Branner B., Fagella N.-Quasiconformal Surgery in Holomorphic Dynamics-CUP (2014)

C A M B R I D G E S T U D I E S I N A DVA N C E D M AT H E M AT I C S 1 4 1
Editorial Board
B . B O L L O B Á S , W. F U LTO N , A . K ATO K , F. K I RWA N ,
P. S A R NA K , B . S I M O N , B . TOTA RO
QUASICONFORMAL SURGERY IN HOLOMORPHIC

DYNAMICS
Since its introduction in the early 1980s, quasiconformal surgery has become a major
tool in the development of the theory of holomorphic dynamics, and it is essential
background knowledge for any researcher in the field.
In this comprehensive introduction the authors begin with the foundations and a
general description of surgery techniques before turning their attention to a wide variety
of applications. They demonstrate the different types of surgeries that lie behind many
important results in holomorphic dynamics, dealing in particular with Julia sets and
the Mandelbrot set. Two of these surgeries go beyond the classical realm of quasicon-
formal surgery and use trans-quasiconformal surgery. Another deals with holomorphic
correspondences, a natural generalization of holomorphic maps.
The book is ideal for graduate students and researchers requiring a self-contained
text including a variety of applications. It particularly emphasizes the geometrical ideas
behind the proofs, with many helpful illustrations seldom found in the literature.
Bodil Branner is Professor Emerita at the Technical University of Denmark, Lyngby.

Her research interests include holomorphic dynamics and complex analysis. She has
published in several renowned international journals and given numerous invited talks
at conferences, workshops and symposia. Branner has served as Vice-President of the
European Mathematical Society, as President of Dansk Matematisk Forening (DMF),
and she was one of the founders of European Women in Mathematics. She is an
honorary member of DMF, and a Fellow of the AMS.
Núria Fagella is currently Associate Professor at Universitat de Barcelona. Her

research is in the area of holomorphic dynamics with an emphasis on the iteration of
transcendental functions. She publishes in renowned international journals and with a
diverse range of collaborators worldwide. Fagella has been invited to deliver talks and
short courses at numerous international conferences and workshops, and has been an
organiser of several such events.
CAMBRIDGE STUDIES IN ADVANCED MATHEMATICS
Editorial Board:
B. Bollobás, W. Fulton, A. Katok, F. Kirwan, P. Sarnak, B. Simon, B. Totaro
All the titles listed below can be obtained from good booksellers or from Cambridge University Press. For a
complete series listing visit: www.cambridge.org/mathematics.
Already published
104 A. Ambrosetti & A. Malchiodi Nonlinear analysis and semilinear elliptic problems
105 T. Tao & V. H. Vu Additive combinatorics
106 E. B. Davies Linear operators and their spectra
107 K. Kodaira Complex analysis
108 T. Ceccherini-Silberstein, F. Scarabotti & F. Tolli Harmonic analysis on finite groups
109 H. Geiges An introduction to contact topology
110 J. Faraut Analysis on Lie groups: An introduction
111 E. Park Complex topological K-theory
112 D. W. Stroock Partial differential equations for probabilists
113 A. Kirillov, Jr An introduction to Lie groups and Lie algebras
114 F. Gesztesy et al. Soliton equations and their algebro-geometric solutions, II
115 E. de Faria & W. de Melo Mathematical tools for one-dimensional dynamics
116 D. Applebaum Lévy processes and stochastic calculus (2nd Edition)
117 T. Szamuely Galois groups and fundamental groups
118 G. W. Anderson, A. Guionnet & O. Zeitouni An introduction to random matrices
119 C. Perez-Garcia & W. H. Schikhof Locally convex spaces over non-Archimedean valued fields
120 P. K. Friz & N. B. Victoir Multidimensional stochastic processes as rough paths
121 T. Ceccherini-Silberstein, F. Scarabotti & F. Tolli Representation theory of the symmetric groups
122 S. Kalikow & R. McCutcheon An outline of ergodic theory
123 G. F. Lawler & V. Limic Random walk: A modern introduction
124 K. Lux & H. Pahlings Representations of groups
125 K. S. Kedlaya p-adic differential equations
126 R. Beals & R. Wong Special functions
127 E. de Faria & W. de Melo Mathematical aspects of quantum field theory
128 A. Terras Zeta functions of graphs
129 D. Goldfeld & J. Hundley Automorphic representations and L-functions for the general linear group, I
130 D. Goldfeld & J. Hundley Automorphic representations and L-functions for the general linear group, II
131 D. A. Craven The theory of fusion systems
132 J. Väänänen Models and games
133 G. Malle & D. Testerman Linear algebraic groups and finite groups of Lie type
134 P. Li Geometric analysis
135 F. Maggi Sets of finite perimeter and geometric variational problems
136 M. Brodmann & R. Y. Sharp Local cohomology (2nd Edition)
137 C. Muscalu & W. Schlag Classical and multilinear harmonic analysis, I
138 C. Muscalu & W. Schlag Classical and multilinear harmonic analysis, II
139 B. Helffer Spectral theory and its applications
140 R. Pemantle & M. C. Wilson Analytic combinatorics in several variables
141 B. Branner & N. Fagella Quasiconformal surgery in holomorphic dynamics
142 R. M. Dudley Uniform central limit theorems (2nd Edition)
143 T. Leinster Basic category theory
Quasiconformal Surgery
in Holomorphic Dynamics
BODIL BRANNER
Technical University of Denmark, Lyngby
N Ú R I A FA G E L L A
Universitat de Barcelona
With contributions by
Xavier Buff, Shaun Bullett, Adam L. Epstein,
Peter Haı̈ssinsky, Christian Henriksen,
Carsten L. Petersen, Kevin M. Pilgrim, Tan Lei
and Michael Yampolsky
University Printing House, Cambridge CB2 8BS, United Kingdom
Published in the United States of America by Cambridge University Press, New York
Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781107042919
© Bodil Branner and Núria Fagella 2014
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2014
Printed and bound in Spain by Grafos SA, Arte Sobre papel
A catalogue record for this publication is available from the British Library
Library of Congress Cataloging-in-Publication Data
Branner, Bodil, author.
Quasiconformal surgery in holomorphic dynamics / Bodil Branner and Núria Fagella ; with
contributions by Xavier Buff & Christian Henriksen, Shaun Bullett, Adam L. Epstein & Michael
Yampolsky, Peter Haı̈ssinsky, Carsten L. Petersen and Kevin M. Pilgrim & Tan Lei.
pages cm
ISBN 978-1-107-04291-9 (hardback)
1. Holomorphic mappings. 2. Differentiable dynamical systems. 3. Kleinian groups.
I. Fagella, Núria, author. II. Title.
QA614.8.B725 2014
5151 .98–dc23 2013017278
ISBN 978-1-107-04291-9 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Dedicated to the memory of Adrien Douady
Contents
List of contributors page ix

Preface xi
Acknowledgements xiii
List of symbols xv
Introduction 1
1 Quasiconformal geometry 7
1.1 The linear case: Beltrami coefficients and ellipses 7
1.2 Almost complex structures and pullbacks 14
1.3 Quasiconformal mappings 20
1.4 The Integrability Theorem 39
1.5 An elementary example 49
1.6 Quasiregular mappings 55
1.7 Application to holomorphic dynamics 60
2 Boundary behaviour of quasiconformal maps:
extensions and interpolations 64
2.1 Preliminaries: quasisymmetric maps and quasicircles 65
2.2 Extensions of mappings from their domains
to their boundaries 69
2.3 Extensions of boundary maps 77
3 Preliminaries on dynamical systems and actions of Kleinian
groups 92
3.1 Conjugacies and equivalences 94
3.2 Circle homeomorphisms and rotation numbers 97
3.3 Holomorphic dynamics: the phase space 105
3.4 Families of holomorphic dynamics: parameter spaces 126
3.5 Actions of Kleinian groups and the Sullivan dictionary 133
vii
viii Contents
4 Introduction to surgery and first occurrences 147

4.1 Changing the multiplier of an attracting cycle 151
4.2 Changing superattracting cycles to attracting ones 162
4.3 No wandering domains for rational maps 169
5 General principles of surgery 179
5.1 Shishikura principles 180
5.2 Sullivan’s Straightening Theorem 184
5.3 Non-rational maps 186
6 Soft surgeries 188
6.1 Deformation of rotation rings
Xavier Buff and Christian Henriksen 189
6.2 Branner–Hubbard motion 207
7 Cut and paste surgeries 218
7.1 Polynomial-like mappings and the Straightening
Theorem 219
7.2 Gluing Siegel discs along invariant curves 224
7.3 Turning Siegel discs into Herman rings 235
7.4 Simultaneous uniformization of Blaschke products 244
7.5 Gluing along continua in the Julia set 248
7.6 Disc-annulus surgery on rational maps
Kevin M. Pilgrim and Tan Lei 267
7.7 Perturbation and counting of non-repelling cycles 282
7.8 Mating a group with a polynomial
Shaun Bullett 291
8 Cut and paste surgeries with sectors 307
8.1 Preliminaries: sectors and opening modulus 308
8.2 Creating new critical points 320
8.3 Embedding limbs of M into other limbs 337
8.4 Intertwining surgery
Adam Epstein and Michael Yampolsky 343
9 Trans-quasiconformal surgery 364
9.1 David maps and David–Beltrami differentials 365
9.2 Siegel discs via trans-quasiconformal surgery
Carsten Lunde Petersen 370
9.3 Turning hyperbolics into parabolics
Peter Haı̈ssinsky 385
References 400
Index 408
Contributors
Bodil Branner
Technical University of Denmark
Xavier Buff
Université Paul Sabatier, Toulouse
Shaun Bullett
Queen Mary, University of London
Adam Epstein
University of Warwick
Núria Fagella
Universitat de Barcelona
Peter Haı̈ssinsky
Université Paul Sabatier, Toulouse
Christian Henriksen
Technical University of Denmark
Carsten Lunde Petersen

Roskilde Universitet
Kevin M. Pilgrim
Indiana University
ix
x Contributors
Tan Lei
Université d’Angers
Michael Yampolsky
University of Toronto
Preface
The firm intention of writing this book was born in the fall of 2003 during the
Ecole Thématique du CNRS Chirurgie holomorphe. This workshop was part
of a trimester organized by Adrien Douady at the Institut Henri Poincaré in
Paris. Douady was one of the fathers of the theory of holomorphic dynamics
and of many of the surgery constructions that the workshop addressed. He
used surgery as a tool in a number of ways, but especially to obtain a better
understanding of different structures in parameter space. As he said: ‘plough
in dynamical spaces and harvest in parameter space’.
Douady’s creative and geometric point of view inspired many to explore
holomorphic dynamics. Furthermore, he encouraged generous collaboration
and believed strongly in the value of sharing ideas. He gathered a large mathe-
matical family around him, and the success of the workshop is a tribute to his
influence.
Many of those who had originally developed holomorphic surgery presented
lectures. It became clear that the content of these wonderful sessions ought to
be the core of a book about surgery. We ourselves felt strongly that the book
should be more than a collection of papers, though: our goal became to enlist
the help of the speakers in creating a comprehensive study of quasiconformal
surgery.
We are delighted that our wishes have come true in the form of this book,
which puts together the foundations of surgery and many of its applications,
with, as we had hoped, contributions by a number of the workshop participants
who themselves had played such an important part in developing the field. We
are grateful for their support. They have our sincere and enthusiastic thanks.
Writing and collecting the material and then unifying it into a book was
not an easy task for us. Douady’s excitement about the project and his con-
stant encouragement were very important. They kept us going until this very
moment. It is to him that we dedicate every word written here, wishing he
could have seen the final result.
xi
Acknowledgements
This book has taken a long time to write, and we have received a lot of assis-
tance along the way. It is our pleasure to thank many friends and colleagues
for their generous help that made it possible.
First and foremost we are extremely grateful to Xavier Buff, Shaun Bullett,
Adam Epstein, Peter Haı̈ssinsky, Christian Henriksen, Carsten Lunde Petersen,
Kevin Pilgrim, Tan Lei and Michael Yampolsky for agreeing to participate in
the project, making the book so much better and special with their contribu-
tions. Among them we owe special gratitude to Buff, Haı̈ssinsky and Henriksen
whose valuable suggestions helped to improve other parts of the book.
There are many others who should also be mentioned. In particular, the
participants in the Quasiconformal Surgery course at the Complex Dynamics
Seminar at Universitat de Barcelona, Antonio Garijo, Xavier Jarque, Helena
Mihaljević-Brandt, Jörn Peter and Jordi Taixés inspired us with questions and
suggestions and read parts of early drafts of the book. There are also those
who made valuable comments to different parts of later drafts. These include
Anna Miriam Benini, Jordi Canela, Matı́as Carrasco, Jonguk Yang and, very
specially, Jonathan Brezin, Albert Clop, Ernest Fontich, Linda Keen, Curtis
McMullen and Caroline Series. We heartfully thank all of them, and others who
helped in various ways, for their effort. Of course, we take full responsibility
for any errors that remain.
Furthermore, we wish to thank our editor, Roger Astley, for his kindness,
enthusiasm and dedication during the whole process. On the technical side we
are grateful to Christian Mannes for creating It, the computer program with
which many of the illustrations were made. Our thanks also go to the Institut
for Matematik og Computer Science at Danmarks Tekniske Universitet, the
Departament de Matemàtica Aplicada i Anàlisi at Universitat de Barcelona,
the IMUB (Institut de Matemàtica de la UB) and the CRM (Centre de Recerca
Matemàtica) for their support.
xiii
xiv Acknowledgements
And finally, but most especially, our sincere gratitude goes to our families,
for their encouragement, support and patience throughout this long, but for us
so very exciting, project.
This book would not have been possible without financial support from
different sources. We first and most wish to thank the Marie Curie project
CODY (MRTN-CT-2006-035651), which included this book as one of its
mathematical training goals. Most of the travel expenses over the last five years
were covered by CODY, together with the grant 272-07-0321 from the Danish
Research Council for Nature and Universe and by the grants MTM2008-01486
and MTM2006-05849 from the spanish Ministry of Science. Most recently, we
were also partially supported by the grant MTM2011-26995-C02-02 from the
same source and the catalan grant 2009SGR-792.
Bodil Branner and Núria Fagella

Kongens Lyngby and Barcelona
Symbols
„ affine conjugate
aff ine
„ hybrid equivalent
hyb
„ quasiconformally conjugate
qc
„ topologically conjugate
top
» conformal equivalence
1A The characteristic function takes the value 1 on A and 0 on CzA
Af pαq The basin of attraction of an attracting p-cycle α “
tα0 , . . . , αp´1 u of f
A˝f pαq The immediate basin of attraction of a cycle α as above
Af p8q The basin of attraction of infinity of a polynomial f
Ar Round open annulus tr ă |z| ă 1u
Ar,R Round open annulus tr ă |z| ă Ru
B The Bryuno numbers or a Bers’ slice
C The complex plane
C˚ The punctured complex plane Czt0u
C
p The extended complex plane C Y t8u
Cf Set of critical points of f
C r pU q, r ě 1 The space of r times differentiable maps on U , whose nth
derivatives are continuous for all 1 ď n ď r
C 8 pU q The spaces of functions which belong to C r pU q for all r ě 1
Ccr pU q The space of functions in C r pU q with compact support, for 1 ď
rď8
Critpf q Set of finite critical points of f , i.e. in C
Bz , Bz Ordinary partial derivatives with respect to z and z
B, B Partial derivatives in the sense of distributions with respect to z
and z
xv
xvi Symbols
A Siegel disc or a linearizing domain around an attracting

periodic point.
D The open unit disc t|z| ă 1u in C
D˚ The punctured unit disc Dzt0u
Dr The open disc t|z| ă ru
Dr pz0 q The open disc t|z ´ z0 | ă ru
D ` pU, V q The set of orientation preserving continuous functions f : U Ñ
V which are differentiable almost everywhere and whose dif-
ferential Du f is non-singular almost everywhere and depends
measurably on u P U
D0` pU, V q The set of functions in D ` pU, V q which are absolutely contin-
uous with resepct to the Lebesgue measure
dCp pz, wq The spherical distance between two points in the Riemann
sphere
Ent The set of entire transcendental maps
Ent˚ The set of holomorphic transcendental self-maps of C˚
extpγ q The domain to the right of γ , an oriented Jordan curve
Ff The Fatou set of f
Fc The Fatou set of Qc
fn f ˝n :“ f ˝ ¨ ¨ ¨ ˝ f , the map f ˇcomposed by itself n times
"ˆ ˙ *
a b ˇˇ
GLp2, Cq General linear group a, b, c, d P C
cd ˇ
H The upper half plane tIm z ą 0u
H The left half plane tRe z ă 0u
Hr The right half plane tRe z ą 0u
H The Herman numbers or, in Section Section 8.4, a hyperbolic
component
Im z The imaginary part of z
intpγ q The domain to the left of γ , an oriented Jordan curve
intpXq Interior of the set X
Jf The Julia set of f
Jc The Julia set of Qc
Kf The filled Julia set of a polynomial or a polynomial-like map f
Kc The filled Julia set of Qc
Lp{q The p{q-limb of the Mandelbrot set
M The Mandelbrot set
Mer The set of transcendental meromorphic maps with at least one
pole which is not omitted
Symbols xvii
Mer8 The set of transcendental maps which are meromorphic outside

a compact countable set of singularities
mod modulus
N The natural numbers t1, 2, . . .u
OpXq The orbit of X, where X is a point or a set
Pf The postsingular set or the postcritical set
Pol The set of polynomials of degree at least two
Pold The set of polynomials of degree d ě 2
Q The rational numbers
Qc The quadratic polynomial Qc pzq “ z2 ` c
R A Riemann map
R The real line
R˚ The punctured real line Rzt0u
R
p The extended real line R Y 8
Rat The set of rational maps of degree at least two
Ratd The set of rational maps of degree d ě 2
Re z The real part of z
Rθ The rigid rotation by θ P R, represented either by z ÞÑ e2π i θ z,
where z P S1 or x ÞÑ x ` θ pmod 1q where x P R
σ pz, wq The chordal distance between two points in the Riemann sphere
S1 The unit circle t|z| “ 1u
S1r The circle t|z| “ ru
Singpf ´1 q Set of singularities of an inverse map
T The quotient space R{Z
Vf Set of critical values of f
Z The integers t. . . , ´2, ´1, 0, 1, 2, . . .u
Introduction
The theory of one-dimensional complex dynamical systems, understood as the

global study of iteration of holomorphic mappings, has its roots in the early
twentieth century with the work of Pierre Fatou and Gaston Julia. Local studies
had been successfully attempted earlier, but it was Fatou and Julia’s seminal
work, inspired by Paul Montel’s notion of normal families of mappings (then
relatively new), that set the basis of what is known today as holomorphic
dynamics. Both Fatou and Julia studied extensively the basic partition of the
dynamical space into the two disjoint, completely invariant subsets, the Fatou
set, which is the open set where tame dynamics occur – the set where Montel’s
normality appears – and its complement, the Julia set, which is the set of initial
values whose orbits are chaotic.
Their greatest achievement, arguably, is their detailed description of the
geometry and the dynamics of the connected components of the Fatou set,
called the ‘Fatou components’. Their Classification Theorem asserts that every
periodic Fatou component of a holomorphic map of the Riemann sphere
(a rational map) is either (i) a component of an immediate basin of attraction
of some attracting or parabolic cycle or (ii) a rotation domain conformally
equivalent to a disc or an annulus.
Their work also left many interesting open questions, such as Fatou’s No
Wandering Domains Conjecture, which states that all Fatou components are
eventually periodic, and waited some 60 years for its resolution. Some of their
other questions remain open today, but despite their compelling interest, the
problems were largely ignored until the late 1970s.
The other side of our story, the theory of quasiconformal functions of the
plane, began at roughly the same time with the work of Herbert Grötzsch.
As the name suggests, quasiconformality is a weakening of the notion of
conformality. It is best understood as a geometric condition: conformal maps
1
2 Introduction
preserve angles, and quasiconformal maps distort angles, but only in a bounded
fashion. Another important difference is that they do not need to be differ-
entiable everywhere, but only almost everywhere. Even so, many important
theorems about conformal mappings, minimally recast, remain true for quasi-
conformal maps.
Grötzsch’s work was soon taken in two different directions. Oswald
Teichmüller showed that there is a deep connection between quasiconfor-
mality and the function theory of Riemann surfaces, and Charles B. Morrey
investigated its relation to the solution of PDEs. Beginning in the 1950s,
the geometric side of the theory was developed by Albert Pfluger and Lars
Ahlfors. This led to a unification of all of the earlier work into a single, general
theory. At the same time, Morrey and Ahlfors, together with Lipman Bers and
Bogdan V. Bojarski formulated the celebrated Measurable Riemann Mapping
Theorem as it is known today – in this book, it is referred to somewhat more
concisely as the Integrability Theorem. As we shall see, it is the essential tool
for quasiconformal surgery.
Around 1980, two remarkable developments added to this account. On the
one hand, computer graphics made it possible to draw pictures of the beautiful
phenomena and fractal structures exhibited by the holomorphic dynamics of
even very simple maps. This clearly awoke an interest in the subject. On the
other hand, and at a much deeper level, in 1981 Dennis Sullivan realized there
was a strong connection between holomorphic iteration and the actions of
Kleinian groups, introducing what has become known as Sullivan’s dictionary
between these two subjects. Inspired by Henri Poincaré’s original (1883)
perturbations of Fuchsian groups into quasi-Fuchsian groups, he injected the
modern theory of quasiconformal mappings into complex iteration to solve
Fatou’s No Wandering Domains Conjecture. With the same arguments, he also
gave a new proof of Ahlfors’ Finiteness Theorem, a cornerstone in the theory
of Kleinian groups.
Sullivan’s technique is now referred to as soft quasiconformal surgery. It
deals with quasiconformal deformations of a given map. However, what we
call cut and paste quasiconformal surgery is more reminiscent of topolog-
ical surgery, which is used to produce one manifold from another in some
‘controlled way’. In holomorphic iteration, quasiconformal surgery is used to
obtain holomorphic maps with prescribed dynamics that arise from certain
model maps which are locally quasiconformal. Often, these model maps are
constructed by cutting and pasting different spaces and maps together, explain-
ing the reference to topological surgery. It is worth noticing that, whereas the
uniqueness of analytic continuation implies that holomorphic maps are rather
Introduction 3
rigid, quasiconformal maps can be pasted together very flexibly without losing
quasiconformality.
Sullivan’s first crucial work was merely the beginning of the uses to which
quasiconformal surgery could be put in holomorphic dynamics. Quasiconfor-
mal surgery has been at the heart of many important advances and is now part
of the essential background knowledge of any researcher in the field.
All of this brings us to the raison d’etre for this book. While the theory
of quasiconformal maps is a mature subject for which an excellent literature
exists, a comprehensive treatment of its application to dynamical systems has,
until now, only been possible by going back to the original papers. Books about
quasiconformal maps, like [Ah5, LV, Le] and [AIM], naturally focus on the
function theory, so the surgery techniques are of mainly anecdotal interest.
Similarly, general texts on holomorphic dynamics, like [Mi1, Bea, CG, St] and
[MNTU], are not the right place to discuss quasiconformal surgery, because
the prerequisite theory of quasiconformality would sit very heavily in such a
general work. Our goal is to present a text where one can learn from the start
about this beautiful, highly geometric and powerful technique and how to apply
it to holomorphic dynamics.
This book is therefore neither a book about holomorphic dynamics nor about
quasiconformal mappings, although it contains introductions to both subjects,
together with a brief introduction to Kleinian groups, in order to understand
the parallels that often come up. These introductions, however, do not present
proofs that can easily be found in other standard texts. Both authors have
a stronger background in dynamics than in analysis and this surely intro-
duces a bias in the way these introductions are presented. While no previous
knowledge about quasiconformal maps or Kleinian groups is assumed, some
familiarity with dynamical systems and especially holomorphic dynamics in
one complex variable is useful, in particular when understanding the purpose
of the different surgery constructions.
While the foundations of the theory and the general description of surgery
techniques occupy a good half of the book, the other half is completely
dedicated to a variety of applications. These applications are more or less
independent of each other, and so each can be read on its own. Our inter-
est is precisely to present the different types of surgery techniques that are
required for attacking the most important results in holomorphic dynamics.
The applications are grouped by similarities between the constructions they
require. Two of them go beyond the classical realm of quasiconformal surgery
and use trans-quasiconformal surgery. Even further, another one deals with
holomorphic correspondences, a natural generalization of holomorphic maps.
4 Introduction
Some of the sections on applications are written by the authors of the original
papers, whom we once again thank.
We should also comment on what this book does not cover. Although
quasiconformal maps and surgery itself are intimately related to Teichmüller
Theory, this deep area of mathematics is not treated in this book. To learn
about it, we strongly recommend the recent volume by John H. Hubbard
[Hu2]. In connection to the above, there is also no reference to quadratic
differentials, although they play an important role in the theory of holomorphic
dynamics.
Actions of Kleinian groups only appear in the book in a sparse way. We
give a brief introduction to the subject so that references to the parallelism
between deformations of Kleinian groups and surgery in holomorphic dynam-
ics, expressed by Sullivan’s dictionary, can be understood.
About the book structure

Chapter 1, Quasiconformal geometry, contains the basic definitions and results
about quasiconformal mappings used in surgery, and those necessary to make
a coherent description. We give a geometrical and an analytical approach, give
precise references to all important results, and prove those (even if simple),
which we could not find proven in the literature. Nevertheless they may be
useful for the reader of this text.
The same comments apply to Chapter 2, Boundary behaviour of quasicon-
formal maps: extensions and interpolations. We study how to extend maps
to the boundary of their domains, depending on the regularity of the ele-
ments involved. Conversely, we see how to extend boundary maps to their
neighbouring domains, depending also on the regularity of the different ingre-
dients. These extensions and interpolations are technical results, which are
used repeatedly in each and every surgery. Some of the results are celebrated
theorems while others are simple exercises; but all are part of the community
folklore. We have made it a point to collect them in this chapter, and to prove
or indicate proofs of those that can be shown with elementary methods.
Chapter 3, Preliminaries on dynamical systems and actions of Kleinian
groups, contains the background in dynamics necessary for understanding
the sections that follow and especially their motivation. We include general
basic background for discrete dynamical systems, circle maps, holomorphic
dynamics, and families of such, as well as dynamics of Kleinian groups. As
in the previous chapters, there are basically no proofs, but all results are well
referenced.
Introduction 5
In Chapter 4, Introduction to surgery and first occurrences, we present

the earliest applications of surgery to holomorphic dynamics, although not
necessarily the simplest. Inspired by quasiconformal deformations of Kleinian
groups, the surgeries lead to conformal parametrizations of hyperbolic compo-
nents of the Mandelbrot set by D, due to Sullivan, Douady and Hubbard, and
to the No Wandering Domains Theorem for rational maps, due to Sullivan. We
have chosen to keep a historical flavour, so the surgeries appear in a form fairly
close to their original presentation.
When applying quasiconformal surgery, there are several paths one may take
to accomplish the same result. More precisely, Chapter 5, General principles of
surgery, is dedicated to establishing three criteria (due to Mitsuhiro Shishikura
and Sullivan) giving conditions under which a surgery can be completed to
result in a holomorphic map. The First and Second Shishikura Principles
are actually corollaries of the third principle, called Sullivan’s Straightening
Theorem. We present all three since their proofs provide different insights.
Sullivan’s Straightening Theorem gives a necessary and sufficient condition to
decide whether a surgery can be completed. However, these criteria are rarely
used in the applications in the book, since certain informations can only be
extracted from the specific details of the proof ‘from scratch’. Needless to say,
the criteria may be useful in many other instances.
Chapters 6 to 9 are entirely dedicated to applications. Sections are pairwise
independent, although they all use the common terminology and background
of the first part of the book. Some of the applications were presented during
the surgery workshop at the Institute of Henri Poincaré in Paris in the fall of
2003 as part of the trimester organized by Adrien Douady. Other applications
have been added for completeness. We have tried to show as many different
constructions as possible. However, the list is by no means exhaustive.
Chapter 6, Soft surgeries, consists of applications where only the com-
plex structure is deformed, so that it is invariant under the given map. The
resulting maps are therefore quasiconformally conjugate to the original one.
The chapter contains a contribution by Xavier Buff and Christian Henrik-
sen. Chapter 7, Cut and paste surgery, is the longest chapter and contains
many groups of applications. The common feature in all of them is that
maps and complex structures are both changed. It contains contributions by
Kevin M. Pilgrim and Tan Lei, and Shaun Bullett, who present a special
surgery application of mating a Kleinian group with a quadratic polynomial.
Chapter 8, Cut and paste surgeries with sectors, contains applications where
the deformation of the maps and the complex structures are concentrated in
sectors. The chapter contains a contribution by Adam Epstein and Michael
Yampolsky.
6 Introduction
Chapter 9, Trans-quasiconformal surgery, is dedicated to the contributions

by Carsten L. Petersen and Peter Haı̈ssinsky. The surgeries require maps, called
David homeomorphisms, that are not quite quasiconformal, but are ‘close
enough’ in a sense made precise there. Many new applications have appeared
using this technique. The two we have included are among the earliest and a
good introduction to the more recent work.
On how to read this book

Readers familiar with quasiconformal geometry and analysis may start in
Chapter 3, Preliminaries on dynamical systems and actions of Kleinian groups,
and continue from there. Those familiar with dynamics, holomorphic dynamics
in particular, may start in Chapter 1, Quasiconformal geometry, read Chapter 2,
Extensions and interpolations, only diagonally to see what type of results it
contains, and then jump directly to Chapter 4, Introduction to surgery and first
occurrences. Chapter 2 can be used as a reference chapter throughout the book.
In both cases, if the first applications in Chapter 4 are too difficult to start
with, we advise to leave them for later and start with those in Chapter 7, Cut
and paste surgery, maybe after having read the introduction to surgery given in
Chapter 4. Chapter 5, General principles of surgery, can be read at any point,
since the principles are not used often in the constructions, and they may be
too abstract for someone not well acquainted with a number of examples.
We hope that experts in both quasiconformal mappings and holomorphic
dynamics, and even in quasiconformal surgery, may still find some of the
applications interesting and in any case a practical collection to have at hand
as a reference. It is clear that such readers may skip Chapters 1 to 5.
The book can be used, and has been used, several times already, for a
graduate course in quasiconformal surgery. If so, it is not recommended to
cover the book in a sequential way but to follow the recommendations given
above. The lecturer may selectively choose some sections in Chapters 6 to 9
and keep the remaining ones as projects for the students who may be asked to
read the original papers, fill in details and do the suggested exercises.
1
Quasiconformal geometry
Since its introduction in the early 1980s quasiconformal surgery has become a
major tool in the development of the theory of holomorphic dynamics.
The goal of this chapter is to collect the basic definitions and results about
quasiconformal mappings used in surgery, and those necessary to make a
coherent description. We give a geometrical and analytical approach and give
precise references to important results in the vast literature about quasicon-
formal mappings. In general we only prove those we could not find proven
elsewhere, even if simple. However, since the Integrability Theorems (Theo-
rems 1.27 and 1.28), also called the Measurable Riemann Mapping Theorems,
are the cornerstone behind every surgery construction we sketch a proof of
those.
Holomorphic maps are very rigid, due to the property of analytic continu-
ation. For this reason it is not possible to paste different holomorphic maps
together along a curve to form a new holomorphic map. However, quasicon-
formal mappings do have this kind of flexibility and can be pasted together to
form new quasiconformal mappings. It is this flexibility that produces the basis
for surgery constructions where we change mappings and sometimes also the
underlying spaces. When the construction is successful the final goal is to end
with a holomorphic map, obtained via the Integrability Theorem.
1.1 The linear case: Beltrami coefficients and ellipses

Let CR denote the complex plane, viewed as the two-dimensional oriented
Euclidean R-vector space with the orthonormal positively oriented standard
basis t1, i u. In CR we shall use as coordinates either px, yq or pz, zq where
z “ x ` iy and z “ x ´ i y.
7
8 Quasiconformal geometry
Any R-linear map L : CR Ñ CR can be written, using the coordinates

pz, zq, in the form
Lpzq “ az ` bz, with a, b, z P C.
The unit square, spanned by 1 and i , is mapped onto the parallelogram spanned
by a ` b and ai ´ bi (see Figure 1.1).
pa ´ b qi
i
L
a`b
Figure 1.1 The unit square is mapped to the parallelogram spanned by

ta ` b, pa ´ bqi u.
The absolute value of the determinant of L is the area of the parallelogram,

i.e. detpLq “ |a|2 ´ |b|2 . We shall restrict to R-linear maps that are invertible
and orientation preserving, i.e. with |a| ą |b|.
We define the Beltrami coefficient of L to be μpLq “ ab , and – for reasons
which will become clear below – we let θ P R{pπ Zq denote half the argument
of μpLq, i.e.
ˇ ˇ
ˇbˇ
μpLq :“ ˇˇ ˇˇ ei 2θ .
a
Note that μpLq P D when L is orientation preserving, and that L is holomor-
phic if and only if b “ 0, which occurs if and only if μpLq “ 0.
Let EpLq denote the inverse image by L of the unit circle. Then EpLq is an
ellipse, and in particular a circle if μpLq “ 0.
In order to determine the ellipse EpLq, we set a “ |a| ei α , where α P
R{p2π Zq, μ “ μpLq and rewrite L as
Lpzq “ ei α |a| z ` |μ| ei 2θ z .

