, FL,
1992.
H. JEFFREYS and B. S. JEFFREYS, Methods of Mathematical Physics, Cambridge
University Press, Cambridge, 1946.
N. E. JOUKOWSKI (H. E. I:KYKOBCKH), Modification of Kirchhoff's method for de-
termination of fluid motion in two dimensions with constant velocity (1890),
in Collected Papers, vol. 2, Gostekhizdat, Moscow, 1950. (Russian)
Uber die Konturen des Tragflachen der Drachenflieger (1910-1912), in Col-
lected Papers, vol. 5, Gl. red. avo lit., Moscow, 1937. (Russian)
Theoretical foundation of aeronautics (1911), in Collected Papers, vol. 6,
Gostekhizdat, Moscow, 1950. (Russian)
Vorticity theory of the propeller (1915), in Collected Papers, vol. 4, Gostekhiz-
dat, Moscow, 1950. (Russian)
G. JULIA, Sur le representation conforme des aires simplement convexes, Compltes
Rendus Acad. Sci. Paris. 182 (1926), 1314-1316.
Sur une serie de polynomes liee a la le representation conforme des aires
simplement convexes, Compltes Rendus Acad. Sci. Paris. 183 (1927), 10-12.
Development en serie de polynomes ou de fonctions rationelles de La fonc-
438 BIBLIOGRAPHY
tion qui fournit la representation conforme d 'une aire simplement convexe sur
un cercle, Annales de l'Ecole normale Sup. 44 (1927), 289-316.
L. V. KANTOROVICH (JI. B. KaHTOpOBHq), Sur quelques methodes de la determination
de la fonction qui effectue une representation conforme, Bull. Acad. Sci. URSS
1 (1933), 229-235.
Sur la representation conforme, Matern. Sbornik 40 (1933), 294-325.
Quelques rectifications d mon memoire "Sur la representation conforme",
Matern. Sbornik 41 (1934), 179-182.
On the approximate computation of some types of definite integrals and
other applications of the method of removing singularities, Matern. Sb. (2) 41
(1934), 235-245. (Russian)
Sur la representation conforme des domaines multiconvexes, Comptes Ren-
dus Acad. Sci. URSS 2 (1934), 441-445.
Conformal mapping of a circle on a simply connected region, in Conformal
Mapping of Simply and Doubly Connected Regions, Gostekhizdat, Leningrad-
Moscow, 1937. (Russian)
and V. I. KRYLOV (B. 11. KPbIJIOB), Methods for the Approximate Solu-
tion of Partial Differential Equations, Gostekhizdat, Leningrad-Moscow, 1936.
(Russian)
and V. I. KRYLOV, Approximate Methods for Higher Analysis, Interscience,
New York, 1958.
V. A. KASIN and V. V. MERKULOV, Determination of the eigenvalues fir waveguide
with complex cross section, Soviet Phys.: Acoustics 11 (1966), 285-287.
A. J. KASSAB, R. T. BAILEY and C. K. HSIEH, CVBEM in simply- and multiply-
connected domains for the solution of heat conduction problems, in Advances in
Boundary Element Methods (M. H. Aliabadi and C. A. Brebbia, OOs.), Compu-
tational Mechanics Publications, Southampton, and Elsevier Applied Science,
London, 1993.
M. V. KELDYS (M. B. KeJl,llblIII), Conformal mappings of multiply connected do-
mains on canonical domains, Uspekhi Matern. Nauk 6 (1939), 90-119 transla-
tion of KOH4JopMble oTo6pa»<eHHHH MHorOUBH3HblX 06J1aCTel:t Ha KaHOHHlJeClilie
06J1aCTH, YcneXH MaT. HayK, 6 (1939), 90-119. (Russian)
O. D. Kellogg, Foundations of Potential Theory, Springer-Verlag, Berlin, 1929;Dover,
New York, 1953.
N. KERZMAN and E. M. STEIN, The Cauchy kernel, the Szego kernel, and the
Riemann mapping function, Math. Ann. 236 (1978), 85-93.
and M. R. TRUMMER, Numerical conformal mapping via the Sgezo kernel,
J. Comput. Appl. Math. 14 (1986), 111-123; also in Numerical Conformal
Mapping (L. N. Trefethen, ed.), North-Holland, Amsterdam, 1986, pp. 111-
123.
G. KHAJALIA (r. XajaJlHH), Sur la theorie de la representation conforme des do-
maines doublement convexes, Matern. Sbornik, New Series 8 (1940), 97-106.
B. V. KHVEDELIDZE (S. B. XBe,lleJlH,ll3e), On the Poincare boundary value prob-
lem of the theory of logarithmic potential for multiply connected regions, Soob-
shcheniya AN Gruz. SSR (Coo6WmeHHl1 AH rpy3. CCP) 2 (1941), 571-578,
865-872. (Russian)
G. R. KIRCHHOFF, Zur Theorie freier FLUssigketsstrhlen, J. Reine Angew. Math.
10 (1869), 289-298.
P. KNUPP and S. STEINBERG, FUndamentals of Grid Generation, CRC Press, Boca
Raton, FL, 1993.
H. KOBER, Dictionary of Conformal Representations, Admiralty Computing Ser-
BIBLIOGRAPHY 439
vice, Dept. of Scientific Research and Experiment, Admiralty, London, Parts

1-5, 1945-1948.
Dictionary of Conformal Representations, Dover, New York, 1952; 2nd
ed., Dover, New York, 1957.
P. KOEBE, Abhandlungen zur Theorie der konformen Abbildung. I. Die Kreisabbil-
dung des allgemeisnten einfach und zweifach zusammenhiingenden schlichten
Bereichs und die Riinderzuordnung bei konformer Abbildung, J. Reine Angew.
Math. 145 (1915), 177-223.
Abhandlungen zur Theorie der konformen Abbildung. IV. Abbildung mehr-
fach zusammenhiingender schlichter Bereiche auf Schlichtbereiche, Acta Math.
41 (1918), 304-344.
W. VON KOPPENFELS, Das hypergeometrische Integral als Periode der Vierecksab-
bildung, Sitzungsberichte Akad. Wissen. Wien 146 (1937), 11-22.
Konforme Abbildung besonderer Kreisbogenvierecke, J. Reine Angew. Math.
181 (1939),83-124.
and F. STALLMANN, Praxis der konformen Abbildung, Springer-Verlag, Berlin,
1959.
V. 1. KRYLOV (B. 11. KPbIJlOB), Concerning a method of constructing a function
which maps a region conformally on a circle, in Conformal Mapping of Sim-
ply and Doubly Connected Regions, Gostekhizdat, Leningrad-Moscow, 1937.
(Russian)
Integral equations of the conformal map of a region on a slit plane, in Con-
formal Mapping of Simply and Doubly Connected Regions, Leningrad-Moscow,
1937. (Russian)
Integral equations of the conformal map of doubly connected regions on a
ring, in Conformal Mapping of Simply and Doubly Connected Regions, Gostekhiz-
dat, Leningrad-Moscow, 1937. (Russian)
An application of integral equations to the proof of certain theorems for
conformal mapping, Matern. Sbornik 4 (1938), 9-30. (Russian)
and N. BOGOLYUBOV, Sur la solution approchee du probleme de Dirichlet,
Comptes Rendus Acad. Sci. URSS (1929), 283-288.
P. P. KUFAREV (n. n. Kyq,apeB), Uber das zweifachzusammenhiingende Minimal-
gebiet, Bull. de I'Institut de Math. et Mecan. Univ. Kouybycheff de Tornsk 1
(1935-1937), 228-236.
On a method of numerical determination of the parameters in the Schwarz-
Christoffel integral, Doklady Akad. Nauk SSSR (new Series) 57 (1947), 535-
537. (Russian)
On conformal mapping of complementary regions, Doklady Akad. Nauk
SSSR 73 (1950), 881-884.
U. KULISCH, Ein Iterationsverfahren zur konformen Abbildung des Einheitskreises
auf ein Stem, ZAMM 43 (1963), 403-410.
P. K. KULSHRESTHA, Distortion of spiral-like mappings, Proc. Royal Irish Acad.
73, Sec. A (1973), 1-5.
Generalized convexity in conformal mappings, J. Math. Anal. Appl. 43
(1973), 441-449.
Coefficients for alpha-convex univalent functions, Bull. Amer. Math. Soc.
80 (1974), 341-342.
Coefficient problem for alpha-convex univalent functions, Archive Ratnl.
Mech. Anal. 54 (1974), 204-211.
Coefficient problem for a class of Mo~anu-Bazilevic functions, Annales
Polon. Math. 31 (1976), 291-299.
440 BIBLIOGRAPHY
Bounded Robertson functions, Rendi. di Matern. 9, Ser. VI (1976), 137-