` ˘
Hence, L is the R-linear map Spzq “ |a| z ` |μ| ei 2θ z post-composed with

` ˘
the rotation Rpzq “ ei α z. We have split L into the composition of a self-

adjoint linear transformation followed by an orthogonal transformation. (It
is easy to check that the 2 ˆ 2 matrix of S in the basis t1, i u is symmetric,
and S therefore is self-adjoint.) It follows that S has two real eigenvalues and,
if b ‰ 0, that their corresponding eigenvector directions are orthogonal (see
Figure 1.2).
i pθ` π
2q
e
|a|p1´|μ|q
S R
ei θ
|a|p1`|μ|q
θ
Figure 1.2 The ellipse EpSq “ EpLq.
It is easy to check that ei θ and ei pθ `π {2q are eigenvectors of S corresponding

to the eigenvalues |a|p1 ` |μ|q and |a|p1 ´ |μ|q respectively. It follows that
1
EpSq is the ellipse with half major axis |a |p1´| μ|q
along the direction ei pθ `π {2q
1
and half minor axis |a |p1`|μ|q
along the orthogonal direction ei θ . The ellipse
EpLq equals EpSq, since the unit circle is preserved by the rotation Rpzq “
ei α z.
We define the dilation KpLq of L as the ratio of the major axis to the minor
axis:
1 ` |μ| |a| ` |b|
KpLq :“ “ ,
1 ´ |μ| |a| ´ |b|
and the complex dilatation of L as the Beltrami coefficient μpLq. The dilatation
KpLq determines the shape of the ellipse up to scaling, but not the position of
its axes. The Beltrami coefficient determines the position and the shape up
to scaling. Conversely, if we happen to start with an ellipse E, the Beltrami
´m i 2θ
coefficient is determined by μpEq “ M M `m e , where M and m are the half
major and half minor axes of E respectively, and θ is the argument of the
direction of the minor axis of E chosen in r0, π q. The dilatation KpEq is
defined in the natural way (see Figure 1.3).
m θ
Figure 1.3 Given an ellipse E, we define μpEq and KpEq in terms of M, m, and θ.
Observe that we choose the argument of the minor axis, θ, to belong to r0, π q.
We shall denote by σ0 the standard conformal structure of CR , that is to

consider CR as a C-vector space with the standard complex scalar multipli-
cation. Any invertible R-linear map L can be used to define a new conformal
structure σ pLq on the domain of L, that is a new operation making CR into a
C-vector space, extending the R-vector space structure. This is done in the
following way: we need to define what it means to ‘multiply’ elements of
CR by complex scalars, which reduces (after imposing all the properties that
must be satisfied) to define ‘multiplication’ by i . That is, we need to choose
an R-linear map J , and define c ˚ z “ Re c z ` Im c J pzq for any c, z P C.
It follows from imposing i 2 ˚ z “ i ˚ i ˚ z, that J pJ pzqq “ ´z. The structure
induced by L is defined by choosing J “ L´1 ˝ I ˝ L, where I pzq “ i z in the
standard way.
We will end this linear discussion by considering how Beltrami coefficients
and dilatations change under inversion and composition of linear maps. We
start with inversion. Given a map L as above, it is easy to check that
1
L´1 pwq “ paw ´ bwq.
|a|2 ´ |b|2
It follows that
μpL´1 q “ ´μpLqei p2 arg a q , (1.1)
and hence |μpL´1 q| “ |μpLq|, which implies that
KpL´1 q “ KpLq. (1.2)
Now suppose j P t1, 2u and we have two R-linear maps Lj pzq “ aj z `

bj z with dilatation Kj and Beltrami coefficient μj . The ellipse defined by the
composition L1 ˝ L2 is the preimage under L2 of the ellipse defined by the
map L1 . Observe that from linear algebra, we can assure that
KpL1 ˝ L2 q ď KpL1 qKpL2 q,
since the maximal possible stretch is the product of the two maximal stretches
of each of the maps, while the corresponding holds for the minimal stretches.
If we want to know the Beltrami coefficient for this new ellipse we compute
the composition
pL1 ˝ L2 qpzq “ pa1 a2 ` b1 b2 qz ` pa1 b2 ` b1 a2 qz,

hence
b2 ` μ1 a2
μpL1 ˝ L2 q “ (1.3)
a2 ` μ1 b2
or, equivalently
μ2 ` μ1 e´2i arg a2
μpL1 ˝ L2 q “ . (1.4)
1 ` μ1 μ2 e´2i arg b2
We will use these expressions later on when we discuss more general maps.
1.1.1 Geometric interpretation

We shall give a geometric interpretation of the conformal structure σ defined
by an ellipse, or rather an infinite family of similar concentric ellipses.
First suppose we have an R-linear map L as above and the ellipse EpLq.
Given a diameter in the unit circle or in the ellipse EpLq, the conjugate
diameter is defined as the one that is made up of the midpoints of the cords
in the circle or in the ellipse parallel to the given diameter. On the circle, this
corresponds to the orthogonal diameter. On a genuine ellipse the only conju-
gate diameters that are orthogonal to each other are the axes (see Figure 1.4).
Alternatively, the conjugate diameter is the one parallel to the line tangent to
the ellipse at the point of intersection with the given diameter. The map L sends
conjugate diameters of the ellipse EpLq to conjugate diameters of the circle,
since parallel lines are mapped to parallel lines and midpoints to midpoints.
z
EpLq Lpzq
J pz q
L
“ L´1 pi Lpzqq
i Lpzq
J J pzq “ ´z
Figure 1.4 E Ă CR so that LpEq is a circle in CR . A pair of conjugate diameters in

E are mapped by L onto the corresponding pair of conjugate diameters in
the circle LpEq.
If z P EpLq, then define J pzq to be the vector in EpLq so that z and J pzq are
on conjugate diameters, turning from z in the positive direction (less than π )
to J pzq. Clearly J “ L´1 ˝ I ˝ L, where I pzq “ i z.
Suppose an ellipse in CR is only defined up to scaling, meaning that we
have an infinite family of similar concentric ellipses (see Figure 1.5). Such a
Figure 1.5 Some similar concentric ellipses. They are equal up to scaling
by a real constant.
family determines a conformal structure σ in CR . For any z ‰ 0, choose the

ellipse that contains z. Let J pzq be the point in the positive direction on that
ellipse and on the diameter conjugate to the one containing z, positioned as
above. The map J clearly satisfies J pJ zqq “ ´z and defines how to multiply
a vector by i .
Exercises Section 1.1

1.1.1 Given a, b P C, consider the R-linear map Lpzq “ az ` bz.
(i) Show that the matrix of L in the standard basis tp1, 0q, p0, 1qu in
R2 , equivalent to t1, i u, is
„ j
Repa ` bq ´ Impa ´ bq
L“ .
Impa ` bq Repa ´ bq
(ii) Justify that the determinant of the matrix is |a|2 ´ |b|2 .
(iii) Consider in particular Spzq “ |a| z ` |μ| ei 2θ z , and confirm that

` ˘
the matrix of S in the standard basis is symmetric.

1.1.2 Given a, b P C with |a| ą |b|, consider the R-linear map Lpzq “ az `
bz. Recall that the norm }L} of L is defined by
}L} :“ max |Lpzq|.
zPS1
Note that LpS1 q “ EpL´1 q. Show that

}L}2
KpLq “ . (1.5)
det L
We apply the content of this exercise in Exercise 1.3.3 in Section 1.3.
1.1.3 Let L : R2 Ñ R2 be an orientation preserving and invertible R-linear
map, which in the standard basis tp1, 0q, p0, 1qu in R2 is given by the
matrix „ j
αβ
L“ ,
γ δ
and let E be the ellipse satisfying LpEq “ S1 .

(i) Show that points px, yq P E satisfy the equation

pαx ` βyq2 ` pγ x ` δyq2 “ 1.
Recall, that the symmetric matrix of the quadratic form in the left-
hand side of this equation in the standard basis is equal to
„ 2
α ` γ 2 αβ ` γ δ
j
T
Q“L L“ .
αβ ` γ δ β 2 ` δ 2
Let λj , j “ 1, 2, denote the eigenvalues of Q. Recall that there
exists an orthonormal change of coordinates, so that the equation of
E in the new coordinates px1 , y1 q takes the form λ1 x12 ` λ2 y12 “ 1.
(ii) Show that both eigenvalues λj , j “ 1, 2, are positive, and label
them so that λ1 ě λ2 ą 0. Express the half major and half minor
axes M and m respectively in terms of λj . Then conclude that the
distortion of L is given as
d
M λ1
KpLq “ “ .
m λ2
1.1.4 (Complex dilatation of an orientation reversing linear map) Given

a, b P C with |b| ą |a|, consider the R-linear orientation reversing map
Apzq “ az ` bz. Let E be the ellipse satisfying ApEq “ S1 . Then μpEq
and KpEq are well defined and we may define (compare with Figure 1.3)
μpAq :“ μpEq and KpAq :“ KpEq.
Show that
á ¯ |b| ` |a|
μpAq “ and KpAq “ . (1.6)
b |b| ´ |a|
Observe that if A is antiholomorphic (i.e. a “ 0) the ellipse is a circle.
Hint: Consider the orientation preserving map Lpzq :“ Apzq and
observe that ApEq “ S1 “ LpEq.
1.1.5 (Composition of two orientation reversing linear maps) For j “ 1, 2,
let Aj pzq “ aj z ` bj z. Set μ1 “ μpA1 q, the Beltrami coefficient of the
1 1
ellipse E1 “ A´ 1 pS q, as defined in the preceding exercise. Observe
that A1 ˝ A2 is orientation preserving and show that
μ1 b2 ` a2
μpA1 ˝ A2 q “ . (1.7)
μ1 a2 ` b2
This formula is the analogue of (1.3) for composition of two orien-
tation preserving linear maps. Note that the Beltrami coefficient of
the complex conjugate ellipse E1 equals μ1 . The formula states the

1
Beltrami coefficient μpA1 ˝ A2 q of the ellipse E “ A´
2 pE1 q “ pA1 ˝
1 ´1
A2 q´1 pS1 q and shows that A´
2 pE1 q “ A2 pE1 q.
1.2 Almost complex structures and pullbacks

Let U Ă C, and let T U “ uPU Tu U be the tangent bundle over U , i.e. the
Ť
collection of the tangent spaces over points u P U , each one viewed as a copy
of CR . An almost complex structure on U is a measurable field of infinitesimal
ellipses E Ă T U . By this we mean an ellipse Eu Ă Tu U defined up to scaling,
for almost every point u P U , such that the map u ÞÑ μpuq from U to D
is measurable (with respect to the Lebesgue measure), where μpuq denotes
the Beltrami coefficient of Eu . Each infinitesimal ellipse defines a conformal
structure σ puq on Tu U , as defined and explained in Section 1.1, making the
tangent space Tu U into a C-linear vector space. We denote the almost complex
structure by σ , and define the dilatation of σ as
1 ` |μpuq|
Kpσ q :“ ess sup Kpuq, where Kpuq :“
uPU 1 ´ |μpuq|
denotes the dilatation of Epuq. Observe that Kpσ q P r1, 8s.
Notice also that any measurable function μ : U Ñ D defines an almost
complex structure in the above sense.
Now let us see how one can obtain almost complex structures from maps
satisfying certain conditions. Consider U, V Ă C and the class D ` pU, V q of
continuous orientation preserving functions f from U onto V which are R-
differentiable almost everywhere, and with a non-singular differential Du f :
Tu U Ñ Tf puq V almost everywhere, depending measurably on u. Since we are
working in tangent spaces we use the infinitesimal coordinates dz and dz. Then
the differential can be written as
Du f “ Bz f puqdz ` Bz f puqdz,
where
ˆ ˙ ˆ ˙
1 Bf Bf 1 Bf Bf
Bz f “ í , Bz f “ ì .
2 Bx By 2 Bx By
Applying the discussion of Section 1.1 to this map, we see that Du f defines
an infinitesimal ellipse in Tu U with Beltrami coefficient equal to
Bz f puq
μf puq “ (1.8)
Bz f puq
or, equivalently, a new conformal structure on this tangent space. The dilatation
can be written as
1 ` |μf puq|
Kf puq :“ KpDu f q “ .
1 ´ |μf puq|
Observe that the Cauchy–Riemann equation Bz f puq “ 0 is satisfied at u if
and only if μf puq “ 0, or equivalently, if and only if the ellipse is a circle.
If we do the same for all points u P U for which f is differentiable, we
obtain a measurable field of infinitesimal ellipses, one in almost every tangent
space or, as defined above, an almost complex structure σf on U , with Beltrami
coefficient μf . We say that σf is the pullback of σ0 (the field of infinitesimal
circles, known as the standard complex structure) by f , or equivalently that
μf is the pullback of μ0 ” 0 by f (see Figure 1.6). We write for almost every
uPU
μf puq “ f ˚ μ0 puq or σf puq “ f ˚ σ0 puq.
The dilatation of this almost complex structure is then the essential supremum
over U of Kf puq, which we denote by Kf .
f
pU, σf q pV , σ0 q
f puq
u
Du f
Figure 1.6 The almost complex structure σf in U is the pullback of σ0 in V , under f .

The fields of infinitesimal ellipses live in the tangent bundle. The infinitesimal ellipse
at Tu U is mapped under Du f (if it exists) to the infinitesimal circle at Tf puq V .
Notice that since the differential does not need to vary continuously with
respect to u, neither do the field of infinitesimal ellipses nor the Beltrami
coefficient. We can only say that μf is a measurable function (it is a quotient
of measurable functions) with respect to u, essentially bounded.
The concept of pullback can be slightly generalized, since we may consider
the pullback of any almost complex structure σ , not necessarily σ0 , under
a map f satisfying an extra condition. We need to require the map f to
be absolutely continuous with respect to the Lebesgue measure, that is, to
require the preimage by f of any measure zero set to be of measure zero.
We denote by D0` pU, V q the subclass of D ` pU, V q consisting of functions
with this property. Let f P D0` pU, V q and let μ be the Beltrami coefficient in
T V corresponding to an almost complex structure σ in V . Let Ev denote the

infinitesimal ellipse defined in Tv V for almost every v P V . By the pullback by
f of the measurable field of infinitesimal ellipses E, we mean the measurable
field of infinitesimal ellipses E 1 with Eu1 “ pDu f q´1 pEf puq q, well defined for
almost every u P U (see Figure 1.7). Indeed, Eu1 is defined for all u such that
Ef puq is defined and Du f exists and is non-singular. The first property is
satisfied on a set of full measure because f is absolutely continuous and so
is the second property by hypothesis.
f
pU, f ˚ μq pV , μq
f puq
u
Du f
Figure 1.7 The pullback of the Beltrami coefficient μ (in V ) under f is denoted by
f ˚ μ (in U ).
When we write
f
pU, μ1 q ÝÑ pV , μ2 q
we mean that f : U Ñ V and f ˚ μ2 “ μ1 , in the sense explained above.

If it happens that μ is given by a certain map g : V Ñ W in the class
`
D pV , W q (i.e. μ “ μg ), then we will be looking at
f ˚ μg “ f ˚ pg ˚ μ0 q “ pg ˝ f q˚ μ0 “ μg ˝f .
From the linear discussion in Section 1.1 it follows that
Kg ˝f ď Kf ¨ Kg ,
and we can write down formulas for Beltrami coefficients (compare with
equations (1.3) and (1.4)):
Bz f puq ` μpf puqqBz f puq

f ˚ μpuq “ , (1.9)
and also
μf puq ` μg pf puqqeí 2 argpBz f puqq

μg ˝f puq “ . (1.10)
1 ` μf puqμg pf puqqeí 2 argpBz f puqq
Furthermore, note that if f is holomorphic then (1.9) reduces to
Bz f puq
f ˚ μpuq “ μpf puqq . (1.11)
Bz f puq
This formula is important for two reasons: first, we shall often extend an almost
complex structure on V with Beltrami coefficient μ by pulling back with a
holomorphic map f : U Ñ V to f ˚ μ on U . Note that if |μ| ă k on V then
|f ˚ μ| ă k on U . Second, when we extend the previous notions to Riemann
surfaces as in Section 1.3.7, everything is expressed in charts and we need
to know how an expression is transferred from one chart to another, having a
holomorphic overlap with the first.
A special and very important case occurs when the pullback of an almost
complex structure σ is again σ , as we will see later.
Definition 1.1 (f-invariant almost complex structure) Let U be an open

subset of C and f : U Ñ U a map in D0` pU, U q. Let σ be an almost complex
structure in U with Beltrami coefficient μ. We say that μ (or σ ) is f -invariant
if f ˚ μpuq “ μpuq for almost every u P U . We also write f ˚ σ “ σ . Equiv-
alently, we say that f is holomorphic (in fact conformal) with respect to μ
or σ .
Almost complex structures that are f -invariant will be of crucial importance

when we perform surgery. Consider the particular situation with mappings F P
D0` pV , V q, G P D0` pU, U q, and f P D0` pU, V q, and suppose the mappings
make the following diagram commutative:
G
U ÝÝÝÝÑ U
§ §
fđ
§ §f
đ
F
V ÝÝÝÝÑ V
Note that for any F -invariant almost complex structure σ on V with Beltrami
coefficient μ, the pullback structure f ˚ σ on U is G-invariant since
G˚ pf ˚ μq “ pf ˝ Gq˚ μ “ pF ˝ f q˚ μ “ f ˚ pF ˚ μq “ f ˚ μ.
We observe that if the inverse map f ´1 is also absolutely continuous then

it makes sense to push forward almost complex structures, which means to
pull them back under f ´1 from the tangent bundle T U to T V . We write f˚ “
pf ´1 q˚ . We also have
Kf ´1 “ Kf .
As a final remark, note that we even may allow f to be non-invertible as

long as it locally belongs to the class D0` , except around a discrete set of
‘critical points’. Around critical points the pushforward operation is not well
defined (because of multiple preimages under f ´1 ) but the pullback by f is.
Critical points are not important since almost complex structures are allowed
to be defined almost everywhere.
1.2.1 Almost complex structures and symmetries

Let μ be the Beltrami coefficient of an almost complex structure defined in C.
We would like to define what it means for μ to be symmetric with respect
to the real axis or the unit circle. Maps that provide these symmetries are
precisely cpzq “ z for R and τ pzq “ 1z for S1 . It might seem reasonable to
say that a Beltrami coefficient which is symmetric with respect to, say the unit
circle, must be τ -invariant. But the reflections c and τ are antiholomorphic and
therefore orientation reversing, and so far we have not defined what it means
to pull back almost complex structures under such maps.
As a general setup, consider U, V Ă C and an orientation reversing map
f P D0´ pU, V q, defined analogously to D0` pU, V q with the differential Du f :
Tu U Ñ Tf puq . Let μ denote a Beltrami coefficient in V .
Definition 1.2 (Pullback under orientation reversing maps) In the setup

above, we define the pullback of μ under f as
˚
f f μ :“ f μ.
In particular
f f μ0 :“ μf and Kf :“ Kf .
Observe that f is orientation preserving and therefore the definitions make

formal sense. We refer to Exercises 1.1.4 and 1.1.5 in Section 1.1 for a
geometric motivation and to Exercises 1.2.1 to 1.2.4 for the deduction of these
and other formulas.
Returning to the initial problem, we obtain the following.
Definition 1.3 (Symmetries) A Beltrami coefficient μ is symmetric with

respect to R if and only if
μpzq “ μpzq,
or equivalently if cf μ “ μ, where cpzq “ z.

We say that μ is symmetric with respect to S1 if and only if

z4
μpzq “ μp1{zq “ μp1{zq e4i arg z .
|z|4
or equivalently if τ f μ “ μ, where τ pzq “ 1z .

1.2.1 (Complex dilatation of an orientation reversing map) Consider
U, V Ă C and an orientation reversing map f P D ´ pU, V q. By apply-
ing Exercise 1.1.4 in Section 1.1 observe that Du f defines an ellipse
Eu :“ pDu f q´1 pS1 q in Tu U for almost all u P U with Beltrami
coefficient
ˆ ˙
Bz f puq
μpEu q “ .
Bz f puq
Recall that Bz f “ Bz f and Bz f “ Bz f , in other words, Du f “

Du f . Show that μpEu q “ μf puq as defined in equation (1.8) and
KpEu q “ Kf puq.
As a consequence, it makes sense to define the pullback of the standard
form by an orientation reversing map as
f f μ0 :“ μf and Kf :“ Kf .
1.2.2 (Pullback under orientation reversing maps) Let f, U and V be as in

the exercise above and let μpvq be a Beltrami coefficient corresponding
to a field of infinitesimal ellipses pEv qv PV . In analogy with formula (1.7)
in Exercise 1.1.5 in Section 1.1, we define the pullback by f as
˚
f f μ :“ f μ.
Deduce from (1.9) that

f f μpuq “ ,
and if f is antiholomorphic, then
Bz f puq
f f μpuq “ μpf puqq .
Bz f puq
If follows from Exercise 1.1.5 that pDu f q´1 pEf puq q “ pDu f q´1
pEf puq q. Hence, the pullback definition is justified by this geometrical
interpretation.
1.2.3 (Reflection with respect to R) Let cpzq “ z and μ a Beltrami coeffi-

cient defined in the upper half plane. Extend μ to the whole plane by
defining
$
&μpzq
’
’ if Impzq ą 0,
r pzq :“ cf μpzq if Impzq ă 0,
μ
’
’
%0 if Impzq “ 0.
Check explicitly that μr pzq “ μpcpzqq for z in the lower half plane.
Check moreover that cf μ r pzq for all z P C.
r pzq “ μ
1.2.4 (Reflection with respect to S1 ) Let τ pzq “ 1z and μ be a Beltrami
coefficient defined in the unit disc. Extend μ to the whole plane by
defining
$
&μpzq
’
’ if |z| ă 1,
r pzq :“ τ f μpzq if |z| ą 1,
μ
’
’
%0 if |z| “ 1.
4
r pzq “ μpτ pzqq |zz|4 for all z P CzD. Check moreover
Check explicitly μ
that τ f μ r pzq for all z P C.
r pzq “ μ
1.2.5 (Pullback of symmetric coefficients under symmetric maps) Let
s denote either the reflection with respect to R (spzq “ z) or with
respect to S1 (spzq “ 1{z). Consider two open sets U, V Ď C satisfying
spU q “ U and spV q “ V . Let μ be a Beltrami coefficient defined on
U symmetric with respect to s, i.e. satisfying s f μ “ μ, and suppose
f P D0` pV , U q is a map which also satisfies f “ s ˝ f ˝ s. Show that
the Beltrami coefficient μr :“ f ˚ μ defined on V also satisfies s f μ r.
r“μ
1.2.6 (Pullback under symmetric maps) Let s and U be as in the exercise
above. Consider f P D0` pU, U q and let μ :“ μf “ f ˚ μ0 . Show that if
we define fr :“ s ˝ f ˝ s, then μfr “ s f μ.
1.3 Quasiconformal mappings

We are now ready to define quasiconformal mappings. There are several
equivalent definitions of K-quasiconformality for an orientation preserving
homeomorphism φ : U Ñ V between domains of C, where 1 ď K ă 8. As
we shall see, a 1-quasiconformal map is conformal. The value of K can be
considered as a measure for how near φ is to being conformal. Although a
quasiconformal mapping is always R-differentiable almost everywhere and
with bounded dilatation, the converse is not true, as illustrated below in the
classical counter example in Section 1.3.3.
We shall not try to give a self-contained treatment of the theory of quasi-
conformal mappings, but instead give an overview which should be sufficient
for the applications of quasiconformal mappings in holomorphic dynamics,
and for quasiconformal surgery in particular. We give precise references to the
literature, mainly to the classical textbooks by Ahlfors [Ah5], and by Lehto and
Virtanen [LV]. However, we also recommend the reader to consult the modern
treatments by Hubbard [Hu2], Lyubich [Ly] and Astala et al. [AIM].
The original idea of a quasiconformal mapping goes back to Grötzsch and
is formulated geometrically in terms of moduli of quadrilaterals Q in U and
φpQq in V , or alternatively, of moduli of annuli A in U and φpAq in V . The
analytic definitions are formulated in terms of distributional derivatives of φ,
or alternatively, of ordinary partial derivatives of φ almost everywhere for a
mapping which is absolutely continuous on lines, abbreviated ACL. It is not at
all obvious why the different definitions are equivalent, but we shall state how
the equivalences are proven.
1.3.1 First analytic definition of quasiconformal mappings

We start with the most modern (not necessarily the most intuitive) approach to
defining quasiconformal mappings, using distributional derivatives. We give a
brief introduction to this concept (see e.g. [Co, Sect. 18.4]).
A test function is a C 8 function with compact support in U , and the space
of test functions is denoted by Cc8 pU q. With the appropriate topology, this is a
topological vector space, and distributions are precisely the continuous linear
functionals of this space, i.e. L : Cc8 pU q Ñ C.
As an example, if φ P L1loc pU q (i.e. is locally integrable like, for example,
any continuous function), then φ defines a distribution
ż
Lφ phq :“ φ ¨ h dm,
U
where m denotes the Lebesgue measure. These are actually the distributions
we will be interested in although not all distributions are of this form. Often
the same symbol is used for the function in L1loc pU q and the distribution it
defines.
Observe that if h is a test function then Bz h and Bz h are also test functions.
It follows that if L is a distribution then h ÞÑ LpBz hq and h ÞÑ LpBz hq are also
distributions.
Definition 1.4 (Distributional derivatives) If L is a distribution we define its

distributional derivatives as the distributions
pBLqphq :“ ´LpBz hq and pBLqphq :“ ´LpBz hq.
The minus signs in this definition are necessary so that, if φ is a C 1 function

then
BLφ “ LBz φ and BLφ “ LBz φ .
But be aware that if φ P L1loc pU q is not C 1 then the distributional derivatives

do not need to be defined by functions, although they are still distributions.
However, if they happen to be, that is, if there exist φ1 , φ2 P L1loc pU q such that
BLφ “ Lφ1 and BLφ “ Lφ2 ,
then φ1 and φ2 are denoted Bφ and Bφ and called B and B derivatives of φ

in the sense of distributions, or the distributional derivatives of φ in L1loc pU q.
Precisely, φ1 and φ2 must satisfy
ż ż ż ż
φ1 ¨ h dm “ ´ φ ¨ Bz h dm and φ2 ¨ h dm “ ´ φ ¨ Bz h dm
(1.12)
p
for any test function h P Cc8 . More generally, for p ě 1, if φ1 , φ2 P Lloc pU q
p
exist we say that φ has distributional derivatives in Lloc pU q. The space of func-
p p
tions in Lloc pU q (or distributions) having distributional derivatives in Lloc pU q
1,p
is called the Sobolev space Wloc pU q.
p
Observe that Lloc pU q Ă L1loc pU q for p ą 1 (see Exercise 1.3.1).
If φ P C , its distributional derivatives are precisely its derivatives in the
1
ordinary sense.
We are ready for the first definition of a quasiconformal mapping.
Definition 1.5 (First analytic definition of K-quasiconformal mapping)

Let U and V be domains in C, and let K ě 1 be given. Set k :“ pK ´ 1q{
pK ` 1q. Then φ : U Ñ V is K-quasiconformal if and only if:
(i) φ is a homeomorphism;
(ii) the partial derivatives Bφ and Bφ exist in the sense of distributions and
belong to L2loc (i.e. are locally square integrable);
(iii) and satisfy |Bφ| ď k|Bφ| in L2loc .
Remarks 1.6
(a) Observe, that if φ is a C 1 homeomorphism then φ is quasiconformal if it

satisfies |Bz φ| ď k|Bz φ| for some k ă 1.
(b) If φ is a diffeomorphism between compact sets then φ is quasiconformal.
To obtain the bound K on the dilatation we simply use the continuity of
the derivatives and the fact that a continuous function attains its maximal
value on a compact set.
1.3.2 Second analytic definition of quasiconformal mappings

A quasiconformal mapping can instead of partial derivatives in the sense of
distribution be characterized in terms of absolute continuity on lines. This is a
condition that insures the existence of partial derivatives in the ordinary sense
almost everywhere. We shall therefore turn to this notion, since it connects to
our previous chapter.
Definition 1.7 (Absolute continuity on an interval) A continuous complex

valued function f defined on an interval I Ă R is said to be absolutely
continuous on I if it satisfies the following: for every ą 0 there exists a
ř
δ ą 0 such that j |f pbj q ´ f paj q| ă for every finite sequence of non-
intersecting intervals paj , bj q whose closure are contained in I and have a
ř
total length i |bj ´ aj | ă δ.
It follows from f being absolutely continuous on I that f is of local

bounded variation, hence that f has a derivative f 1 pxq in the ordinary sense
for almost all x P I . See [LV, Sect. 2.7, Chapt. III].
The above notion of absolute continuity on an interval generalizes to a
complex valued function on a domain U as follows.
Definition 1.8 (ACL, absolute continuity on lines) A continuous function

f : U Ñ C is said to be absolutely continuous on lines if for any family of
parallel lines in any disc D compactly contained in U (that is, D Ă U ), f is
absolutely continuous on almost all of them.
From the interval-discussion above it follows that if f is absolutely con-

tinuous on lines, then f has partial derivatives in the ordinary sense almost
everywhere in U . The existence of partial derivatives at a point does not
necessarily imply differentiability of f at the point. However, the follow-
ing result by Gehring and Lehto is true (see e.g. [Ah5, pp. 17–18] and in
[LV, p. 128]).
Theorem 1.9 (Differentiability almost everywhere) Let f : U Ñ V be a

continuous open mapping. If f has partial derivatives fx and fy in the ordi-
nary sense almost everywhere then f is R-differentiable almost everywhere.
Note that for the partial derivatives fx and fy to exist almost everywhere it
is sufficient for f to be absolutely continuous along almost every horizontal
and almost every vertical line in every rectangle compactly contained in U .
Therefore Bz f and Bz f are defined almost everywhere. Since the integral of
the Jacobian over a set is bounded above by the area of the image of this set
(see [LV, Lem. III.3.3]), it follows that the Jacobian Jac f “ |Bz f |2 ´ |Bz f |2
is in L1loc .
But to really relate the ACL property to the definition of quasiconformality
we need to understand the relation of the ordinary derivatives and the dis-
tributional ones. It turns out that these two coincide when they are locally
integrable. More precisely (see [Ah5, p. 19] and [AIM, Lem. A.5.2]):
Lemma 1.10 (ACL characterization of Sobolev spaces) A continuous func-
1,1
tion f : U Ñ C is in Wloc (i.e. has distributional derivatives in L1loc ) if and
only if f is ACL and its ordinary partial derivatives are in L1loc . When this is
the case, the distributional and the ordinary derivatives coincide.
We are ready to formulate the second analytic definition.
Definition 1.11 (Second analytic definition of K-quasiconformal mapping)
A mapping φ : U Ñ V is K-quasiconformal if and only if:
(1) φ is a homeomorphism;
(2) φ is ACL;
(3) |Bz φ| ď k|Bz φ| almost everywhere where k :“ K ´1
K `1 .
To conclude that the two definitions are equivalent we must add something
to the statements above. Namely, that since φ is ACL, the Jacobian is in L1loc
and (3) is fulfilled, it follows that
Jac φ
|Bφ|2 ď |Bφ|2 ď ,
1 ´ k2
and therefore the partial derivatives are locally square integrable.
With this definition we can now relate to our previous section on almost
complex structures and pullbacks by a quasiconformal map φ. Since we have
that φ is R-differentiable almost everywhere, and as we shall see in Theo-
rem 1.15 with a non-singular differential defined almost everywhere, condition
(3) in the second analytic definition requires that the Beltrami coefficient μφ “
Bz φ
Bz φ has modulus bounded away from 1 almost everywhere, i.e. }μφ }8 ă 1.
Equivalently, in the language of pullbacks, we are asking that the almost

complex structure φ ˚ σ0 , induced by φ as the pullback of the standard one,
have bounded dilatation, i.e. Kφ “ K ă 8. The larger K is, the further the
ellipses are from being circles, that is the further the function φ is from being
conformal.
However, be aware that this condition alone is not enough. Even if φ ˚ σ
has bounded dilatation a.e., ACL is needed to ensure quasiconformality. We
will see an example of this fact once we have a geometric definition of
quasiconformal mappings. Nevertheless, we will introduce the example at this
point to see a mapping that is not ACL.
1.3.3 A devil’s staircase example

The following classical example defines a homeomorphism F : CR Ñ CR ,
which is R-differentiable almost everywhere and with a non-singular R-
differential almost everywhere. However, it is not absolutely continuous on
horizontal lines.
Let Ă r0, 1s denote the middle third Cantor set. The Cantor function f :
r0, 1s Ñ r0, 1s, also called a devil staircase function, is defined below in two
alternative ways. Its graph is shown in Figure 1.8.
1
7
8
3
4
5
8
1
2
3
8
1
4
1
8
1 2 1 2 7 8
9 9 3 3 9 9
1
Figure 1.8 The graph of the Cantor function, a devil’s staircase.
Recall that the Cantor set consists of the points x P r0, 1s which in base 3
have a representation of the form
x1 x2 xn
x “ .x1 x2 ¨ ¨ ¨ xn ¨ ¨ ¨ “ ` 2 ` ¨¨¨ ` n ` ¨¨¨ with xn P t0, 2u.
3 3 3
Define f as the unique non-decreasing function, which on x P as above

takes the value with the following representation in base 2
x1 x2 xn
f pxq “ .x 1 x 2 ¨ ¨ ¨ x n ¨ ¨ ¨ “ ` 2 ` ¨¨¨ ` n ` ¨¨¨
2 2 2
where x n equals 0 if xn “ 0 and 1 if xn “ 2. The Cantor set has measure zero

and f pq “ r0, 1s.
Alternatively, f is defined as the limiting function f “ lim fn of the
uniformly convergent sequence of continuous functions fn : r0, 1s Ñ r0, 1s,
defined inductively and starting with f0 pxq “ x, and for n ě 0:
$
1
& 2 fn p3xq
’
’ if 0 ď x ď 13 ,
fn`1 pxq :“ 2
1
if 1
3 ď x ď 23 ,
’
% 1 ` 1 f p3x ´ 2q
’
if 2
ď x ď 1.
2 2 n 3
The function f is equal to the constant 21n on any gap of the Cantor set of
the form r 3jn , j3`n1 s with j appropriate. Hence the derivative is f 1 pxq “ 0 on
r0, 1sz, i.e. almost everywhere. However, f is not absolutely continuous on
r0, 1s, since the Cantor set can be covered by a finite set of disjoint open
intervals of arbitrarily small total length, but so that the corresponding sum
ř 1
j |f pbj q ´ f paj q| is bigger than, say 2 .
Now extend the f function to f : R Ñ R by f pxq “ 0 for x ď 0 and
f pxq “ 1 for x ě 1. Let F : CR Ñ CR be defined as F px ` i yq “ x `
i py ` f pxqq for x ` i y P CR (see Figure 1.9). Then F is an orientation
preserving homeomorphism, with Bz F px ` i yq “ 1 and Bz F px ` i yq “ 0
for x ` i y P CR zp ` i Rq, i.e. almost everywhere. It follows from the
one-dimensional example that F is not absolutely continuous on horizontal
lines, and therefore F is not ACL.
1.3.4 Geometric definitions of quasiconformal mappings

Classically quasiconformal mappings are defined geometrically, without refer-
ence to distributions.
A quadrilateral Q “ Qpz1 , z2 , z3 , z4 q is a Jordan domain in C with an
ordered sequence of boundary points pz1 , z2 , z3 , z4 q called the vertices of Q,
with their order agreeing with the positive orientation of Q. Any quadrilateral
Q can be mapped conformally onto a rectangle in such a way that vertices
map to vertices. This rectangle is unique up to similarity (see for instance
Figure 1.9 The image of the unit square under F , a two-dimensional

devil’s staircase function.
[LV, p. 15]). If we denote by ϕ such a conformal map, then the conformal

modulus of Q is defined as
|ϕpz1 q ´ ϕpz2 q|
mod Qpz1 , z2 , z3 , z4 q :“ .
|ϕpz2 q ´ ϕpz3 q|
See Figure 1.10. Two quadrilaterals Q and Q1 are conformally equivalent if
and only if they have the same modulus. By rearranging the vertices we obtain
1
mod Qpz2 , z3 , z4 , z1 q “ . (1.13)
mod Qpz1 , z2 , z3 , z4 q
Let U, V be domains in C and φ : U Ñ V be an orientation preserving
homeomorphism. Then for any quadrilateral Q, compactly contained in U , its
mod φ pQq
image φpQq is a quadrilateral, compactly contained in V . The ratio mod Q
is called the dilatation of Q under φ. The maximal dilation of φ is defined as
mod φpQq
Kφ :“ sup .
QĂU mod Q
z1 z4 ϕ pz4 q ϕ pz3 q
ϕ
b
z2
z3 ϕ pz1 q a ϕ pz2 q
Figure 1.10 The modulus of the quadrilateral Qpz1 , z2 , z3 , z4 q is a{b.

Using equation (1.13) we see that for every dilatation, we also have its inverse,
just by considering the same quadrilateral with the vertices rearranged. Thus
we conclude that Kφ ě 1.
Definition 1.12 (First geometric definition of K-quasiconformal mapping)
Let U and V be domains in C, and let K ě 1 be given. Then φ : U Ñ V is K-
quasiconformal if and only if φ is an orientation preserving homeomorphism
satisfying
1
mod Q ď mod φpQq ď K mod Q,
K
for all quadrilaterals Q that are compactly contained in U ; or equivalently
satisfying
Kφ ď K.
A proof of the equivalence of the first geometric definition (Definition 1.12)

of K-quasiconformality and the second analytic definition (Definition 1.11)
can be found in [Ah5, pp. 20–21].
It is a fact proven by Grötzsch ([Ah5, p. 8]) that the affine map from a
rectangle Q1 to a rectangle Q2 has the least maximal dilatation (see also
Exercise 1.3.5).
The second geometric definition of K-quasiconformality refers to annuli
instead of quadrilaterals. Let A denote an open annulus in C, i.e. a doubly
connected domain in C. p Then A can be mapped conformally onto a standard
annulus Ar,R :“ tz P C | 0 ď r ă |z| ă R ď 8u, which is unique up to mul-
tiplication by a real constant. If we denote by ϕ such a conformal map, then
the conformal modulus of A is defined as the modulus of Ar,R , i.e.
#
1
log Rr if r ą 0 and R ă 8,
mod A :“ mod Ar,R :“ 2π
8 if r “ 0 or R “ 8.
Observe that the two concepts are related in the following way. For a quadri-
lateral of finite modulus M choose the conformally equivalent rectangle so
that the ordered vertices are p0, 2π M, 2π pM ` i q, 2π i q. The rectangle is
mapped by the exponential z ÞÑ ez onto the standard annulus A1,R , where
log R “ 2π M, identifying points on the horizontal edges of the rectangle.
See Figure 1.11, which also shows the straight cylinder obtained from this
identification. Note that the modulus of a straight cylinder is the height of the
cylinder divided by the circumference of the circle.
Definition 1.13 (Second geometric definition of K-quasiconformal map-
ping) Let U and V be domains in C, and let K ě 1 be given. Then φ : U Ñ V
2π i
z ÞÑ ez
2π »
1 R
log R
0 log R
Figure 1.11 The exponential maps a rectangle of the given form onto a standard
annulus, and their moduli are the same.
is K-quasiconformal if and only if φ is an orientation preserving homeomor-

phism satisfying
1
mod A ď mod φpAq ď K mod A (1.14)
K
for all annuli A that are compactly contained in U .
1.3.5 Revisiting the devil’s staircase example

Consider again the devil’s staircase homeomorphism F : CR Ñ CR . We have
seen that F is not ACL although it is differentiable and with Bz F “ 0 almost
everywhere! This means that the pullback of the standard complex structure by
F is the standard complex structure almost everywhere. One would be tempted
to say that this is a condition for conformality but, using the second geometric
definition of quasiconformality, we show why F is not even quasiconformal.
For each n ě 0 we form the two double-infinite cylinders
Cn “ pr0, 1{3n s ` i Rq {pi y „ 1{3n ` i yq,
Cn1 “ pr0, 1{3n s ` i Rq {pi y „ 1{3n ` i py ` 1{2n qq,
and let Fn : Cn Ñ Cn1 denote the homeomorphism induced by F . Let An Ă Cn

denote the annulus
´ ¯n
An “ pr0, 1{3n s ` i r0, 1{2n sq {pi y „ 1{3n ` i yq with mod An “ 32 .
Note that the annulus Fn pAn q is contained in the round annulus Bn Ă Cn1 (see
Figure 1.12) bounded by the two line segments of slope p 32 qn through the points
Ln p2{3n`1 ` i {2n`1 q and Un p1{3n`1 ` i 3{2n`1 q respectively.
1
2n´1
Bn
3
Un
2n`1
1
3¨2n`1
1
Fn pAn q
2n
1
2n`1
Ln
1 2 1
3n`1 3n`1 3n
´1
3¨2n`1
Figure 1.12 The parallelogram inducing the round cylinder Bn in Cn1 .
If F were K-quasiconformal, so would Fn be for any n ě 0. This would

imply
1
mod An ď mod Fn pAn q ď mod Bn ,
K
where the left inequality corresponds to the left inequality of (1.14) and the
right inequality follows from the Grötzsch inequality (see e.g. [Mi1, Cor. B.6]),
using that Fn pAn q is embedded in Bn and of the same homotopy type. Hence
K would satisfy
mod An
Kě for any n ě 0.
mod Bn
The modulus of the round cylinder Bn is the height divided by the circumfer-
ence of the circle, hence
p 12 q2n ´ p 13 q2n`1 1 ´ 13 p 23 q2n

mod Bn “ “ .
p 12 q2n ` p 13 q2n 1 ` p 23 q2n
Since mod Bn tends to 1 and mod An tends to 8 when n tends to 8, there is

no such K.
1.3.6 Properties of quasiconformal mappings

Let φ : U Ñ V be an orientation preserving homeomorphism. We have given
four definitions of quasiconformality. However, in practice it is not easy to
check any of them. In this section we state some useful properties of quasi-
conformal mappings. In many instances they help deciding on the quasicon-
formality of a mapping. In general we refer to [Ah5] for the proofs.
We start with some properties which follow easily from the geometric
definitions:
P1 If φ is K-quasiconformal, then φ ´1 is K-quasiconformal.