150.
Convex combinations of some regular locally univalent functions, Bul!.
Inst. Math., Acad. Sinica 12 (1984), 37-50.
P. K. KYTHE, An Introduction to Boundary Element Methods, CRC Press, Boca
Raton, FL, 1995.
Fundamental Solutions for Differential Operators and Applications, Birk-
hauser, Boston, 1996.
P. PURl and M. R. SCHAFERKOTTER, Partial Differential Equations and
Mathematica, CRC Press, Boca Raton, FL, 1997.
E. LANDAU, Einige Bemerkungen iiber schlichte Abbildung, Jahresbericht Deut.
Math.-Vereinigung 34 (1926), 239-243.
P. A. A. LAURA, Calculations of eigenvalues for uniform fluid waveguide with
complicated cross section, J. Acoust. Soc. Am. 42 (1967), 21-26.
A survey of modem applications of the method of conformal mapping,
Revista de la Union Mathematica Argentina 27 (1975), 167-179.
and M. CHI, Approximate method for the study of heat conduction in bars
of arbitrary cross section, J. Heat Transfer, Trans. ASME 86, Ser. C (1964),
466-467.
and M. CHI, An application of conformal mapping to a three-dimensional
unsteady heat conduction problem, Aeronaut. Quart. 16 (1965), 221-230.
and A. J. FAULSTICH, Unsteady heat conduction in plates of polygonal shape,
Int. J. Heat Mass Transfer 11 (1968), 297-303.
and P. A. SHAHADY, Longitudinal vibrations of a solid propellant rocket mo-
tor, Proc. Third Southeastern Conf. on Theor. App!. Mech. (1966), Pergamon
Press, New York, 623-633.
E. ROMANELLI and M. J. MAURIZI, On the analysis of waveguides of doubly-
connected cross section by the method of conformal mapping, J. Sound Vibra-
tion 20 (1972), 27-38.
M. A. LAVRENT'EV (M. A. JIanpeHTben), K Teopmi KOHljlOpM6bIX OTo6pa)f(eHHll,
TpY.llbI ljlH3.-MaT. H-Ta. HM. B. A. CTeKJIOna, 1934. (Russian)
Boundary problems in the theory of univalent functions, Amer. Math. Soc.
Trans!. (2) 32 (1963), 1-35; translation of M. A. JIanpeHTen, 0 HeKOTopHX
rpaHH'lXbIX 3a.lla'IaX n TeopHH O.llHOJIHCTHbIX ljlYHKI.\Hll, MaT. C6., 1 (43), 1936,
815-844 (Russian)
Conformal Mappings with Applications to Some Problems of Mechan-
ics, Gostekhizdat, Moscow-Leningrad, 1947 translation of KOHljlopMble OTO-
6pameHYH ~ npHJIO)f(eHHHMH K HeKOTopHH nonpo~aM MeKaHHKH, rOCTeXH3.llaT,
MO~Kna-JIeHHHrpa.ll, 1947. (Russian)
On the theory of conformal mappings, Amer. Math. Soc. Trans!. (2) 122
(1984), 1-69; translation of M. A. JIanpeHTen, K TeopHH KOHljlOpMHbIX OTO-
6pameHHH, TpY.llbI <PH3.-MaT. I1HCT. CTeKJIOna, 5 (1934), 159-245 (Russian)
M. A. LAVRENTJEW and B. W. SCHABAT (M. A. JIanpeHTben Y B. B. IIIa6aT),
Methoden der komplexen Funktiontheorie, VEB Deutscher Verlag der Wissen-
schaften, Berlin, 1967; translation of MeTO.llH TeopHH ljlYHKI.\Hll KOMnJIeKCHOro
nepeMeHHoro, 113.llaT. «HaYKa», MO~Kna, 1965. (Russian)
N. N. LEBEDEV (H. H. JIe6e.llen), On the theory of conformal mappings of a circle
onto nonoverlapping regions, Doklady Akad. Nauk SSSR 103 (1955), 553-555.
(Russian)
R. S. LEHMAN, Development of the mapping function at an analytic comer, Pacific
J. Math. 7 (1957), 1437-1449.
BIBLIOGRAPHY 441
O. LEHTO and K. I. VIRTAANEN, Quasiconformal Mappings in the Plane, Springer-

Verlag, Berlin, 1973.
F. LEJA, Une methode de construction de la fonction de Green appartenant Ii un
domain plan quelconque, Comptes Rendus Acad. Sci. Paris 198 (1934), 231-
234.
Construction de la fonction analytique effectuant la representation con-
forme d'un domaine plan quelconque sur Ie cercle, Math. Ann. 111 (1935),
501-504.
Sur une suite de polyn6mes et la representation conforme d'un domain plan
quelconque sur Ie cercle, Annales Soc. Polonaise Math. 14 (1936), 116-134.
A. F. LEONT'EV (A. <P. JIeoHTl'IeB), Representation of functions in convex domains
by generalized exponential series, Amer. Math. Soc. Transl. (2) 140 (1988),
121-130; translation of IIpe,llcTaBJIeHl'Ie <pbIHKuyi1 B Bl'InyKJIbIX 06JIaCTJIX 0606-
Il\eHHbIMI'I pJI,llaMI'I 3KcnOHeHT, Acta Sc. Math. 45 (1983), 305-315.
D. LEVIN, N. PAPA MICHAEL and A. SIDERIDIS, The Bergman kernel method for
an improved conformal mapping of simply-connected domains, J. Inst. Math.
Appl. 22 (1978),171-187.
H. LEWY, Developments at the confluence of analytic boundary conditions, Univ.
Calif. Publ. Math. 1 (1950),247-280.
L. LICHTENSTEIN, Uber die konforme Abbildung ebener analytischer Gebiete mit
Ecken, J. Reine Angew. Math. 140 (1911), 100-119.
Zur konformen Abbildung einfach zusammenhiingender schlichter Gebiete,
Archiv Math. Phys. 25 (1917), 179-180.
H. LIEBMANN, Die angeniiherte Ermittlung harmonischer Funktionen unf konformer
Abbildungen (nach Ideen von Boltzmann und Jacobi), Sitzungberichte math.-
phys. Klasse Akad. Wissen. Miinchen (1918), 385-416.
K. L6wNER, Untersuchungen iiber schlichte konforme Abbildungen des Einheit-
skreises, I., Math. Ann. 89 (1923), 103-121.
A. J. LURYE (A. H. JIypbe), Zur Schwarz-Christoffelschen Formel, Annales Inst.
Polytech. Nom M. I. Kalinin, Leningrad 30 (1927), 113-120. (Russian)
J. Lu, A class of quadrature formulas of Chebyshev type for singular integrals, J.
Math. Anal. Appl. (2) 100 (1984), 416-435.
Boundary value problems for analytic functions and integral equations with
transformations, in Complex Analysis and Its Applications (C. C. Yang, G. C.
Wen, K. Y. Li and Y. M. Chiang, eds.), Pitman Res. Notes in Math. Series #
305, Longman Scientific and Technical, Essex, UK, 1994, pp. 318-323.
L. A. LYUSTERNIK (JI. A. JIlOcTepHI'IK), Uber einige Anwendungen der direkten
Methoden in der Variationsrechnung, Matern. Sbornik 33 (1926), 173-201.
(Russian)
Remarks on the numerical solution of boundary problems for Laplace's
equation and the calculation of characteristic values by the method of networks,
Trudy Mathern. Inst. V. A. Steklova 20 (1947), 49-64. (Russian)
R. E. MADER, Programming in Mathematica, 2nd ed., Addison-Wesley, Redwood
City, CA, 1991.
M. MAlT!, A note on the integral equation methods in potential theory, Quart. Appl.
Math. 25 (1968), 480--484.
P. B. MALENTIEV (II. B. MaJIeHTl'IeB), Approximate conformal mapping, in Confor-
mal Mapping of Simply and Doubly Connected Regions, Gostekhizdat, Leningrad'
Moscow, 1937. (Russian)
A. I. MARKUSHEVICH (A. H. MapKymeBI'I'I), TeopHJI aHaJII'ITI'I'IeCKI'IX <PYHK~I'Ii1, roc-
TeXI'I3,llaT, MocKBa-JIeHI'IHrpa,ll, 1950. (Russian)
442 BIBLIOGRAPHY
J. E. MARSDEN and M. J. HOFFMAN, Basic Complex Analysis, 2nd ed., W. H.