P2 If φ is K-quasiconformal, then any composition on the left or right with a
conformal mapping is K-quasiconformal.
P3 The composition of a K1 -quasiconformal mapping and a K2 -
quasiconformal mapping is K1 K2 -quasiconformal.
The next statement comes directly from the first analytic definition, although
it is less obvious when we think about it in geometric terms:
P4 The homeomorphism φ is K-quasiconformal if and only if φ is locally

K-quasiconformal.
The next property follows also from the analytic definitions:
P5 If φ is K-quasiconformal and of class C 1 then the dilatation of the infinites-

imal ellipse Ez in Tz U is bounded by K for all z P U .
Indeed, if the R-differential changes continuously, then so does the dilatation

of Ez . If the dilatation would be larger than K at some point, this would be
true in a neighbourhood of the point, i.e. on a set of positive measure.
P6 If φ is K-quasiconformal, then φ satisfies a uniform Hölder condition
|φpz1 q ´ φpz2 q| ď M|z1 ´ z2 |1{K
on every compact subset of U .

P7 If φ is bilipschitz, i.e. if there exist L ą 0 such that
L´1 |z1 ´ z2 | ď |φpz1 q ´ φpz2 q| ď L|z1 ´ z2 |,
then φ is quasiconformal. The converse is not true. (See also [Roh].)

The following is a very useful result in surgery applications (see [Ah5,

p. 16]).
Theorem 1.14 (Weyl’s Lemma) If φ is 1-quasiconformal, then φ is confor-

mal. In other words, if φ is quasiconformal and Bz φ “ 0 almost everywhere,
then φ is conformal.
The first statement is proved by using the geometric definition of qua-

siconformality. The second statement, which is generally known as Weyl’s
Lemma, follows from the equivalence of the definitions. In fact the condition
of quasiconformality could be substituted by the mapping being an ACL
homeomorphism.
Notice that the devil’s staircase example illustrates that the quasiconformal-
ity (or at least ACL) is a necessary condition in this statement. Indeed, the
mapping F in the example satisfies Bz F “ 0 almost everywhere, but F is not
conformal.
Another important property is that quasiconformal mappings are absolutely
continuous with respect to Lebesgue measure, see [Ah5, p. 22]. More precisely,
Theorem 1.15 (Quasiconformal maps and measure) If φ is quasiconformal,

then it maps sets of measure zero to sets of measure zero. Moreover, for every
measurable set E
ż
mpφpEqq “ Jac φ dm.
E
Corollary 1.16 If φ is quasiconformal, then Bz φ ‰ 0 almost everywhere.

Moreover, Jac φpzq ą 0 almost everywhere.
If φ : U Ñ V is K-quasiconformal then Jac φ ě p1 ´ k 2 q|Bz φ|2 , where k “

pK ´ 1q{pK ` 1q. It follows that Jac φ ą 0 almost everywhere and that φ is
orientation preserving.
In the analytic definitions of quasiconformality we did not require the home-
omorphism φ to be orientation preserving. It is a consequence. In the geomet-
rical definition we needed to add orientation preserving as a requirement. So
only now we have the equivalence of the analytic and geometric definitions of
quasiconformality.
Remark 1.17 Given a Beltrami coefficient μ defined a.e. in a set U Ă C,
the pullback φ ˚ pμq under a quasiconformal map φ : U 1 Ñ U is well defined
a.e. in U 1 . This is due to Theorem 1.15 and its Corollary. Indeed, if μ is defined
on U zX with mpXq “ 0, then φ ˚ pμq is well defined in U 1 zpφ ´1 pXq Y Y q,
where Y is the set where φ is not differentiable, hence a set of measure zero.
Since φ ´1 pXq also has measure zero, it follows that φ ˚ μ is defined a.e.
Finally, the following area distortion estimate due to Astala will be useful
later on, see [As].
Theorem 1.18 (Quasiconformal maps and area) Let φ : D Ñ D be a K-

quasiconformal mapping such that φp0q “ 0 and E any Borel measurable set.
Then,
AreapφpEqq ď MpAreapEqq1{K ,
where M “ MpKq “ 1 ´ OpK ´ 1q.
Some facts about quasiconformal mappings are useful, for instance as men-
tioned above that a C 1 -diffeomorphism between compact sets is quasiconfor-
mal (see Remark 1.6). If f : U Ñ V is a C 1 -diffeomorphism between subsets
of C that are not assumed to be compact, there are criterions for quasiconfor-
mality formulated in Exercises 1.3.3 and 1.3.4. Another useful, although not
so simple, fact is that certain sets are removable in terms of quasiconformality.
More precisely, a closed set X is quasiconformally removable if any homeo-
morphism which is defined on a neighbourhood U of X and is quasiconformal
on U zX, is quasiconformal on U . It is a fact that quasiconformally removable
sets have measure zero, and hence the dilatation of f in U is the same as in
U zX.
Notice that sets of measure zero do not affect the dilatation condition in
the definitions of quasiconformality. But being ACL in U zX does not imply
necessarily that the map is ACL in U , even if X has zero measure. This is the
main difficulty when showing that a certain set is quasiconformally removable.
We shall need the following removability result, see [Ly] or [Hu2, Prop. 4.2.7
and 4.9.9].
Theorem 1.19 (Quasiconformal removability of quasiarcs) If is a

quasiarc (the image of a straight line under a quasiconformal mapping) and
φ : U Ñ V a homeomorphism that is K-quasiconformal on U z, then φ
is K-quasiconformal on U , and hence is quasiconformally removable. In
particular, points, lines and smooth arcs are quasiconformally removable.
As a consequence, we may construct global quasiconformal mappings by

gluing together different quasiconformal mappings along a finite number of
boundary pieces, as long as they are sufficiently nice. This is a common
technique in holomorphic dynamics as we shall see repeatedly.
Another result of this type is Rickman’s Lemma, also known as Bers’s
Sewing Lemma, see [DH1].
Lemma 1.20 (Rickman’s Lemma) Let U Ă C be open, C Ă U compact, φ

and two mappings U Ñ C which are homeomorphisms onto their images.
Suppose that φ is quasiconformal, that is quasiconformal on U zC, and that
φ “ on C. Then is quasiconformal and Bzφ “ Bz almost everywhere
on C.
The following property deals with families of quasiconformal mappings, see

[Hu2, p. 131].
Theorem 1.21 (Compactness) The set of K-quasiconformal homeomor-

phisms φ : D Ñ D with φp0q “ 0 is compact in the topology of uniform con-
vergence on compact subsets.
1.3.7 Mappings and Beltrami forms on Riemann surfaces

Up to now we have defined quasiconformal mappings, Beltrami coefficients
and pullbacks of Beltrami coefficients on subsets of C. However, we shall
need the notions also on the Riemann sphere C. p Therefore, we briefly discuss
how the different terms are extended to Riemann surfaces. They are defined
locally in charts, where they agree with the previous definitions. The extra
work involved consists in checking that the definitions are independent of
the charts in which they are expressed. The checking is easy, we just have
to rephrase facts we already know.
First some notation: recall that a Riemann surface S is a one-dimensional
complex manifold, i.e. a two-dimensional real manifold endowed with a com-
plex analytic structure. Such a structure is given by a collection of charts
ϕ : US Ñ U which forms an atlas. For each such chart, ϕ is a homeomorphism
from an open subset US of S onto an open subset U of C. The union of all the
domains US equals S, and for any pair of overlapping charts the transition map
is conformal, i.e. if ϕr : U r is a chart such that US X U
rS Ñ U rS ‰ H then
ϕr ˝ ϕ ´1 : ϕpUS X U
rS q Ñ ϕrpUS X U
rS q
is conformal.
We remark that a topological surface S can be endowed with a quasi-
conformal (resp. smooth) structure, which means with a collection of charts
ϕ : US Ñ U Ă C which forms an atlas with quasiconformal (resp. smooth)
transition maps.
When dealing with maps between Riemann surfaces, its regularity is always
defined in terms of the regularity of the maps between the open sets of
C induced by the charts. In this sense, a conformal isomorphism between
Riemann surfaces is an isomorphism which is locally conformal (between open

sets of C) when expressed in charts.
A remarkable theorem by Koebe and Poincaré states that there are only three
possible simply connected Riemann surfaces, up to conformal isomorphisms.
For a proof, see for instance [Ah3, Chapt. 10], [FK, IV.4.] or [Hu2, Chapt. 1].
Theorem 1.22 (Uniformization Theorem) Let S be a simply connected

Riemann surface. Then S is conformally equivalent to D, C or C.
p
We now turn to analyze quasiconformal mappings between Riemann sur-

faces.
Definition 1.23 (K-quasiconformal mapping between Riemann surfaces)

Let S and S 1 be two Riemann surfaces, and let φ : S Ñ S 1 be a homeomor-
phism. Then φ is quasiconformal if and only if there exists a K ě 1 so that φ
is locally K-quasiconformal when expressed in charts.
For an arbitrary point s P S choose an arbitrary chart ϕ : US Ñ U defined on

a neighbourhood US of s and choose an arbitrary chart ϕ 1 : US 1 Ñ U 1 defined
on a neighbourhood US 1 of φpsq. We may assume that φpUS q “ US 1 . By defi-
nition φ|US : US Ñ US 1 is K-quasiconformal if ϕ 1 ˝ φ ˝ ϕ ´1 : U Ñ U 1 is K-
quasiconformal. The notion is well defined since it is independent of the charts
in which we locally check K-quasiconformality. This follows from property P2
in Section 1.3.6: the composition of a K-quasiconformal mapping with con-
formal mappings left and right is again K-quasiconformal. Changing charts in
the domain and the range of φ corresponds exactly to composing ϕ 1 ˝ φ ˝ ϕ ´1
left and right with conformal mappings, i.e.
ϕ 1 ˝ ϕ 1´1 q ˝ pϕ 1 ˝ φ ˝ ϕ ´1 q ˝ pr
pr ϕ ˝ ϕ ´1 q´1 “ ϕr1 ˝ φ ˝ ϕr´1 ,
where ϕr and ϕr1 are charts around s and φpsq respectively (see Figure 1.13).
Next we see how Beltrami coefficients on subsets of C are generalized to
Beltrami forms on Riemann surfaces.
Definition 1.24 (Beltrami form on a Riemann surface) A Beltrami form

μ, also called a Beltrami differential, on a Riemann surface S is a p´1, 1q-
differential on S.
Locally a p´1, 1q-differential is expressed as μpzqdz{dz “ μpzqdz´1 dz.

Formally, this means that if ϕ and ϕr are two overlapping charts defined on
Figure 1.13 Sketch of a K-quasiconformal mapping φ between Riemann surfaces.

The composition ϕ 1 ˝ φ ˝ ϕ ´1 : U Ñ U 1 is K-quasiconformal, while the transition
´1
mappings ϕ ˝ ϕr´1 and ϕ 1 ˝ ϕr1 between the coloured areas are conformal.
S with the conformal transition map h “ ϕr ˝ ϕ ´1 then the Beltrami form

satisfies the following transformation rule:
h1 pzq
zq 1 ,
μϕ pzq “ μϕrpr (1.15)
h pzq
where z “ ϕpsq and r z “ ϕrpsq. In other words, μϕ pzq and μϕrprzq are the Bel-
trami coefficients of the Beltrami form when expressed in the z-coordinate in
U , respectively the r
z-coordinate in Ur , in the two overlapping charts. Note that
1
since h is holomorphic, h pzq “ Bz hpzq, and discover the similarity between
(1.15) and the pullback formula (1.11) of a Beltrami coefficient by a holomor-
phic map. Moreover, note that |μϕrpr zq| “ |μϕ pzq| so that }μ}8 is well defined.
To justify the transformation rule we consider the geometric information
given by a Beltrami form on S with }μ}8 ă 1. The Beltrami form defines
in this case an almost complex structure σ on S, i.e. a measurable field of
infinitesimal ellipses on the tangent bundle T S. The Beltrami coefficient μϕ pzq
1`|μ pzq|
defines a.e. an infinitesimal ellipse at Tz U with dilatation equal to 1´|μϕ pzq| .
ϕ
It follows from (1.15) that the infinitesimal ellipse is mapped by the transition
map Dz h : Tz U Ñ Trz U r of the charts on the tangent bundles to the infinitesimal
ellipse defined by the Beltrami coefficient μϕrpr zq at Trz U
r . More precisely, let E
be an ellipse representing the infinitesimal ellipse in Tz U , which is well defined
up to scaling. Then E is mapped by Dz h onto the ellipse E r in Trz U
r , which is
1 1
E scaled by |h pzq| and rotated by the argument of h pzq, see Figure 1.14. The
two ellipses have the same dilatation:
Figure 1.14 The ellipse E in Tz U is mapped by Dz h : w ÞÑ h1 pzq ¨ w onto

the ellipse E z . Hence the ellipses are equal up to multiplication by
r in T U
r
r
a complex number.
KpEq “ KpEq,
r (1.16)
and twice the argument from the real axis to the minor axis of E
r is exactly
equal to the argument of μϕrpr
zq:
arg μϕ pzq ` 2 arg h1 pzq “ arg μϕrpr
zq. (1.17)
Therefore, the ellipse E r represents the infinitesimal ellipse defined by the

Beltrami coefficient μϕrprzq. It follows that the measurable field of infinitesimal
ellipses in T S is well defined and, hence, that the Beltrami form with bounded
dilatation determines an almost complex structure on S.
The last thing we define is the pullback of a Beltrami form by a quasicon-
formal mapping.
Definition 1.25 (Pullback of Beltrami forms) Let S and S 1 be two Riemann
surfaces, and let φ : S Ñ S 1 be a quasiconformal mapping. If μ1 is a Beltrami
form on S 1 then φ ˚ μ1 is defined as the Beltrami form on S, which when
expressed in charts fits with the previous pullback definition.
To make this more precise, let ϕ : US Ñ U and ϕ 1 : US 1 Ñ U 1 be arbi-
trary charts around s P S and φpsq P S 1 , and set z “ ϕpsq and z1 “ ϕ 1 pφpsqq.
Let μ1ϕ 1 pz1 q be the Beltrami coefficient of μ1 in the chart ϕ 1 . Then φ ˚ μ1 is
the Beltrami form on S, which in the chart ϕ has the following Beltrami
coefficient:
Bz f pzq ` μ1ϕ 1 pf pzqqBz f pzq

pφ ˚ μ1 qϕ pzq “ ,
Bz f pzq ` μ1ϕ 1 pf pzqqBz f pzq
where f pzq “ ϕ 1 ˝ φ ˝ ϕ ´1 pzq. Compare with (1.9). Also this definition is

independent of charts (see Figure 1.15).
S S1
Us Us1
φ
s
φ ps q
ϕ
ϕ1
U U1
f “ ϕ 1 ˝ φ ˝ ϕ ´1
z z1
Dz f
Figure 1.15 The ellipse in Tz U is mapped by Dz f onto the ellipse in Tz1 U 1 . These
ellipses are in general different since Dz f is just a linear map of CR .
Recall that the universal covering space of the Riemann sphere minus three
points is the unit disc. From this fact, the analogue to Theorem 1.21 follows.
Theorem 1.26 (Compactness) The set of K-quasiconformal homeomor-

phisms φ : C
p ÑC p fixing three points is compact in the topology of uniform
convergence on compact subsets.

p
1.3.1 Prove that Lloc pU q Ă L1loc pU q.
Hint: Apply for q satisfying 1{p ` 1{q “ 1 the Hölder inequality
ż ˆż ˙1{p ˆż ˙1{q
|f ¨ g| dm ď |f |p dm |g|q dm ,
U U U
p q
where f P Lloc pU q and g P Lloc pU q.
1.3.2 Let a “ t0 ă t1 ă ¨ ¨ ¨ ă tn “ b be a finite partition of the interval ra, bs.
Prove that if f : ra, bs Ñ R is absolutely continuous in every interval
rtj , tj `1 s for 0 ă j ă n ´ 1, then f is absolutely continuous in ra, bs.
1.3.3 Let f : U Ñ V be an orientation preserving C 1 -diffeomorphism
between subsets of C, which are not assumed to be compact. For each
u P U consider the linear map Dz f puq : Tu U Ñ Tf puq V . In the follow-

ing we formulate a criterion, which guaranties quasiconformality of f .
We shall use the result from Exercise 1.1.2 in Section 1.1.
Assume that the norm of Dz f puq is uniformly bounded and that the
determinant of Dz f puq is uniformly bounded away from zero and infin-
ity. Then show that f is quasiconformal by showing that the dilatation
KpDz f puqq is uniformly bounded.
1.3.4 Let f : U Ñ V be given as in Exercise 1.3.3 above. The inequality
below gives another criterion that guaranties quasiconformality of f .
We shall use notation similar to that introduced in Exercise 1.1.3 in
Section 1.1. For each u P U let the linear map Lu “ Dz f puq : Tu U Ñ
Tf puq V be represented by the matrix, which in the standard basis in Tu U
and in Tf puq V is of the form
„ j
αpuq βpuq
Du f “ .
γ puq δpuq
Let λj puq, j “ 1, 2, denote the eigenvalues of the symmetric matrix

Qu “ pLu qT pLu q, labelled so that λ1 pzq ě λ2 pzq ą 0.
Show that if the eigenvalues λj puq, j “ 1, 2, are uniformly bounded
away from zero and infinity then f is quasiconformal.
1.3.5 Let Q1 and Q2 be two rectangles of vertical sides of length vj and hori-
zontal sides of length hj , for j “ 1, 2. Let A be the affine map taking Q1
modpQ2 q
to Q2 . Use the discussion in Section 1.1 to show that KpAq “ modpQ1 q
.
1.4 The Integrability Theorem

Let U Ă C be an open set which is either the whole plane or conformally
equivalent to the unit disc. In previous sections we have seen how a quasicon-
formal homeomorphism φ : U Ñ C gives rise to an almost complex structure
σφ on U of bounded dilatation, or equivalently, to a Beltrami coefficient
μφ “ Bz φ{Bz φ defined a.e. with }μφ }8 “ k ă 1.
The following natural question arises. Given an almost complex structure σ
in U , under which conditions can one find a quasiconformal homeomorphism
φ such that it induces σ almost everywhere? In the pullback language that
means φ ˚ σ0 “ σ a.e. Or equivalently, given a Beltrami coefficient μ on U
under which conditions can one find φ : U Ñ C quasiconformal (hence in
D0` ) such that
Bz φpzq “ μpzqBz φpzq (1.18)

for almost every z P U ? Such function φ is said to integrate μ and is therefore

called an integrating map. This equation is called the Beltrami equation and the
answer to the question is given by the remarkable Integrability Theorem (also
known as the Measurable Riemann Mapping Theorem or simply the Mapping
Theorem) due to Morrey [Mo], Bojarski [Bo1] and Ahlfors and Bers [AB]. The
local version reads as follows.
Theorem 1.27 (Integrability Theorem – local version) Let U Ă C be an

open set such that U » D (respectively U “ C). Let σ be an almost com-
plex structure on U corresponding to the Beltrami coefficient μ. Suppose the
dilatation of σ is uniformly bounded, that is, Kpσ q ă 8 or, equivalently, the
essential supremum of |μ| on U is
}μ}8 “ k ă 1.
Then μ is integrable, i.e. there exists a quasiconformal homeomorphism φ :

U Ñ D (respectively onto C) which solves the Beltrami equation, i.e. such
that
Bz φpzq
μpzq “
Bz φpzq
for a.e. z P U . Moreover, φ is unique up to post-composition with automor-
phisms of D (respectively C).
In the pullback language we would say that there exists a quasiconformal

homeomorphism φ such that μ “ φ ˚ μ0 a.e.
The proof of this theorem is involved and can be found in many classical
texts as well as in some more recent ones. See for example [LV, Ah5] or [AIM,
Bo2, DB, Hu2, Ly], and Section 1.4.2 for a sketch of the proof.
As stated above, the Integrability Theorem applies to Beltrami coefficients in
subsets of the complex plane. However, we often deal with rational mappings
on the Riemann sphere and are interested in a global change to an almost com-
plex structure on C.
p Moreover, we build abstract Riemann surfaces isomorphic
either to the disc, to the complex plane or to the Riemann sphere. Therefore,
we are also interested in a global version of the Integrability Theorem.
Theorem 1.28 (Integrability Theorem – global version) Let S be a simply

connected Riemann surface and σ be an almost complex structure on S with
measurable Beltrami form μ. Suppose the dilatation of σ is uniformly bounded,
i.e. Kpσ q ă 8 or, equivalently, the essential supremum of |μ| on S is
}μ}8 “ k ă 1.
Then μ is integrable, i.e. there exists a quasiconformal homeomorphism φ :

S Ñ D (respectively onto C or C)p which satisfies
φ ˚ μ0 “ μ.
If S is isomorphic to D (respectively to C or C)
p then φ : S Ñ D (respectively
onto C or C) is unique up to post-composition with automorphisms of D
p
(respectively of C or C).
p
In this case we also say that φ solves the Beltrami equation and write μ “
Bz φ{Bz φ a.e. (where the equation should be expressed in charts if necessary).
The global theorem easily follows from the local one as explained below.
Proof The case of the disc or the complex plane follows almost immediately
from the local theorem. Indeed, let X be either D or C and let : S Ñ X be a
conformal isomorphism. Then for any chart ϕ : US Ñ U in a given atlas for S
the map ˝ ϕ ´1 : U Ñ X is holomorphic. Using these conformal maps, we
can push forward the Beltrami form μ on S to a Beltrami form μ1 on X. Since
the Beltrami coefficients in overlapping charts on S transform as explained
in (1.15), the Beltrami form μ1 is well defined on X. It satisfies }μ1 }8 “
}μ}8 :“ k ă 1. The Local Integrability Theorem provides a quasiconformal
homeomorphism φ : X Ñ X such that φ ˚ μ0 “ μ1 . Then φ ˝ : S Ñ X is the
required integrating map.
Now suppose S is isomorphic to C p and μ is the Beltrami form on S,
j
satisfying }μ}8 :“ k ă 1. Choose a finite atlas tϕj : US Ñ U j u for S so that
U j Ă C is conformally isomorphic to D. By definition, μ induces Beltrami
coefficients μj on the sets U j , compatible under the transition maps between
overlapping charts. We apply the Local Integrability Theorem to each of
these Beltrami coefficients and obtain Kj -quasiconformal homeomorphisms
φj : U j Ñ D, such that φj˚ μ0 “ μj (see Figure 1.16).
It follows from Weyl’s Lemma that, wherever well defined, the maps φj ˝
ϕj ˝ ϕi´1 ˝ φi´1 are conformal, since they are quasiconformal and preserve the
j
standard complex structure. Thus the maps tφj ˝ ϕj : US Ñ Du form a new
atlas for S with conformal transition maps.
By the Uniformization Theorem there exists a conformal isomorphism :
S Ñ C, p meaning that it is conformal when expressed in charts of the new atlas
for S and charts for C.p Such an expression is j :“ ψj ˝ ˝ ϕ ´1 ˝ φ ´1 :
j j
j
D Ñ Vj , where ψj : pU q Ñ Vj is a chart on C.
S
p
Since the integrating map φj is Kj -quasiconformal, it follows that the maps
φj ˝ j : Uj Ñ Vj for any j are K-quasiconformal, where K “ maxj Kj .
Figure 1.16 The Integrability Theorem on Riemann surfaces follows from the local
version on open sets of the complex plane. In the drawing, tpϕ, i, j q :“ ϕj ˝ ϕi´1
stands for the transition map with respect to the charts tϕu. Likewise, tpφ ˝ ϕ, i, j q is
with respect to the new charts tφ ˝ ϕu. These transition maps are only defined on the
shaded regions. On the other hand, the map j is defined on the whole disc, and the
integrating maps j ˝ φj are defined on the whole Uj .
Hence, with respect to the original atlas on S, the map is K-quasiconformal,

and ˚ μ0 “ μ, so : S Ñ C p is the required integrating map.
Remark 1.29 Note that we can use the Integrability Theorem to endow a
surface S with a complex analytic structure, making it into a Riemann surface,
provided it admits a quasiconformal structure. Indeed, suppose there exists an
atlas tϕj u with quasiconformal transition maps and a Beltrami form μ which is
compatible with the atlas and with }μ}8 ă 1. Let φj be the integrating maps.
Then tφj ˝ ϕj u is a new atlas with holomorphic transition maps, and therefore
it defines a complex analytic structure on S.
1.4.1 Dependence on parameters

In surgery constructions, we are often interested in applying the Integrability
Theorem to a family of Beltrami forms depending on parameters. In these
cases, it is necessary to know how the solution of the Beltrami equation varies
in terms of parameters. Note that an integrating map φ is uniquely determined
if it is normalized. If s1 , s2 and s3 are distinct points in S then, in the case of
D, a normalization may be chosen by requiring φps1 q “ 0 and φps2 q P R` ; in

the case of C by requiring that φps1 q “ 0 and φps2 q “ 1; and in the case of Cp
by additionally requiring that φps3 q “ 8. Other kinds of normalizations will
occur. For the following theorem to hold it is important to refer to some chosen
normalization.
Theorem 1.30 (Integrability Theorem – dependence on parameters) Let

be an open subset of CN (or RN ) for some N ě 1.
(a) Let S be a Riemann surface isomorphic to C (or C), p and let pμλ qλP
be a family of measurable Beltrami forms on S. Suppose λ ÞÑ μλ psq is
holomorphic (respectively continuous, differentiable, real analytic) in λ
for each fixed s P S (whenever defined). Moreover, assume there exists a
k ă 1 such that }μλ }8 ď k for all λ P . Choose an appropriate nor-
malization and let φλ : S Ñ C (or onto C) p be the unique quasiconformal
homeomorphism which integrates μλ . Then for any fixed s P S the map
λ ÞÑ φλ psq is holomorphic (respectively continuous, differentiable, real
analytic).
(b) Let U be a Jordan domain in C, p and let pμλ qλP be a family of measurable
Beltrami forms on U . Suppose λ ÞÑ μλ psq is continuous (respectively
differentiable, real analytic) in λ for each fixed s P S (whenever defined).
Moreover, assume there exists a k ă 1 such that }μλ }8 ď k for all λ P .
Choose an appropriate normalization and let φλ : S Ñ D be the unique
quasiconformal homeomorphism which integrates μλ . Then for any fixed
s P S the map λ ÞÑ φλ psq is continuous (respectively differentiable, real
analytic).
Remark 1.31 (On D one cannot expect holomorphic dependence) Observe

that in the case of the unit disc, we cannot expect holomorphic dependence.
Indeed, if s P BU then necessarily φλ psq P S1 . The mapping λ ÞÑ φλ psq cannot
be holomorphic, since it is not an open mapping.
Remark 1.32 (Dependence of the inverse maps on parameters) It is impor-

tant to observe that even if the integrating maps depend holomorphically on a
parameter λ, their inverses do not necessarily have the same property. Indeed,
implicit differentiation shows that we need differentiability of φλ pzq with
respect to z in order to ensure differentiability of φλ´1 pwq with respect to λ,
and this is something which is not true in general. See the exercises below and
at the end of Section 1.7. See also [Hu2, Example 4.8.18].
1.4.2 On the proof of the Integrability Theorem

The Integrability Theorem is the cornerstone behind every surgery construc-
tion. There are several methods to prove the existence of solutions to the
Beltrami equation (1.18), depending on the domain and the regularity of μ.
We always assume that }μ}8 “ k ă 1.
We shall give a short account on different approaches with proper references
to the huge available literature.
The earliest occurrence of the Beltrami equation, though expressed in real
variables, appeared in the work of Gauss (see [Ga]). In 1822 he proved the
existence of isothermal coordinates on any surface with a real analytic Rieman-
nian metric. For an orientable surface the existence of isothermal coordinates
ensures that the surface can be equipped with the structure of a Riemann
surface. Half a century later, Beltrami in his studies of non-Euclidean geometry
also considered the equation that now bears his name (see [Bel]).
In order to obtain isothermal coordinates one has to solve a system of first-
order partial differential equations (PDEs), which in complex notation is equiv-
alent to solving the Beltrami equation. To ensure the existence of isothermal
coordinates the condition on the Riemannian metric was later relaxed by Korn
and Lichtenstein. They showed it was enough to require μ to be only Hölder
continuous (see [Ko, Li, C]). Lavrentiev [Lavr] showed the same result for
continuous μ using approximation by simple functions. See [LV, V.1.3] for a
similar approach.
Morrey was the first to prove the existence of a homeomorphic solution to
the Beltrami equation for μ measurable (see [Mo]). His methods were based
on PDEs.
The important results accomplished in the 1950s to prove the existence of
solutions were all obtained by using singular integral operators: the Cauchy
transform denoted by P and the Beurling transform denoted by T (to be
defined below). Bojarski was the first to interpret the Beurling transform as
an operator on Lp pCq by using the Calderón–Zygmund Inequality (see [Bo1]
and [Ah5, Chapt. V]). In the paper by Ahlfors and Bers [AB] they proved the
parameter dependence of normalized homeomorphic solutions of the Beltrami
equation in terms of the dependence of μ on the parameter.
More recently Douady gave a proof of the Integrability Theorem for every
measurable μ, using only classical techniques for L2 and Fourier transforms
(see [DB]).
We now proceed to sketch a proof of the Integrability Theorem.
Proof of uniqueness The uniqueness part of Theorems 1.27 and 1.28 is quite
simple. Suppose φ1 and φ2 are two solutions of the Beltrami equation (1.18).
Then φ1˚ μ0 “ φ2˚ μ0 a.e. and therefore the quasiconformal homeomorphism

ϕ :“ φ1 ˝ φ2´1 satisfies ϕ ˚ μ0 “ μ0 a.e. It follows by Weyl’s Lemma (Theo-
rem 1.14) that ϕ is conformal and hence φ1 “ ϕ ˝ φ2 as desired.
Proof of existence We now sketch a proof of the existence part of the Inte-
grability Theorem (Theorem 1.27) following mainly the survey in [GRSY]
and the proofs in [Ah5] and [LV]. In this proof we shall write φz :“ Bz φ and
φz :“ Bz φ, for any function φ for which the derivatives make sense.
As we shall see, a solution will be obtained as the fixed point of a strictly
contracting operator acting on an appropriate complete metric space, a com-
mon tecnique in dynamical systems.
Before starting the argument, we introduce the two operators P and T which
are used later on (see [LV, III 7.2] for details). The first one is the inverse of
the operator Bz , and is given by the Cauchy transform P , defined for h P Lp
with p ą 2 as
1 hpuq
ż
P hpzq “ ´ dmpuq,
π C u´z
where dm denotes the Lebesgue measure. This is an absolutely convergent
integral. The operator P satisfies
pP hqz “ h,
and if hz , hz P Lp then
P hz “ h. (1.19)
The second linear operator is the derivative of P defined by

1 hpuq
ż
T hpzq “ pP hqz pzq “ ´ dmpuq
π C pu ´ zq2
as a principal value. The Calderón–Zygmund inequality [Ah5, Sect. V.D]
ensures that T is well defined in Lp for all p ą 1 and satisfies
}T h}p ď Cp }h}p
for some constant Cp that fulfills Cp Ñ 1 as p Ñ 2.

Case 1: μ has compact support
Let μ be a measurable Beltrami coefficient defined on a open set U » D or
U “ C, and assume that μ has compact support. We extend μ to C by setting
μ “ 0 outside U . The extended μ is then measurable, with compact support
and has the same L8 norm k ă 1.
Suppose φ is a homeomorphic solution to the Beltrami equation

φ z “ μ φz . (1.20)
Then φ is conformal in CzU (since μ “ 0 on this set) and 8 is a removable
singularity. Therefore it can be extended to a function φ : Cp ÑC p setting
φp8q “ 8. Observe that any post-composition of φ with an affine map is also
a solution of (1.20). Hence, without loss of generality, we may assume that
φ 1 p8q “ 1 or equivalently
φpzq “ z ` gpzq,
where g is conformal in a neighbourhood of infinity. Since φ solves (1.20) then
φz “ gz “ μ p1 ` gz q.
Note that, because of the behaviour of gz in a neighbourhood of infinity (the
Laurent series of gz has lowest term 1{z2 ), the derivative gz belongs to Lp pCq,
for every p ą 1, and so does gz . Hence, we may apply the operator P to gz
and obtain from (1.19)
g “ P gz “ P μ p1 ` gz q. (1.21)
Differentiating now with respect to z, we have
gz “ T μ p1 ` gz q.
Therefore, the function h :“ gz satisfies the equation
h “ T μ p1 ` hq,
or equivalently h is a fixed point of the operator
Tr h “ T μ ` T μ h.
We now reverse the process and start by showing that the operator Tr does
indeed have a fixed point, i.e. there exists h P Lp satisfying Tr h “ h. This
follows from the Banach Fixed Point Theorem [Kre, Thm. 5.1-2], since Tr is
a strict contraction in the Lp norm for p sufficiently close to 2, and Lp is a
complete metric space. More precisely
}Tr h1 ´ Tr h2 }p “ }T μph1 ´ h2 q}p ď }T }p }μ}8 }h1 ´ h2 }p
ď Cp k}h1 ´ h2 }p ,
where k “ }μ}8 ă 1 and Cp Ñ 1 as p Ñ 2. Choose p ą 2 so that kCp ă 1,
and observe that this is possible because we require k ă 1. Then Tr is a
strict contraction in Lp . By the Fixed Point Theorem there exists a unique

function h P Lp which is a fixed point of Tr . It can be obtained by successive
applications of the operator with an arbitrary initial condition. To obtain g and
φ from h “ gz we go back to equation (1.21). Hence
φpzq “ z ` gpzq “ z ` P μp1 ` hq
is a solution of the Beltrami equation (1.20).