Freeman and Company, New York, 1987.
A. MAYO, The fast solution of Poisson's and the biharmonic equations on irregular
regions, SIAM J. Numer. Ana!. 21 (1984), 285-299.
Rapid methods for the conformal mapping of multiply connected regions,
J. Comput. App!. Math. 14 (1986), 143-153; also in Numerical Conformal
Mapping (L. N. Trefethen, ed.), North-Holland, Amsterdam, 1986, pp. 143-
153.
D. 1. MEIRON, S. A. ORSZAG and M. ISRAELI, Applications of numerical conformal
mapping, J. Compo Phys. 40 (1981), 345-360.
P. V. MELENT'EV (II. B. MeJIeHTbeB), HecKoJI'bKO Ho6bIX MeTO,l(OB II npHeMoB npH-
6JIIIlKeHHblx BbIqHCJIeHH~ (Several New Methods and Devices for Approximate
Computations), OHTH, JIeHIIHf'pa,l(-MoCKBa, 1937. (Russian)
R. MENIKOFF and C. ZEMACH, Methods for numerical conformal mapping, J. Compo
Phys. 36 (1980), 366-410.
Rayleigh-Taylor instability and the use of conformal maps for ideal fluid
flow, J. Comput. Phys. 51 (1983),28-64.
S. G. MIKHLIN (C. r. MHXJIHH), Integral Equations and Their Applications, Perg-
amon, London, 1957.
V. N. MONAKOV, Boundary- Value Problems with Free Boundaries for Elliptic Sys-
tems of Equations, Amererican Mathematical Society, Providence, RI, 1983.
M. MULLER, Zur konforme Abbildung angeniihert kreisformiger Gebiete, Math. Zeit.
43 (1938), 628-636.
R. MUNIPALLI and D. A. ANDERSEN, An adaptive grid scheme using the boundary
element method, J. Comput. Phys. 127 (1996), 452-463.
M.1. MURATOV (M. H. MypaToB), Conformal mapping of a half-plane on a region
close to it, in Conformal Mapping of Simply and Doubly Connected Regions,
Gostekhizdat, Leningrad-Moscow, 1937. (Russian)
W. MURRY, Failure, the Causes and Cures, Chap. 7 in Numerical Methods for
Unconstrained Optimization (W. Murray, ed.), Academic Press, London, 1972,
p.118.
N. 1. MUSKHELISHVILI (H. H. MycXeJIHWBHJIH), Some Basic Problems of the Math-
ematical Theory of Elasticity, Noordhoff, Groningen, 1963.
Singular Integral Equations, Noordhoff, Leiden, 1953; 2nd ed., Dover, New
York, 1992; translation of CHHryJIJIpHble HHTerpaJIbHbIe YpaBHeHJI, MocKBa,
1946. (Russian)
P. S. NAIDU and K. O. WESTPHAL, Some theoretical considerations of ion optics of
the mass spectrometer ion source - I, Brit. J. App!. Phys. 17 (1966), 645-651.
Some theoretical considerations of ion optics of the mass spectrometer ion
source - II, Brit. J. App!. Phys. 17 (1966), 653--656.
Z. NEHARI, The Schwarzian derivative and schlicht functions, Bull. Am. Math. Soc.
55 (1949), 545-551.
The kernel function and canonical conformal maps, Duke Math. J. 16
(1949),165-178.
Conformal Mapping, Dover Publications, New York, 1952.
C. NEUMANN, Untersuchungen iiber das logarithmische und Newtonsche Potential,
Leipzig, 1877.
R. NEVANLINNA, Uber das alternierende Verfahren von Schwarz, J. Reine Angew.
Math. 180 (1939), 121-128.
and V. PAATERO, Introduction to Complex Analysis, Addison-Wesley Pub-
lishing Co., Reading, MA, 1969 Einfiihrung in die Funktionentheorie Birkhauser,
BIBLIOGRAPHY 443
Basel, 1964.
F. NOETHER, Uber eine Klasse singuliirer Integralgleichungen, Math. Ann. 82
(1921), 42-63.
E. J. NYSTROM, Uber die praktische Aufiosung von Integralgleichungen mit Anwen-
dung en auf Randwertaufgaben, Acta Math. 54 (1930), 185-204.
G. OPFER, New extremal properties for constructing conformal mappings, Numer.
Math. 32 (1979), 423-429.
Conformal mappings onto prescribed regions via optimization techniques,
Numer. Math. 35 (1980), 189-200.
Solving complex approximation problems by semiinfinite-finite optimiza-
tion techniques: A study on convergence, Numer. Math. 39 (1982),411-420.
A. OSTROWSKI, Mathematische Miszellen XV. Zur konformen Abbildung einfachzu-
sammenhiingender Gebiete, Jahresbericht Deut. Math.-Vereiningung 38 (1929),
168-182.
Uber konforme Abbildungen anniihrend kreisfOrmiger Gebiete, Jahresbericht
Deut. Math.-Verieinigung 39 (1930), 78-81.
On a discontinuous analogue of Theodorsen's and Garrick's model, Sympo-
sium on the construction and application of conformal maps, National Bureau
of Standards Appl. Math. Ser. 18 (1952), 165-174.
A. M. OSTROWSKI, On the convergence of Theodorsen's and Garrick's method of
conformal mapping, National Bureau of Standards, Appl. Math. Ser. 18 (1952),
149-164.
Conformal mapping of a special ellipse on the unit circle, Appl. Math.
Ser., National Bureau of Standards 42 (1955).
M. N. OZISIK, Finite Difference Methods in Heat Transfer, CRC Press, Boca Raton,
FL,1994.
N. PAPAMICHAEL and J. WHITMAN, A cubic spline technique for the one-dimensional
heat conduction equation, J. Inst. Math. Appl. 11 (1973), 111-113.
and C. A. KOKKINOS, Two numerical methods for the conformal mapping
of simply-connected domains, Comput. Meth. Appl. Mech. Engg. 28 (1981),
285-307.
and C. A. KOKKINOS, Numerical conformal mapping of exterior domains,
Comput. Meth. Appl. Mech. Engg,. 31 (1982), 189-203.
and C. A. KOKKINOS, The use of singular functions for the approximate
conformal mapping of doubly-connected domains, SIAM J. Sci. Stat. Comput.
5 (1984),93-106.
and M. K. WARBY, Pole-type singularities and the numerical conformal
mapping of doubly-connected domains, J. Comput. Appl. Math. 10 (1984),
93-106.
M. K. WARBY and D. M. HOUGH, The determination of the poles of the
mapping function and their use in numerical conformal mapping, J. Comput.
Appl. Math. 9 (1983), 155-166.
M. K. WARBY and D. M. HOUGH, The treatment of comer and pole-type
singularities in numerical conformal mapping techniques, J. Comput. Appl.
Math. 14 (1986), 163-191; in Numerical Conformal Mapping (L. N. Trefethen,
ed.), North-Holland, Amsredam, 1986, pp. 163-191.
L. L. PENNISI, L. I. GORDON and S. LASHER, Elements of Complex Variables, Holt,
Rinehart and Winston, New York, 1963.
E. G. PHILLIPS, FUnctions of a Complex Variable with Applications, Interscience,
New York, 1943.
Some Topics in Complex Analysis, Pergamon Press, Oxford, 1966.
444 BIBLIOGRAPHY
H. B. PHILLIPS and N. WIENER, Nets and the Dirichlet problem, J. Math. Phys.,
MIT 2 (1923), 105-124.
H. T. H. PIAGGIO and M. N. STRAIN, The conformal transformation Z = (lz2 +
2mz + n)/(pz2 + 2qz + r), J. London Math. Soc. 22 (1947), 165-167.
E. PICARD, Lei;ons aux quelques types simples d'equations aux derivees partielles,
Paris, 1927.
J. PLEMELJ, Ein Ergiinzungssatz zur Cauchyschen Integraldarstellung analytischer
Funktionen, Monatshefte fiir Math. Phys. 19 (1908), 205-210.
G. POLYA and G. SZEGO, Isoperimetric Inequalities in Mathematical Physics, Prince-
ton University, 1951.
I. PRIVALOFF, Sur les fonctions conjuguees, Bull. Soc. Math. France 44 (1916),
100-103.
I. I. PRIVALOV (11. 11. IIpHBaJIoB), On the boundary problem in the theory of ana-
lytic functions, Matern. Sbornik, N. S. 41 (1934),519-526.
Randeigenschaften Analytischer Funktionen, 2nd ed., Deutscher Verlag der
Wissenschaften, Berlin, 1956. (Russian)
G. N. PUKHTEEV (r. H. IIyxTeeB), Exact Methods for Calculation of Cauchy-type
Integrals, Nauka, Novosibirsk, 1980.
P. RABINOWITZ, Numerical experiments in conformal mapping by the method of
orthogonal polynomials, J. Assoc. Compo Mach. 13 (1966), 296-303.
J. RADON, Uber die Randwertaufgaben beim logarithmischen Potential, Sitz.-Ber.
Wien. Akad. Wiss., Abt. IIa 128 (1919), 1123-1167.
E. REICH and S. E. WARSCHAWSKI, On canonical conformal maps of regions of
arbitrary connectivity, Pacific J. Math. 10 (1960), 965-989.
L. REICHEL On polynomial approximation in the complex plane with application to
conformal mapping, Math. Compo 44 (1985), 425-433.
A fast method for solving certain integral equations of the first kind with
application to conformal mapping, J. Comput. Appl. Math. 14 (1986), 125-
142; also in Numerical Conformal Mapping (L. N. Trefethen, ed.), North-
Holland, Amsterdam, 1986, pp. 125-142.
M. K. RICHARDSON, A numerical method for the conformal mapping of finite doubly
connected regions with application to the torsion problem for hollow bars, Ph.
D. Thesis, University of Alabama, 1965.
B. RIEMANN, Grundlagen fur eine allgemeine Theorie der Funktionen einer veriind-
erlichen komplexen Grosse (1851), in Collected Works, Dover, New York, 1953.
F. and M. RIESZ, Uber die Randwerte einer analytischen Funktion, Comptes Ren-
dus du Quatrieme Congres des Matlrematiciens Scandinaves it Stockholm (1916),
27-44;Math. Zeitschrift 18 (1923), 95.
G. O. ROBERTS, Computational meshes for boundary layer problems, Lecture Notes
in Physics, vol. 8, Springer-Verlag, New York, 1971, pp. 117-177.
J. A. ROBERTSON, Hydrodynamics in Theory and Application, Prentice-Hall, En-
glewood Cliffs, NJ, 1965.
A. ROSENBLATT, Sur la representation conforme de domaines plans, Comptes Ren-
dus Acad. Sci. Paris 202 (1936), 1398-1400.
Sur la representation conforme de domaines bornes par des courbes genera-
les, Comptes Rendus Acad. Sci. Paris 202 (1936), 1832-1834.
On Mr. L. Kantorovic's method in the theory of conformal mapping and on
the application of that method to aerodynamics, Actas de la Academia Nacional
de Ciencias Exactas, Fisicas y Naturales de Lima 6 (1943), 199-219.
and S. TURSKI, Sur la representation conforme de domaines plans, Comptes
Rendus Acad. Sci. Paris 202 (1936), 899-901.
BIBLIOGRAPHY 445
H. ROTHE, tiber das Grundtheorem und die Obertheoreme der automorphen Funk-
tionen im Faile der Hermite-Lameschen Gleichung mit vier singuliiren Punk-
ten, Monatshefte Math. Phys. 19 (1908), 258-288.
E. B. SAFF and A. D. SNIDER, Fundamentals of Complex Analysis for Mathematics,
Science and Engineering, Prentice-Hall, Englewood Cliffs, NJ, 1976.
P. G. SAFFMAN, The structure and decay of trailing vortices, Arch. Mech. (Archivum
Mechaniki Stosowanej) 26 (1974), 423-439.
G. SANSONE and J. GERRETSEN, Lectures on the Theory of a Complex Variable,
Noordhoff, Groningen, 1960.
Lectures on the Theory of a Complex Variable, II, Wolters-Noordhoff,
Groningen, 1969.
A. L. SCHAGINYAN (A. JI. IIIanlHJIH), Sur les polynomes extremaux qui presentent
l'approximation d'une fonction relisant la representation conforme d'un do-
maine sur un cercle, Comptes Rendus (Doklady) de I'Academie des Sciences de
I'URSS, New Series 45 (1944), 50-52.
M. SCHIFFER, The kernel function of an orthonormal system, Duke Math. J. 13
(1946), 529-540.
Hadamard's formula and variation of domain functions, Am. J. Math. 68
(1946), 417-448.
An application of orthonormal functions in the theory of conformal map-
ping, Am. J. Math. 70 (1948), 147-156.
H. A. SCHWARZ, tiber einige Abbildungsaufgaben, J. Reine Angew. Math. 70 (1869),
105-120; also in Gesammelte Mathematische Abhandlung, II (1890), Springer-
Verlag, Berlin, 65-83.
Notizia sulla rappresentazione conforme di un'area ellitica sopra un'area
circolare (1869,); in Gesammelte Mathematische Abhandlung, II (1890), Springer-
- ~u ~u
Uber die Integration der partiellen Glechung --2 +--2 unter vorgeschriebe-
ax ay
nen Bedingungen, M. B. PreuBische Akademie (1980), 767-795;Werke 2, 144-
171.
tiber diejenigen Fiille, in welchen die Gaussische hypergeometrische Reihe
eine algebraische Funktion ihres vierten Elementes darstellt (1872,)in Gesam-
melte Mathematische Abhandlung, II (1890), Springer-Verlag, Berlin, 211.
Formeln und Lehrsiitze zum Gebrauch der elliptischen Funktionen, Springer-
W. SEIDEL, tiber die Riinderzuordnung bei konformen Abbildungen, Math. Ann.
104 (1931), 182-243.
Bibliography of numerical methods in conformal mapping, National Bureau
of Standards Appl. Math. Sel. 8 (1952), 269-280.
M. SHIFFMAN, The Plateau problem for non-relative minima, Annals of Math. 40
(1939), 834-854.
R. SIEGEL, M. E. GOLDSTEIN and J. M. SAVINO, Conformal mapping procedure
for transient and steady state two-dimensional solidification, Heat Transfer,
Elsevier, Amsterdam, 1970, 1970.
R. A. SILVERMAN, Introductory Complex Analysis, Prentice-Hall, Englewood Cliffs,
NJ,1967.
V. SMIRNOV (B. CMHpHOB), Sur la theeorie des polynomes orthogonaux a une vari-
able complexe, J. Soc. Phys.-math. Leningrade 2 (1928), 155-179.
tiber die Riinderzuordnung bei konformen Abbildungen, Math. Ann. 107
(1932), 313-323.
446 BIBLIOGRAPHY
R. V. SOUTHWELL, Relaxation Methods in Theoretical Physics, Oxford University