We now address the issue of injectivity of φ. First assume that μ P Cc1 .
Then its partial derivatives are continuous and in Lp for p ą 2. Under these
conditions it can be shown that φ is a C 1 homeomorphism (see [Ah5, Lem.
V.B.3]). Indeed, injectivity follows from showing that Jac φ is continuous
and positive in C.p This implies that φ is locally injective and hence a global
homeomorphism in C. p
If μ is only measurable, we approximate μ by a sequence of functions μn P
Cc1 such that μn Ñ μ a.e., }μn }8 ď k and μn “ 0 outside a fixed disc. Then
the K-quasiconformal homeomorphisms φn , obtained as above and normalized
so that they fix 0, 1 and 8, form a normal family and therefore must have a
convergent subsequence which converges uniformly to some limit φ. Then, φ is
K-quasiconformal since it is the uniform limit of K-quasiconformal maps (see
Theorem 1.21). By theorem [LV, IV.5.2] or lemma [AIM, 5.3.5] the complex
dilatation of φ coincides with μ almost everywhere, so φ is the desired solution
to the Beltrami equation.
Case 2: μ “ 0 in a neighbourhood of the origin
To get rid of the hypothesis of compact support consider first the case where
μ “ 0 in a neighbourhood of 0. Equalities are from now on, if necessary,
understood to hold a.e. Set I pzq :“ 1{z, and consider
ˆ ˙ 2
1 z
r pzq :“ I ˚ μ “ μ
μ .
z z2
The formula can be computed using (1.11). Then μ r has compact support in C
so it can be integrated by a quasiconformal map φ,
r normalized so that it fixes
0, 1 and 8. Then,
1
φpzq :“ pI ˝ φr ˝ I ´1 qpzq “
φp1{zq
r
is a quasiconformal map which fixes 0, 1 and 8 and integrates the given

Beltrami form μ. Indeed,
φ ˚ μ0 “ pI ´1 q˚ φr˚ I ˚ μ0 “ pI ´1 q˚ φr˚ μ0 “ pI ´1 q˚ μ
r “ I˚ μ
r “ μ,
where we used that I is injective and the pushforward is well defined. This is
summarized in the following commutative diagram:
φr
r q ÝÝÝÝÑ pC, μ0 q
pC, μ
§ §
§ §
Iđ đI
φ
pC, μq ÝÝÝÝÑ pC, μ0 q
Case 3: the general case

Finally, in the general case, where μ is a measurable Beltrami form on C,
we can obtain a solution to the Beltrami equation by combining solutions that
correspond to cases 1 and 2 above. Set μ “ μ1 ` μ2 where μ1 “ μ ¨ 1D and
μ2 “ μ ¨ 1CzD , with 1A being the characteristic function over the set A Ă C.
Since μ2 “ 0 on D there exists, as explained in Case 2, a quasiconformal
homeomoprhism φ2 normalized to fix 0, 1 and 8, such that φ2˚ μ0 “ μ2 . Now
define
r :“ pφ2 q˚ μ,
μ
which is well defined since φ2 is injective, and observe that μ

r has compact
support. Indeed, on CzD we have μ “ μ2 and, therefore, on this set
r “ pφ2 q˚ μ2 “ μ0 ” 0.
μ
Hence there exists a unique integrating map φr fixing 0, 1 and 8 so that φr˚ μ0 “
r . The map defined as
μ
φ :“ φr ˝ φ2
is a quasiconformal homeomorphism that integrates μ. Indeed,
φ ˚ μ0 “ φ2˚ φr˚ μ0 “ φ2˚ μ

r “ φ2˚ pφ2 q˚ μ “ μ.
We can summarize the construction in the following commutative diagram:
φ2
pC, μq / pC, μ rq
HH
HH
HH
HH φr
φ H$
pC, μ0 q
This concludes the proof of Theorem 1.27.

Dependence on parameters The chosen approach for the existence proof,

using the singular integral operators, has the advantage that it allows to intro-
duce parameters into the Beltrami equation. If μ “ μλ depends with a certain
regularity on a parameter λ it can be shown that the normalized solution φλ
of the Beltrami equation φz “ μλ φz depends with the same regularity on the
parameter λ, proving Theorem 1.30. For the proof we refer to [Ah5, V.C],
[Hu2, Sect. 4.7] or [AB].
1.4.1 (Integration of symmetric coefficients) Suppose μ is a Beltrami coef-

ficient defined in C satisfying the hypothesis of the Integrability Theo-
rem 1.27 and additionally being symmetric with respect to the real line
(resp. unit circle) (see Section 1.2.1). Prove that the integrating map φ
normalized to fix 1 and 0 is also symmetric with respect to the real line
(resp. unit circle).
Hint: Use the unicity of φ and Exercises in Section 1.2, especially
Exercise 1.2.6.
1.4.2 For any R ą 1 set R “ t|λ| ą Ru. Consider the family of maps w “
Lλ pzq “ λz ` z with λ P R . Show that this is a family of quasiconfor-
mal maps Lλ : C Ñ C which depend holomorphically on λ while the
1
family of inverses z “ L´ λ pwq do not.
1.4.3 Consider a family of φλ : C Ñ C of homeomorphisms, which depend
holomorphically on λ P . Show that if w “ φλ pzq also depends holo-
morphically on z, then the inverses z “ φλ´1 pwq depend holomorphi-
cally on both z and λ.
1.5 An elementary example

In the example explained in this section we use a few dynamical concepts. They
are surveyed in greater generality in Chapter 3. Later on we introduce what is
called soft surgery. The elementary example is a special case of soft surgery.
The example is elementary in the sense that everything can be computed
explicitly.
We start with a simple linear contracting map M0 : D Ñ D defined by
M0 pzq “ λ0 z for some 0 ă |λ0 | ă 1. We shall define new almost complex
structures in D, induced by certain quasiconformal mappings that are all M0 -
invariant.
The left half plane, H , is the universal covering space of D˚ “ Dzt0u under
the exponential map. Notice that the exponential map semi-conjugates M0 on
H
2π i D˚
exp λ0
ζ ÞÑ ζ ` ν0 z ÞÑ λ0 z
ν0
0
Figure 1.17 The exponential map semi-conjugates translation by ν0 on the left half
plane H to multiplication by λ0 “ eν0 on D˚ .
D˚ to a translation by ν0 “ log λ0 on H . We choose ν0 so that 0 ď Impν0 q

ă 2π (see Figure 1.17).
We have the following commutative diagram:
zÞÑλ z
D˚ ÝÝÝÝ0Ñ D˚
İ İ
exp§
§ §exp
§
ζ ÞÑζ `ν
H ÝÝÝÝÝÑ
0
H
Choose any parameter ν in the left half plane H . Consider the R-linear map
Lν : H Ñ H which maps the basis tν0 , i u onto the basis tν, i u, and there-
fore the parallelogram P pν0 q “ ttν0 ` i s | t P p0, 1q, s P p0, 2π qu onto the
parallelogram P pνq “ ttν ` i s | t P p0, 1q, s P p0, 2π qu (see Figure 1.18). If
we write Lν pζ q “ aζ ` bζ̄ it follows that
aν0 ` bν0 “ ν and a ´ b “ 1,
hence
ν ` ν0 ν ´ ν0
a“ , b“ ,
ν0 ` ν0 ν0 ` ν0
and the diagram
ζ ÞÑζ `ν
H ÝÝÝÝÝÑ
0
H
§ §
Lν đ
§ §L
đ ν
ζ ÞÑζ `ν
H ÝÝÝÝÝÑ H
H H
D˚ 2π i 2π i
Pν0
log Lν
ν0
Pν
0 0
ν
ξν
Figure 1.18 The standard complex structure in H is first pulled back by Lν and then
by the logarithm, to an almost complex structure on D˚ with Beltrami coefficient μν .
commutes. Note that Dζ Lν “ Lν for any ζ , so that the Beltrami coefficient of

Lν , is constantly equal to
BLν M BLν b ν ´ ν0
“ μpLν q “ “ .
Bζ Bζ a ν ` ν0
Set ξν “ Lν ˝ log, where log may be chosen as the branch of the logarithm
that maps D˚ onto the half strip ttν ` i s | t P p0, 8q, s P r0, 2π qu. Then
we can easily calculate the Beltrami coefficient μν of the pullback ξν˚ μ0 “
log˚ μpLν q at an arbitrary point z P D˚ . From ξν pzq “ a log z ` blog z we
obtain
1 1
Bz ξν “ a , Bz ξν “ b ,
z z
and therefore
ν ´ ν0 z
μν pzq “ . (1.22)
ν ` ν0 z
The modulus of the complex ´ dilatation
¯ μν is constant, but the argument
ν ´ν0
varies. At z ‰ 0 it is 2 arg z ` arg ν `ν0 . This means that the argument of the
´ ¯
minor axes of the infinitesimal ellipse in Tz D is equal to arg z ` 12 arg νν ´ ν0
`ν0 .
Hence, along any circle |z| “ r in D˚ the ellipses will make one full turn,
and along any radial line´the argument
¯ of the minor axis to the radial line is
constantly equal to 12 arg νν ´ ν0
`ν0 .
For a given ν P H we have constructed an almost complex structure σν
on D˚ corresponding to the Beltrami coefficient μν defined above; we set
μν p0q “ 0. The map φν : D Ñ D defined by
#
pexp ˝Lν ˝ logqpzq if z P D˚
φν pzq :“
0 if z “ 0
is an integrating map, which by construction makes the following diagram

commute:
M0 :zÞÑλ0 z
pD˚ , μν q / pD˚ , μν q
O O
exp exp
ζ ÞÑζ `ν0
pH , μpLν qq / pH , μpLν qq
φν Lν Lν φν
ζ ÞÑζ `Êν
pH , μ0 q / pH , μ0 q
exp exp
zÞÑeν z

pD˚ , μ0 q / pD˚ , μ0 q
In other words, M0˚ μν “ μν , and the map Mν “ φν ˝ M0 ˝ φν´1 , obtained as

the conjugate to M0 by the integrating map φν , is holomorphic. The commu-
tative diagram also shows that this composition φν ˝ M0 ˝ φν´1 on D˚ is equal
to Mν : z ÞÑ eν z, hence it holds on D (see Figure 1.19).
Note that whenever ν equals any one of the values of log λ0 , the composition
φν ˝ M0 ˝ φν´1 is just M0 again. Hence, it may happen that although we have
constructed an M0 -invariant almost complex structure very different from the
standard complex structure, we may get the map M0 back when conjugating
with the integrating map.
The integrating map φν maps the annulus A0 :“ t|λ0 | ď |z| ă 1u onto the
annulus Aν :“ t|eν | ď |z| ă 1u. Note that the moduli of the two annuli are
1 1 1 1
mod A0 “ log and mod Aν “ log ν .
2π |λ0 | 2π |e |
See also Figure 1.20.
1.5.1 Spreading μν by the dynamics

In this section we comment on the extension of the almost complex structure
given by the M0 -invariant Beltrami coefficient μν in D to an M0 -invariant
H H
2π i 2π i
Lν
ν0
0 0
ν
ζ ÞÑ ζ ` ν0 ζ ÞÑ ζ ` ν
Log exp
D˚ D˚
φν
M0 Me ν
Figure 1.19 The composition exp ˝Lν ˝ Log is precisely the integrating map φν on
D˚ .
A0 Aν
| λ0 |
φν |e ν |
Figure 1.20 The dynamically defined annuli A0 and Aν in the unit disc. For visual
reasons, the drawing is out of proportion, compared to the values of ν and
ν0 “ Log λ0 in the other drawings.
Beltrami coefficient in C, which we shall also denote by μν . It is geometrically

obvious that the almost complex structure given by
ν ´ ν0 z
μν pzq “ if z P C˚
ν ` ν0 z
is M0 -invariant. (It is also an easy exercise, using (1.11), to show that M0˚ μν “
μν on all of C.)
However, we shall use this example to show how this extension can be
defined recursively by pullbacks of μν on the annulus A0 by M0n for increasing
n P N. We say that the Beltrami coefficient is spread by the dynamics. This way
of extending a Beltrami coefficient is very common, and therefore worth

illustrating in this example. Set
" *
1 1
An :“ M0ń pA0 q “ 1
ď |z| ă ,
|λ0 |n ´ |λ0 |n
and
#
μν pzq if z P D,
r ν pzq :“
μ
pM0n q˚ μν pzq if z P An .
Then μ r ν is M0 -invariant by construction and with unchanged bound, i.e.

kpμν q “ kpr μν q, since all pullbacks are done by holomorphic maps (see
Figure 1.21).
A2
A1
Figure 1.21 We spread the M0 -invariant Beltrami coefficient μν on D to an

M0 -invariant Beltrami coefficient on all of C.
In this example where everything can be calculated explicitly we can also

extend the integrating map φν : D Ñ D conjugating M0 to Mν , to an integrat-
ing map φrν : C Ñ C, still conjugating M0 to Mν .
We set
#
φν pzq if z P D,
φrν pzq :“ ` ń
Mν ˝ φν ˝ M0n pzq
˘
if z P An .
Note that Kφν “ Kφrν , since composing by holomorphic maps does not change
the dilatation. Observe that φrν fixes 0 and 1. Hence φν is the uniquely deter-
mined integrating map, normalized in the standard way. The global Beltrami
r ν which we obtained in this way is the pullback of the standard
coefficient μ
Beltrami coefficient μ0 under the map φrν . In other words, the following
diagram commutes:
zÞÑλ z
pC, μ r ν q ÝÝÝÝ0Ñ pC, μ
rν q
§ §
§ §r
φrν đ đφν
zÞÑeν z
pC, μ0 q ÝÝÝÝÑ pC, μ0 q
1.5.2 Dependence with respect to parameters

We end this section by observing how the construction varies with respect to
parameters ν P H . Note that for any z P C˚ the map
ν ´ ν0 z
ν ÞÑ μ
r ν pzq “
ν ` ν0 z
is holomorphic in ν. Hence the integrating map φrν , normalized by fixing 0 and

1, also varies holomorphically with the parameter ν.
For z “ 0:
ν ÞÑ φrν p0q “ 0.
For z P D˚ :
˜ ¸ ˆ ˙
ν0 log z ´ ν0 log z 2 log |z|
ν ÞÑ φrν pzq “ exp exp ν .
ν0 ` ν0 ν0 ` ν0
For z P An :
ν ÞÑ exppńνqφrν pλn0 zq.
1.6 Quasiregular mappings

Pullback is a local concept. Global injectivity is not really necessary in order
to be able to pull back almost complex structures. This leads to the concept of
quasiregular mappings. In this section all quasiregular mappings are denoted
by g.
Definition 1.33 (First definition of K-quasiregular mapping) Let U Ď C

be an open set and K ă 8. A mapping g : U Ñ C is K-quasiregular if and
only if g can be expressed as
g “ f ˝ φ,
where φ : U Ñ φpU q is K-quasiconformal and f : φpU q Ñ gpU q is holo-

morphic.
It follows immediately that g is locally K-quasiconformal except at a

discrete set of points φ ´1 pCritpf qq, where Critpf q is the set of points with
vanishing derivative, i.e. the set of critical points of f . Around these points the
map is not injective.
This fact leads to the following alternative definition.
Definition 1.34 (Second definition of K-quasiregular mapping) Let U Ď

C be an open set and K ă 8. A continuous mapping g : U Ñ C is K-
quasiregular if and only if g is locally K-quasiconformal, except at a discrete
set of points in U .
Proof The first definition implies the second immediately.

To see the converse, let be the discrete set of points for which g is not
K-quasiconformal in any neighbourhood of such a point. Cover the open set
U z by a countable collection of open sets on which g is K-quasiconformal.
Such countable collections exists, since any open cover of a subset of R2
has a countable subcover (Lindelöf’s Theorem [Ke, p. 49]). It follows that
Bz g and Bz g are well defined almost everywhere in each of the open sets in
the subcover and locally belong to L2loc . Since a countable union of sets of
measure zero is a set of measure zero, Bz g and Bz g are well defined almost
everywhere in U . Moreover, the Beltrami coefficient μpzq “ Bz gpzq{Bz gpzq
´1
satisfies }μ}8 ď k “ K K `1 ă 1 in U . If U is simply connected we can apply
the Local Integrability Theorem immediately to obtain a K-quasiconformal
homeomorphism φ : U Ñ D (or onto C) integrating the Beltrami coefficient
μ. If U is not simply connected we first extend μ to μ p in C by defining
p pzq “ 0 if z R U , then apply the Global Integrability Theorem to obtain a
μ
K-quasiconformal homeomorphism φp : C Ñ C integrating the Beltrami coef-
ficient μ p U : U Ñ φpU
p . Set φ :“ φ| p q. Then the composition f :“ g ˝ φ ´1
is locally K-quasiconformal except at the discrete set of points φ ´1 pq.
Moreover
f ˚ pμ0 q “ pφ ´1 q˚ g ˚ pμ0 q “ pφ ´1 q˚ μ “ μ0 ,
as shown by the following diagram:
g
pU, μq / pgpU q, μ0 q
q8
qqqqq
q
qqq f
φ

pφpU
p q, μ0 q
Hence, by Weyl’s Lemma, f is locally conformal except at a discrete set of

points where it is continuous. This implies that f is holomorphic, and it follows
that g “ f ˝ φ as we wanted to show.
Quasiregular mappings can also be characterized locally by the following

equivalent definition.
Definition 1.35 (Third definition of K-quasiregular mapping) Let U Ď C

be an open set and K ă 8. A mapping g : U Ñ C is K-quasiregular if
and only if for every z P U , there exist neighbourhoods Nz and Ng pzq of z
and gpzq respectively, a K-quasiconformal mapping ψ : Nz Ñ D and a con-
formal mapping ϕ : Nf pzq Ñ D, such that pϕ ˝ g ˝ ψ ´1 qpzq “ zd , for some
d ě 1.
We prove that this definition is indeed equivalent to the two previous ones.
Proof Suppose g : U Ñ C satisfies the third definition of quasiregularity.

Then g is locally K-quasiconformal at those points where d “ 1. If d ą 1
on Nz , then z is the only point in this neighbourhood where the mapping is not
locally invertible. Hence, such points are isolated, and g satisfies the second
definition of quasiregularity.
Suppose g : U Ñ C satisfies the first definition of quasiregularity and there-
fore can be written as g “ f ˝ φ on U . We shall prove that g satisfies the third
definition of quasiregularity. We distinguish between two cases.
Case 1. If φpzq is not in Critpf q, then there exist neighbourhoods Nφ pzq and
Ng pzq of φpzq and gpzq respectively such that f : Nφ pzq Ñ Ng pzq is invertible
with a conformal inverse. Let Nz “ φ ´1 pNφ pzq q. Then we have the following
commutative diagram:
g
Nz ÝÝÝÝÑ Ng pzq
§ §
φđ
§ § ´1
đf
zÞÑz1
Nφ pzq ÝÝÝÝÑ Nφ pzq
and we are done.

Case 2. If φpzq is in Critpf q then there exist neighbourhoods Nφ pzq and Ng pzq
of φpzq and gpzq respectively such that f : Nφ pzq Ñ Ng pzq is a branched cov-
ering of degree d ą 1, ramified only at φpzq. All such coverings are equivalent
to z ÞÑ zd , in the sense that if ϕ : Ng pzq Ñ D is a Riemann map sending gpzq
to 0, then ϕ can be lifted to a conformal map ϕr : Nφ pzq Ñ D, and we have the
following commutative diagram:
g
Nz ÝÝÝÝÑ Ng pzq
§ §
φđ
§ §
đId
f
Nφ pzq ÝÝÝÝÑ Ng pzq
§ §
§ §ϕ
ϕrđ đ
zÞÑzd
D ÝÝÝÝÑ D
Since ϕr ˝ φ is K-quasiconformal and ϕ is conformal, we are done.
Finally for completeness we mention a fourth definition, which is common

in functional analysis. Compare to the first analytic definition (Definition 1.5)
of quasiconformality.
Definition 1.36 (Fourth definition of K-quasiregular mapping) Let U Ď C

be an open set and K ă 8. Set k :“ pK ´ 1q{pK ` 1q. A mapping g : U Ñ C
is K-quasiregular if and only if the partial derivatives
ˇ Bgˇ andˇ Bgˇ exist in the
sense of distributions and belong to Lloc and satisfy ˇBg ˇ ď k ˇBg ˇ in L2loc .
2 ˇ ˇ ˇ ˇ
The continuity of the map g is a consequence of the definition, since any

1,2
solution of the Beltrami equation in Wloc factorizes as in Definition 1.33. See
[AIM, Sect. 5.5] for details.
In what follows, we state some straightforward properties of quasiregular
mappings. We have actually used some of them already, but we state them
here for future reference.
Proposition 1.37 (Properties of quasiregular mappings) Let U, U 1 be open

subsets of C.
(i) If g1 : U Ñ U 1 and g2 : U 1 Ñ C are K1 - and K2 -quasiregular respec-

tively, then g2 ˝ g1 is K1 K2 -quasiregular.
(ii) A mapping g : U Ñ C is holomorphic if and only if g is 1-quasiregular.
(iii) If f : U Ñ U 1 is holomorphic and φ : U 1 Ñ C is K-quasiconformal,
then g :“ φ ˝ f is K-quasiregular.
(iv) If in the third definition, we allow ϕ to be K 1 -quasiconformal (instead of
conformal), then g is KK 1 -quasiregular.
(v) If g : U Ñ C is quasiconformally conjugate to f : U Ñ C, and f is
holomorphic, then g is quasiregular.
(vi) (Variant of Weyl’s Lemma) If g : U Ñ C is quasiregular and g ˚ μ0 “ μ0
almost everywhere in U , then g is holomorphic.
Proof
(i) Follows from the second definition.
(ii) If g is holomorphic it is also 1-quasiregular. If g is 1-quasiregular, then
g “ f ˝ φ, where f is holomorphic and φ is 1-quasiconformal. By Weyl’s
Lemma φ is conformal, and hence g is holomorphic.
(iii) Follows from (i).
(iv) Locally g can be written as ϕ ´1 ˝ pz ÞÑ zd q ˝ ψ. By the second definition
this implies that g is KK 1 -quasiregular.
(v) Let φ be the K-quasiconformal conjugacy, so that g “ φ ˝ f ˝ φ ´1 . Since
f ˝ φ ´1 is K-quasiregular and φ is K-quasiconformal, it follows that g is
K 2 -quasiregular.
(vi) We can write g “ f ˝ φ, where f is holomorphic and φ is quasiconfor-
mal. Then
μ0 “ g ˚ μ0 “ φ ˚ f ˚ μ0 “ φ ˚ μ0 ,
almost everywhere in U . It follows from Weyl’s Lemma that φ is confor-
mal. Hence g is holomorphic.

For mappings between Riemann surfaces, these definitions generalize in

the natural way: a mapping is quasiregular if it is locally quasiregular when
expressed in charts.
We end this section with an observation that is crucial for surgery.
Lemma 1.38 (Pullbacks by quasiregular mappings) Quasiregular map-
pings and their inverse branches send sets of measure zero to sets of measure
zero. Consequently, the pullback of a Beltrami form defined a.e., by a quasireg-
ular map, is well defined a.e.
Proof A quasiregular map g factors as g “ f ˝ φ with f holomorphic and
φ quasiconformal. Since quasiconformal maps (and holomorphic maps in
particular) are absolutely continuous with respect to the Lebesgue measure
(see Theorem 1.15), it follows that quasiregular maps also have that property.
The same arguments apply to inverse branches wherever defined. Arguing as
in Remark 1.17, we conclude that pullbacks of Beltrami forms defined a.e., are
well defined a.e.
In general, when we write

f
pS1 , μ1 q ÝÑ pS2 , μ2 q,
we mean that f is a quasiregular map between the Riemannn surfaces S1 and

S2 and that f ˚ μ2 “ μ1 a.e., where μj for j “ 1, 2 is a Beltrami form defined
on Sj .
1.7 Application to holomorphic dynamics

Before giving an introduction in Chapter 4 to what quasiconformal surgery is
about, we end the current chapter by getting a flavour of how to apply what
we have learned so far to dynamical systems. The application is at the heart of
surgery constructions.
Suppose S is a Riemann surface isomorphic to C or C. p We are interested in
dynamical systems generated by the iterates of a mapping f : S Ñ S, which
we assume for the moment to be at least quasiregular. In other words, given
an initial point s P S we are interested in the asymptotic behaviour of its orbit
Opsq “ ts, f psq, f 2 psq, . . . , f n psq, . . .u.
In the qualitative study of such dynamical systems, conjugacies play a
crucial role, see Section 3.1. We are interested in finding a ‘holomorphic
dynamical copy’, i.e. a holomorphic mapping which is quasiconformally con-
jugate to the given one. The following lemma states conditions under which
this is possible.
Lemma 1.39 (Key Lemma for surgery)
(a) Let S be a Riemann surface isomorphic to C or C, p and let g : S Ñ S

be quasiregular. Let μ be a g-invariant Beltrami form on S such that
}μ}8 :“ k ă 1. Then there exists a holomorphic mapping f : X Ñ X
where X P tC, Cu,
p such that g and f are quasiconformally conjugate.
(b) Let S 1 Ă S » D be an open set, and let g : S 1 Ñ S be quasiregular. Let

μ be a g-invariant Beltrami form on S such that }μ}8 :“ k ă 1. Then
there exists a holomorphic mapping f : D 1 Ñ D where D 1 Ă D is open
and such that g and f are quasiconformally conjugate.
Proof (a) Let φ : S Ñ X be a quasiconformal mapping that integrates μ,

given by the Integrability Theorem. Then, if we define f :“ φ ˝ g ˝ φ ´1 , the
following diagram commutes:
g
pS, μq ÝÝÝÝÑ pS, μq
§ §
φđ
§ §φ
đ
f
pX, μ0 q ÝÝÝÝÑ pX, μ0 q
The mapping f is quasiregular and satisfies f ˚ μ0 “ μ0 almost everywhere.

Hence by Weyl’s Lemma f is holomorphic and is quasiconformally conjugate
to g via φ (see Figure 1.22).
pC, μq pC, μq
φ φ
pC, μ0 q pC, μ0 q
Figure 1.22 Sketch of the Key Lemma for surgery where S “ C.
(b) The proof is analogous to (a) taking D 1 :“ φpS 1 q.
As we shall see in later surgery constructions, such procedures may be

performed with dependence on a parameter, say λ P Ă C. The mappings
gλ : S Ñ S and the Beltrami forms μλ , which are gλ -invariant, may both
depend holomorphically on λ. The Integrability Theorem depending on param-
eters then provides a family of integrating maps φλ , uniquely determined when
normalized appropriately, and varying holomorphically with λ. Applying the
Key Lemma to each of these mappings, we obtain a family of holomorphic
mappings fλ : X Ñ X satisfying
fλ ˝ φλ ” φλ ˝ gλ
on S. It is natural to ask whether fλ depends holomorphically on λ. Often,

one is able to deduce the form of the resulting family, for example a family of
polynomials with coefficients in C, so that the question is equivalent to asking
if the coefficients are holomorphic functions of λ.
However, the answer is not always simple. As mentioned in Remark 1.32,
the inverses of the integrating maps might not depend holomorphically on the
parameter. However, we present a particular case where it is possible to prove
the holomorphic dependence (cf. Section 4.1.2 and Claim 4.8). We refer to
[BC, Prop. 13] for another computation of this type.
Lemma 1.40 (Dependence on parameters) Let g be given as in

Lemma 1.39(a), and let the Beltrami forms μλ and therefore the integrating
mappings φλ , appropriately normalized, depend holomorphically on λ. Then
the holomorphic mappings fλ pzq :“ φλ ˝ g ˝ φλ´1 pzq depend holomorphically
on λ.
Proof Consider the expression fλ ˝ φλ ” φλ ˝ g. We want to show that

Bfλ {B λ̄pφλ pzqq “ 0 for all z. The mapping at the left-hand side can be
written as
pλ, λ̄, z, zq ÞÑ fλ pλ, λ̄, φλ pzq, φλ pzqq,
and the mapping at the right-hand side as
pλ, λ̄, z, zq ÞÑ φλ pλ, λ̄, gpzq, gpzqq.
Applying the chain rule when computing the derivatives in the sense of distri-
butions we obtain that
ˇ ˇ ˇ ˇ ˇ ˇ
Bfλ ˇˇ Bfλ ˇˇ Bφλ ˇˇ Bfλ ˇˇ Bφλ ˇˇ Bφλ ˇˇ
` ¨ ` ¨ ˇ “ .
B λ̄ ˇφλ pzq Bz ˇφλ pzq B λ̄ ˇz Bz ˇφλ pzq B λ̄ ˇ B λ̄ ˇg pzq
z
Since BBφλ̄λ ” 0 and BBfzλ ” 0 we have

ˇ
Bfλ ˇˇ
“ 0.
B λ̄ ˇφλ pzq
It follows that fλ is varying holomorphically with λ.
The justification for the chain rule in the sense of distributions follows from
the fact that quasiconformal mappings can be approximated by C 8 functions
(cf. [AIM, Cor. 5.5.8]).
We emphasize that in the lemma above, the quasiregular map g is indepen-
dent of the parameter. However, if it depends holomorphically on the param-
eter, then the resulting family of maps fλ need not do so. See Exercise 1.7.1
below.
In Chapter 8 we shall nevertheless encounter several examples of surgery
constructions with a given family of quasiregular mappings that depend holo-
morphically on a parameter, and so the resulting family of mappings also
depend holomorphically on the parameter.

1.7.1 In the setup of Lemma 1.40, assume furthermore that the quasiregular
map g depends holomorphically on λ. Prove that in such a case
ˇ ˇ ˇ
Bfλ ˇˇ Bφλ ˇˇ Bgλ ˇˇ
“ ¨ .
B λ̄ ˇφλ pzq B z̄ ˇgλ pzq B λ̄ ˇz
2
Boundary behaviour of quasiconformal maps:
extensions and interpolations
In surgery constructions we often need to interpolate (quasiconformally)

between two different maps defined on distinct regions of the plane. This
involves finding quasiconformal extensions of given boundary maps, whenever
possible, but also knowing whether a map defined on a certain domain extends
continuously to its boundary. We address these two types of problems in this
chapter.
We recall some elementary definitions about curves and their regularity
(see [Pom] for details and proofs). A curve γ in the complex plane is the
image set of a parametric representation or parametrization. In the following
we use γ both as a name for a continuous function γ : ra, bs Ñ C, parametriz-
ing the curve, and for its image γ pra, bsq. The curve γ is closed if γ paq “
γ pbq, in which case it can be reparametrized as γ : T Ñ C; it is a Jordan arc if
it has an injective parametrization, and a Jordan curve if in addition it is closed.
The Jordan Curve Theorem states that the complement of any Jordan curve γ in
the complex plane has exactly two connected components. A bounded domain
whose boundary is a Jordan curve is called a Jordan domain.
A curve has different degrees of regularity depending on the regularity of
a parametrization. For n ě 1 a curve γ is C n if there exists a parametrization
of γ which is n times continuously differentiable and satisfies γ 1 ptq ‰ 0 for
all t. Curves which are C 1 are often referred to as smooth curves. The highest
degree of regularity is achieved when a parametrization can be chosen so that
it extends to a holomorphic univalent map in some horizontal band containing
ra, bs or, in the closed case, in some annular neighbourhood of the unit circle.
The set T “ R{Z is identified with the unit circle S1 :“ tz P C | |z| “ 1u
by the parametrization t ÞÑ e2π i t . We often use complex or angular notation
indistinctively, if it does not lead to confusion.
64
In the problem of extending a given map defined on a domain to a map

defined on its closure we use in general the following notation. Let G denote
a Jordan domain bounded by a Jordan curve γ , and let f : D Ñ G be a
homeomorphism. In general, f does not extend continuously to the boundary
of D (see Exercise 2.2.1), but under certain assumptions on the regularity of f
and γ it does, and we denote its extension by fp : D Ñ G. In such cases, we
shall discuss the regularity of fp |S1 in terms of the regularity of f and γ . Note
that whenever a continuous extension exists it is unique.
In the problem of extending a given boundary map to the domain we use in
general the following notation. Let γ be a Jordan curve bounding the Jordan
domain G, and let f : S1 Ñ γ be a homeomorphism. Then f can always be
extended continuously to a homeomorphism fp : D Ñ G in different ways. We
shall discuss the maximal regularity of the extension in terms of the regularity
of f and γ .
Finally, we consider the same type of problems for annuli and quadrilaterals,
as well as (briefly) covering maps of degree higher than one.
The goal of this chapter is to collect all the results about extensions that will
be used throughout the surgeries.
2.1 Preliminaries: quasisymmetric maps and quasicircles

When dealing with quasiconformal mappings, we are led to the concept of
quasisymmetry.
Definition 2.1 (First definition of quasisymmetry) A map h : S1 Ñ C is

quasisymmetric if h is injective and, for z1 , z2 , z3 P S1 ,
1 |hpz1 q ´ hpz2 q|
0 ‰ |z1 ´ z2 | “ |z2 ´ z3 | ñ ď ďM (2.1)
M |hpz2 q ´ hpz3 q|
for some M ě 1. Sometimes h is called M-quasisymmetric or quasisymmetric
of modulus M. In angular notation, i.e. writing H ptq :“ hpe2π i t q, the condition
takes the form
1 |H px ` tq ´ H pxq|
ď ď M, (2.2)
M |H pxq ´ H px ´ tq|
for all x, x ` t, x ´ t P T, with t ą 0 (see Figure 2.1).
Note that if one of the inequalities is satisfied, the other follows by inter-
changing the roles of z1 and z3 , or the sign of t.
The following is a formally stronger but actually equivalent definition of
quasisymmetry, which is some times easier to apply. See [Pom, Chapt. 5].
66 Extensions and interpolations
z2 hpz3 q
z3 B
z1
Þ e2π i x 2π t 2π t h
0 1 xÑ
hpz2 q A
x´t x x`t
hpz1 q
Figure 2.1 The definition of a quasisymmetric map of the circle. The quotient of the
length of the two segments A and B must be bounded for all triples of equidistant
points in S1 , or equivalently for all x P T and all t ą 0.
Definition 2.2 (Second definition of quasisymmetry) A map h : S1 Ñ C

is quasisymmetric if h is injective and if there exists a strictly increasing
continuous function λ : r0, `8q Ñ r0, `8q such that
ˆ ˙
1 |hpz1 q ´ hpz2 q| |z1 ´ z2 |
´ ¯ď ďλ for z1 , z2 , z3 P S1 .
|z ´z | |hpz2 q ´ hpz3 q| |z2 ´ z3 |
λ 2 3
|z1 ´z2 |
(2.3)
Note that if one of the inequalities in (2.3) is satisfied, the other follows by
interchanging the roles of z1 and z3 .
Note that if hpS1 q “ S1 and h is quasisymmetric then h´1 is also quasisym-
metric.
As in the case of quasiconformal maps, the existence of a continuous non-
zero derivative is enough to guarantee quasisymmetry.
Lemma 2.3 If h : S1 Ñ hpS1 q is a C 1 -diffeomorphism, then it is quasisymmet-

ric.
Proof Let H ptq “ hpe2π i t q. We shall prove that H satisfies a bilipschitz

condition
1{K |x ´ y| ď |H pxq ´ H pyq| ď K |x ´ y| (2.4)
for some K ą 0 and all x, y P T. From this it follows that H is K 2 -

quasisymmetric.
Note that, independently of x, t P T we have
ˇż t ˇ
1 1 1
ˇ ˇ
|H px ` tq ´ H pxq ´ tH pxq| “ ˇ rH px ` τ q ´ H pxqsdτ ˇˇ ď k|t|
ˇ
0
for some k ą 0 since H1 is uniformly continuous on T.

Therefore G : T ˆ T Ñ R defined by
# |H px q´H py q|
if x ‰ y,
Gpx, yq :“ |x ý |
| H 1 pxq | if x “ y,
is continuous. Since the torus T ˆ T is compact, G attains both its maximal

value and minimum value. The minimal value must be positive since H is
injective and H 1 does not vanish on T. Hence, the inequality (2.4) is satisfied
for some K ą 0.
As we shall see, quasisymmetric mappings on S1 turn out to be precisely

the boundary values (i.e. continuous extensions) of quasiconformal homeo-
morphisms on the unit disc. And, conversely, any quasisymmetric mapping on
S1 can be extended to a quasiconformal mapping on D. It is in this sense that
quasisymmetric mappings can be thought of as the one-dimensional analogue
of quasiconformal mappings. But this analogy is not quite accurate, since qua-
sisymmetric mappings are not as well behaved as quasiconformal mappings.
For example, quasisymmetric mappings cannot in general be pasted together to
form a quasisymmetric mapping. Consider, for instance, the homeomorphism
H : T Ñ S1 defined as
2
#
e4π i t if 0 ď t ď 1{2,
H ptq :“ 2π i t
e if 1{2 ď t ă 1.
It is C 1 on pRz 12 Zq{Z but not quasisymmetric. Indeed, for x “ 0 and 0 ă t ď

|H px `t q´H px q| sinp2π t 2 q
1{2 the fraction |H px q´H px ´t q| “ sinpπ t q
is not bounded below when t Ñ 0
(see Figure 2.2).
aptq
1
2
2t 2
H ptq
4π t 2
1 t H p0q
´ 12 2
´2π t
H p´tq
´ 12
Figure 2.2 Two quasisymmetric maps pasted together do not always produce a
1 argpH ptqq.
quasisymmetric map. The left graph shows the map aptq “ 2π
Proposition 2.4 (Compositions of quasisymmetric maps) Let hj : S1 Ñ C,

for j “ 1, 2, be quasisymmetric.
(a) If h1 pS1 q “ S1 , then h2 ˝ h1 : S1 Ñ C is quasisymmetric.