Press, Oxford, 1946.
E. M. SPARROW and A. HAJI-SHEIKH, Flow and heat transfer in ducts of arbi-
trary shape with arbitrary thermal boundary conditions, J. Heat Transfer, Trans.
AS ME 88 (1966), 351-356.
N. P. STEIN (H. II. IIlTellH), The determination of parameters in the Schwarz-
Christoffel formula, in Conformal Mapping of Simply and Doubly Connected
Regions, Gostechizdat, Leningrad-Moscow, 1937. (Russian)
E. STIEFEL, On solving Fredholm integral equations. Applications to conformal map-
ping and variational problems of potential theory, J. Soc. Indust. Appi. Math.
4 (1956), 63-85.
A. G. SVESHNIKOV and A. N. TIKHONOV (A. r. CBemHHKOB H A. H. THXOHOB,
The Theory of Functions of a Complex Variable, Mir Publishers, Moscow,
1978; translation of TeopHH 4>yHKnHlt KOMIIJIeKCHOlt IIepeMeHHolt, HaYKa, Moc-
KBa, 1974. (Russian)
G. T. SYMM, Integral equation methods in potential theory. II., Proc. Roy. Soc. A
215 (1963), 33-46.
An integral equation method in conformal mapping, Numer. Math. 9 (1966),
250--258.
Numerical mapping of exterior domains, Numer. Math. 10 (1967), 437-
445.
Conformal mapping of doubly-connected domains, Numer. Math. 13 (1969),
448-457.
The Robin problem for Laplace's equation, in New Developments in Bound-
ary Element Methods (C. A. Brebbia, ed.), CML Publications, South'lmpton,
1980.
G. SZEGO, Uber orthogonale Polynome, die zu einer gegeben Kurve der komplexen
Ebene gehoren, Math. Zeit. 9 (1921), 218-270.
Orthogonal Polynomials, AMS Colloquium Publications 23, American Math-
ematical Society, New York, 1939.
Conformal mapping of the interior of an ellipse onto a circle, Amer. Math.
Monthly 51 (1950), 474-478.
O. TEICHMULLER, Untersuchungen iiber konforme und quasikonforme abbildung,
Deutsche Math. 3 (1938), 621-678.
T. THEODORSEN, Theory of wing sections of arbitrary shape, National Advisory
Committe on Aeronautics, Tech. Rep. 411 (1931).
and I. E. GARRICK, General potential theory of arbitrary wing sections,
National Advisory Committe on Aeronautics, Tech. Rep. 452 (1933).
J. F. THOMPSON, Numerical Grid Generation, North-Holland, Amsterdam, 1982.
Elliptic Grid Generation, Elsevier Publishing Co., Inc., New York, 1982.
F. C. THAMES and C. W. MASTIN, Boundary-fitted coordinate system for
solution of partial differential equations on fields containing any number of
arbitrary two-dimensional bodies, NASA, CR-2729 (1976).
F. C. THAMES and C. W. MASTIN, TOMCAT - A code for numerical
generation of boundary-fitted curvilinear coordinate system on fields containing
any number of arbitrary two-dimensional bodies, J. Comput. Phys. 24 (1977),
274-302.
Z. U. A. WARSI and C. W. MASTIN, Boundary-fitted coordinate systems
for numerical solution of partial differential equations - A Review, J. Comput.
Phys. 41 (1982), 1-108.
Z. A. WARSI and C. W. MASTIN, Numerical Grid Generation: Foundations
BIBLIOGRAPHY 447
and Applications, North-Holland, New York 1985.