1
(b) If γ “ h1 pS1 q “ h2 pS1 q, then h´2 ˝ h1 : S Ñ S is quasisymmetric.
1 1
Proof For j “ 1, 2, let λj : r0, `8q Ñ r1, `8q denote a strictly increasing
homeomorphism so that the inequality (2.3) is satisfied for hj .
In case (a), set h “ h2 ˝ h1 . Then λ “ λ2 ˝ λ1 is a strictly increasing homeo-
morphism so that (2.3) is satisfied for h. Hence
´ h is quasisymmetric.
¯
´1 1
In case (b), set h “ h2 ˝ h1 . Then λ “ 1{ λ´ 2 p1{λ1 q is a strictly increas-
ing homeomorphism. From
1 |h1 pz1 q ´ h1 pz2 q| ` |z1 ´ z2 | ˘
´ ¯ď ď λ1
|hpz q´hpz q| |h1 pz2 q ´ h1 pz3 q| |z2 ´ z3 |
λ2 |hpz2 q´hpz3 q|
1 2
it follows that
|hpz1 q ´ hpz2 q| 1
ď ˜ ¸,
|hpz2 q ´ hpz3 q|
1
λ´
2
´ 1
|z1 ´z2 |
¯
λ1 |z2 ´z3 |
´ ¯
1
so that (2.3) is satisfied for h with λ “ 1{ λ´
2 p1{λ1 q . Hence h is quasisym-
metric.
Definition 2.5 (Quasiarc, quasicircle and quasidisc) A Jordan arc (resp. Jor-
dan curve) γ in C is a quasiarc (resp. quasicircle) if for some C ą 0
diam γ pz1 , z2 q ď C|z1 ´ z2 | for z1 , z2 P γ ,
where γ pz1 , z2 q is the arc of smaller diameter of γ joining z1 and z2 . A Jordan

domain bounded by a quasicircle is a quasidisc.
A piecewise smooth Jordan arc is a quasiarc if and only if it has no cusps

(pointing inwards or outwards). See Exercises 2.1.1, 2.1.2 and Figure 2.3.
The geometric property associated with quasisymmetric parametrizations is
the following, which in fact follows from Theorem 2.9 in the next section.
Proposition 2.6 (Quasiarcs and quasicircles are curves with quasisymmet-

ric parametrizations) If h : ra, bs Ñ C (resp. h : S1 Ñ C) is quasisymmet-
ric, then its image γ is a quasiarc (resp. a quasicircle). Conversely, if γ is a
quasiarc or a quasicircle, then it is the image of an interval or the unit circle
under some quasisymmetric map.
2.2 Extensions of mappings from their domains to their boundaries 69
Figure 2.3 Left: The Koch snowflake is a well-known quasicircle, even if it is

nowhere differentiable. Right: A cardioid is not a quasicircle due to its cusp.

2.1.1 Let γ “ tpt, |t|α q| ´ 1 ď t ď 1u with α ą 0. Show that γ is a quasiarc
if and only if α ě 1, by using the definition.
2.1.2 Let γ “ tpt, |t|α sinp2π {tqq| t P r´1, 0q Y p0, 1su Y tp0, 0qu with
α ą 0. Show that γ is not a quasiarc if 0 ă α ă 2.
1
Hint: Take points of the form an “ 2n and bn “ 2n1`1 and apply the
definition of quasiarc.
2.2 Extensions of mappings from their domains to their

boundaries
2.2.1 Extensions of conformal maps
We recall the following theorem (cf. [Pom, Chapt. 2]), which is an essential
tool in holomorphic dynamics. We first recall the definition of a locally con-
nected set.
Definition 2.7 (Locally connected) A set X Ă C is locally connected if

for every point in x P X and any arbitrarily small ą 0, the intersection
D pxq X X is connected.
Theorem 2.8 (Carathéodory Theorem) Let f : D Ñ G be a conformal iso-

morphism and G a bounded domain in C. Then f has a continuous extension
fp : D Ñ G if and only if BG is locally connected. Moreover, f has a continu-
ous and injective extension to D if and only if BG is a Jordan curve.
Maps on the unit disc We start by assuming that f : D Ñ G is a conformal

isomorphism, in other words a Riemann mapping, and G Ă C is a Jordan
domain with boundary γ :“ BG. It follows from Carathéodory’s Theorem
that the map f has a continuous and injective extension to fp : D Ñ G. The
extended map fp |S1 : S1 Ñ γ is called a conformal parametrization of γ .
Given a Jordan curve γ , its conformal parametrization is uniquely determined
up to pre-composition with a Möbius transformation preserving the unit disc.
If we assume more regularity for γ we obtain more regularity for the extension,
as explained in the following theorem.
Theorem 2.9 (Extension of conformal maps on D) Let G Ă C be a Jordan

domain with γ :“ BG, and let f : D Ñ G be a conformal isomorphism. Then:
(a) f has a quasisymmetric extension to S1 if and only if γ is a quasicircle;

(b) if γ is C n`1 for some n ě 0 then the maps f, f 1 , . . . , f pnq have continuous
extensions to D;
(c) f can be extended conformally to some disc of radius r ą 1 if and only if
γ is analytic.
f
D G
Figure 2.4 If a homeomorphism f from D to a Jordan domain G extends

continuously to the boundary, the regularity of the extension depends on the regularity
of f and the one of γ :“ BG.
Remark 2.10 For part (b) to hold, it is enough that γ is of class C n,α for some
0 ă α ď 1; this means that γ is C n and γ pnq satisfies a Hölder condition of
exponent α.
The proofs of the different parts of the theorem can all be found in [Pom,
Chapt. 3 and 5], more precisely for (a): Proposition 5.10 and Theorem 5.11; for
(b) and the remark above: Theorem 3.6; and for (c): Proposition 3.1, a simple
application of the Schwarz Reflection Principle.
A consequence of this theorem is that if there exists a quasisymmetric
(resp. differentiable or analytic) parametrization of γ , then for any Rie-
mann mapping f : D Ñ G, the corresponding conformal parametrization is
quasisymmetric (resp. differentiable or analytic). The conformal parametriza-

tion is as good as it gets.
Remark 2.11 (Extensions of coverings of degree d) Part (c) of Theorem 2.9
is an application of Schwarz Reflection Principle. Lifting the map and using
exactly the same arguments, it can be shown that if f is a holomorphic covering
of degree d ą 1, it also extends to a larger disc as a holomorphic covering of
the same degree. In particular, the restriction to the boundary fp : S1 Ñ γ is an
analytic covering map.
Remark 2.12 (Piecewise regular curves) If the Jordan curve BG in The-
orem 2.9 consists of finitely many arcs, which are either of class C n`1 or
analytic, then a conformal map f : D Ñ G extends to the boundary as a
piecewise smooth map of class C n or a piecewise analytic map respectively.
To see this, observe that the Schwarz Reflection Principle applies to open sets
that intersect the unit circle, hence the extension is local.
We are also interested in doubly connected domains (annuli) bounded by
Jordan curves and simply connected domains such as half strips that are not
Jordan domains. We start with the doubly connected case.
Jordan annuli A bounded annulus will be called a Jordan annulus (or a

quasiannulus) if its outer and inner boundary curves are both Jordan curves
(or both quasicircles resp.).
For any 0 ă r ă 1 ă R set
Ar,R :“ tr ă |z| ă Ru, Ar :“ Ar,1 “ tr ă |z| ă 1u and S1r :“ t|z| “ ru.
Recall that any bounded open annulus A of finite modulus is conformally

equivalent to a unique standard annulus Ar under a conformal map f : Ar Ñ
A, which is uniquely determined up to pre-composition by a rotation. The
following is a straightforward corollary of Theorem 2.9.
Corollary 2.13 (Extension of conformal maps on annuli) Let A be a Jordan

annulus and let f : Ar Ñ A be a conformal equivalence. Let γ i , γ o denote
the inner and outer boundary of A. Then:
(a) f has a quasisymmetric extension to S1 (S1r resp.) if and only if γ o (γ i

resp.) is a quasicircle;
(b) If γ o and γ i are C n`1 for some n ě 0 then the maps f, f 1 , . . . , f pnq have
continuous extensions to Ar ;
(c) f can be extended conformally to some annulus Ar,R where 1 ă R (resp.
to Ar 1 where 0 ă r 1 ă r) if and only if γ o (resp. γ i ) is analytic.
Proof Let Go denote the Jordan domain of γ o , and let R : D Ñ Go be a

conformal map which extends continuously to the boundary as R p : D Ñ Go .
The boundary map R : S Ñ γ has the properties described in Theorem 2.9.
p 1 o
The composition R´1 ˝ f : Ar Ñ R´1 pAq is conformal. By the Schwarz

Reflection Principle this map extends to an analytic map from S1 to itself. The
boundary map ψ o : S1 Ñ γ o is obtained as the composition of this analytic
map and R.p Therefore ψ o has the same properties as R. p The properties for
the extended map to the inner boundaries is obtained similarly by using the
exterior domain of γ i in C.
p
Half strips We continue by considering special horizontal half strips. The

result is applied in Chapter 8. It suffices to consider the following case. Let
γ be a C 2 -Jordan arc, parametrized by y P r0, 1s so that Im γ pyq is strictly,
increasing between Im γ p0q “ 0 and Im γ p1q “ 1. A standard half strip of
height 1 and bounded by γ is defined by
ă
1,γ :“ tz “ x ` i y | 0 ď y ď 1, x ď Re γ pyqu.
See Figure 2.6. Considered as a subset on the Riemann sphere the half strip
has three vertices. We denote the vertex at the left end by ´8. Any Riemann
mapping, between the interiors of two such half strips, bounded by γ1 and γ2
respectively, extends continuously to the boundaries as piecewise analytic on
the horizontal curves and C 1 from γ1 to γ2 (see Remark 2.12). In order to state
the property of the horizontal boundary maps we need the following definition.
The notion of near translation and its use was introduced by Bielefeld in [Bi1].
Definition 2.14 (Near translation) For j “ 1, 2, let j denote C 1 -Jordan

arcs, which are backward invariant under translation by σ where Re σ ą 0,
i.e. j ´ σ Ă j as sets. Let h : 1 Ñ 2 be a C 1 -diffeomorphism. Then h is
said to be a near translation or C 1 -bounded if there exists a constant C ą 1
such that
ˇ ˇ
1 ˇ d hpzptqq ˇ
ˇ dt
|hpzq ´ z| ă C for any z P 1 and ăˇ d ˇ ă C, (2.5)
ˇ
C ˇ
dt zptq
ˇ
where t ÞÑ zptq is a C 1 -parametrization of 1 (see Figure 2.5).
d d
Note that dt hpzptqq{ dt zptq is independent of the chosen parametrization of
1 . Moreover, the composition of two near translations and the inverse of a
1 h
σ
2
C
z
Figure 2.5 Two backwards σ -periodic Jordan arcs j for j “ 1, 2, and a

C 1 -diffeomorphism h : 1 Ñ 2 . If h is a near translation then for any point z P 1
the image hpzq P 2 belongs to the open disc of radius C around z.
near translation are near translations, and any σ -periodic map, i.e. hpz ` σ q “
hpzq ` σ , is a near translation (see Exercise 2.2.2).
Proposition 2.15 (Extension of conformal maps on standard half strips

of equal height) Let f : 1,γ ă Ñ ă be the Riemann mapping between
1 1,γ2
the interior of the half strips whose continuous extension to the bound-
aries satisfies fpp´8q “ ´8, fppγ1 p0qq “ γ2 p0q and fppγ1 p1qq “ γ2 p1q. Set
fr1 pxq :“ fppxq for x ď γ1 p0q and fr2 pxq :“ fppx ` i q ´ i for x ď Re γ1 p1q.
Then frj , j “ 1, 2, are near translations.
Proof Let 1 :“ tz “ x ` i y | 0 ď y ď 1u denote the full horizontal strip

of height 1. The exponential map : z ÞÑ w “ u ` i v :“ eπ z maps 1 onto
Hzt0u, the closure of the upper half plane minus the origin, 1,γ
ă onto D X
1
γ1
Hzt0u, and 1,γă onto D X Hzt0u, where D may be chosen as the open
2
γ2 γj
topological disc, which is symmetric with respect to R and bounded by j “
eπ γj in H. Then f is conjugate by to a map F : Dγ1 X Hzt0u Ñ Dγ2 X
Hzt0u, which extends to a conformal isomorphism F : Dγ1 Ñ Dγ2 , first by
reflection with respect to R (the Schwarz Reflection Principle) and then, using
that fpp´8q “ ´8, by continuity, setting F p0q “ 0 (see Figure 2.6).
The derivative F 1 p0q is some positive number, since F is mapping an
interval of R around 0 diffeomorphically onto an interval of R around 0 with
the orientation preserved. By using the expansion of F around 0: F pwq “
F 1 p0qw ` Opw 2 q and by differentiating the equation
F ppzqq “ pf pzqq,

xì γ1 p1q f px ` i q “ fr2 px q ` i

γ2 p1q
f
ă ă
1,γ
1,γ1 2
γ1 p0q f px q “ fr1 px q γ2 p0q

x
: z ÞÑ eπ z : z ÞÑ eπ z
1 2
Dγ1 Dγ2
0 u “ px q F púq 0 F puq

F
ú “ px ` i q
Figure 2.6 Sketch of the mappings and sets used in the proof of Proposition 2.15.
ă
we obtain for z P 1,γ1
F 1 pwqw F 1 pwq
f 1 pzq “ “ 1 Ñ 1 as z Ñ ´8 so that w Ñ 0.
F pwq F p0q ` Opwq
Similarly, we obtain for x ď Re γ1 p0q, and x ď Re γ1 p1q respectively, that
e π f px q F puq
p
eπ rf1 px q´x s “ “ Ñ F 1 p0q and
r
e π x u
éπ f px ì q ´F púq
p
eπ rf2 px q´x s “ “ Ñ F 1 p0q
r
e π x u
as x Ñ ´8 so that u “ eπ x Ñ 0.
Hence, for any choice of a constant C ą 1 the inequalities
1
Ć ă frj pxq ´ x ă C and ă frj1 pxq ă C
C
are satisfied for x sufficiently close to ´8, say for x ă x0 . On the remain-
ing compact intervals rx0 , γ1 p0qs and rx0 , Re γ1 p1qs the continuous functions
frj1 pxq and frj pxq ´ x attain their maximal and minimal values. Hence, it is easy
to adjust the constant C ą 1 such that the inequalities are satisfied.
Remark 2.16 (Half strips of different height) Note that if the half strips do
not have the same height (here chosen to be 1) then fr1 and fr2 would not be
near translations, since |frj pxq ´ x| would be unbounded as x tends to ´8 (see
Exercise 2.2.4).
We shall see in Lemma 2.26, to be applied in Section 8.1, why it is essential

to assume certain boundary mappings to be near translations.
2.2.2 Extensions of quasiconformal maps

Next we change the assumption on f , requiring it to be quasiconformal. In
this case, we only deal with discs and annuli. First observe that a continu-
ous extension to the boundary may not exist a priori. Indeed Carathéodory’s
Theorem does not apply in this more general setting. However, the extension
does exist, and can be obtained as a simple and beautiful application of the
Integrability Theorem in Section 1.4. The proof can be found for example in
[Hu2, Sect. 4.9], but we also include it here.
Theorem 2.17 (Homeomorphic extension of a quasiconformal self-map

of D) Every K-quasiconformal homeomorphism f : D Ñ D (resp. f : H Ñ
H) extends continuously as a homeomorphism from S1 to S1 (resp. R
p to R).
p
Proof Suppose f : H Ñ H is K-quasiconformal, and let μf denote the Bel-

trami coefficient in H induced by f . Extend μf to C by reflection as follows:
$
&μf pzq
’
’ if Impzq ą 0,
μpzq :“
’
’
%μ pzq
f if Impzq ă 0.
Then μ is symmetric with respect to R and has dilatation bounded by K. Let

φ : pC, μq Ñ pC, μ0 q, fixing 0, 1, be the quasiconformal homeomorphism,
which integrates μ. Observe that it must preserve the real line because of sym-
metry (compare with Exercise 1.4.1). Now consider the map g : pH, μ0 q Ñ
pH, μ0 q defined as gpzq :“ f ˝ φ ´1 . By Weyl’s Lemma (Theorem 1.14), g is
conformal. Applying the Schwarz Reflection Principle, g can be extended to a
conformal map gp : C Ñ C, hence a homeomorphism when restricted to R. We
therefore define the extension of f to R by setting
fp :“ gp ˝ φ : R
p Ñ R,
p
which is a homeomorphism as we wanted to prove.

The same result holds for the unit disc, as seen by conjugating with the
ì which maps H to D.
i
Möbius transformation T pzq “ zz´
Using a Riemann map, we obtain the following corollary.

Corollary 2.18 (Homeomorphic extension of a quasiconformal map on D)

If G is a Jordan domain and f : D Ñ G is quasiconformal, then f extends
continuously to a homeomorphism from D to G.
Proof Consider a Riemann map R : D Ñ G which extends to the boundary
because BG is locally connected. Apply the theorem above to the map R´1 ˝ f
and then compose the extensions.
In the proof of Theorem 2.17, we could have deduced right away that g
is a Möbius transformation and hence globally defined. We used instead the
Schwarz Reflection Principle to emphasize that it is not necessary to have the
original map defined on the whole upper half plane. Indeed, the following is a
corollary of the proof.
Corollary 2.19 (Homeomorphic extension of a quasiconformal map on
an annulus) Suppose U and V are one-sided neighbourhoods of S1 and
f : U Ñ V is a quasiconformal homeomorphism so that |f pzq| Ñ 1 when
|z| Ñ 1. Then f extends continuously to a homeomorphism of S1 to itself.
Consequently, any quasiconformal map from a standard annulus Ar,R to an
arbitrary annulus A extends to a homeomorphism from Ar,R to A.
A more delicate problem is to study the regularity of the homeomorphic
extension. The following proposition is a well-known result whose proof can
be found for example in [Ah5, p. 40] (see also Exercise 2.2.5).
Proposition 2.20 (Quasisymmetric extension of a quasiconformal self-map
of D) Let f : D Ñ D be quasiconformal. Then f extends continuously to a
quasisymmetric map fp : S1 Ñ S1 .
Combining all results and ideas above we have the following statement.
Corollary 2.21
(a) Let G Ă C be a quasidisc and f : D Ñ G a quasiconformal map. Then f
extends continuously as a quasisymmetric map fp : S1 Ñ BG.
(b) Let A be a quasiannulus with boundaries γ o and γ i , and let f : Ar Ñ A
be a quasiconformal homeomorphism. Then f extends continuously to the
boundaries as quasisymmetric maps.
2.2.1 Prove that there exists a homeomorphism from D to itself which does
not have a continuous extension to the unit circle.
Hint: Consider for instance f prei θ q “ rei pθ `s pr qq , where s : r0, 1q Ñ
r0, `8q denotes some increasing homeomorphism.
2.2.2 Let j , j “ 1, 2, 3, be three Jordan arcs which are backwards invari-

ant under translation by σ for some σ with Repσ q ą 0. Show that
the composition h2 ˝ h1 of two near translations: hj : j Ñ j `1 for
j “ 1, 2 is a near translation, and that the inverse of a near translation is
a near translation. Show that any σ -periodic map h : 1 Ñ 2 is a near
translation.
2.2.3 Let 1 “ 2 “ R´ . Give two examples of diffeomorphisms h : 1 Ñ
2 which are not near translations: one where |hpxq ´ x| is unbounded
as x tends to ´8 but the derivative condition is satisfied, and another
one where |hpxq ´ x| is bounded but the derivative condition is violated.
2.2.4 Go through the proof of Proposition 2.15 in the case where the two strips
do not necessarily have the same height. In other words, substitute the
ă by ă , where the height m ą 0 is arbitrary. As before
strip 1,γ 2 m,γ2
let f denote the unique Riemann mapping between the interior of the
half strips whose continuous extension to the boundaries maps corners
to corners. Observe that we then need two different exponential maps
π
1 pzq :“ eπ z and 2 pzq :“ e m z in order to define the map F . Set F ˝
1 “ 2 ˝ f .
Prove that only for m “ 1, the mappings frj , j “ 1, 2, are near transla-
tions. More precisely, prove that:
(a) f 1 pzq Ñ m as z Ñ ´8,

(b) eπ rf1 px q´x s Ñ F 1 p0q ¨ lim1 px qÑ0` rF p1 pxqqsm´1 as x Ñ ´8,
r
and
(c) eπ rf2 px q´x s Ñ F 1 p0q ¨ lim1 px qÑ0` r´F p´1 pxqqsm´1 as
r
x Ñ ´8.
Deduce that |frj pxq ´ x| is unbounded as x Ñ ´8 if and only if m ‰ 1.

2.2.5 Theorem 1 in [Ah5, p. 40] states that every K-quasiconformal self–map
of H, extends continuously to R as a quasisymmetric homeomorphism.
Deduce Proposition 2.20 from this result.
Hint: Lift to the upper half plane under an exponential map and
compare the (chordal) distance between two points of the circle, with the
length of the arc of S1 that joins them. Relate the latter to the Euclidian
distance between two points (angles) on the real line.
2.3 Extensions of boundary maps

We now proceed to consider the boundary value problem, first for homeomor-
phisms of the unit circle and later for general curves. Given an orientation pre-
serving homeomorphism f : S1 Ñ S1 we want to find a continuous extension
fp : D Ñ D such that fp “ f on S1 . Clearly an extension, which is a homeo-

morphism, always exists: just extend radially fppre2π i t q :“ rf pe2π i t q). How-
ever, this extension has very bad properties at z “ 0, since the whole varia-
tion of fp is concentrated at this point. There are many other possible ways
to extend f to a homeomorphism of the disc to itself. The most sophisti-
cated extensions by Beurling–Ahlfors or by Douady–Earle have very good
properties.
Before addressing the general boundary value problem, we consider the
easier cases where the boundaries are circles and the boundary maps are
differentiable. Then the extensions can be constructed by interpolation without
referring to more sophisticated tools.
2.3.1 Extensions of C 1 -boundary maps

Strips and annuli We start by considering the annulus case.
Lemma 2.22 (C 1 -interpolation on standard annuli)
(a) Suppose 0 ă r1 , r2 ă 1 and f1 : S1 Ñ S1 and f2 : S1r1 Ñ S1r2 are two

orientation preserving C 1 -diffeomorphisms. Then there exists a C 1 - dif-
feomorphism f : Ar1 Ñ Ar2 , which equals f1 on S1 and f2 on S1r1 . In
particular, f is quasiconformal.
(b) Moreover, suppose r2 pλq is a continuous function of a complex parameter
λ P and that the C 1 -diffeomorphisms f2λ : S1r1 Ñ S1r pλq are such that
2
for any fixed z P S1r1 the map λ ÞÑ f2λ pzq is continuous. Then for any fixed
z P Ar1 the map λ ÞÑ fλ pzq is continuous, where fλ is the quasiconformal
extension above.
Proof The extension f can be constructed by linear interpolation when lifted

to the universal covering spaces. Consider the exponential map pzq :“ e2π i z
which maps the horizontal strip r 1 :“ t0 ď Im z ď r11 “ ´ log r1
2π u onto the
1
annulus Ar1 , and likewise, the strip r 1 onto Ar2 with r21 “ ´ log r2
2π . Then
2
f1 and f2 can be lifted to maps F1 and F2 defined on the two respec-
tive boundaries of the horizontal strip, satisfying Fj pz ` 1q “ Fj pzq ` 1, for
j “ 1, 2. Let us write Fr1 pxq :“ F1 pxq and Fr2 pxq :“ F2 px ` i r11 q ´ i r21 , where
Frj pxq, j “ 1, 2, are strictly increasing C 1 -diffeomorphisms of R.
We define the extension map F on the strip as the one sending every vertical
segment with real part x linearly to the segment joining the points F1 pxq and
Ar1 f1 Ar2
f
r1 1 r2 1
f2

F2
i r11 x ` i r11
i r21 F2 px ` i r11 q
x ` iy F
r21 y
i F px ` i y q
r11
0 x 0 F1 p x q
F1
Figure 2.7 Sketch of the proof of Lemma 2.22: definition of the interpolating map
F : r 1 Ñ r 1 on the strips which projects to the interpolating map f : Ar1 Ñ Ar2
1 2
on the annuli with boundary values f1 and f2 .
F2 px ` i r11 q, i.e. scaling the imaginary part accordingly. More precisely (see
Figure 2.7),
r1 y
ˆ ˙
y y
F px ` i yq :“ 1 ´ 1 Fr1 pxq ` 1 Fr2 pxq ` i 21 . (2.6)
r1 r1 r1
It is easy to check that F is C 1 with the Jacobian

r21
ˆˆ ˙ ˙
y r1 y r1
pJac F qpx ` i yq “ 1 1 ´ 1 F1 pxq ` 1 F2 pxq ą 0
r1 r1 r1
for all x ` i y P r 1 .
1
From the fact that Fr1 and Fr2 are strictly increasing we get that F is injective.
To see that F is surjective observe that F preserves horizontal lines. Since the
image of a horizontal line in r 1 must be a connected set, the image is the
1
entire corresponding horizontal line in r 1 .
2
It follows that F : r 1 Ñ r 1 is a C 1 -diffeomorphism. Clearly F pz `
1 2
1q “ F pzq ` 1, and hence F projects to a C 1 -diffeomorphism f : Ar1 Ñ Ar2
satisfying ˝ F “ f ˝ and with boundary values f1 and f2 . Since f is C 1

on a compact set, f is quasiconformal.
Now suppose we have one fixed boundary map f1 : S1 Ñ S1 and other maps
f2 “ f2λ : S1r1 Ñ S1r pλq depending continuously on a parameter λ P . Then,
2
equation (2.6) becomes
r 1 pλqy
ˆ ˙
y y
Fλ px ` i yq :“ 1 ´ 1 Fr1 pxq ` 1 Fr2λ pxq ` i 2 1 .
r1 r1 r1
The mappings Fλ : r 1 Ñ r 1 pλq depend continuously on λ, and so do the
1 2
projections fλ : Ar1 Ñ Ar2 pλq satisfying ˝ Fλ “ f ˝ .
Remark 2.23 (Differentiability of higher order) Observe that if the original

boundary maps in Lemma 2.22 are C n , n ě 1, then the resulting map on the
annuli is also.
Rectangles and quadrilaterals Using the idea in the proof of the Interpolation
Lemma 2.22 above we can extend piecewise C 1 -boundary maps of rectangles
to the inside as explained in part (a) of the lemma below; in part (b) the
extension is generalized to quadrilaterals.
Lemma 2.24 (C 1 -interpolation on rectangles and quadrilaterals)
(a) Let R1 , R2 be two rectangles and let f : BR1 Ñ BR2 be an orientation

preserving piecewise C 1 -diffeomorphism, mapping corners to corners.
Then f extends continuously to a C 1 -diffeomorphism fp : R 1 Ñ R 2 . In
particular, the extension is quasiconformal.
(b) Let Qj :“ Qj pzj,1 , zj,2 , zj,3 , zj,4 q, j “ 1, 2, be two quadrilaterals with
piecewise C 2 boundary curves and let f : BQ1 Ñ BQ2 be an orientation
preserving piecewise C 1 -diffeomorphism, mapping z1,k to z2,k . Then f
extends continuously to a C 1 -diffeomorphism fp : Q1 Ñ Q2 . In particular,
the extension is quasiconformal.
Proof
(a) Compare with Figure 2.8. Assume without loss of generality that R j “
r0, aj s ˆ r0, bj s for j “ 1, 2. The extension of f is constructed so that
every vertical segment with real part x is mapped to the segment joining
f pxq and f px ` i b1 q, and every horizontal segment with imaginary part
y is mapped to the segment joining f pi yq and f pa1 ` i yq. It is easy to
check that the extension is invertible and that both fp and its inverse are C 1 .
f px ` i b1 q
i b2
x ` i b1
i b1
f pi yq
iy x ` iy a1 ` i y fppx ` i yq
f pa1 ` i yq
0 x a1 0 f pxq a2
Figure 2.8 Illustrating how the map fp : R 1 Ñ R 2 is constructed by mapping line

segments to line segments.
Since the closures of the rectangles are compact the map is quasiconfor-
mal. (see Exercise 2.3.4).
(b) Choose rectangles Rj of the same modulus as Qj and similar to those
above, and let Rj : Rj Ñ Qj denote the unique Riemann mapping which,
when extended continuously to the boundary, maps vertices to vertices.
From Remark 2.12 it follows that the maps R p j : BRj Ñ BQj are piece-
wise C -diffeomorphisms. By applying part (a) to the boundary map
1
g :“ R p ´1 ˝ f ˝ R
p 1 we obtain a C 1 -extension gp : R 1 Ñ R 2 , so that fp “
2
´ 1
Rp 2 ˝ gp ˝ R
p is a C 1 -extension of f .
1
The unit disc Using the C 1 -interpolation on annuli (Lemma 2.22), we can
solve the extension problem to the disc in the differentiable case.
Proposition 2.25 (Piecewise C 1 -extension to D) Let f : S1 Ñ S1 be an ori-

entation preserving C 1 -diffeomorphism. Then there exists an extension fp :
D Ñ D, which is a piecewise C 1 -diffeomorphism. In particular, the map fp is
quasiconformal on D, and it may be chosen to equal the identity on a disc
around z “ 0.
Proof Choose any 0 ă r ă 1 and define fppzq :“ z for all z P Dr . Extend fp

to the annulus Ar as in Lemma 2.22, with boundary values f1 “ f on S1 and
f2 “ Id on S1r .
We remark that as r approaches 1, the quasiconformal constant may get larger

and larger.
Half strips In the proof of Lemma 2.22 (C 1 -interpolation on annuli) we obtain

a quasiconformal map from one horizontal strip onto another as an extension
of two boundary maps. These boundary maps are lifts of circle maps and hence
invariant under translation by 1. We shall need a slight generalization of this
lemma where we consider half horizontal strips and horizontal boundary maps
which are near translations.
Consider two half strips of heights r1 , r2 that are bounded by Jordan arcs
γ1 , γ2 respectively, that is
răj ,γj “ tz “ x ` i y | 0 ď y ď rj , x ď Re γj pyqu,
where the Jordan arcs are parametrized by C 2 -functions with y P r0, rj s and so
that Im γj pyq is strictly increasing from Im γj p0q “ 0 to Im γj prj q “ rj .
Lemma 2.26 (C 1 -interpolation on half strips) Let
fr1 : p´8, γ1 p0qs Ñ p´8, γ2 p0qs
and
fr2 : p´8, Re γ1 pr1 qs Ñ p´8, Re γ2 pr2 qs
be increasing C 2 -diffeomorphisms, which are near translations. The boundary

map from Bră1 ,γ1 to Bră2 ,γ2 consisting of the three diffeomorphisms f1 pxq “
fr1 pxq, f2 px ` i r1 q “ fr2 pxq ` i r2 , and fγ pγ1 pyqq “ γ2 p rr21 yq extends to a
quasiconformal mapping f : ră1 ,γ1 Ñ ră2 ,γ2 .
Proof Choose x0 ď γ1 p0q so that 1ă “ tz “ x ` i y | 0 ď y ď r1 , x ď

x0 u is contained in ră1 ,γ1 and 2ă “ tz “ x ` i y | 0 ď y ď r2 , x ď p1 ´
y r y r
r1 qf1 px0 q ` r1 f2 px0 qu is contained in r2 ,γ2 . Define f : 1 Ñ 2 by
ă ă ă
ˆ ˙
y r y r2
f px ` i yq “ 1 ´ f1 pxq ` fr2 pxq ` i y.
r1 r1 r1
This is a C 2 -diffeomorphism between the non-compact horizontal half strips
extending the boundary map restricted to the boundary of 1ă . Its Jacobi matrix
is
« ff
p1 ´ ry1 qfr11 pxq ` ry1 fr21 pxq r11 pfr2 pxq ´ fr1 pxqq
Dz f “ r2 .
0 r1
By assumption it follows that there exists a constant C ą 1 so that C1 ă

p1 ´ ry1 qfr11 pxq ` ry1 fr21 pxq ă C for all z P 1ă . The Jacobian therefore satisfies
r2 1 r2
r1 C ď det Dz f pzq ď r1 C, i.e. it is uniformly bounded away from 0 and 8.
Moreover, since all elements in the Jacobi matrix are uniformly bounded it fol-
lows that the norm }Dz f pzq} is uniformly bounded. Hence f : 1ă Ñ 2ă is
quasiconformal (recall Exercise 1.3.3 in Section 1.3). To obtain the full exten-
sion we consider the remaining quadrilaterals, i.e. the closures of răj ,γj zjă .
The boundary curves are piecewise C 2 and the boundary map is piecewise
C 1 , so we can extend quasiconformally to the interior, as stated in part (b)
of Lemma 2.24.
2.3.2 Extensions of quasisymmetric boundary maps

Simple approaches as in the preceding section do not work in the general case
when the boundary maps are not differentiable. However, it is remarkable that
if f is quasisymmetric on the unit circle, an extension can be constructed
which is quasiconformal and even real analytic in the interior of the disc, as
established in this section.
The unit disc The following important result deals with extensions to the unit
disc, of quasisymmetric maps of the unit circle, and is due to Beurling–Ahlfors
[BA] and Douady–Earle [DE] (cf. [Pom, Thm. 5.15] or [Hu2, Thm. 4.9.5 and
5.1.2]).
Theorem 2.27 (Quasiconformal extensions of quasisymmetric maps) Sup-

pose f : S1 Ñ S1 is an orientation preserving quasisymmetric homeomor-
phism. Then there is an extension fp : D Ñ D of f which is real analytic in
D and has the following properties:
(a) if σ, τ P MöbpDq then the extension of σ ˝ f ˝ τ is given by σ ˝ fp ˝ τ

(b) fp is quasiconformal in D.
The Beurling–Ahlfors extension does not satisfy (a) but it is simple to con-
struct. We shall extract some general facts about Beurling–Ahlfors extensions,
which are useful when extending quasisymmetric maps to annuli.
The Beurling–Ahlfors extension Given an M-quasisymmetric map f : R Ñ

R, which by definition satisfies equation (2.2) for all x, t P R, then (one of) the
(possible) Beurling–Ahlfors extension Bf of f to the upper half plane H is
given by
ż1
1
Bf px ` i yq :“ tf px ` tyq ` f px ´ tyqudt
2 0
ż1
ì tf px ` tyq ´ f px ´ tyqudt,
0
for all x P R and y ě 0. Clearly its restriction to R is precisely f . Other vari-

ants of the extension are obtained by keeping the real part of Bf and changing
the imaginary part by multiplying it by any positive constant. Compare with
[LV, p. 83], where the constant is 1{2. Our normalization is chosen so that, as
we shall see, the horizontal line through i is invariant under Bf when f is a
lift of a quasisymmetric map from the circle T “ R{Z onto itself.
The Beurling–Ahlfors extension Bf can be rewritten as
ż x `y ˆż x ` y żx ˙
1 1
Bf px ` i yq “ f psq ds ` i f psq ds ´ f psq ds ,
2y x ý y x x ý
which shows that it is continuously R-differentiable, using that f 1 pxq exists

almost everywhere since f is quasisymmetric. By direct estimation, using the
quasisymmetry of f , one can check that Bf is K-quasiconformal with a K
which depends only on M.
If f is a lift of a quasisymmetric map from the circle T onto itself, then it
satisfies f px ` 1q “ f pxq ` 1. In this case Bf pz ` 1q “ Bf pzq ` 1. Further-
more, one can check that the imaginary part of Bf px ` i q is
żx
tf pt ` 1q ´ f ptqu dt “ 1,
x ´1
so that the horizontal line tIm z “ 1u is invariant. It also follows that f pxq “
x ` θ pxq, where θ pxq is a periodic function of period 1, i.e. θ px ` 1q “ θ pxq.
We have that the real part of Bf px ` i q is
ż x `1
1
pt ` θ ptqq dt “ x ` α,
2 x ´1
where α is the integral of θ over one period. The map x ` i ÞÑ Bf px ` i q is

therefore the translation Tα by α on that horizontal line.
Standard annuli A variation of the Beurling–Ahlfors extension can be used

to interpolate between quasisymmetric boundary maps of standard annuli, as
we show in the following proposition.
Proposition 2.28 (Extension of quasisymmetric maps between boundaries

of standard annuli) Let 0 ă r1 , r2 ă 1 and suppose f1 : S1 Ñ S1 and f2 :
S1r1 Ñ S1r2 are two orientation preserving quasisymmetric homeomorphisms.
Then there exists an extension f : Ar1 Ñ Ar2 which is quasiconformal in the
interior.
Proof We use the notation in the proof of Lemma 2.22. First we lift the
annuli to the strip 3 :“ tz | 0 ď Im z ď 3u. We denote by r1 : 3 Ñ Ar1
and r2 : 3 Ñ Ar2 the respective projections, which consist of the exponen-
tial composed with a rescaling of the imaginary part. Observe that these
projections are not conformal but locally quasiconformal.
The maps f1 and f2 lift to maps on the boundaries of the strip denoted by
F1 and F2 respectively. Let Fr1 and Fr2 denote the corresponding real mappings
defined so that Fr1 pxq :“ F1 pxq and Fr2 pxq :“ F2 px ` 3i q ´ 3i . Further, we
may write Frj pxq “ x ` θj pxq for j “ 1, 2, where θj are periodic functions of
ş1
period 1. In both cases, set αj “ 0 θj ptq dt.
We define the extension map F in three substrips of 3 : one in the strip
1 “ tz | 0 ď Im z ď 1u, another in 1,2 “ tz | 1 ď Im z ď 2u and a third
in 2,3 “ tz | 2 ď Im z ď 3u (see Figure 2.9). The idea is as follows: On
the bottom substrip 1 we define F :“ B Fr1 , which satisfies F px ` i q “
x ` α1 ` i . On the top substrip 2,3 we define an appropriate modification of
the Beurling–Ahlfors extension of Fr2 , namely F pzq :“ c ˝ B Fr2 ˝ c´1 , where
c is the map cpzq “ z ` 3i . Hence F px ` 2i q “ x ` α2 ` 2i .
Finally, on the middle strip 1,2 we extend F by an affine map, which
fits with the boundary values x ` i ÞÑ x ` α1 ` i and x ` 2i ÞÑ x ` α2 ` 2i
on its lower and upper boundary respectively. The three pieces then match
continuously and are quasiconformal on their domains. Hence the map F is
quasiconformal on 3 . Since F satisfies F pz ` 1q “ F pzq ` 1, it projects to
f : Ar1 Ñ Ar2 satisfying ´ 1
r2 ˝ F “ f ˝ r1 . We have obtained a quasicon-
formal map on the annulus with boundary values f1 and f2 as desired.
Quasidiscs and quasiannuli Next, we consider similar extension problems

from given boundary maps, where the domains are quasidiscs or quasiannuli.
First we generalize the definition of a quasisymmetric map.
Definition 2.29 (Third definition of quasisymmetry) For j “ 1, 2, let γj

be two quasicircles, Gj the corresponding Jordan domains, and Rj : D Ñ
Gj two Riemann mappings, giving rise to the conformal parametrizations
R
p j : S1 Ñ γj . An orientation preserving homeomorphism f : γ1 Ñ γ2 is
Ar1 f1 Ar2
f
r1 1 r2 1
f2
r1 r2
F2 px ` 3i q “ Fr2 px q ` 3i
3i 3i
B Fr2 x ` α2
2,3 x0 ` 2i
2i 2i x0 ` α2 ` 2i
F
1,2
i i
x ì x0 ` α1 ` i
1 B Fr1 0 x ` α1
0 0
F1 “ Fr1
Figure 2.9 The quasiconformal extension F is defined in three pieces with boundary
values F1 , Tα1 , Tα2 and F2 on the horizontal lines of height 0, 1, 2 and 3 respectively.
On 1 and 2,3 , the extension F is the Beurling–Ahlfors extension, appropriately
defined. On 1,2 we interpolate linearly between Tα1 and Tα2 .
quasisymmetric if and only if f ˝ R

p 1 : S1 Ñ γ2 is quasisymmetric, and equiv-
´ 1
p ˝f ˝R
alently if and only if R 2
p 1 : S1 Ñ S1 is quasisymmetric.
Observe that the notion is independent of the choices of conformal parame-

trizations. The proof that the statements are equivalent is left as Exercise 2.3.2.
With this definition, the following proposition is straightforward.
Proposition 2.30 (Extension of quasisymmetric maps between boundaries

of quasidiscs and quasiannuli)
(a) Suppose G1 and G2 are quasidiscs bounded by γ1 and γ2 , and let f :

γ1 Ñ γ2 be quasisymmetric. Then f extends to a quasiconformal map
fp : G1 Ñ G2 .
(b) For j “ 1, 2, suppose Aj are open quasiannuli bounded by the quasicir-
cles γji , γjo . Let f i : γ1i Ñ γ2i and f o : γ1o Ñ γ2o be quasisymmetric maps
between the inner and outer boundaries respectively. Then there exists
a quasiconformal map f : A1 Ñ A2 extending the boundary maps f i
and f o .
Proof Start by choosing Riemann mappings Rj : D Ñ Gj . Their extensions

p j : S1 Ñ γj are quasisymmetric, and so is the composition
to the boundaries R
´ 1
g :“ R p ˝f ˝R
2
p 1 : S1 Ñ S1 . It follows from Theorem 2.27 that g can be
extended to a quasiconformal map gp : D Ñ D. The map fp “ R2 ˝ gp ˝ R´1 : 1
G1 Ñ G2 is quasiconformal and is an extension of the boundary map f (see
also Figure 2.10).
f
γ1 γ2
G1 fp G2
R
p1 R1 R2 R
p2
D gp D
Figure 2.10 Sketch of the proof of Proposition 2.30(a). Quasisymmetric maps

between quasicircles extend to a quasiconformal map between the interior domains.
In the proof of (b) start by choosing ϕj : Arj Ñ Aj to be conformal iso-

morphisms where rj is uniquely determined by the moduli of the annuli.
Let ϕpjo : S1 Ñ γjo and ϕpji : S1rj Ñ γji be the continuous extensions of ϕj to
the outer and inner boundaries respectively. It follows from Corollary 2.13(a)
that these maps are quasisymmetric, and therefore it follows from Proposi-
tion 2.28 that there exists a quasiconformal map g : Ar1 Ñ Ar2 extending the
boundary maps g o :“ pp ϕ2o q´1 ˝ f o ˝ ϕp1o and g i :“ pp
ϕ2i q´1 ˝ f i ˝ ϕp1i . Hence
´1
f :“ ϕ2 ˝ g ˝ ϕ1 : A1 Ñ A2 is quasiconformal and is a continuous extension
of the boundary maps f o and f i (see also Figure 2.11).
Remark 2.31 In both cases, if the curves are C 2 and the maps are C 1 , then
the extensions can also be (piecewise) C 1 in G1 or A1 respectively. See
Theorem 2.9, Corollary 2.13, Proposition 2.25 and Lemma 2.22.
We end this section by considering an interpolation depending on parame-
ters, analogous to the case in Lemma 2.22.
Proposition 2.32 (Quasiconformal interpolation on quasiannuli) For λ P

a complex parameter, let A1 and Aλ2 be bounded annuli of finite modulus,
fo
A1
γ1i γ2i
f
γ1o γ2o
A2
fi
ϕp1i ϕp1o ϕp2o ϕ2i

ϕ1 ϕ2
go
Ar1 Ar2
g
r1 1 r2 1
gi
Figure 2.11 Sketch of the proof of Proposition 2.30(b). Quasisymmetric maps

between quasiannuli boundaries extend to quasiconformal maps of the annuli.
bounded by quasicircles γ1o , γ2o , γ1i and γ2i pλq, where the boundary γ2i pλq
depends continuously on λ. Let fλi : γ1i Ñ γ2i pλq and f o : γ1o Ñ γ2o be qua-
sisymmetric between the inner and outer boundaries respectively such that,
for any fixed z P γ1i , the map λ ÞÑ fλi pzq is continuous. Then there exists a
quasiconformal map fλ : A1 Ñ Aλ2 with boundary values fλi and f o , such
that for any fixed z P A1 , the map λ ÞÑ fλ pzq is continuous.
To prove this proposition we use the following lemma, which we prove at

the end of the section.
Lemma 2.33 (Continuity of the modulus and uniformizing maps) Let

Ă C be open and bounded. Let tAλ uλP be a family of Jordan annuli, with
common outer boundary γ o and inner boundary γλi depending continuously on
λ. Let 0 ă rpλq ă 1 be so that Aλ – Ar pλq . Then rpλq depends continuously
on λ and consequently so does the modulus of Aλ .
Moreover, choose an arbitrary point z0 P γ o and normalize the uniformiz-
ing maps ϕλ : Ar pλq Ñ Aλ so that their extensions to the boundaries satisfy
ϕpλ p1q “ z0 for all λ P . Then λ ÞÑ ϕλ pzq depends continuously on λ for any
fixed z whenever it makes sense.
Proof of Proposition 2.32 Let r1 and r2 pλq denote the real numbers such
that A1 – Ar1 and Aλ2 – Ar2 pλq respectively. Since γ2o is fixed and γ2i pλq
depends continuously on λ so does r2 pλq due to the first part of Lemma 2.33.
Choose conformal isomorphisms ϕ1 : Ar1 Ñ A1 and ϕ2λ : Ar2 pλq Ñ Aλ2 . These
mappings extend to the boundaries. Normalize the mappings ϕ2λ so that the
extensions satisfy ϕp2λ p1q “ z0 for some chosen point z0 P γ2o . Then, due
to the second part of Lemma 2.33, the map λ ÞÑ ϕ2λ pzq depends continu-
ously on λ for any fixed z whenever it makes sense. As in the proof of
Proposition 2.30 define the induced boundary maps from the boundaries
of Ar1 to the boundaries of Ar2 pλq . Then apply Lemma 2.22 to obtain the
quasiconformal extension gλ : Ar1 Ñ Ar2 pλq , depending continuously on λ.
Finally, compose with the uniformizing maps ϕ1´1 and ϕ2λ and the statement
follows.
We now proceed to prove Lemma 2.33. This lemma can be deduced from
[Com, Thm. 3.2] (see also [Ep, Lem. 6]), where the convergence of bounded
multiply connected pointed domains in C and the relation to their normalized
uniformizing maps is treated in general. However, because of simplicity, we
have chosen to give a direct proof in the doubly connected case. 1
Proof of Lemma 2.33 Fix any λ P , and let λn be a sequence of parameters

in that converges to λ. We shall suppress λ from the notation and set
γni :“ γλin , γ i :“ γλi , An :“ Aλn , A :“ Aλ , rn :“ rpλn q, r :“ rpλq
and
ϕn :“ ϕλn : Arn Ñ An , ϕ :“ ϕλ : Ar Ñ A, with ϕn p1q “ z0 “ ϕp1q.
It follows from the assumptions that γni converges to γ i in the Hausdorff

topology. To prove the lemma it suffices to prove that rn converges to r and
that ϕn converges to ϕ uniformly on every compact subset of Ar as n tends to
infinity.
Choose an arbitrary r 1 P pr, 1q. Then ϕpAr 1 q Ă An for n ě N pr 1 q, where
Npr 1 q is some constant large enough. Then the composition
ψn :“ ϕn´1 ˝ ϕ : Ar 1 Ñ Arn .
is well defined and univalent on Ar 1 . The extended boundary map on the outer
boundary is mapping S1 homeomorphically onto itself, fixing 1. It follows that
1 We thank Xavier Buff for providing this proof.

mod Arn ě mod Ar 1 and therefore that rn ď r 1 for all n ě N pr 1 q. Since r 1 P

pr, 1q is arbitrary, we have lim supn rn ď r.
Extracting a subsequence if necessary we may assume that rn tends to
r ˚ P r0, rs as n tends to infinity and that ϕn : Arn Ñ An converges uniformly
on every compact subset of Ar ˚ to some map ϕ8 : Ar ˚ Ñ A. The latter is
obtained by the following argument. On any open subset compactly contained
in Ar ˚ , the sequence tϕn u forms a normal family (see Montel’s Theorem 3.23).
Hence, it has a convergent subsequence. A diagonal argument shows that there
exists a subsequence that converges uniformly on compact subsets of Ar ˚ .
In order to prove that r ˚ “ r and ϕ8 “ ϕ we consider again the mappings
ψn “ ϕn´1 ˝ ϕ and extend them by reflection in S1 to
ψn : Ar 1 ,1{r 1 Ñ Arn ,1{rn .
The restriction of ψn to S1 is a homeomorphism, fixing 1. Recall that ψn is

defined on Ar 1 ,1{r 1 for all n ě N pr 1 q where r 1 P pr, 1q is arbitrary. By choosing
a decreasing sequence rk1 Ñ r as k Ñ 8 and restricting to a subsequence nk ě
Nprk1 q with nk Ñ 8 as k Ñ 8, we have
ψnk : Ar 1 ,1{r 1 Ñ Arn .

k k k ,1{rnk
Extracting once again a subsequence if necessary, we may assume that the

sequence ψnk converges uniformly on every compact subset of Ar,1{r to
ψ8 : Ar,1{r Ñ Ar ˚ ,1{r ˚ .
The restriction of ψ8 to S1 is a homeomorphism, fixing 1. Moreover, it is

univalent on Ar,1{r . On Ar the map ϕ8 ˝ ψ8 is the limit of ϕnk ˝ ψnk “
ϕ : Ar 1 Ñ Ank , and therefore ϕ8 ˝ ψ8 “ ϕ : Ar Ñ A. In particular, ϕ8 :
k
Ar ˚ Ñ A is not constant, thus univalent. It follows that mod Ar ˚ ď mod A
and therefore that r ˚ ě r, hence r ˚ “ r. Moreover, ψ8 : Ar,1{r Ñ Ar,1{r is an
isomorphism, fixing 1, and therefore equal to the identity. Since ϕ8 ˝ ψ8 “ ϕ
on Ar we conclude that ϕ8 “ ϕ : Ar Ñ A as required.
2.3.1 (a) In the case where f1 : S1 Ñ S1 and f2 : S1r1 Ñ S1r2 are two ori-
entation preserving C 1 -maps of degree n, construct a C 1 -extension
f : Ar1 Ñ Ar2 which is a covering map of degree n. Compare with
Lemma 2.22.
(b) Let Ar1 and Ar2 be two annuli, compactly contained in C and
whose inner and outer boundaries are C 2 curves. Suppose the inner
and outer boundary maps f i : BAir1 Ñ BAir2 and f o : BAor1 Ñ BAor2 are
given as orientation preserving C 1 -maps of degree n. Then construct a
C 1 -extension f : Ar1 Ñ Ar2 , which is a covering map of degree n (see
also Exercise 2.3.3).
2.3.2 Using Proposition 2.4, show that Definition 2.29 is independent of the
choices of conformal parametrizations, and that the two statements are
equivalent.
2.3.3 Let γjo for j “ 1, 2 and γ1i , and γ2i pλq be C 2 curves, with γ2i pλq depend-
ing continuously on λ, and being outer and inner boundaries of annuli
A1 and Aλ2 in C. Suppose fλi : γ1i Ñ γ2i pλq and f o : γ1o Ñ γ2o are given
as orientation preserving C 1 -maps of degree n. Then construct a C 1 -
extension fλ : A1 Ñ Aλ2 , which is a covering map of degree n, such that
for any fixed z P A1 , the map λ ÞÑ fλ pzq is continuous. Compare with
Proposition 2.32.
2.3.4 Write an explicit expression for the extension of the boundary map f :
BR1 Ñ BR2 in Lemma 2.24 to a map fp : R 1 Ñ R 2 , and check that it is
a C 1 -diffeomorphism.
3
Preliminaries on dynamical systems
and actions of Kleinian groups
In this book we apply the technique of quasiconformal surgery to holomorphic

dynamical systems, or more precisely to the dynamical systems generated by
the iteration of a holomorphic map
f : S Ñ S,
where S is a Riemann surface, i.e. a complex manifold of (complex) dimension
1. For any given initial condition z0 P S we consider the (forward) orbit of z0
under f , that is the sequence
Opz0 q “ O` pz0 q “ tz0 , z1 “ f pz0 q, . . . , zn “ f pzn´1 q “: f n pz0 q, . . .u.
To understand the dynamical system generated by the iterates of f means to
understand the fate of all orbits, in terms of their initial condition, i.e. their
asymptotic behaviour when the time n tends to infinity. To fix notation, we
mention some examples of long-term behaviour. Under f , the point z0 (or its
orbit) can be fixed if f pz0 q “ z0 ; periodic of period p ą 1, or p-periodic, if
f p pz0 q “ z0 and p is the smallest number for which this occurs; preperiodic if
f k pz0 q is periodic for some k ą 0, but z0 is not; ‘converging’ if f np pz0 q Ñ z˚
as n Ñ 8 for some p ě 1 (then, z˚ is p-periodic if p is minimal with respect
to this property). A p-periodic orbit is also called a p-cycle.
A point z0 P S is called recurrent if the orbit of z0 intersects any arbitrarily
small neighbourhood of z0 . Likewise, U Ă S is a recurrent set if f n pU q
intersect U for infinitely many values of n.
As a consequence of Montel’s Theorem 3.23, the Uniformization Theo-
rem 1.22, and the study of dynamics on D and on the torus, it can be shown
that this theory is interesting and non-trivial in only three cases (up to normal-
ization) [Mi1, §2,5 and 6], namely C p (rational maps), C (entire transcendental
maps) and C “ Czt0u (transcendental self-maps of the punctured plane).
˚
92
Preliminaries on dynamical systems and actions of Kleinian groups 93
Although polynomials are entire maps we classify them here as particular

examples of rational maps since they can be extended to the Riemann sphere.
In contrast, entire transcendental maps have an essential singularity at infinity
and cannot be extended to the whole sphere. The exponential map and the
elementary trigonometric functions are examples. Transcendental self-maps
of C˚ have two essential singularities, one at infinity and one at the origin.
Examples of such are ez`1{z or complexifications of certain maps of the
circle, as we will see later. Meromorphic transcendental maps (for example
the tangent map) also present interesting dynamics. The basic definitions take
different forms for these maps, since orbits of poles and prepoles are truncated.
For this reason, we shall not include them in the general exposition although
we might occasionally treat them separately (for example in Chapter 5).
A holomorphic map f : C p ÑC p is a rational map. Hence f pzq “ P pzq ,
Qpzq
where P and Q are polynomials with no common factors. The degree of f
is defined as
d “ maxtdegpP q, degpQqu.
This is then the topological degree of f in the sense that every point in C
p has
exactly d preimages under f when counted with multiplicity. For d “ 1 the
map f is a Möbius transformation and it is a conformal isomorphism. Inter-
esting dynamics on C p require a rational map of degree d ě 2. The topological
degree of a transcendental map is always infinite.
To simplify notation we name the following classes of maps:
• Rat “ tf : C
p ÑC p | f is rational of degree at least twou;
• Ent “ tf : C Ñ C | f is entire transcendentalu;
• Ent˚ “ tf : C˚ Ñ C˚ | f is holomorphic transcendentalu;
• Mer “ tf : C Ñ C p | f is transcendental meromorphic with at least one
pole, which is not omitted u.
A subset of Rat is the class of polynomials which we denote by Pol.

For any degree d ě 2 we shall also consider the classes Ratd and Pold ,
which denote respectively rational maps and polynomials of degree d.
Holomorphic dynamical systems are special cases of discrete dynamical
systems f : X Ñ X, where f is only required to be continuous and X is any
topological or metric space. Occasionally we shall consider non-holomorphic
dynamics like, for example, iteration of functions on the unit circle (see Sec-
tion 3.2 below).
When we work on the Riemann sphere C p we consider its standard repre-
sentation as the unit sphere in R , with the complex plane C as the horizontal
3
94 Preliminaries on dynamical systems and actions of Kleinian groups
equator plane. Points in C are projected to the sphere by stereographic pro-

jection. On C
p we use the spherical metric and the chordal metric, which are
equivalent to each other. For completeness and later use we give an explicit
expression (see [Bea, §2.1]) for the chordal distance σ between two points in
C
p represented by corresponding points z, w P C, namely
2|z ´ w|
σ pz, wq “ , (3.1)
p1 ` |z|2 q1{2 p1 ` |w|2 q1{2
and
2
σ pz, 8q “ lim σ pz, wq “ . (3.2)
wÑ8 p1 ` |z|2 q1{2
A useful relation between σ and the spherical distance dCp is given by
2
d pz, wq ď σ pz, wq ď dCp pz, wq. (3.3)
π Cp
In this chapter we give a brief survey of the basic theory of holomorphic
dynamics and analytic circle maps focusing on results that are used in the book.
It is not our intention to reproduce the excellent texts which can be found in
the literature – instead we refer to them for proofs and for a more detailed
exposition than the one given here.
The last section contains a brief survey of the actions of Kleinian groups
and the Sullivan dictionary, summarizing the striking analogies between holo-
morphic dynamics and the dynamics of Kleinian groups. Quasiconformal
deformations were used in the context of Kleinian groups long before they
appeared in holomorphic dynamics. The section should give the reader suffi-
cient background to appreciate the references to the analogies that appear in
several places in the book.
As general standing references for the basic theory of holomorphic dynam-
ics we recommend [Bea, CG, Mi1, MNTU, St] and also [Bl]; for the back-
ground on the iteration of entire transcendental maps (and also meromorphic
maps) [Ber, EL1] and [HY]. For more advanced issues we refer to the books
and collections [Dev1, FSY, HP, K, Kri, McM2, RS, Sc] and [Ta2]. References
for Kleinian groups are mentioned in the dedicated section.
3.1 Conjugacies and equivalences

Conjugacies are the main tool used for comparison and classification of
dynamical systems. In this section, X and Y are topological or metric
spaces and f : X Ñ X and g : Y Ñ Y are continuous maps, unless otherwise
3.1 Conjugacies and equivalences 95
specified. Most of the facts which follow can easily be checked, and those
which are not obvious can be found for example in [Ro].
Definition 3.1 (Conjugate dynamical systems) We say that f and g are

topologically conjugate if there exists a homeomorphism h : X Ñ Y such that
h ˝ f “ g ˝ h, i.e. the following diagram commutes:
f
X ÝÝÝÝÑ X
§ §
§ §
hđ đh
g
Y ÝÝÝÝÑ Y
h
We write f „ g, or f „ g if the topological conjugacy h needs to be speci-
top top
fied.
If h can be chosen to be C r with 0 ă r ď 8 (resp. linear, affine) we say
that f and g are C r -conjugate (resp. linearly conjugate, affine conjugate). If
X, Y Ă C p we shall also use the terms conformally (resp. quasiconformally)
conjugate with the obvious definition. These definitions also make sense locally.
Observe that when X “ Y “ C, conformal conjugacies are affine

conjugacies.
Conjugate systems are ‘the same’ up to a change of variables. The regularity
of the conjugacy makes this identification more or less restrictive.
Conjugacies preserve orbits. Indeed, if f „ g then f n „ g n . Therefore,
top top
Of px0 q is mapped bijectively onto Og phpx0 qq by h. In particular, periodic
orbits are mapped onto periodic orbits of the same period.
Properties which are preserved under conjugacies are important and receive
a special name.
Definition 3.2 (Conjugacy invariants) A property or a quantity associated to

a dynamical system which is preserved under topological (resp. C r , quasicon-
formal, conformal, etc.) conjugacy is called a topological (resp. C r , quasicon-
formal, conformal, etc.) invariant.
An example of a C 1 -invariant is what is known as the multiplier of a periodic

orbit.
Lemma 3.3 Suppose X, Y Ă R and f, g P C 1 pRq or X, Y Ă C and f, g are

holomorphic. If h is a real (resp. complex) C 1 -conjugacy between f and g and
x0 is a p-periodic point then y0 :“ hpx0 q is p-periodic and
pf p q1 px0 q “ pg p q1 py0 q.
We shall see that many properties of dynamically defined sets are preserved
under topological conjugacy. As an example, we mention the following.
Definition 3.4 (Invariant sets) A set U Ă X is f -invariant or (forward)

invariant under f : X Ñ X if f pU q Ă U . We say that U is backward invariant
under f if f ´1 pU q Ă U . Finally, U is totally invariant or completely invariant
under f if f pU q “ U “ f ´1 pU q.
Lemma 3.5 Topological conjugacies between f and g map f -invariant sets

onto g-invariant sets. The same statement is true for backwards and totally
invariant sets.
Some dynamical systems which are not conjugate may still be semi-
conjugate. This happens when there exists a continuous function h : X Ñ Y
which satisfies h ˝ f “ g ˝ h but is not a homeomorphism. When this is the
case, many properties are still transferred from f to g. For example, orbits
of f are mapped onto orbits of g, but a p-cycle of f may be mapped onto a
p1 -cycle of g, where p1 |p.
A weaker concept is that of equivalence between dynamical systems.
Definition 3.6 (Equivalent dynamical systems) We say that f and g are

topologically equivalent if there exist homeomorphisms h1 and h2 such that
the following diagram commutes:
f
X ÝÝÝÝÑ X
§ §
h1 đ
§ §h
đ 2
g
Y ÝÝÝÝÑ Y
Equivalences other than topological ones are defined in the obvious way.
Equivalences do not preserve orbits and therefore the dynamics of f and

g may be very different. However, if f and g are topologically equivalent,
there is a bijection between the critical points of f and the critical points of
g (a critical point is a point where a map fails to be locally injective), and the
same is true for critical values (images of critical points). As we shall see, the
number of critical points and values play a very important role in holomorphic
dynamics. In this context, being equivalent means, in some sense, to belong to
the same ‘family’ of maps.

3.1.1 Let f and g be topologically conjugate by a homeomorphism h. Show
that critical points of f (points where f is not locally injective) are sent
to critical points of g by h.
3.1.2 Let f : X Ñ X and g : Y Ñ Y be topologically conjugate by a home-
omorphism h. Suppose U Ă X is dense in X. Show that hpU q is dense
in Y . Deduce that if periodic points of f are dense in X then periodic
points of g are dense in Y .
3.1.3 Show that a polynomial of degree d ě 2
cd zd ` cd ´1 zd ´1 ` cd ´2 zd ´2 ` ¨ ¨ ¨ ` c1 z ` c0 , with ci P C
for i “ 0, . . . , d,
is centred (i.e. the sum of the critical points is equal to zero when
counted with multiplicity) if and only if cd ´1 “ 0. Show that any poly-
nomial of degree d ě 2 is globally conformally conjugate to a monic
(cd “ 1) centred polynomial. Note that for d ą 2 the coefficients are
not uniquely determined.
In particular, determine the affine conjugacy classes of monic centred
cubic polynomials.
3.1.4 Prove that the quadratic family tQc pzq “ z2 ` c | c P Cu has exactly
one representative in each affine conjugacy class.
3.2 Circle homeomorphisms and rotation numbers

One-dimensional systems generated by the iteration of maps from the unit
circle to itself play an important role in general dynamical systems. If a system
possesses a simple closed invariant curve, then the dynamics restricted to that
curve is conjugate to a circle map. For the content of this section we refer to
[Dev2, Sect. 1.14] and [dMvS, Chapt. I].
Consider the unit circle S1 “ tz P C | |z| “ 1u, or equivalently the quo-
tient space T “ R{Z, identified via the exponential map z “ e2π i x . Let π :
R Ñ T denote the projection π pxq “ x pmod 1q, and : R Ñ S1 be pxq “
e2π i x . These mappings induce a metric and an orientation on the circle. We
study orientation preserving homeomorphisms f : S1 Ñ S1 , or equivalently
f : T Ñ T when this is more convenient.
The simplest such maps are the rigid rotations Rθ with θ P R,
Rθ pzq “ e2π i θ z or equivalently Rθ pxq “ x ` θ pmod 1q.

For a rigid rotation, either θ P Q and then all points of S1 are periodic of the
same period, or θ P RzQ and no points are periodic. In fact something stronger
is true.
Theorem 3.7 (Jacobi’s Theorem) If θ P RzQ, all orbits under Rθ are dense
in S1 .
See Exercise 3.2.2. An orientation preserving homeomorphism of S1 can

always be lifted to an orientation preserving homeomorphism of R, via the
projection map.
Definition 3.8 (Lifts of homeomorphisms) Let f : S1 Ñ S1 be an orienta-

tion preserving homeomorphism. A continuous map F : R Ñ R is a lift of f if
˝ F “ f ˝ .
Note that this is a semi-conjugacy of infinite degree, since is an infinite

degree covering map.
As an example, observe that the maps Rk,θ pxq “ x ` θ ` k, for k P Z,
denote all the different lifts of the rigid rotation Rθ . This is a general fact:
every integer translation of a lift is again a lift, and vice versa. Moreover, if F
is a lift then F px ` 1q “ F pxq ` 1 for all x P R and F n is a lift of f n .
The rotation number is an important topological invariant associated to a
circle map. It measures the average asymptotic rate of rotation of points of the
circle.
Definition 3.9 (Rotation number) Let f : S1 Ñ S1 be an orientation preserv-

ing homeomorphism, and let F : R Ñ R be a lift of f . For x P R we define
the rotation number of f as
F n pxq ´ x
ˆ ˙
rotpf q “ lim pmod 1q.
nÑ8 n
This is a number in r0, 1q independent of the choices of F and x [Poi].
If f is not a homeomorphism the limit still exists, but different points could
result in different rotation numbers. Observe that the rigid rotations have
rotpRθ q “ θ pmod 1q.
The rotation number is very robust. It is in fact invariant both under C 0 -

perturbations of f (see e.g. [Dev2, Cor. 14.7]) and under topological conjuga-
cies (see Exercise 3.2.3).
It is straightforward to check that if f has a fixed point then rotpf q “ 0. In
fact, Poincaré showed that much more is true.
Proposition 3.10 (Rotation number and periodic points) Let f : S1 Ñ S1

be an orientation preserving homeomorphism, and choose F to be the lift of f
so that F p0q P r0, 1q.
(1) f has irrational rotation number if and only if f has no periodic points.
(2) rotpf q “ p{q with gcdpp, qq “ 1 if and only if f has a p-periodic point
z0 so that F q px0 q “ x0 ` p, with px0 q “ z0 .
One may ask whether every circle map with an irrational rotation number
is actually conjugate to a rigid rotation. This would imply that every circle
map with no periodic points has all of its orbits dense in S1 (since homeo-
morphisms – and therefore conjugacies – map dense sets onto dense sets, see
Exercise 3.1.2). The statement turns out to be false. There exist examples,
known as Denjoy counterexamples, of circle maps with irrational rotation
number, having wandering intervals (intervals which never meet themselves
under iteration). This prevents many orbits from being dense in S1 . Hence such
maps cannot be conjugate to a rigid irrational rotation.
It was Denjoy [Den] who in 1932 specified some additional conditions on
the regularity of f to ensure the existence of a conjugacy to a rigid rotation (cf.
[dMvS, I.2]).
Theorem 3.11 (Denjoy’s Theorem) Let f : S1 Ñ S1 be a diffeomorphism
which is C 2 and has an irrational rotation number θ . Then f is conjugate to
the rigid rotation Rθ .
We are especially interested in analytic circle maps with analytic conjuga-
cies to rigid rotations, because of the applications in holomorphic dynamics.
Definition 3.12 (Linearizable circle maps) We say that an analytic circle
diffeomorphism without periodic points is (analytically) linearizable if it is
analytically conjugate to a rigid rotation.
In 1961, Arnol’d [Ar1] found examples of analytic circle diffeomorphisms
without periodic points, whose conjugacy to an irrational rotation was not
even absolutely continuous. To obtain more regularity for the conjugacies it is
necessary to impose further conditions on the rotation number. To state these
results we need to introduce some special subsets of irrational numbers.
3.2.1 Continued fractions and subsets of irrational numbers

We collect some facts about continued fractions and define certain subsets of
irrational numbers which are relevant in the study of circle maps. For more
details we refer to [Kin, Lan, Mi1, PM].
For θ P pRzQq X p0, 1q, consider the continued fraction expansion

1
θ“ “ ra1 , a2 , a3 , . . .s
1
a1 `
1
a2 `
a3 ` ¨ ¨ ¨
where the coefficients an are uniquely determined positive integers depending
on θ . The nth convergent of θ is defined as the irreducible fraction
pn
:“ ra1 , a2 , . . . , an s.
qn
Each convergent pn {qn is the best approximation to θ by fractions with denom-

inator at most qn [Mi1, Thm. 11.8].
Definition 3.13 (Diophantine numbers) An irrational number θ is said to be

Diophantine of order ď k, or to belong to Dpkq, if there exists ą 0 such that
ˇ ˇ
ˇθ ´ p ˇ ą , for every rational number p .
ˇ ˇ
ˇ q ˇ qk q
For example, every algebraic number is Diophantine. The set D “
k ě2 Dpkq
Ť
of Diophantine numbers has full measure in p0, 1q. In fact,
Dp2`q “ k ą2 Dpkq has full measure in p0, 1q [Mi1, Lem. 11.7], while the
Ş
set Dp2q has measure zero [Mi1, Lem. C.8].