R. TIMMAN, The direct and the inverse problem of airfoil theory. A method to obtain
numerical solutions, Nat. Luchtv. Labor. Amsterdam, Report F.16 (1951).
J. TODD (ed.), Experiments in the Computation of Conformal Maps, U.S. National
Bureau of Standards, Appl. Math. Ser. 42, U.S. Govt. Printing Office, 1955.
and S. E. WARSCHAWSKI, On the solution of the Lichtenstein-Gershgorin
integml equation in conformal mapping, II, Computation experiments, National
Bureau Standards Appl. Math. Series 42 (1955), 31-44.
L. N. TREFETHEN, Numerical computation of the Schwarz-Christoffel tmnsforma-
tion, SIAM J. Sci. Stat. Comput. 1 (1980),82-102.
SCPACK Version 2 User's Guide, Internal Report 24, Institute for Com-
puter Applications in Science and Engineering, NASA Langley Research Center,
1983.
Analysis and design of polygonal resistors by conformal mapping, ZAMP
35 (1984), 692-704.
(ed.), Numerical Conformal Mapping, North-Holland, Amsterdam, 1986.
and R. J. WILLIAMS, Conformal mapping solution of Laplace's equation on
a polygon with oblique derivative boundary conditions, J. Comput. Appl. Math.
14 (1986), 227-249; also in Numerical Conformal Mapping (L. N. Trefethen,
ed.), North-Holland, Amsterdam, 1986, pp. 227-249.
M. R. TRUMMER, An efficient implementation of a conformal mapping method
based on the Szego kernel, SIAM J. Numer. Anal. 23 (1986), 853-872.
B. A. VERTGEIM (B. A. BepTreHM), Approximate construction of some conformal
mappings, Doklady Akad. Nauk SSSR 119 (1958), 12-14. (Russian)
S. VLADIMIRSKY, Sur la representation conforme des domaines limites interieure-
ment par des segments rectilignes et arcs circulaires, Comptes Rendus Acad.
Sci. Paris 212 (1941), 379-382.
T. VON KARMAN and E. TREFFTZ, Potential-stromung um gegebene Tragftachen-
querschnitte, Zeitschrift fiir Flugtechnische MoforJuftsch 9 (1918), 111-116.
J. L. WALSH, Interpolation and Approximation by Rational Functions in the Com-
plex Domain, 5th ed., American Mathematical Society, Providence, RI, 1969.
S. E. WARSCHAWSKI, Uber das Randverhalten der Ableitung der Abbildungsfunktion
bei konformer Abbildung, Math. Zeit. 35 (1932), 321-456.
Uber einen Satz von O. D. Kellogg, Nachr. Akad. Wiss. Giittingen, Math.
Phys. Kl. (1932), 73-86.
On the higher derivatives at the boundary in conformal mapping, Trans.
Am. Math. Soc. 38 (1935), 326.
On Theodorsen's method of conformal mapping of nearly circular regions,
Quart. Appl. Math. 3 (1945), 12-28.
On conformal mapping of nearly circular regions, Proc. Amer. Math. Soc.
1 (1950), 562-574.
On the solution of the Lichtenstein-Cershgorin integml equation in con-
formal mapping, I, Theory, National Bureau Standards Appl. Math. Series 42
(1955), 7-30.
Recent results in numerical methods of conformal mapping, in Proc. Symp.
Appl. Math., vol. 6, McGraw-Hill, New York, 1956, pp. 219-250.
On differentiability at the boundary in conformal mapping, Proc. Am.
Math. Soc. 12 (1961), 614-620.
On conformal mapping of certain classes of Jordan domains, Arch. Ratnl.
Mech. Anal. 22 (1966), 201-209.
and G. E. SCHOBER, On conformal mapping of certain classes of Jordan
448 BIBLIOGRAPHY
domains, Arch. Ratnl. Mech. Anal. 22 (1966), 201-209.

H. WAYLAND, Complex Variables Applied in Science and Engineering, Van Nos-
trand Reinhold, New York, 1970.
R. WEGMANN, Ein Iterationsverfahren zur konformen Abbildung, Numer. Math. 30
(1978), 453-466.
Ein Iterationsverfahren zur konformen Abbildung, ZAMM 59 (1979), T85-
86.
Convergence proofs and error estimates for an iterative method for con-
formal mapping, Numer. Math. 44 (1984), 435-461.
An iterative method for conformal mapping, J. Comput. App!. Math. 14
(1986), 7-18; also in Numerical Conformal Mapping (L. N. Trefethen, ed.),
North-Holland, Amsterdam, 1986, pp. 7-18.
An iterative method for the conformal mapping of doubly connected re-
gions, J. Comput. App!. Math. 14 (1986), 79-98; also in Numerical Conformal
Mapping (L. N. Trefethen, ed.), North-Holland, Amsterdam, 1986, pp. 79-98.
G.-C. WEN, Conformal Mappings and Boundary Value Problems, Translations
of Mathematical Monographs, Vol. 106, American Mathathematical Society,
Providence, RI, 1992.
E. T. WHITTAKER and G. N. WATSON, A Course of Modern Analysis, 4th ed.,
Cambridge University Press, Cambridge, UK, 1927.
H. B. WILSON, A method of conformal mapping and the determination of stresses in
solid propellant rocket grains., Rep. S-38 (1963), Rohm and Haas Co., Huntsville,
AL.
W. WIRTINGER, Uber die konforme Abbildung der Halbebene auf ein Kreisbogen-
dreieck, Atti Pontif. Accad. Sci. Nuovi Lincei 80 (1927), 291-309.
H. WITTICH, Konforme Abbildung einfach zusammenhiingender Gebiete, ZAMM
25/27 (1947), 131-132.
L. VON WOLFERDORF, Zur Unitiit der Losung der Theodorsenschen Integralgle-
ichung der konformen Abbildung, Z. Anal. Anwendungen 3 (1984), 523-526.
S. WOLFRAM, The Mathematica Book, 2nd and 3rd eds., Wolfram Media, Cham-
paign, IL, 1991, 1996.
K. ZARANKIEWICZ, Sur la representation conforme d'un domaine doublement con-
vexe sur un anneau circulaire, Comptes Rendus Acad. Sci. Paris 198 (1934),
1347-1349.
Uber ein numerisches Verfahren zur konformen Abbildung zweifach zusam-
menhiingender Gebiete, ZAMM 14 (1934), 97-104.
C. ZEMACH, A conformal map formula for difficult cases, J. Comput. Appl. Math.
14 (1986), 207-215; also in Numerical Conformal Mapping (L. N. Trefethen,
ed.), North-Holland, Amsterdam, 1986, pp. 207-215.
V. A. ZMOROVICH (B. A. 3MOPOBIfQ), Deux problemes du domaine des representations
conformes, J. de I'Institut Math. de I'Academie des Sciences de l'Ukraine 3/4
(1935), 215-222.
A. ZYGMUND, Trigonometric Series, Monografie Matematyczne, Warsaw, 1935; also
Cambridge University Press, London, 1959.
Notation
(The number within parentheses represents the page number where the notation
is first introduced.)
A closure of a set A (15)

A\B complement of a set B with respect to a set A (15)
AxB product of sets A and B (15)
A(z, a) Kerzman-Stein kernel (200)
A(PI' pz) annulus {PI < Iwl < pz} (296)
Area (G) area of a region G (38)
B(a,r) open disk {Iz - al < r} (16)
B(a,r) closed disk {Iz - al ~ r} (16)
B(O,l) open unit disk {izi < I} (16)
B(m,n) beta function (55)
cap(D) capacity of a region D (30)
comp(B) complement of a set B
cond(Mn ) condition of the map M n (252)
CO(D) class of functions continuous on a region D (15)
Ck(D) class of continuous functions with k-th continuous derivative
on a region D, 0 ~ k < 00 (15)
Coo (D) class of continuous functions infinitely differentiable on
a region D (15)
C-function same as a CO-function (15)
C complex plane (16)
Coo extended complex plane (16)
diam(E) transfinite diameter of a set E (18)
d ({Si}' {ti}) Frechet metric (21)
dn a Jacobian elliptic function (325)
D region (16)
449
450 NOTATION
fJ closure of D (16)
D* complement of a set D (29);
= Ext (r) region exterior to r (239)
Ext (r) region exterior to r (18)
I analytic function (19);
mapping of D = Int (r) onto U = {Iwl < I}, 1(0) = 0 (237)
mapping of D* = Ext(r) onto U* = {Iwl > I}, h(O) = 1,
h(oo) = 00 (239)
In mapping of a doubly connected region n onto the annulus
A(pll P2) (296)
(J,g) inner product (22)
11/1100 norm of IE Loo (23)
1I/1I~,r (line)-norm of I (22)
1I/11~,D (surface)-norm of 1(22)
p+,p- limit values of P from right and left of r (27)
F operator (235)
F' P -derivative of F (235)
9 complex conjugate of an analytic function 9 (22)
gd Gudermannian (40)
G region (31)
Q(-,o) Green's function (29)
HI Lipschitz condition (20)
HO: Holder condition of order 0: (20)
H(z, zo) Cauchy kernel (25)
Int(r) region interior of r (18)
I(f, zo) index of a contour r (winding number) (18)
~{-} imaginary part of a complex quantity (17)
K(k) elliptic integral (57)
K(z,a) Bergman kernel (94)
K conjugation operator (234)
KO(D) class offunctions IE L 2 (D), I(a) = 0, a E D (92)
K1(D) class offunctions I E L 2 (D), I(a) = 1, a E D (92)
lP a vector space (21)
lk(Z) Lagrange's interpolation functions (254)
L2 Hilbert space of square-integrable functions (21)
Loo Hilbert space of 21l"-periodic and bounded functions (23)
L 2 (D) class of square-integrable functions defined on
a region D (21)
NOTATION 451
£1 (f) class of functions 1 E £2(f), I(a) = 1, a ED (104)

M = P2/Pl conformal modulus of a doubly connected region n (296)
N(s,t) Neumann kernel (169)
N(-, .) Neumann function (160)
Pn(Z) complex polynomials (41)
p partition of an interval [a, b] (19)
Qn(Z) orthonormal polynomials (117)
~{-} real part of a complex quantity (17)
jRn Euclidean n-space (16)
jR+ set of nonnegative real numbers (16)
sn a Jacobian elliptic function (57)
S(z, a) Szego kernel (105)
Tn(z) Chebyshev polynomials of the first kind (185)
U open unit disk (16)
U* region exterior of the unit circle (239)
Un(z) Chebyshev polynomials of the second kind (119)
V(Zll' .. ,zn) Vandermonde determinant (17)
W = I(z) = u + iv = pei<J>, image of z under 1 (16)
{w,z} Schwarzian derivative (65)
W Sobolev space of 21T-periodic and absolutely continuous
functions 1 with f' E £2[0, 21T] (23)
(W,II·II) Banach space (23)
Z = x + i Y = r eiO , complex variable (16)
z* point symmetric to z (47)
(Zll Z2, Z3, Z4) cross-ratio (46)
{zdf a set of distinct points in D (29)
Z(z) fundamental solution (414)
"( path, contour (18); Robin's constant (30)
"(s) a point z on r for a given s E [a, b] (18,169)
'Y reparametrization of"( (19)
r=aD boundary of a region D (a simple closed curve or
a Jordan contour) (18)
a,a partial differential operators 01 = ~~, al = ~~ (16)
aB(a,r) circle Iz - al = r (16)
a=D boundary of Din C= (16)
Neumann constant (181);
index of a Hilbert problem (411)
J1(() density function (238)
452 NOTATION
1r n (Z) Bieberbach polynomial (100)

TIn(z) orthogonal polynomials (116)
CTk arcs (18)
CTn(Z) Szego polynomials (Ill)
<Pn (Z) minimal polynomials (96)
n doubly connected region (295);
multiply connected region (359)
" . 0 .0
'V gradlent vector = 1 ox + J oy (17)
('VU, 'Vv) inner (scalar) product (17)
Index
A boundary comer 325ff

Ahlfors estimate 185 boundary correspondence function
airfoil(s) 174,207, 269ff, 394 187ff, 192,202,208,212,
Joukowski 275, 293 255,259,296,316,321,
main 287ff 323, 325ff, 329, 347, 358
multi-element 8, 284, 291ff
secondary 287ff C
single-element 8, 279ff, 284ff capacity 29ff, 33
symmetric Joukowski 276ff cardioid 48,307,318,351
two-element 279, 288ff cascade of blades 379, 390ff
analytic continuation 15,34,327,331, pitch 400
403 Cassini's oval(s) 157, 165,203, 243ff,
annulus 296ff, 299, 317, 347, 360ff, 250, 267, 319, 372ff
367,379 Cauchy-Goursat theorems 24
approximation(s), Chebyshev 11 Cauchy formula 9, 109, 233, 315
nearly circular 277ff principal value 7, 10, 26, 170,
polynomial 92ff, US, 251 205,209,234,321,323
Ritz 11 p.v. integrals 206, 234, 321,
spline 237, 258ff, 337 401,406ff
successive 3, 123ff, 138, 144ff, Cauchy's inequality 25
425 integral(s) 26, 111, 132, 367,
VM 347 370, 374ff
argument principle 185 integral formula 24, 148, 164,
315
B kernel 11, 15,25, 198ff
Banach space 23, 235 theorem 15, 23, 365
basis 259, 340, 345, 346ff Cauchy-Riemann equations 16ff, 27,
augmented (AB) 281, 327ff, 239,273,371
342ff, 347ff, 354ff, 357 Cauchy-Schwarz inequality 21
monomial 252 centerline 396ff
orthonormal 347, 354 chain property 33ff
well-conditioned 251ff Chebyshev approximation 11
bean-shaped curve 165, 344 polynomials 118, 185,252
bilinear transformation 45ff, 65ff, 293, circle(s) 16,83, 191,379
398 eccenetric 190
binomial expansion 131 in a cross 356ff
453
454 INDEX
in a square 348ff 190,309, 358ff, 364ff, 368,

circular cuts 359ff 379,423ff
cylinder 294 for the annulus 302ff
region 361ff for the unit disk 423
tube 395 modified 359, 369
circulation 273 discontinuity 259, 329, 375, 397
closure 15 disk(s) 16,361
collocation points 263 divergence 381
complement 15 domain 16
complete set 95, 102
orthonormal set 101, 118 E
conformal mapping(s) 30ff, 41ff electromagnetics 12
equivalence 38, 296 ellipse 119, 123ff, 131, 134, 141,
invariants 296ff 145ff, 155,203,242,246,
module 384 251ff, 257, 266ff, 333ff, 378
modulus 295ff, 299ff, 318, 345, cambered 282
368ff coaxial 318
radius 33 confocal 270
conjugation operator 234 inverted 203, 234
convergence factor 184 nondegenerate 254
convergence of iterations 179, 387, elliptic arc 338
392 rings 307
continued fraction 67 epitrochoid 203, 307
corner 235, 260, 321, 325ff, 328, 348 equation(s), Bernoulli's 83
re-entrant 259,329 Cauchy-Riemann 16ff, 27, 239,
singularities 246, 265, 282, 273,371
320,328,348 collocation 260ff
Cramer's rule 193 continuity 262, 381ff
cross in a square 350 Laplace 4ff, 12, 370, 399
cross-ratio 46 error estimate 194, 242ff, 245ff, 250,
curve, nonrectifiable 19 257,263, 265ff, 306
rectifiable 18, 255 expansions, asymptotic 327ff, 351
curve-fit coefficients 281ff Laurent series 126
cusp 275, 277 small parameter 121ff
exterior maps 177ff
D t-condition 235
de Branges estimate 40
dilatation 41 F
diameter 16 fast Fourier transform (see FFI')
transfinite 17, 33, 245ff fast Poisson solver II, 358, 369ff,
difference operator(s) 371 373ff,377
dipole density 309ff, 371 Fatou's lemma 20
distribution 190, 205, 308ff F-derivative 235
strength 190, 205 Fejer points 254ff
Dirichlet problem Iff, 6, 10,31, 147ff, FFf, 1, 10, 235, 269, 285ff
INDEX 455
flow(s), aerodynamic 8 366ff, 373, 378ff, 399

fluid 12, 272 harmonic 3,15,27,31, 147ff,
boundary layer 382 228,239,267, 280ff, 283,
formula(s), Cauchy 9, 1109,233, 315 302,342,364,367,370,373
Cauchy's integral 24, 148, 164, hyperbolic 387
315 hypergeometric 301
Chaplygin's 273 implicit 132
Green's 23, 101,345 incomplete beta 82
Hadamard 5 Jacobian elliptic 57, 116, 187,
interpolation 202 325,391
modified Scwarz-Christoffel 86 Joukowski 269ff, 277
Plemelj 26,403,412 kernel 6,162,319, 341ff, 406
quadrature 369ff, 376, 403ff Koebe40
Schwarz 158ff, 208, 423 Lagrange's interpolation 254ff,
Schwarz-Christoffel 5, 12,51, 257ff
64,68,259 meromorphic 361,367,377
force flux 204 minimal 93
Fourier analysis 387 minimizing 5
coefficients 108, 134, 163, 229, Neumann 160ff
303,347,398 periodic Green's 314
series 99, 110, 133, 146,194, potential 204, 309
229, 303ff, 347, 392, 397 rational 41
sum 340 Riemann mapping 1 I, 119ff
transform, discrete 232, 286 Schottky 367
Frechet metric 21 sectionally analytic 418
function(s), arc-wise linear 404ff singular 344, 347ff, 356ff
Bergman kernel 6, 11, 94ff, source density 238,240, 260
119,328,340ff spline 11,259,264
beta 55 Szego kernel 105ff
biharmonic 3 trigonometric 141
boundary correspondence univalent 32, 39, 318, 330, 333,
187ff, 192,202,208,212, 377
255,259,296,316,321, weight 402
323, 325ff, 329, 347, 358 fundamental solution 412ff, 418, 423
collocation 261 fundamental theorem of line integrals
complete elliptic 297, 394 19
density 238, 294, 329, 351,
368ff, 374, 376ff G
dipole density 309ff, 368 Gauss quadrature 102, 264
elliptic 6, 267, 340 Gauss-Chebyshev type quadrature
elliptic sine 57, 257 403ff
elliptic theta 314ff Gautschi criterion 251
exponential 397 Glaubert's mapping 276
Green's 4, 29ff, 33, 147ff, 160, modification 275ff
237,267,313,325,341, gradient 380
456 INDEX
theorem 407 Minkowsky's 21, 217, 219,