The set Dp2q plays an important role in dynamics (not only in holomorphic
dynamics and circle maps). It is known as the set of numbers of bounded type
(sometimes also referred to a constant type). The motivation for these names
comes from the alternative definition of Dpkq in terms of continued fractions
(cf. [Mi1, Cor. 11.9]).
Proposition 3.14 (Alternative definition of Dpkq) An irrational number θ

belongs to Dpkq if and only if qn`1 ă Cqnk ´1 for all n ą 0 and for some
constant C independent of n. In particular,
" *
qn`1
θ P Dp2q ô ă C ô tan un is bounded.
qn n
Irrational numbers which are not Diophantine are often called Liouville
numbers. It is a startling fact that the set of Liouville numbers (and many
proper subsets thereof) is both generic, i.e. is a countable intersection of open
dense sets, and at the same time extremely small – not only a set of measure
zero but also of Hausdorff dimension zero (cf. [Mi1, Lem. C.7]).
There are other subsets which play an important role in holomorphic dynam-
ics and circle maps. The set B of Bryuno numbers is a superset of D, and is
important in the problem of the linearization of fixed points of holomorphic
maps (see Section 3.3.2).
Definition 3.15 (Bryuno numbers) With tqn un as above, the set of Bryuno
numbers is defined as
# ˇ +
ˇ ÿ logpq
n`1 q
B “ θ P RzQ ˇ ă8 .
ˇ
ˇ n qn
Finally we mention that Yoccoz [Yo3, Yo4] discovered the so-called set of
Herman numbers H when studying the linearization of analytic circle maps.
This completed the work started by Herman. The definition is somewhat
involved and can be found in [PM, p. 249]. The relative positions of these
subsets of irrational numbers are
Dp2q Ă D Ă H Ă B Ă pRzQq.
3.2.2 Real analytic circle maps

In this section we concentrate on real analytic circle maps f : S1 Ñ S1 , with
an irrational rotation number, say θ , hence with no periodic points. Sometimes
we assume that f is a diffeomorphism while in other cases we consider critical
circle maps, i.e. real analytic circle homeomorphisms with a unique (double)
critical point.
Due to Denjoy’s Theorem, all analytic diffeomorphisms are conjugate to
a rigid rotation Rθ . We are interested in conditions that ensure a certain
regularity of the conjugacy such as quasisymmetry (see Definition 2.1), ana-
lyticity, etc.
The first result is due to Herman [He1] and Yoccoz [Yo3] and gives a precise
and optimal condition on the rotation number.
Theorem 3.16 (Analytic linearization) Let f : S1 Ñ S1 be a real analytic

circle diffeomorphism and θ “ rotpf q. If θ P H then f is analytically lineariz-
able. For an arbitrary θ R H, there exists a real analytic circle diffeomorphism
with that rotation number, which is not analytically linearizable.
For critical circle maps with irrational rotation number, a theorem of Yoccoz
in [Yo2] establishes that the conjugacy to the rigid rotation always exists. It is
not difficult to see that it can never be real analytic. This is in fact explained
using holomorphic dynamics (see Exercises 3.3.1 and 3.3.2 in Section 3.3).
However, in some cases we can obtain quasisymmetric conjugacies, as shown

in the Herman–Świa̧tek Theorem below (cf. [He3, Sw2] and [Pe2]).
Theorem 3.17 (Quasiconformal linearization for critical circle maps) Let

f : S1 Ñ S1 be a critical circle map with rotpf q “ θ . Then, θ is of bounded
type if and only if f is quasisymmetrically conjugate to the rigid rotation Rθ .
We shall see in Section 7.2 that quasisymmetric linearization is all we need

in order to perform surgery.
The theorem above has been generalized by Petersen [Pe3] for holomorphic
critical quasicircle maps, that is holomorphic maps defined on a neighbour-
hood of a quasicircle having a critical point on the curve (see the section below
for the definition of rotation number on a general curve).
3.2.3 Rotation number on invariant curves

We mentioned at the beginning of the section that circle maps arise in general
dynamics when studying invariant simple closed curves. The rotation number
is also defined for such invariant curves. Indeed, suppose γ is an oriented
Jordan curve (see the beginning of Chapter 2) on the Riemann sphere and f :
γ Ñ γ is a homeomorphism. Let ϕ : S1 Ñ γ denote a conformal parametriza-
tion of γ which must be at least a homeomorphism (see Section 2.2.1). Then,
fr :“ ϕ ´1 ˝ f ˝ ϕ is a homeomorphism of the unit circle, conjugate to f .
We define the rotation number of f on γ , denoted by rotpf, γ q, as the rota-
tion number of fr on the unit circle. Observe that this number is independent
of the chosen parametrization ϕ, because the rotation number is a topological
invariant.
Observe that higher regularity of the curve γ is equivalent to higher regular-
ity of the parametrization ϕ, since conformal parametrizations (boundary val-
ues of Riemann maps) are always ‘as good as they can be’ (see Section 2.2.1).
In most of our applications, f : γ Ñ γ is the restriction of a holomorphic
map defined on a neighbourhood of γ . In this context, the map fr is always
real analytic (Exercise 3.2.4). Moreover, if we require some condition on the
rotation number then the curve must also be real analytic and f analytically
linearizable, as shown by Ghys [Gh] in the following theorem.
Theorem 3.18 (Rotation numbers in H and linearizability) Suppose γ is a

Jordan curve and F is a univalent map defined on a neighbourhood of γ such
that F pγ q “ γ . Let f “ F |γ and θ “ rotpf, γ q P H. Then γ is analytic and
F and Rθ are conformally conjugate in a neighbourhood of γ .
3.2.4 Families of circle homeomorphisms

We end this section by considering families of circle maps in the sense of
Arnol’d and Herman. More precisely, let f : S1 Ñ S1 be a circle homeomor-
phism and define the family
fα pzq “ pRα ˝ f qpzq “ e2π i α f pzq, for α P r0, 1q,
with the following non-degeneracy condition
fαm ı Id |S1 for all α P r0, 1q and for all m ě 0. (3.4)
This is not a very restrictive condition. It is satisfied, for example, if f is a real
analytic homeomorphism of S1 that extends to a non-affine holomorphic map
of C˚ (cf. [dMvS, Lem. 4.3(I)]). This is in general the setup in the surgery
applications we consider.
Given fα as above, define the rotation function
ρpαq “ rotpfα q.
Since the rotation number varies continuously under C 0 -perturbations, ρpαq is
continuous. This function has many other remarkable properties. It is in the
same family as the one presented in Section 1.3.3 and is also called a devil’s
staircase (cf. [dMvS, Sect. I.4]) (see Figure 3.1).
Proposition 3.19 (Properties of the rotation functions) Let f, fα and ρpαq
be as above. Then:
(1) ρ : T Ñ T is continuous, orientation preserving and onto.
(2) If ρpα ˚ q P p0, 1qzQ, then ρ is strictly increasing at α ˚ . Otherwise ρ is
locally constant at α ˚ (maybe only on one side). More precisely, the set
ρ ´1 pr{sq has non-empty interior for every rational r{s P r0, 1q.
(3) The set tα P p0, 1q | ρpαq P p0, 1qzQu is nowhere dense.
An immediate consequence of the properties above is the following
statement.
Theorem 3.20 (Adjusting α to obtain a given rotation number) Let f :
S1 Ñ S1 be a circle homeomorphism and fα “ Rα ˝ f the associated family
of circle maps satisfying (3.4). Then, for any θ P r0, 1qzQ, there exists a unique
α P r0, 1q such that rotpfα q “ θ .
To end this section, we state a result related to the linearization problem.
This theorem is due to Herman [He2], who constructed exotic examples
showing that quasisymmetric conjugacies to rigid rotations cannot always be
upgraded.
ρ pα q
1.0
Ò
0.8
0.6
0.4
0.2
0.2 0.4 0.6 0.8 1.0Ñ α
Figure 3.1 A devil’s staircase, the graph of the rotation function for
fα pxq “ α ` x ` 2π 1 sinp2π xq pmod 1q.
Theorem 3.21 (Quasisymmetric linearization which is not C 2 ) Suppose

f : S1 Ñ S1 is a C 8 circle diffeomorphism and fα satisfies (3.2.1). Then there
exists α P r0, 1q such that fα is quasisymmetrically conjugate to an irrational
rigid rotation, but the conjugacy is not C 2 .
3.2.1 Assume θ P p0, 1q is a Bryuno number. Using the continued fraction
expansion and the expression for the denominators of the convergents
denominators prove that if 1 ´ θ is also a Bryuno number.
Hint: For θ P p0, 1{2q and θ “ ra1 , a2 , a3 , . . . , an , . . .s show that
1 ´ θ “ r1, a1 ´ 1, a2 , . . . , an´1 , . . .s. Conclude afterwards that if
tpn {qn u are the convergents of θ , and tpn1 {qn1 u are the convergents of
1 ´ θ , then qn1 “ qn´1 for n ě 2.
3.2.2 (Jacobi’s Theorem) Let θ P RzQ. Prove that all orbits under the rigid
rotation Rθ pxq “ x ` θ pmod 1q, are dense in T.
Hint: First show that every orbit is infinite and therefore there exist two
iterates which are arbitrarily close. Rotate to obtain k such that Rθk pxq is
arbitrarily close to x, and then iterate to obtain the denseness.
3.2.3 (Rotation numbers are topological invariants) Suppose f, g : S1 Ñ
S1 are orientation preserving homeomorphisms and f „ g. Prove that
top
rotpf q “ rotpgq.
Hint: Let F, G and H be lifts of f, g and h respectively, where h is the

H
conjugacy. Observe that F „ G. In the definition of rotpgq substitute
top
Gn by H ˝ F n ˝ H ´1 and add and subtract terms conveniently. Use that
H pxq ´ x is bounded for all x and that so is H ´1 pxq ´ x.
3.2.4 (Invariant curves under univalent maps) Let γ be a Jordan curve in
C
p and f a univalent map defined on a neighbourhood of γ such that
f pγ q “ γ . Show that fr : S1 Ñ S1 as defined in Section 3.2.3 is real
analytic.
Hint: Use the Schwarz Reflection Principle.
3.3 Holomorphic dynamics: the phase space

From now on we consider holomorphic mappings f : S Ñ S with the phase
space or dynamical space S equal to the Riemann sphere C,p the complex plane
C or the punctured plane C . Most of the basic theory of holomorphic dynam-
˚
ics, especially the local theory, is common for the three settings, although
the proofs may vary from one to another. Most texts deal with rational maps,
although there are survey references like [Ber, EL1, HY, MNTU], where the
differences with transcendental maps are treated. The exposition in this section
follows mainly the book by Milnor [Mi1] but see also [Bea] or [Ber].
3.3.1 The dynamical partition

A special property of holomorphic dynamical systems is the splitting of the
phase space induced by the concept of normal families.
Definition 3.22 (Normal family of holomorphic maps) Let U Ă C p be a

domain and F a family of holomorphic maps from U to C. We say that F
p
is a normal family in U if any infinite sequence of elements of F contains
a subsequence which converges uniformly on compact sets of U to some
limit map.
The condition of normality can be phrased in terms of equicontinuity, asking

the family to be locally equicontinuous in U with respect to the spherical
metric in C
p (cf. the Arzelà–Ascoli Theorem).
An easy way to check normality is to apply Montel’s Theorem.
Theorem 3.23 (Montel’s Theorem) Let U Ă C p be a domain and F a family

of holomorphic maps from U to C. If there exist three distinct points a, b, c P C
p p
such that f pU q Ă Czta,
p b, cu for all f P F, then F is a normal family in U .
In particular if |f pU q| ď R for all f P F and for some R ą 0, then F is a

normal family in U .
In holomorphic dynamics the concept of normality is applied to the family

of iterates of the given map f . If tf n uně0 is normal in a domain U , orbits of
points in U behave in a similar manner. The concept of normality defines the
dynamical partition explained in the following definition.
Definition 3.24 (Fatou set and Julia set) Given a holomorphic map f : S Ñ
S as above, we define the Fatou set of f as
Ff “ tz P S | tf n un forms a normal family in a neighbourhood of zu,
and the Julia set of f as its complement
Jf “ SzFf .
For f P Ent Y Ent˚ , the essential singularities are sometimes considered

part of the Julia set, which is then defined in C.
p
The Fatou set is open and the Julia set is closed in S, and both sets are totally
invariant (i.e. their orbits do not mix). In fact, J pf q is the smallest closed set
which is totally invariant under f . We think of the Fatou set as the set of orbits
of f which are in some sense stable or tame, while the Julia set is the set of
chaotic orbits.
Figures 3.2–3.4 illustrate the splitting of the phase space into the Fatou set
and the Julia set of a map f in Pol, Rat and Ent respectively.
Figure 3.2 The dynamical plane of a cubic polynomial. The Fatou set consists of the
basin of attraction of infinity (white) and the basin of attraction of an attracting 2-cycle
(yellow). The common boundary between these two basins is the Julia set (black).
Figure 3.3 The Julia set (in white) of a rational map, which is Newton’s method
applied to a cubic polynomial. The Fatou set (in colour) consists of three basins of
attraction, one for each root of the cubic polynomial.
Figure 3.4 The Julia set (in black) of z ÞÑ λ exppzq with λ “ 0.64 ` 2.3i consists of a
Cantor set of unbounded curves (see [DT]). The Fatou set (white) is the basin of
attraction of an attracting 3-cycle.
The Fatou and the Julia sets are preserved under topological conjugacy
in the sense that if h is a homeomorphism such that h ˝ f “ g ˝ h, for f, g
holomorphic, then J pgq “ hpJ pf qq and Fpgq “ hpFpf qq.
As a basic example, consider f pzq “ zd for d ě 2. Then the Fatou set is
p 1 , while Jf “ S1 . Indeed, any orbit inside D converges to 0 while
equal to CzS
every orbit in CzD
p converges to infinity. Orbits of points in S1 stay in S1 and
follow the dynamics of the map Md : T Ñ T defined by t ÞÑ d ¨ t pmod 1q.
Points arbitrarily close to the unit circle have radically different asymptotic
behaviour than those on the circle. Although Julia sets In general are fractals,
the dynamics on a Julia set share most of the properties of the map Md on the
unit circle.
In the example above, 0 and 8 are attracting fixed points. In general, if
Opz0 q “ tz0 , z1 , . . . , zp´1 u form a p-cycle, we define the multiplier of the
cycle as
λ “ pf p q1 pzi q “ f 1 pz0 q ¨ f 1 pz1 q ¨ ¨ ¨ f 1 pzp´1 q,
for any i P t0, . . . , p ´ 1u, if the orbit lies in C. If the orbit includes the point
at infinity, the multiplier is defined as above after a change of variables that
moves the orbit into C. The periodic orbits are classified as:
• attracting if |λ| ă 1 (superattracting if λ “ 0);

• repelling if |λ| ą 1, and
• neutral or indifferent if |λ| “ 1,
where the last case splits into the following:
• parabolic or rationally indifferent if λ “ e2π i p{q with p{q P Q, or

• irrationally indifferent if λ “ e2π i θ with θ P RzQ.
In Section 3.3.2 we describe how these definitions relate to the local dynam-
ics around the periodic orbits. Observe that the multiplier is zero if and only if
the derivative vanishes at one of the points in the cycle. The points z for which
f 1 pzq “ 0 are called critical points. The dynamical behaviour of the critical
points plays an important role in the study of the dynamics of f .
In the basic example the two attracting fixed points at 0 and 8 belong to
open sets of orbits whose iterates converge towards the fixed point. Such sets
are in general called basins or attraction.
Definition 3.25 (Basin of attraction of an attracting cycle) Given an attract-

ing p-cycle Opz0 q of f , we define its basin of attraction A “ Af as the set of
points z P S such that f np pzq converges to some zi P Opz0 q as n Ñ 8.
The following theorem summarizes some of the basic properties of Julia sets
and Fatou sets. All proofs can be found in [Mi1, §4, §12, §14] or [Ber]. In the
statement we use the concept of grand orbit.
Definition 3.26 (Grand orbit and exceptional set) Given f : S Ñ S and a

point z P S, its grand orbit consists of all points in S which are related forwards
or backwards with z under iteration of f . More precisely,
p | f p pzq “ f q pwq for some p, q P N} .
GOpzq “ tw P C
We define the exceptional set Epf q as the set of points with a finite grand orbit.
Grand orbits are totally invariant.
Theorem 3.27 (Properties of Julia sets and Fatou sets) Let the map
f : S Ñ S be in Rat, Ent or Ent˚ .
(1) For any k ą 0, the Julia set Jf k coincides with Jf .

(2) Every attracting cycle and its basin of attraction belongs to Ff .
(3) If A is the basin of attraction of an attracting cycle, then Jf “ BA.
(4) Every repelling and parabolic cycle belongs to Jf .
(5) Jf ‰ H. If f P Rat then Jf contains a repelling fixed point or a
parabolic fixed point of multiplier 1.
(6) The set Epf q has at most two, one or no points if f is in Rat, Ent or Ent˚
respectively.
(7) If z0 P Jf and U is a neighbourhood of z0 disjoint from Epf q, then
SzEpf q Ă n pf n pU qq. Consequently, if K is a compact set disjoint from
Ť
Epf q, there exists N ą 0 such that K Ă f n pU q for all n ě N . If f P Rat,

then f n pJf X U q “ Jf for all n ě N .
(8) Either Jf has no interior point, or Jf “ S.
If z0 P SzEpf q, then Jf Ă ną0 f ń pz0 q.
Ť
(9)
(10) Jf has no isolated points.
(11) Jf is either connected or has uncountably many components.
(12) For z in a generic set of points in Jf , the forward orbit of z is everywhere
dense in Jf .
(13) Repelling periodic points are dense in Jf .
We end this section defining a special type of conjugacy which is often used
in holomorphic dynamics.
Definition 3.28 (J -conjugacy) Let f, g P Rat Y Ent X Ent˚ . We say that

f and g are J -conjugate if there exists a homeomorphism h : Jf Ñ Jg
such that
h˝f “g˝h
on the respective Julia sets of f and g.
We shall discuss a related concept, J -stability in Section 3.4.

In some cases the conjugacy h can be extended to a neighbourhood of the
Julia set. We shall see examples of this in Section 4.2.2.
3.3.2 Local theory of fixed points

In this section we describe the local dynamics of f in a neighbourhood of
a p-cycle. After replacing f by f p , we may assume that the orbit is fixed.
Furthermore we have Jf p “ Jf . We may also assume that the fixed point is
at 0, conjugating by a translation if necessary.
Hence, let f be a holomorphic map defined in a neighbourhood of the origin
with Taylor expansion
f pzq “ λz ` a2 z2 ` a3 z3 ` ¨ ¨ ¨ ,
where λ “ f 1 p0q is the multiplier of f at the fixed point 0.

We are interested in finding normal forms of f . More precisely, we look
for conformal changes of variables which conjugate f to a map with a finite
number of terms in its power series. The solution to the problem depends on
the value of the multiplier λ.
(a) Attracting and repelling fixed points (|λ| ‰ 0,1) In the attracting and
repelling cases, f is locally conformally conjugate to its linear part z ÞÑ λz.
The precise statement is from 1884, and due to Kœnigs [Kœ] (cf. [Mi1, §8]).
Theorem 3.29 (Kœnigs’ Linearization) If the multiplier satisfies |λ| ‰ 0, 1,

then there exists a neighbourhood U of 0 and a local conformal conjugacy
w “ ϕpzq where ϕ : U Ñ ϕpU q satisfies ϕp0q “ 0 and ϕ ˝ f ˝ ϕ ´1 pwq “ λw
in ϕpU X f ´1 pU qq. The conjugacy ϕ is called a linearizing map of f at the
fixed point and is unique up to multiplication by a non-zero constant.
Moreover, if fα pzq “ λpαqz ` a2 pαqz2 ` ¨ ¨ ¨ depends holomorphically on
a complex parameter α, satisfying |λpαq| ‰ 0, 1, then the linearizing map ϕα ,
appropriately normalized (for instance by ϕα1 p0q “ 1), also depends holomor-
phically on α.
Kœnigs’ Linearization Theorem is clearly a local result. But in the attracting

case, the linearizing map can actually be extended to the whole basin of
attraction of the fixed point, at the expense of its bijectivity.
Theorem 3.30 (Global Linearization) Suppose f is globally defined and has

an attracting fixed point z0 with 0 ă |λ| ă 1. Let A be the basin of attraction
of z0 . Then there exists a holomorphic map ϕ : A Ñ C with ϕpz0 q “ 0, so that
the diagram
f
A ÝÝÝÝÑ A
§ §
ϕđ
§ §ϕ
đ
wÞÑλw
C ÝÝÝÝÑ C
commutes. Moreover, ϕ is biholomorphic in a neighbourhood of z0 (where it
determines a local conformal conjugacy). The map ϕ is unique up to multipli-
cation by a non-zero constant.
The global linearizing map is constructed as follows. Let ϕ : U Ñ C denote

a linearizing map for f in a neighbourhood U of z0 . Then for any z P A define
ˆ ˙ ´
1 ¯
ϕpzq “ k
ϕ f k pzq ,
λ
where k is chosen so that f k pzq P U . The value of ϕpzq is independent of

the choice of k. Note that the globally defined map is no longer injective. For
example, every preimage of z0 under f is mapped to 0 by ϕ.
(b) Superattracting fixed points (λ “ 0) The superattracting case corresponds

to a holomorphic map f with a fixed critical point at the origin. Then f is
locally conformally conjugate to the normal form w ÞÑ wm , where m ´ 1 is
the multiplicity of the critical point (its order as a zero of f 1 ). The precise
statement is from 1904 and is due to Böttcher [Bö] (cf. [Mi1, §9]).
Theorem 3.31 (Böttcher coordinates) Let f pzq “ am zm ` am`1 zm`1 ` ¨ ¨ ¨

with m ě 2 and am ‰ 0. Then there exists a neighbourhood U of 0 and a
local conformal conjugacy ϕ : U Ñ C that conjugates f to w ÞÑ wm . The
conjugacy ϕ is unique up to multiplication by an pm ´ 1qst root of unity.
The map ϕ is called a Böttcher map or coordinate. We can not extend ϕ to

the whole basin of attraction as we did in the attracting case where λ ‰ 0 since
w ÞÑ wn is not an invertible map. However, it is still possible to extend the real
function |ϕ| : U Ñ R` Y t0u to the whole basin.
This theorem has important applications to the dynamics of polynomials
near infinity, which is a superattracting fixed point for any polynomial of
degree ě 2 (see Section 3.3.4).
(c) Parabolic fixed points (λ “ e2π i p{q ) The parabolic case corresponds to a
holomorphic map f with a fixed point at the origin, where the multiplier λ is
a qth root of unity, say e2π i p{q . Taking the qth iterate of the map we reduce to
the case where λ “ 1. Hence we consider maps of the form
f pzq “ z ` azm`1 ` Opzm`2 q, with m ą 0 and a ‰ 0.
The integer m ` 1 is called the multiplicity of the parabolic point with λ “ 1,

which is the order of the zero of f ´ Id at the fixed point.
To describe the dynamics around the parabolic point, we need the following
definition.
Definition 3.32 (Attracting and repelling petals) Suppose f is defined and

univalent in a neighbourhood U of the origin. An open set P Ă U is called an
attracting petal for f at the fixed point if:
(a) f pPq Ă P Y t0u;

(b) n f n pPq “ t0u.
Ş
An open set P Ă f pU q is called a repelling petal for f at the fixed point if P

is an attracting petal for f ´1 : f pU q Ñ U , where f ´1 denotes the branch of
the inverse of f fixing the origin.
The following theorem is due to Leau [Lea], Fatou [Fat] and Julia [J] in
successive approximations.
Theorem 3.33 (Parabolic Flower Theorem) Suppose f has a parabolic fixed

point with multiplier λ “ 1 at the origin of multiplicity m ` 1. Then there are
j “1 , numbered cyclically around the origin and such that Pj
2m petals tPj u2m
is attracting or repelling according to whether j is odd or even. Each petal
Pj intersects only its two immediate neighbours Pj ´1 and Pj `1 (indices are
taken mod 2m), and are disjoint from the rest. The petals can be chosen so that
the union
P1 Y ¨ ¨ ¨ Y P2m Y t0u
forms an open neighbourhood of the origin (see Figure 3.5).
Suppose f is globally defined and has a parabolic fixed point at the origin
of multiplier λ “ 1. If the orbit of z ‰ 0 is infinite and converges to 0, it needs
to belong to one of the attracting petals P in the Flower Theorem from some
iterate and onwards. We say that the orbit converges to 0 through P. Each P
has an associated parabolic basin (of attraction).
P4 P3 P3 P2
P2
P5
P1 P4 P1
P6
P10
P7
P9 P5 P6
P8
Figure 3.5 The distribution of the invariant petals around a parabolic point with
multiplier λ “ 1 and multiplicity 5 (left) and 3 (right).
Definition 3.34 (Parabolic basin of attraction) If z0 is a parabolic fixed point

of f with multiplier λ “ 1 and P is an attracting petal at 0, we define the
parabolic basin (of attraction) of z0 associated to P as
ď
AP “ tz P Sz f ń pz0 q | f n pzq ÝÑ z0 through Pu.
nÑ8
ną0
Observe that if the parabolic fixed point has multiplicity m ` 1 then it has
exactly m disjoint parabolic basins.
Although f is not conjugate to its linear part nor to any normal form in
a neighbourhood of the parabolic fixed point, it turns out that some kind of
linearization is possible inside each of the petals (cf. [Mi1, Thm. 10.9]).
Theorem 3.35 (Parabolic Linearization: Fatou coordinates) For every

attracting and for every repelling petal P, there is a conformal embedding
ϕ : P Ñ C called the Fatou coordinate in P which conjugates f to the trans-
lation w ÞÑ w ` 1 on P X f ´1 pPq. The following diagram commutes:
f
P X f ´1 pPq ÝÝÝÝÑ P
§ §
ϕđ
§ §ϕ
đ
wÞÑw`1
C ÝÝÝÝÝÑ C
If 0 is a parabolic fixed point of f with multiplier λ “ e2π i p{q ‰ 1 and if f q
has multiplicity m ` 1 at 0, then the number of attracting and repelling petals
in the Flower Theorem is m “ kq for some k P N. The m attracting petals
are forward invariant under f q and can be chosen so that they form k forward
invariant cycles of petals of period q under f . The petals in a cycle are mapped
among themselves with combinatorial rotation number p{q (see Figure 3.6).
P3
P5
P1
P7
P9
Figure 3.6 A parabolic fixed point of multiplier λ “ e2π i 2{5 and multiplicity 5 for
f 5 (hence k “ 1). There is one cycle of attracting petals with combinatorial rotation
number 2{5.
The k cycles of petals of period q give rise to k disjoint parabolic basins of

attraction for the parabolic fixed point.
As in the case of basins of attraction, the parabolic ones are subsets of the
Fatou set of f . Their boundaries are part of the Julia set (cf. [Mi1, Lem. 10.5]).
(d) Irrationally indifferent fixed points: Cremer points and Siegel discs The
irrationally indifferent case corresponds to a holomorphic map f with a fixed
point at the origin where the multiplier is λ “ e2π i θ and θ P RzQ. A theorem
of Năishul’ [N] asserts that the number θ is a topological invariant. It is called
the rotation number of the fixed point.
The fundamental question is which conditions on θ assure that f is locally
conformally conjugate to its linear part z ÞÑ λz. If so, f is (conformally)
linearizable around the fixed point. In contrast to the cases treated above, the
complete answer is still open in some cases.
We first observe that a local holomorphic linearization is possible if and
only if the fixed point belongs to the Fatou set (cf. [Mi1, Lem. 11.1]). If this
is the case, a neighbourhood of the fixed point is foliated by simple closed
invariant curves on which all orbits are dense (since this is the case for the
linear map z ÞÑ λz). The maximal neighbourhood of the fixed point where
the linearization is defined is called a Siegel disc and the fixed point a Siegel
point. Fixed points that are not Siegel are called Cremer points and they always
belong to the Julia set.
In 1927, Cremer proved that for a generic choice of rotation numbers,
linearization is not possible. It was not until 1942 that Siegel [Si] showed that
this is not always the case. Indeed he showed that if θ is Diophantine then
there is a Siegel disc around the fixed point. This condition was later improved
by Bryuno [Bry1] and Rüssman [Rü] in the 1960s, and the improvement was
shown to be the best possible by Yoccoz [Yo1] in 1988 (cf. [Mi1, Thm. 11.10,
11.11]).
Theorem 3.36 (Existence of Siegel discs) Let f pzq “ e2π i θ z ` Opz2 q be

defined in a neighbourhood of the origin with θ P RzQ. If θ P B then f is holo-
morphically linearizable in a neighbourhood of 0. Conversely, for any θ R B,
the quadratic polynomial f pzq “ e2π i θ z ` z2 is not locally holomorphically
linearizable.
In other words, for the family of quadratic polynomials Pol2 , the Bryuno
condition is optimal for linearization. The condition is also optimal for the
family of entire maps z ÞÑ λzez [Ge2], but for instance it is an open question
whether there exists a cubic polynomial with a Siegel disc whose rotation
number is not Bryuno. It is conjectured by Douady that the Bryuno condition
is optimal for linearization in the whole class Rat.
The local dynamics around Cremer points is complicated and interesting.
When linearization is not possible, there is always a compact invariant set with
non-trivial topology (known as a hedgehog) around the fixed point, see [PM].
3.3.3 Global theory: periodic cycles, Fatou components

and singular values
In this section f : S Ñ S is again assumed to be in Rat, Ent or Ent˚ . We
shall describe the structure of the Fatou set. We assume that Jf ‰ S, so that
the Fatou set is open and non-empty. In general, it has infinitely many con-
nected components, called Fatou components. If f P Rat a Fatou component
is mapped onto a Fatou component, since f is an open map and the Fatou and
Julia sets are completely invariant. If f P Ent Y Ent˚ , and U and U 1 are Fatou
components so that f pU q Ă U 1 , then U 1 zf pU q may consist of at most one
point (cf. [Ber, Sect. 4]). Therefore a Fatou component U is:
• p-periodic, if f p pU q Ă U for some minimal p ą 0;

• (strictly) preperiodic, if f k pU q is periodic for some k ą 0 but U is not; or
• wandering, if f k pU q X f m pU q “ H for all k, m ą 0, k ‰ m.
A Fatou component which is 1-periodic is said to be fixed or invariant.

If U is p-periodic, we denote by OpU q the cycle of Fatou components to
which U belongs.
The classification of periodic Fatou components is well understood and is

essentially due to Cremer [Cr] and Fatou [Fat] (cf. [Mi1, §16] and [Ber]).
Theorem 3.37 (Classification of periodic Fatou components) Let f be as

above and U be a p-periodic Fatou component of f . Then one and only one
of the following possibilities occurs:
(1) U contains an attracting p-periodic point z0 and f np pzq Ñ z0 as n Ñ 8

for all z P U . The cycle OpU q is called the immediate basin of attraction
of the attracting cycle Opz0 q (denoted by A˝ ).
(2) BU contains a periodic point z0 and f np pzq Ñ z0 as n Ñ 8 for all
z P U . Then z0 is a parabolic fixed point of multiplier 1 for f p . The
cycle OpU q is called the immediate parabolic basin of attraction of the
parabolic cycle Opz0 q.
(3) There exists a conformal map ϕ : U Ñ D so that pϕ ˝ f p ˝ ϕ ´1 qpzq “
e2π i ω z for some ω P RzQ. The cycle OpU q is called a p-cycle of Siegel
discs.
(4) There exists 0 ă r ă 1 and a conformal map ϕ : U Ñ Ar :“ tr ă
|z| ă 1u so that pϕ ˝ f p ˝ ϕ ´1 qpzq “ e2π i ω z for some ω P RzQ. The
cycle OpU q is called a p-cycle of Herman rings.
(5) There exists z0 P BU so that f np pzq Ñ z0 as n Ñ 8 for all z P U , but
f p pz0 q is not defined. Then z0 is an essential singularity and the cycle
OpU q is called a p-cycle of Baker domains.
Baker domains do not exist for rational maps since any cycle of Baker
domains contains an essential singularity on their boundary. Herman rings do
not exist for polynomials due to the Maximum Modulus Principle, nor do they
exist for entire transcendental maps, since they require the existence of poles.
An example of a rational map with a Herman ring can be seen in Figure 3.8.
Figure 7.8 shows the dynamical plane of an entire map with a Siegel disc.
It was an open question for a long time whether all Fatou components for
a rational map are preperiodic. An affirmative answer was given by Sullivan
in 1985 [Su2]. It is known as the Theorem of No Wandering Domains for
rational maps. A proof is given in Section 4.1 as one of the earliest examples
of quasiconformal surgery.
Singularities of the inverse map A global inverse is never well defined for the
maps f P Rat Y Ent Y Ent˚ we consider, but local inverse branches often are.
A point v P S is called regular if all possible inverse branches of f are well
defined in some neighbourhood of v. Otherwise v is called a singular value of
f . Its orbit is called a singular orbit.
The set of singular values, denoted by Singpf ´1 q, may contain three types
of points:
(1) critical values, defined as the images of critical points;

(2) asymptotic values. A point a P S is an asymptotic value of f if there
is an unbounded curve γ ptq ÝÑ 8 such that f pγ ptqq ÝÑ a. Morally,
t Ñ8 t Ñ8
asymptotic values have some of their preimages at infinity. A typical
example is a “ 0 for the exponential map, with γ ptq being any path
whose real part tends to ´8; or
(3) limits of the above.
If V is a set which does not meet Singpf ´1 q, then f : U Ñ V is a covering

for every connected component U of f ´1 pV q.
The set of critical points of f is denoted by Cf while Vf denotes its set of
critical values. All singular values of maps in Ratd are critical values, and there
are only a finite number of them, exactly 2d ´ 2 counted with multiplicity.
Transcendental maps may have infinitely many singular values of all three
types. For a deeper treatment of singularities of transcendental maps we refer
to [BerE].
In holomorphic dynamics, singularities of the inverse map and their orbits
play an important role. This is due to the fact that every cycle of Fatou
components and every non-repelling cycle is associated to the orbit of some
singular value.
Definition 3.38 (Postsingular and postcritical set) The postsingular set Pf

(also called the postcritical set if f P Rat), is defined as
ď ď
Pf “ f n pvq.
v PSingpf ´1 q ně0
A map f is called Misiurewicz if all singular orbits are preperiodic to

repelling cycles.
We summarize the relations between Pf and Fatou components in the
following theorem (see e.g. [Ber, Sect. 4.3], [Bea, §9.3] or [Mi1, §8-11]).
Theorem 3.39 (Cycles and singular values) Let f P Rat Y Ent Y Ent˚ , and
let OpU q denote a p-cycle of Fatou components.
(a) If A˝ :“ OpU q is the immediate basin of an attracting or parabolic cycle

then there exists U 1 P OpU q which contains either a critical point or an
asymptotic value.
(b) If OpU q is a cycle of Siegel discs or Herman rings, then for every U 1 P
OpU q we have
BU 1 Ă Pf .
(c) If Opz0 q is a Cremer cycle then for every z P Opz0 q we have z P Pf .
The following lemma is attributed to Fatou and can be proven by similar

arguments as those in part (b) of the theorem above (see [Mi1, Cor. 14.4]).
Lemma 3.40 (Small preimages) Let f P Rat Y Ent Y Ent˚ and let U be
a simply connected neighbourhood of z0 P Jf which does not intersect the
postsingular set Pf . Then, diampf ń pU qq Ñ 0 for any sequence of inverse
branches of f defined on U .
Indeed, one can show that otherwise, the given sequence forms a normal
family and therefore has a subsequence converging to some non-constant
holomorphic limit function g : U Ñ V . Hence, for some V 1 Ă V containing
gpz0 q and infinitely many n’s, f n pU 1 q Ă U , contradicting that gpz0 q is in the
Julia set (see Property (7) in Theorem 3.27).
As a consequence of Theorem 3.39, the number of attracting and parabolic
periodic points is always bounded above by the number of singular values
of the map. (One full singular orbit needs to be contained in each basin and
different basins are pairwise disjoint). More is known for rational maps: it is
not possible for an orbit of a singular value to accumulate on the boundaries
of two different Siegel cycles, Herman rings or Cremer cycles. This is known
as the Fatou–Shishikura inequality for rational maps [Sh1], and its proof is a
classical example of surgery which we explain in Section 7.7.
Theorem 3.41 (Bound on non-repelling cycles and Herman rings) Let f P

Ratd . Let natt , np , nsd , nhr and nc denote the number of attracting cycles,
parabolic cycles, cycles of Siegel discs, cycles of Herman rings and Cremer
cycles respectively. Then
natt ` np ` nsd ` 2nhr ` nc ď 2d ´ 2.
The generalization of this result for f P Ent with a finite number of singular
values is addressed in [Ep2, Ep3]. Analougously to rational maps, maps in this
class do not have wandering domains [EL2, GK].
In some cases singular orbits may lie on the boundary of Herman rings or
Siegel discs. The proof of this type of results often uses surgery. Some are
discussed in Sections 7.2 and 7.3. However, there is a general result which
does not involve surgery. It follows from Ghys’ Theorem (Theorem 3.18)
about rotation numbers.
Theorem 3.42 (Siegel discs and critical points) Let be a Siegel disc
with rotation number θ P H, of a holomorphic map f defined at least on
a neighbourhood of . If B is a Jordan curve, then it contains a critical
point.
Indeed, if B did not contain a critical point, then the map f would be
injective in a neighbourhood of B, and f would have rotation number θ on
the curve B. It then follows from Theorem 3.18 that B is analytic and that
f is conformally conjugate to the rigid rotation Rθ in a neighbourhood of B.
But this contradicts that the Siegel disc is the maximal domain with that
property.
Hyperbolicity and related concepts Hyperbolic maps play an important role

in dynamics, because of their ‘stability’ under perturbation (the precise concept
will be discussed in Section 3.4). The following statement summarizes their
properties (see [McM3, Thm. 3.13]).
Theorem 3.43 (Definitions of hyperbolicity) A map f P Ratd with d ě 2 is

called hyperbolic if any of the following equivalent conditions is satisfied:
(a) Pf is disjoint from the Jf ;

(b) there are no critical points or parabolic cycles in Jf ;
(c) every critical point tends to an attracting cycle under forward iteration;
(d) there is a smooth conformal metric ρ defined on a neighbourhood of the
Jf such that ||f 1 pzq||ρ ą C ą 1 for all z P Jf ;
(e) there is n P N such that f n strictly expands the spherical metric on Jf .
The last two conditions are to say that f is expanding on its Julia set. The
concept of expanding dynamical systems is very general. In this context, it is
equivalent to hyperbolicity.
For general transcendental maps, those with infinitely many singular values,
it is not clear a priori that hyperbolicity makes much sense. In particular,
statements (a) and (b) are not equivalent in the presence of Baker or wan-
dering domains. However, for transcendental maps of finite type (those with a
finite number of singular values), hyperbolicity is defined analogously. See for
example [EL2].
The following two concepts are weaker than hyperbolicity.
Definition 3.44 (Subhyperbolic and geometrically finite rational maps) A

rational map is subhyperbolic if every critical point is either attracted to an
attracting cycle or is preperiodic. It is geometrically finite if all critical points
in the Julia set are preperiodic.
Observe that geometrically finite rational maps are allowed to have parabolic
cycles, while subhyperbolic ones are not. It was proven in [DH2, Chapt. 3] that
if a polynomial is subhyperbolic and its Julia set is connected, then it is also
locally connected. Tan Lei and Yin [TY] and Pilgrim and Tan Lei [PT2] have
proven the much sharper result: if a rational map is geometrically finite then
each connected component of its Julia set is locally connected.
The concept of geometrically finite rational maps will appear several times
in the book (e.g. Section 9.3).
Blaschke products A special type of rational maps are called Blaschke

products. We shall use them repeatedly throughout the book.
Definition 3.45 ((Finite) Blaschke product) A (finite) Blaschke product is a
rational map of the form
d
ź z ´ aj
Bpzq “ ei θ ,
j “1
1 ´ aj z
for some |aj | ă 1, 1 ď j ď d and some θ P r0, 2π q.