Gram-Schmidt process 101,117 221,225
Green's formula 23, 101,345 Schwarz 179, 181,213,221
identities 407ff triangle 19
theorem 407 infinity, exterior 396ff
reciprocity theorem 408 interior 396ff
grid concentration 379, 383, 390ff initial guess 70, 79, 183
adaptive scheme 399 inlet 395
generation 12, 13, 379ff, 382 configuration(s) 379, 384ff
lines 371ff, 378, 389ff, 399 exterior 396
orthogonal 384 interior 396ff
uniformization 383 nose 396ff
Gudermannian 40 inner product 22
integral(s), Bernoulli 237
H Cauchy 7, 10,26, Ill, 132,
half-line 352 267,360,374ff
Harnack theorem 29 Cauchy p.v. 206, 234, 401ff,
hermitian 98, 170 406
quadratic forms 110 complete elliptic 267
Hilbert problem, homogeneous 7, elliptic, of first kind 57, 311
41Off,419 hyperelliptic 6
nonhomogeneous 7ff, 415ff improper 40lff
Hilbert space 21, 99,106,170 of Cauchy type 25
Holder condition 7,20,27, 168, 222ff, Poisson 158,214,313
401, 41Off, 417, 424 Schwarz-Christoffel 52, 64,
inequality 21 68ff, 79, 86
Holder-continuous 20, 209, 232 Stieltjes 320ff
horizontal strip 60 integral equation(s), Arbenz's 320ff
hyperbola 334, 35lff Banin's 176
hyperbolic arcs 165, 353 Carrier's 173
hypertrochoids 307 coupled 240
Fredholm 7, 10, 12, 193,358,
I 368
identity, Green's first 408 Gershgorin's 172, 188,313,
Green's second 149,408 314, 316, 320ff
Green's third 150,409 Lichtenstein's 17lff
mapping 36lff Mikhlin's 259, 358, 367ff, 377
theorem 29 modified Symm's 255ff
inclination 359ff singular 231
index of a curve 24 Stie1tjes 321 ff
of Hilbert problem 411, 415, Symm's 1, 10ff, 237ff, 261,
418,423 267,295,306,325ff
inequality, Cauchy's 25 Symm's, for interior regions
Cauchy-Schwarz 21 238ff
Holder 21 Symm's, modified 255ff
INDEX 457
Theodorsen's 1,3,8,10, 207ff, Laplace equation 4ff, 12,370,399

234ff, 237,410 Laplacian 11,358,370, 372ff, 381ff
Warschawski-Stiefel's 177 399
integral representation 222ff analytic 374
interior maps 177ff discrete 372ff, 375ff, 399
interpolation error 257 lima~on 165,265,307
formula 202 Pascal's 307
Lagrange 237, 25 1ff, 257ff linear programming 11
polynomials 233 Lipschitz condition 20
trigonometric 194, 228ff Lipschitz-continuous 20, 236
inverse points 395 Liouville theorem 25, 163
inversion 45,177,339,424 logarithmic singularity 265, 313, 366
spiral(s) 359ff
J log-hodograph plane 83
Jacobian 380ff variable 85
Jordan contour 17, 181
Joukowski airfoil 275, 293 M
function 269f[, 277 magnification 41
generalized mapping 274ff matrix, hermitian 98, 170
theorem 274 skew-hermitian 202
jump discontinuity 259, 329, 374ff maximum modulus theorem 25ff, 28,
242,246,306
K principle 28ff
kernel 6, 162,318, 341ff, 369ff, 373, mean-value property 25
375,406 theorem 28, 147ff
Bergman 11, 94ff, 118, 328, mesh points 370ff, 376ff
340ff,406 size 373, 378
Cauchy 11, 15,25, 198ff metric spaces 20ff
conjugate 178 method, Bergman kernel (BKM) 95,
degenerate 192ff 101ff, 280,327, 340ff
iterated 178 Bieberbach 3
Kerzman-Stein 200ff Carleman's 7
Neumann 168ff, 180ff, 192ff classical hodograph 85ff
Szego 11, 92, 105ff, 113, 198ff, classical iterative 207ff
201 collocation 261
Kirchhoff flow problem 13, 83ff conjugate gradient 206
knots 261 contour integration 403
Kronecker delta 110 discretization 323ff
Elcrat-Trefethen 86ff
L fast Fourier transform (FFT) 1,
Lagrange functions 254ff, 257ff 10, 235, 269, 285ff
interpolation 237, 25 Iff, 257ff finite difference 379ff, 399
polynomials 5, 254 finite element 379ff
Lame differential equation 6 fixed-point iteration 10
Landen transformation 58ff, 394 Fourier 10
458 INDEX
Fourier-Galerkin 11 orthogonal 383ff

Fredholm 3 orthonormal polynomial (ONP)
Gaier's variational (VM) 4ff, 240ff, 265ff, 280, 306
92,341,346 Picard's 7
Gakhov's 7 polynomial approximation 1
Galerkin 377 quadratic convergence 229
Garrick's, of conjugate func- Radon's 320ff
tions 8, 13,269, 288ff relaxation 6
Gaussian elimination 377 Ritz (RM) 95ff, 280, 327, 340ff
generalized conjugate gradient small parameters 7
(GeM) 202, 206 Symm's 367
Green's function 5 Theodorsen's iterative 209ff,
Hadamard variational 4 269, 279ff, 294
integral equation 9, 168ff two-step 388ff
iteration 1,230,390,399 variational (VM) 4ff, 92, 341,
iterative 178ff, 202, 207, 233, 346
269, 285ff, 311, 317, 387ff, Wegmann's 207, 229ff, 236
391 Minkowsky's inequality 21
Ives-Liutermoza 390 Mercator's projection 1,40
James's 8, 269, 279ff, 281ff, Mobius transformation 45ff, 69, 257,
294 288ff
Kantorovich 71 ff
Lagrange's interpolation 237, N
25lff near circle 210ff, 213, 229, 288, 290,
Mayo's 358, 370ff, 377 390
modified ONP 247ff nearly circular approximations 277ff
modified Symm's 255 boundary 133
modified Theodorsen's 284 contour 208, 267, 284
Newton's 1, 10,60,68,70, region 2ff, 120ff, 145, 181,
72ff, 77, 207, 231, 234ff 209,277
Newton-like 207 nearly convex 182
Newton-Raphson iteration 292 Neumann conditions 304
Nystrom 202, 369ff constant 181
of images 341ff function 160ff
of infinite systems 125ff, 132, kernel 168ff, 180ff, 192ff
141,207 lemma 181
of networks 6 problem 3ff, 7, 159ff
of reduction of connectivity problem for the annulus 303
311 series 178
of separation of singularities n-gon 51, 83
402ff nodes 240ff, 245, 246, 307, 369
of successive approximations
123, 137, 139, 144ff o
of successive conjugates 280ff oblique angle 360
one-step 384ff octagonal curve 165
INDEX 459
p 147ft, 190,309, 358ft, 364ff,

parabola 351ft 368, 379, 423ft
parabolic arcs 351, 353 Dirichlet, for the unit disk 423
parallel cuts 359, 361 Hilbert 7, 410ft, 419
strip 296, 313 homogeneous Hilbert 7, 410ft,
perturbations 251 417
plane, computational 381 ft Kirchhoff flow 13, 84ft, 90
physical 83, 381ft minimum area 92ft
Plemelj formula 26 minimum boundary 103ft
point of inflection 337 modified Dirichlet 359, 369
Poisson grid generation 399 Neumann 3ft, 7, 159ft
integral 158, 214, 313 nonhomogeneous Hilbert 7ff,
pole(s) 63,235, 330ft, 335ft,342ft 415ft
polygon 69 parameter 76ft
polygonal obstacles 83ff Plateau 5
polynomial(s), Bieberbach 100, 107 Riemann-Hilbert 8, 10, 147,
Bochner-Bergman type 4 207,235, 410ft, 417ff
Chebyshev 118, 185,252, Schwarz-Christoftel 87
403ft transonic flow 394
complex 41 variational 345
Fourier 228 projection, Mercator's 1, 39
harmonic 153 orthogonal 199
interpolation 233 stereographic 1, 30
Lagrange 5, 254 property, mean-value, 25
minimal 96, 98,100,106
orthogonal 109ft, 116ft Q
orthonormal 111, 117,237,295 quadrature 309, 369ft, 377,404
Szego 4, 110ft formula 370
potential 190,204 Gaussian 102, 264
complex 272ff quadric functional 170
complex velocity 84, 394 Gauss-Chebyschev type 403ft
flow 174, 269,279 quadrilaterl60ff,72, 165,354
function 204, 309 quasiconformal map 384
logarithmic 2
single-layer 2 R
single-layer logarithmic 305 rectangle(s) 56, 242, 251, 264, 267,
principle, argument 185 297,333,337, 341ft, 352,
Bieberbach minimizing 92 370,384
maximum 28, 30 in a circle 356
maximum modulus 210 rectifiable curve 18
minimum 4 reflection 46
Schwarz reflection 15, 35ft, principle 331
330ft region 16
symmetry 49ft, 417 bean-shaped 343ff
problem, Dirichlet Iff, 6, 10, 31, canonical 5, 359ft
460 INDEX
circular 361 ff 64,68,259

computational 379ff integral(s) 52, 64, 68ff, 86
cross-shaped 350, 356 modified formula 87
doubly connected 295ff, 345ff, problem 87
367 transformation 1, 13, 5lff, 66,
exterior 140ff, 174ff, 226ff, 83,86,90,116,326
239ff, 245ff, 339ff Schwarzian derivative 65
interior 17lff, 237ff, 240ff semi-infinite strip 62, 66, 389
lens-shaped 66 series, Fourier 99, 110, 132, 145, 194,
multiply connected 358ff, 368 230, 290, 303ff, 350, 392,
nearly circular 2ff, 12lff, 146, 397
181,209,277 Laurent 125,302
nearly convex 181 Neumann 178
nose 284, 289 Taylor70ff, 132, 163, 194,201,
octagonal 354 37lff, 376, 378, 397
physical 379ff trigonometric 121, 123, 138,
rectangular 356 141,288
starlike 207, 214, 224ff Simpson's rule 72, 241
triply connected 373, 378 single-valuedness condition 239, 306
relaxation constant 391ff singularity behavior 328ff
reparametrization 19 problem 11
resolvent 194ff singularities 257, 320ff, 336, 340, 345,
Riemann-Hilbert problem 8, 10, 147, 348ff
207,235, 41Off, 417ff branch point 326ff, 333ff, 348
for the unit disk 417ff corner 246, 265, 282, 320, 328,
homogeneous 410 348
Riemann mapping theorem 2, 4, 9, 15, curvature 397
3 Iff, 70, 93, 208, 384 logarithmic 265,313,366
Riemann surface(s) 2, 16,230,270, pole-type 330ff, 358
299 removal of 81ff
Ritz approximations 11 slit(s) 378
Robin's constant 29ff, 33, 240, 326 small parameter expansions 120ff
rotation 41,298 Smirnov class 107
Sobolev space 23, 235
S solidification 12
S-condition 107, 108, 118 solidity 390
scaling factor 252 source strength 205
Schwarz formula 158ff, 208 density function 238, 260,
differential operator 65 305ff
inequality 179, 181,213,221 sources and sinks 205, 392ff
lemma 39, 187 spiral cut(s) 359ff
reflection principle 15, 35ff, spline, approximations 237, 259ff, 337
330ff,416 augmented 264
symmetry principle 417 augmented cubic 264
Schwarz-Christoffel formula 5, 12,51, B- 264
INDEX 461
cubic 285ff Harnack 29

function 11, 264 identity 29
periodic cubic 285, 288, 291 Joukowski 274
quintic 377 Lehman's 350
square(s) 89, 100, 136, 143ff, 154, Liouville 25, 163
242,324,348,379 maximum modulus 25ff, 28,
family of 137 242,246,306
in a circle 319 mean vlaue 28, 147ff
with round corners 165, 258 ofF. and M. Riesz 214
stadium 203 Osgood-Caratheodory 4
stagnation point 84, 393ff residue 364, 366, 403
streamline 395 Riemann mapping 2, 4, 9, 15,
star of a region 299ff 31ff, 70, 93,208,384
stereographic projection 1,30 Stokes' 407ff
Stieltjes integral 320ff three-element loop 283
integration 321, 324 tolerance 184
streamlines 84, 392ff torsion problem 5
stretching parameter 382 trailing edge 285ff, 288ff
Symm's integral equation I, lOff, trapezoid 78
237ff, 261, 267, 295, 306, trapezoidal rule 11, 202, 369ff, 377
325ff transfinite diameter 17, 33, 245ff
for interior regions 238ff transformation, bilinear 45ff, 65ff,
modified 255ff 293,398
symmetric points 36, 46ff, 289, 331 coordinate 380ff
symmetry principle 49ff, 417 Landen 58ff, 394
property 367 linear 69, 362
Szego kernel II, 92, 105ff, 113, I98ff, linear-fractional 45ff
201 Mobius 45ff, 69' 257, 288ff
polynomials 4, II Off Schwarz-Christoffel I, 13, 51ff,
66, 83, 85, 90, 116, 326
T Theodorsen-Garrick 285
Taylor series 70ff, 132, 163, 194,201, von Karman-Trefftz 8, 269,
37Iff, 376, 378, 397 283ff, 288, 291
Theodorsen's integral equation 1,3,8, translation 41, 71, 284
10, 207ff, 234ff, 237,410 triangle 54ff, 62, 89,
uniqueness 222ff isosceles 250ff
theorem, area 2, 38, 298 inequality 19
Cauchy-Goursat 24 trigonometric interpolation 228ff, 231
Cauchy's 15,23,365
conformal mapping 30 U
divergence 407ff unit disk 16, 51, 80, 115, 119, 132,
gradient 407ff 134, 145,214,279
Green's 407 slit 368
Green's reciprocity 408 square 203, 378
Hally's 322 univalent function 32, 39, 318, 330,
462 INDEX
333, 377 W
mapping 32, 244, 366 Wallis criterion 67
upper half-plane 51ff, 56, 60, 63, 66, Warschawski's sufficient condition
69,72,78,83,89,165 283,289
Weddle's rule 190
V Wing profile, elliptic 294, 394
Vandermonde determinant 16 tear-drop 294
velocity, complex 83
von Karman-Trefftz transformation 8, Z
269, 283ff, 288, 291 zeros 63

Computational Conformal Mapping - Prem K. Kythe

Uploaded by

Copyright:

Available Formats

Computational Conformal Mapping - Prem K. Kythe

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Computational Conformal Mapping - Prem K. Kythe

Uploaded by

Copyright:

Available Formats

FILIO • P R I M O

Springer Science+Business Media, L L C

Library of Congress Cataloging-in-Publication Data

I S B N 978-1-4612-7376-9 I S B N 978-1-4612-2002-2 (eBook)

QA360.K97 1998 98-4748

Printed on acid-free paper

Chapter 1. Basic Concepts, 15

Chapter 2. Conformal Mappings, 41

Chapter 3. Schwarz-Christoffel Integrals, 68

Chapter 4. Polynomial Approximations, 92

Chapter 5. Nearly Circular Regions, 120

Chapter 6. Green's Functions, 147

Chapter 7. Integral Equation Methods, 168

Chapter 8. Theodorsen's Integral Equation, 207

Chapter 9. Symm's Integral Equation, 237

Chapter 10. Airfoils, 269

Chapter 11. Doubly Connected Regions, 295

Chapter 12. Singularities, 320

Chapter 13. Multiply Connected Regions, 358

Chapter 14. Grid Generation, 379

Appendix A. Cauchy P. V. Integrals, 401

Appendix B. Green's Identities, 407

Appendix C. Riemann-Hilbert Problem, 410

Appendix D. Successive Approximations, 425

Appendix E. Catalog of Conformal Mappings, 428

*Mathematica is a registered trade mark of Wolfram Research, Inc.

The main purpose of this book is to provide a self-contained and systematic

Besides an exposition on the history of the subject, there are 14 chapters in

Various integral equation formulations of the conformal mapping problem

A detailed account of airfoils is presented in Chapter 10. Various methods,

The book is intended to contribute to an effective study program at the

numerical treatment of Fredholm integral equations is dealt with in depth in

It has been my experience that Mathematica and a programming language, like

The help provided by some of my colleagues and students is gratefully ac-

New Orleans, Louisiana

Current research in computational conformal mapping has taken two major

0.1. Historical Background

The oldest transformation, known as the stereographic projection of the

P. K. Kythe, Computational Conformal Mapping

constructed by the method. These two projections, however, do not produce

After the basic theoretical aspects of the theory of functions of a com-

The first integral equation method was developed by Lichtenstein (1917)

Successive approximations in the integral equation method for simply and

Julia (1927) determined a sequence of polynomials which converges to

ries K(a, z) Pn(a) Pn(z) converges uniformly and absolutely in every

The problem of minimizing a functional was first solved by Hadamard

In 1931 Grotsch solved the problem of conformally mapping a multiply

Green's function method for an arbitrary region was established by Leja

The Schwarz-Christoffel formula was used by Bergman in 1923-24 to

the existence of a complete orthononnal system with respect to a region is

The first systematic construction of conformal maps by the method of net-

Relaxation methods were first applied to the problems of conformal map-

The small parameters method developed by Kantorovich (1933) was later

Special conformal mappings were studied by Rothe (1908) who constructed

The theory of homogeneous boundary problems for analytic functions was

boundary problem which is now known as the Riemann boundary problem.

One of the widely investigated practical aspects of conformal mapping

0.2. Modern Developments

One major approach in developing methods for numerical conformal map-

ries expansion f(z) = Cl (z - zo) Cn (z - zot, with Cl nonzero real