Since Bpzq is a product of Möbius transformations which preserve the unit
disc, it follows that B is a degree d rational map which preserves D and also
CzD.
p By Montel’s Theorem, the Julia set must then be contained in the unit
circle. It is easy to check that B is symmetric with respect to S1 in the sense
that it commutes with the reflection with respect to the unit circle τ pzq “ 1{z.
If aj “ 0 for some j (i.e. if z is one of the terms in the product), then 0 and
8 are attracting fixed points and it follows from the Schwarz Lemma that the
Julia set is exactly S1 . This is also the case whenever there is a fixed point in
CzS
p 1 . In this cases, B is expanding on the Julia set.
Blaschke products are useful in dynamics in view of the following fact.
Proposition 3.46 (Proper maps in simply connected domains)
(a) If f : D Ñ D is a proper holomorphic map, then f is a Blaschke product
of degree d ě 1.
(b) If U Ă Cp is a simply connected domain and f : U Ñ U a proper holo-
morphic map, then f is conformally conjugate on U to a Blaschke
product of degree d.
Observe that, in the proposition above, f is not required to be rational or
even globally defined.
In holomorphic dynamics, and also in this book, the concept of Blaschke
product is often extended to allow for the points aj to be anywhere in CzS1 .
These maps are still symmetric with respect to S1 , but in general they do not
preserve the unit disc, which may contain proper preimages of itself and of
its complement. In this situation the Julia set is not bound to lie on the unit
circle. To distinguish these from the classical Blaschke products we sometimes
call them generalized Blaschke products. Examples can be seen in Figures 3.7
and 3.8.
Figure 3.7 The Julia set of the (generalized) Blaschke product fλ,a pzq “ z3 1z´ á
az
with a “ 2.05133 ` i0.490272. The points z “ 0 and z “ 8 are superattracting fixed
points while the other two critical points escape to infinity. The Julia set is symmetric
with respect to the S1 and has infinitely many components. The basin of z “ 0
(shaded) is infinitely connected.
The following lemma is related to the proposition above (see [Mi1,

Lem. 15.5]).
Lemma 3.47 (Rational maps which preserve the unit circle) If a rational
map of degree d preserves the unit circle then it is a generalized Blaschke
product of degree d.
3.3.4 Polynomial dynamics

For the contents of this section we refer to [Mi1, §18].
The general theory explained in the sections above takes a very special form
when we deal with polynomial dynamics.
For d ě 2, let f P Pold , that is
f pzq “ ad zd ` ad ´1 zd ´1 ` ¨ ¨ ¨ ` a1 z ` a0
with leading coefficient ad ‰ 0. Note that f p8q “ 8 “ f ´1 p8q.

`4
Figure 3.8 The Julia set of the (generalized) Blaschke product fλ,a pzq “ λz2 1z` 4z
with λ “ e2π i t , t “ 0.61517 . . .. The Julia set is symmetric with respect to the unit
circle. The Fatou set consists of the basins of 0 and 8 together with an invariant
Herman ring (shaded) containing the unit circle, and all its preimages.
Then f has d ´ 1 critical points in C when counted with multiplicity and

one critical point of multiplicity d ´ 1 at infinity. We let Critpf q denote the set
of finite critical points. The point at infinity is always a superattracting fixed
point. It turns out that the basin of attraction of infinity
Af p8q :“ tz P C
p | f n pzq ÝÑ 8u
nÑ8
is always connected, since f has no poles, and hence Af p8q does not have
any preimage other than itself. The complement
Kf :“ CzA
p f p8q
is called the filled Julia set. It is completely invariant and compact. It is also full
which means that CzK p f is connected, or equivalently, that all bounded Fatou
components are simply connected. This is a consequence of the Maximum
Modulus Principle.
The Julia set is the common boundary of these two complementary sets, i.e.
Jf “ BAf p8q “ BKf ,
while the Fatou set Ff consists of the connected component Af p8q and all
connected components of the interior of Kf , if any. Since Kf is full, it follows
that Kf is connected if and only if Jf is connected.
The name ‘filled Julia set’ for Kf refers to the fact that Kf is obtained from
the Julia set Jf by filling-in all bounded Fatou components, if any.
It turns out that the topological property of connectivity is strongly related

to the dynamical behaviour of the critical orbits (cf. [Mi1, Thm. 9.5]).
Theorem 3.48 (Connectivity of polynomial Julia sets) Let f P Pold . Then:
(a) Kf is connected if and only if Critpf q Ă Kf . In this case, the restriction

p f is conformally conjugate to z ÞÑ zd on CzD;
of f to CzK p
(b) Kf is totally disconnected if Critpf q Ă Af p8q. In this case Jf “ Kf
and it is a Cantor set;
(c) if at least one critical point of f belongs to Af p8q then both Kf and Jf
are disconnected and have uncountably many connected components.
Note that items (a) and (c) in the theorem above imply that Af p8q is either
simply connected or infinitely connected.
Remark 3.49 (The Böttcher coordinate around infinity) Since the point
at 8 is a superattracting fixed point for any f P Pold , it follows from The-
orem 3.31 that there exists a neighbourhood U of 8 and a local conformal
conjugacy ϕ “ ϕf : U Ñ C p that conjugates f to z ÞÑ zd . If the polynomial is
monic then we shall refer to the Böttcher coordinate around 8 as the conjugacy
that is uniquely determined by the requirement that f pzq{z Ñ 1 as z Ñ 8.
Green’s function, equipotentials and external rays Assume that f is monic,

and let ϕ : U Ñ CzD
p r denote the Böttcher coordinate that conjugates f to
z ÞÑ z in the domain U , which is mapped outside Dr for the maximum r ě
d
1. If Kf is connected then r “ 1 as stated in Theorem 3.48(a), and if Kf is

disconnected then r ą 1 and BU contains a critical point.
Set A˚f p8q “ Af p8qzt8u and U ˚ “ U zt8u. The Green’s function for
Kf is the real continuous harmonic function g : A˚f p8q Ñ R` that extends
log |ϕ| : U ˚ Ñ plog r, `8q as follows:
#
log |ϕpzq| if z P U ˚ ,
gpzq “ 1
dk
gpf k pzqq if f k pzq P U ˚ .
Since gpf pzqq “ d ¨ gpzq for z P U ˚ the function g is well defined on all of
A˚f p8q. It may also be extended to the whole plane by setting g ” 0 on Kf .
For ρ ą 0, the level set
gρ :“ g ´1 pρq “ tz P A˚f p8q | gpzq “ ρu
is called the equipotential of potential ρ. If eρ ą r, it is a simple closed curve

that surrounds the Julia set.
Suppose Kf is connected. Then for t P T “ R{Z, the curve
Rf ptq :“ tz P CzKf | argpϕpzqq “ 2π tu
is called the (external) ray of (external) argument t. Note that Rf ptq is mapped
bijectively under f to Rf pd ¨ tq with arguments in T. In particular, if d p ¨ t ” t
pmod 1q then the ray Rf ptq is p-periodic (with minimal p). For example,
Rf p0q is always fixed, while Rf pn{dq is always prefixed for any 1 ď n ă d
(see Figure 3.9).
Rf p 27 q
2
7
Rf p 17 q 1
7
ϕ ´1

Rf p 12 q
ϕ
- 1
2
0
Rf p 47 q Rf p0q
4
7
Figure 3.9 The filled Julia set of f pzq “ z2 ` c with c „ ´0.12256 ` 0.74486i , also
known as the Douady rabbit. This is an example of a locally connected Julia set where
all rays land. Some rays and equipotentials are drawn, in particular the fixed ray
Rf p0q landing at one fixed point of f , the prefixed ray Rf p 12 q and the 3-cycle of rays
landing at the other fixed point of f .
If the limit
´ ¯
γ ptq “ lim ϕ ´1 re2π it
r Ñ 1`
exists, we say that Rf ptq lands at the point γ ptq which necessarily belongs
to the Julia set. All rays of rational argument land at a repelling or parabolic
periodic point (see [Mi1, Thm. 18.10]). If Rf ptq lands at γ ptq, then Rf pd ¨ tq
lands at f pγ ptqq. Hence, the landing points of a k-cycle of rays must form a
periodic orbit of a period dividing k.
Conversely, every repelling or parabolic periodic point z0 is the landing
point of at least one periodic ray (see [Mi1, Thm. 8.11] or [Hu1, Thm. I.A]).
If k is the period of z0 , only finitely many rays, say q 1 , land at a z0 , and
these rays are all periodic of the same period. They are transitively permuted
by f k , and this permutation must preserve their circular order since f is
a local homeomorphism at z0 ; thus the permutation must send each ray to
the one which is p1 further counterclockwise for some p1 ă q 1 . We call p{q

the combinatorial rotation number of f at z0 , where p{q “ p1 {q 1 written in
lowest terms.
There exist polynomials for which not all rays land. But Fatou showed that
the set of arguments t P T for which Rf ptq does not land has measure zero.
Furthermore, M. and F. Riesz proved that for any given z0 P Jf the set of
arguments for which γ ptq “ z0 has measure zero (cf. [CL]).
The following theorem summarizes the landing properties of external rays
(cf. [Mi1, Thm. 18.3]).
Theorem 3.50 (Continuous landing of rays) For any polynomial f with

connected Julia set the following conditions are equivalent:
(i) every external ray Rt lands at a point γ ptq which depends continuously
on the argument t;
(ii) the Julia set Jf is locally connected;
(iii) the filled Julia set Kf is locally connected;
(iv) the inverse of the Böttcher coordinate ϕ ´1 : CzD Ñ CzKf extends con-
tinuously to BD and hence induces a continuous parametrization γ :
T Ñ Jf of the Julia set, given as γ ptq “ ϕ ´1 pe2π i t q.
Remark 3.51 (Carathéodory semi-conjugacy) If the conditions in Theo-

rem 3.50 are satisfied then the map γ : T Ñ Jf is a semi-conjugacy between
the map t ÞÑ d ¨ t pmod 1q on T and f on Jf . It is called the Carathéodory
semi-conjugacy of Jf (sometimes the Carathéodory loop). This is yet another
example of how strongly topological and dynamical properties interplay.
Julia sets which are not locally connected are for example those with a
Cremer periodic point, or with a Siegel disc whose boundary does not contain
a critical point (cf. [Mi1, Cor. 18.6]). Maps with this property are constructed
by surgery as explained in Section 7.2.
Hybrid equivalences We end this section by mentioning another important

type of conjugacy, although it is called equivalence for historical reasons.
Definition 3.52 (Hybrid equivalence) Two polynomials f and g in Pold are

hybrid equivalent if there exist neighbourhoods Uf and Ug of Kf and Kg
respectively, and a quasiconformal conjugacy φ : Uf Ñ Ug between f and g,
satisfying Bφ “ 0 almost everywhere on Kf .
Observe that by Weyl’s Lemma (Theorem 1.14) this means that φ is holo-
morphic in the interior of Kf , if this is non-empty.
Hybrid equivalences are the strongest possible type of conjugacy when the
Julia sets are connected, as shown by the following theorem [DH3, Prop. 21].
Theorem 3.53 (Hybrid classes are affine classes in the connected case) Let
f, g P Pold with connected Julia sets. If f and g are hybrid equivalent then
they are affine conjugate.
3.3.1 Let f be holomorphic in a neighbourhood U of S1 . Assume S1 is
invariant under f and f |S1 is analytically conjugate to the rigid rotation
Rθ for some θ P RzQ. Show that f has a Herman ring (or an annular
subset of a Herman ring) containing the unit circle.
More precisely, let ϕ : S1 Ñ S1 denote an analytic conjugacy and prove
that there exists a neighbourhood U 1 Ă U of S1 such that f |U is con-
formally conjugate to the rigid rotation Rθ on some standard annulus
containing S1 and such that the conformal conjugacy ϕr is equal to ϕ
on S1 .
3.3.2 Show that a critical circle map can not be analytically conjugate to an
irrational rigid rotation.
Hint: Use Exercise 3.3.1 above.
3.3.3 Let γ be a Jordan curve that is invariant under a univalent map f defined
in a neighbourhood of γ , and assume that rotpf, γ q P H. Prove that f
has a Herman ring (or an annular subset of a Herman ring) around γ .
Hint: Apply Theorem 3.18
3.4 Families of holomorphic dynamics: parameter spaces

We consider holomorphic families of maps in Ratd , where d is fixed, that is
tfλ : C
p Ñ C,
p λ P u Ă Ratd ,
where the parameter space is a complex manifold of dimension N ě 1 so

that the map ˆ C p ÑC p defined by pλ, zq ÞÑ fλ pzq is holomorphic, i.e. so
that the coefficients of fλ depend holomorphically on λ.
In order to define the notion of J -stability we need to state the general
concept of holomorphic motion introduced by Mañé, Sad and Sullivan in
[MSS].
Definition 3.54 (Holomorphic motion) A holomorphic motion of a set A Ă
C
p over is a mapping H : ˆ A Ñ C
p satisfying the following:
(1) for any fixed z P A the mapping λ ÞÑ H pλ, zq is holomorphic;

(2) for any fixed λ P the mapping z ÞÑ Hλ pzq :“ H pλ, zq is injective;
(3) the mapping Hλ0 “ IdA , for a base point λ0 P .
Note that continuity of H is not required in the definition. However, this
property follows as a consequence, as shown in the λ-lemma proved in [MSS].
Theorem 3.55 (The λ-lemma) A holomorphic motion of A as above has
a unique extension to a holomorphic motion of A. The extended map H :
ÂÑC p is continuous, and for each λ P , Hλ : A Ñ Cp extends to a
quasiconformal self-map of C.
p
For further results about holomorphic motion and the λ-lemma, see for
instance, papers by Bers and Royden [BR] and Słodkowski [Sl], and the survey
by Astala and Martin [AM].
Definition 3.56 (J -stability) Consider as above a holomorphic family f :
ˆC p ÑC p of rational maps in Ratd . For a λ0 P the Julia set Jf is J -
λ0
stable if there exists a neighbourhood Uλ0 Ă of λ0 over which the Julia sets
move holomorphically, i.e. H : Uλ0 ˆ Jλ0 Ñ C p such that Hλ pJλ q “ Jλ and
0
furthermore Hλ ˝ fλ0 “ fλ ˝ Hλ0 .
Hyperbolic rational maps are J -stable. An even stronger statement is proven
in Section 4.1 by surgery.
The following theorem is also proved in [MSS].
Theorem 3.57 (J -stable set of parameters) The J -stable set of parameters
J -stable Ă is an open dense subset of .
Our goal is to identify sets of parameter values in for which the corre-
sponding dynamical systems are conjugate or share some given properties. In
particular, we shall be interested in conformal and quasiconformal conjugacies,
either global or on (neighbourhoods of) the Julia set.
3.4.1 The parameter plane of quadratic polynomials:

the Mandelbrot set
The best-understood holomorphic family is Pol2 . In the parametrization
tQc pzq “ z2 ` c | c P Cu,
all quadratic polynomials are uniquely represented up to conformal conjugacy
(see Exercise 3.1.4 in Section 3.1).
We refer to dedicated chapters in [DH2, Bra1, CG, St] for extensions of the
contents of this section.
For quadratic polynomials of the form Qc with c P C, we set Kc :“ KQc ,

Ac :“ AQc p8q and Jc :“ JQc .
In the parameter plane C the Mandelbrot set M is defined as
M :“ tc P C | Qnc p0q Ñ 8 as n Ñ 8u,
i.e. the set of c-values for which the critical orbit tQnc p0qun is bounded. It
follows from Theorem 3.48 that if c P M then Kc (and hence also Jc ) is
connected, and if c R M then Kc “ Jc is a Cantor set. This means that
M “ tc P C | Kc is connectedu.
To emphasize this property M is also called the connectedness locus of the

quadratic family tQc ucPC (see Figure 3.10).
The first basic properties of M were proven by Douady and Hubbard in
1982 (cf. [DH2]).
Theorem 3.58 (Basic properties of M) The Mandelbrot set is compact,

connected and full. Moreover, it is contained in D2 .
Hyperbolic components of M A polynomial Qc can have at most one attract-

ing cycle in C (see Theorem 3.39). It follows that Qc is hyperbolic if and only
if Qc has an attracting cycle in C, or if the orbit of the critical point is attracted
to the superattracting fixed point at 8 (see Theorem 3.43). In this case the
c-value is called a hyperbolic parameter.
If Qc0 has an attracting p-cycle in C, then it follows from the Implicit Func-
tion Theorem that the p-cycle moves holomorphically for parameter values
close to c0 and that the cycle is still attracting. As a consequence, c0 P intpMq.
The maximal open set containing c0 , , of parameters for which a p-cycle
exists and remains attracting is called a hyperbolic component (of period p)
of M.
One can check that B Ă BM so that a hyperbolic component is actually
a connected component of the interior of M. It is not known if all connected
components of intpMq are hyperbolic. It is, however, conjectured to be so.
Conjecture 3.59 (Hyperbolicity conjecture)

ď
intpMq “
hyp.comp.
Equivalently, hyperbolic parameters are dense in C.
The conjecture has been proven for parameters on the real line [GS, KSvS].
Figure 3.10 The Mandelbrot set can be seen as a catalogue of Julia sets. Filled Julia
sets are shaded and attracting cycles are marked. The upper-left picture shows a Siegel
disc with some interior orbits drawn.
Given a hyperbolic component of period p, the multiplier map is

defined as
: ÝÑ D,
p
c ÞÝÑ pcq “ pQc q1 pzpcqq,
where zpcq for each c P denotes a point in the attracting p-cycle so that
c ÞÑ zpcq is holomorphic. It follows that is holomorphic. One of the very
first surgery constructions was done in order to prove the following theorem as
we explain in Section 4.1.
Theorem 3.60 (Multiplier map) Let be a hyperbolic component of M of
period p. Then the multiplier map is a conformal isomorphism which extends
continuously to : Ñ D.
It follows that all hyperbolic components are topological discs, each with
a centre, the unique parameter value for which the attracting cycle is superat-
tracting, i.e. ´1 p0q. The boundary of has a unique point, called the root of
, for which the multiplier of the p-cycle is exactly 1.
Attached to every boundary parameter for which the multiplier of the cycle
has derivative e2π i r {s we find a new hyperbolic component of period p ¨ s,
which explains part of the fractal structure observable near any point of the
boundary of the M (see Figure 3.11).
Figure 3.11 If is a hyperbolic component, for every rational number there is a

smaller hyperbolic component attached to B.
The boundary of M The boundary of the Mandelbrot set is an intricate

and fascinating object which is still the focus of intense research. It has a
remarkable fractal structure of Hausdorff dimension equal to 2 [Sh2], the
highest possible in the plane.
The boundary of the Mandelbrot set can also be characterized in terms of
normal families (see Exercise 3.4.1 or [McM2, Thm. 4.6]) as
BM “ tc P C | tgn pcq :“ Qnc p0qun is not normal in any neighbourhood of cu.

We say that c P M is Misiurewicz if Qc is a Misiurewicz polynomial. This

characterization of BM is used to prove the following properties.
Theorem 3.61 (Parameters in BM)
(a) Misiurewicz points are dense in BM.
(b) For each c P BM there exists a sequence tcn un of centres of hyperbolic
components such that cn Ñ c as n Ñ 8.
(c) If Qc has a parabolic, Siegel or Cremer cycle, then c P BM.
The convergence of centres to the boundary points is very fast. This is a

property used to compute very accurate approximations to the Mandelbrot set
as the ones shown in this section.
J -stability can be formulated in many different equivalent ways, see cf.
[McM2, Thm. 4.2]. For the quadratic family tQc ucPC , J -stability at c is
equivalent to the family of maps gn pcq :“ Qnc p0q being normal at c. Hence
CJ -stable “ tc P C | tgn pcqunPN normal in a neighbourhood of cu “ CzBM.
The most important conjecture about the Mandelbrot set concerns a topo-
logical property.
Conjecture 3.62 (MLC) M is locally connected.
Conjugacy classes We already mentioned (Exercise 3.1.4 in Section 3.1) that

the family tQc ucPC has exactly one representative in each conformal (therefore
affine) conjugacy class.
Some facts concerning other types of conjugacy are summarized in the
following theorem.
Theorem 3.63 (Conjugacy classes)
(a) Any two polynomials in CzM are hybrid equivalent, but no two different
polynomials in M are (see Theorem 3.53).
(b) All polynomials in a given hyperbolic component with the exception of
its centre are globally quasiconformally conjugate. If we only require the
conjugacies to hold in a neighbourhood of the Julia set, then the centre
is also included (see Section 4.1 and Figure 3.12).
(c) If c P BM and c1 P C and Qc is quasiconformally conjugate to Qc1 , then
c “ c1 . In other words, all parameters c P BM are unique representa-
tives of their quasiconformal conjugacy class.
It is a central question in holomorphic dynamics whether a topological

conjugacy implies a quasiconformal one.
Figure 3.12 Any two polynomials in the same hyperbolic component, different from
the centre polynomial, are quasiconformally conjugate. The quasiconformal conjugacy
sends one Julia set to the other. In the figure, the basins of attraction are shaded and the
periodic cycles are marked.
Conjecture 3.64 (Rigidity conjecture) Two rational maps which are topolog-
ically conjugate are quasiconformally conjugate.
In the particular case of quadratic polynomials:
Rigidity conjecture ùñ MLC ùñ Hyperbolicity conjecture.
A weaker type of conjugacy between Julia sets (David conjugacy) is treated

in Section 9.3.
3.4.2 Other parameter spaces

There are systematic studies of other parameter spaces for particular holomor-
phic families of polynomials, rational maps or transcendental maps. Here we
mention those that are treated in later sections in this book.
For a holomorphic family F “ tfλ | λ P u of complex polynomials of a
fixed degree d ě 2, the connectedness locus of the family tfλ uλP is
CpFq “ tλ P | Kfλ is connectedu.

The structure of parameter planes of unicritical polynomials,

tz ÞÑ zd ` c | c P C, d ą 2u,
is very similar to that of quadratic polynomials.
The connectedness loci Md of such families are known as Multibrot sets
or generalized Mandelbrot sets. One difference with respect to M is that
the multiplier map defined on a hyperbolic component is a degree pd ´ 1q
branched covering of the disc, ramified over 0 in D. We refer to Section 4.1.2
for details.
In general, any polynomial of degree d ě 2 is affine conjugate to a monic,
centred polynomial of the same degree. Hence, the family of such polynomials
can be parametrized by pd ´ 1q complex coefficients. The parameter space is
therefore Cd ´1 .
For cubic polynomials we may choose a parametrization of monic, centred
polynomials with marked critical points
z ÞÑ z3 ´ 3a 2 z ` b.
The critical points are ˘a and the parameter space is C2 . Cubic polynomials
are treated in Sections 6.2 and 7.5.
We shall also deal with the uniparametric families of polynomials of degree
pq ` 1q ě 3
z q
ˆ ˙
Pq,λ pzq “ λz 1 ` , with λ P C˚ .
q
The connectedness loci of tPq,λ uλPC˚ presents many similarities to the Man-
delbrot set, some of which are explored and used in Section 8.2.

3.4.1 Prove that BM is the set of non-normality of the family G :“ tc ÞÑ
gn pcq “ Qnc p0qu.
Hint: Using Montel’s Theorem (Theorem 3.23), show that for c R BM
there is a neighbourhood of c in which the family is normal. Addi-
tionally, show that for c P BM the family can not be normal in any
neighbourhood of c.
3.5 Actions of Kleinian groups and the Sullivan dictionary

Fatou and Julia were aware of the analogy between iteration of holomorphic
mappings and Poincaré’s work on Kleinian groups. But it was Sullivan in 1985
[Su2, MS] who deeply exploited the interplay between these two theories and
established a first version of a dictionary between them. The analogy suggested

that quasiconformal maps could play a role in holomorphic iteration, as it did in
Kleinian groups. This is exemplified by the celebrated No Wandering Domains
Theorem for rational maps and a new proof of its analogue for Kleinian groups,
Ahlfors’ Finiteness Theorem. In this section we present a brief survey of
the theory of Kleinian groups, emphasizing the parallelism with holomorphic
iteration. As in the other sections in the current chapter we include mainly
statements without proofs. We mostly follow [MNTU, Chapt. 5]. We also refer
to [MT] and [Mar] for deep expositions of the subject.
3.5.1 Basic definitions

Let MöbpCqp denote the group of Möbius transformations of the Riemann
sphere, i.e.
" ˇ *
p “ z ÞÑ az ` b ˇ a, b, c, d P C, ad ´ bc “ 1 ,
MöbpCq
ˇ
cz ` d ˇ
which is identified with PSLp2, Cq “SLp2, Cq{t˘I u.

The dynamics of Möbius transformations are fairly simple since every ele-
ment in MöbpCq,
p except the identity, is globally conformally conjugate (i.e.
conjugate by a Möbius transformation) to one of three fundamental types.
Proposition 3.65 (Classification of Möbius transformations) Let γ P

MöbpCq,
p γ ‰ Id. Then γ is conformally conjugate to one of the following:
(a) z ÞÑ z ` 1, and we call γ Parabolic;

(b) z Ñ
Þ λz, |λ| “ 1, λ ‰ 1, and we call γ Elliptic;
(c) z ÑÞ λz, |λ| ą 1, and we call γ Loxodromic or Hyperbolic if λ P R.
Parabolic transformations have only one fixed point, while elliptic and lox-
odromic have two. The only transformations of finite order (γ n “ Id for some
n ą 1), different from the identity, are elliptic and conjugate to z ÞÑ e2π i p{q z
with p, q P Z.
Using that the trace of a matrix is invariant under matrix conjugation it
follows that the type of a Möbius transformation γ ‰ Id is also classified by
the trace of a representing matrix:
• γ is parabolic (resp. elliptic) iff tr = ˘2 (resp. ´2 ă tr ă 2);

• γ is loxodromic (in particular hyperbolic) iff | tr |ą 2 (tr P R).
We are now ready to define a Kleinian group.

Definition 3.66 (Kleinian group) A subgroup of MöbpCq

p is a Kleinian group
if its identification in PSLp2, Cq is discrete.
From the definition it follows that Kleinian groups are countable subsets
of PSLp2, Cq and do not contain any elliptic elements of infinite order. It can
be shown that a Kleinian group which consists only of elliptic transformations
apart from the identity is a finite group. One may also observe that if a Kleinian
group contains a loxodromic element γ , then it cannot contain a parabolic one,
say η, that shares a fixed point with γ since γ ń ˝ η ˝ γ n Ñ Id as n Ñ 8 (or
n Ñ ´8), contradicting the discreteness of the group.
3.5.2 The action of a Kleinian group and the dynamical partition

A Kleinian group acts on the Riemann sphere as
ˆC p ÝÑ C, p
pγ , zq ÞÝÑ γ pzq.
We set pzq :“ tγ pzq | γ P u.

The dynamics generated by the action of a Kleinian group (also known as
conformal dynamics) are different from those generated by iteration of a single
holomorphic (usually non-invertible) map. If we apply to a point z we obtain
a set of points pzq which can be thought of as the grand orbit of z, but without
the order relation that comes with iteration. Invariant sets are still meaningful
but forward and backward invariance are the same, since a group contains the
inverse of all its elements.
Definition 3.67 (Group invariance) We say that U Ă C p is invariant under

the Kleinian group if pzq Ă U for all z P U . In this case pU q “ U .
Another familiar concept is that of conformal or quasiconformal

conjugation.
Definition 3.68 (Quasiconformally conjugate groups) We say that the

Kleinian groups and 1 are conformally (resp. quasiconformally) conju-
gate if there exists a Möbius transformation (resp. a quasiconformal map)
φ:C pÑC p such that 1 “ φφ ´1 :“ tφ ˝ γ ˝ φ ´1 | γ P u. If is given by
a set of generators (called a marking of ), then φ takes the marking of to
the marking of 1 .
As in iteration theory, normality is the property which induces the dynamical

partition in C.
p In the case of families of Möbius transformations, Montel’s
Theorem has a special formulation (compare with Theorem 3.23).
Theorem 3.69 (Montel’s Theorem adapted) Let U Ă C p be a domain and F

a family of Möbius transformations. If there exist two points a, b P C
p such that
γ pU q Ă Czta,
p bu for all f P F, then F is a normal family on U .
We are now ready to define the analogues to the Fatou and the Julia sets.
Definition 3.70 (Ordinary set and limit set) Let be a Kleinian group. We
define the ordinary set of , often also called the regular set, as
pq :“ tz P C
p | is a normal family in some neighbourhood of zu,
and the limit set of as its complement, pq :“ Czpq

p
It follows from the definition that pq is open and consequently that pq
is closed. Both sets are invariant under and, in fact, if #ppqq ě 3, then
pq is the smallest non-empty, -invariant closed subset of C. p The definition
above is chosen to emphasize the analogy between ordinary set and limit set
and Fatou set and Julia set respectively. Classically, the ordinary set is defined
as the complement of the limit set, which is defined as
pq “ tz P C
p | z is a limit pointu.
Note that z is called a limit point if there exists ζ P C p and an infinite

sequence of distinct elements γn P so that z “ limnÑ8 γn pζ q. Compare with
Theorem 3.72,
Definition 3.71 (Proper discontinuity) We say that a Kleinian group acts
properly discontinuously at a point z P C
p if there exists a neighbourhood U of
z such that the number of elements of satisfying γ pU q X U ‰ H is finite.
It can be shown that the ordinary set is precisely the set of points at which
acts properly discontinuously. For this reason pq is often called the region
of discontinuity.
The limit set and the ordinary set are both preserved under conjugation in
the sense that if and 1 are Kleinian groups conjugate by a quasiconformal
map φ, then p 1 q “ φppqq and p 1 q “ φppqq.
As an example consider a loxodromic transformation γ and set “ă γ ą,
the group generated by γ . We may assume that γ pzq “ λz with |λ| ą 1, so
that γ has fixed points at 0 and 8. Given any open set U Ă Czt0u, the
sequence tpγ |U qn uně0 converges to the constant function equal to 8, while
the sequence tpγ |U qń uně0 converges to the constant function equal to 0. If,
on the other hand, V is a neighbourhood of 0 then γ n pV q “ C and if V is
Ť
a neighbourhood of 8 then γ pV q “ Czt0u. It follows that pq “ Czt0u

Ť n p
and pq “ t0, 8u. One can observe the difference with respect to iteration
theory where the Julia set of γ consists of the repelling fixed point t0u, while
the point at infinity is attracting and hence belongs to the Fatou set.
A Kleinian group whose limit set consists of at most two points is called
elementary. The group above is an elementary Kleinian group. It can be
shown that a non-elementary Kleinian group contains infinitely many loxo-
dromic elements, and that for any loxodromic γ P there exists a loxodromic
γ 1 P so that the two transformations have no fixed points in common. In our
discussions, Kleinian groups will in general be assumed to be non-elementary.
In addition, we shall also assume they contain no elliptic elements, since it
can be shown that any non-elementary Kleinian group with elliptic elements
always contains a subgroup 1 of finite index in , with no elliptic elements
and such that pq “ p 1 q.
From the example above it is easy to see that both fixed points of every
loxodromic element in a Kleinian group belong to the limit set. Similarly,
one can show that the fixed point of every parabolic element also belongs to
the limit set. In the following theorem we summarize some of the properties of
limit sets and ordinary sets, some of which were already mentioned. Compare
with Theorem 3.27.
Theorem 3.72 (Properties of limit sets and ordinary sets) Let be a

Kleinian group.
(1) Fixed points of loxodromic or parabolic elements of belong to pq.

Hence pq ‰ H.
(2) Fixed points of elliptic elements of belong to pq.
(3) If 1 is a component of pq which is invariant under , then pq “
B1 .
(4) The number of components of pq is 0, 1, 2 or 8. There are at most two
invariant components.
(5) Either pq has no interior point or pq “ C. p
(6) Let z, w P C.
p If there exists a sequence of distinct elements γn P such
that γn pwq Ñ z, then z P pq.
Assume furthermore that is non-elementary.
(7) If U is an open set intersecting pq, then there exist finitely many
elements γ1 , . . . , γN P such that
N
ď
pq Ă γn pU q.
n“1
(8) If z P pq and w P C, p then there exists a sequence tγn u of such that
γn pwq Ñ z as n Ñ 8.
(9) pq has no isolated points.
(10) For all z P pq, the set pzq is dense in pq.
(11) The set of fixed points of all loxodromic elements are dense in pq.
(12) If contains parabolic transformations, then the set of fixed points of
all parabolic elements is dense in pq.
Special examples where pq ‰ C p so as to see the analogy with Julia sets
In what follows we give four examples of Kleinian groups and their limit
sets. They will motivate the concepts in the rest of the section and serve as
references throughout.
In all four cases is a Kleinian group with two generators “ă α, β ą,
which are both loxodromic (except in the last example). In each case there exist
four circles in the plane Cα , Cα1 , Cβ and Cβ1 with pairwise disjoint interiors
such that:
• αpCα q “ Cα1 and αpintpCα qq “ extpCα1 q;

• βpCβ q “ Cβ1 and βpintpCβ qq “ extpCβ1 q,
where int and ext respectively denote the bounded and unbounded compo-
nents of the complement. Let D :“ intpCα q Y intpCα1 q Y intpCβ q Y intpCβ1 q,
and observe that pq Ă D. Indeed, CzD
p maps to D by α, α ´1 , β and β ´1 .
Thus one can check that every orbit in is completely contained in D or has
at most one point in the complement.
Consequently one can construct the limit set as the limit of images of the
four circles by all possible words in the group formed with α, β and their
inverses. For a beautiful exposition about these type of constructions we refer
to the book Indra’s Pearls by Mumford, Series and Wright [MSW]. The four
examples below follow their recipes.
Example 3.73 (pq is a Cantor set)

ˆ? ˙ ˆ? ˙
2?´2 2 ?i
Let α “ and β “ . One can check that Cα and Cα1
´ 12 2 í 2
can be chosen to be the circles centred at ˘ ?5 with radius ?1 . Likewise,
6 6
3 1
Cβ and Cβ1 can be chosen as the circles centred at ˘i ? with radius ? .
2 2 2 2
The four circles have disjoint closures, and therefore their images form an
Cα1
Cβ1 Cβ
Cα
Figure 3.13 Example 3.73. On the left, the four circles Cα , Cα1 , Cβ , Cβ1 and their
images under α, α ´1 , β, β ´1 . The small circles at level 1 are coloured depending of
which large circle they come from. On the right, several more steps are shown. The
limit set is a Cantor set.
uncountable collection of nested circles. The circles are called Schottky circles.
Figure 3.13 shows on the left the first step of the sequence, while on the right
seven steps are shown. One can show that all ends consist of a single point.
Hence the limit set is a Cantor set.
Example 3.74 (pq is the unit circle)

˜ 2 1 ¸
? ?
ˆ ? ˙
3 3 2? i 3
Let α “ and β “ (see Figure 3.14). In this case
?1 ?2 í 3 2
3 3 ?
the circles Cα and Cα1 are chosen to be centred at ˘2 with radius 3 while
the circles Cβ , Cβ1 are centred at ˘i ?2 with radius ?1 . In this case the unit
3 3
circle is invariant under both α and β. The?example is chosen so that the four
circles touch pairwise at the points ˘ 12 ˘ i 23 . These points are parabolic fixed
points of the different commutators formed by α, β, α ´1 , β ´1 . The limit set is
the unit circle.
Cα Cβ1 Cα1
Cβ
Figure 3.14 Similar to the previous figure, illustrating the group in Example 3.74.
The limit set is the unit circle.
Example 3.75 (pq is a quasicircle)

˜? ¸ ˜? ¸
5 1 89
2 ?2 5 i 16
?5
Let α “ and
β“ (see Figure 3.15). The circles
1
2 2
5
í 45 589
?
Cα and Cα1 are centred at ˘ 5 with radius 2, and the circles Cβ and Cβ1
?
are centred at ˘i 489 with radius 54 . The four circles are touching pairwise
at parabolic points of the different commutators formed by α, β, α ´1 , β ´1 .
The four points are not on a circle. The limit set is a quasicircle.
Cβ1
Cα Cα1
Cβ
Figure 3.15 Similar to the previous figure, illustrating the group in Example 3.75 The
limit set is a quasicircle.
Example 3.76 (pq has infinitely many pinch-points)

In this example we have chosen β so that the circles Cβ and Cβ1 meet at the
origin at a fixed point (see Figure 3.16). This can be seen as a limit case of
Example 3.75, where the repelling and attracting fixed points of β, previously
in the interior of the circles, have collapsed at the origin (common boundary
point) and thus β has become a parabolic transformation. The transformation α
is loxodromic and chosen, as before, so that the four common boundary points
(different from the origin) are parabolic fixed points of the commutators. This
collapse induces the limit set to ‘pinch’ at infinitely many points. This example
is known as ‘Mickey Mouse’.
3.5.3 Fuchsian, quasi-Fuchsian and Schottky groups

Example 3.74 above is an example of a special type of Kleinian group called
Fuchsian. They are the analogues of Blaschke products in complex iteration.
Figure 3.16 Similar to the previous figure, illustrating the group in Example 3.76.
The ordinary set has infinitely many bounded components.
Definition 3.77 (Fuchsian group) We say that a Kleinian group is Fuchsian

if there is a circle S Ă C
p such that both components of CzS
p are invariant under
(hence so is S).
Since pq is minimal, it follows that pq Ă S. We say that is a

Fuchsian group of the first kind if pq “ S. Otherwise it is of the second
kind. In greater generality, if the limit set of a finitely generated Kleinian group
is a Jordan curve, and both complementary components are invariant, then
is called a quasi-Fuchsian group (see Example 3.75 above). These are the
analogues of maps which are quasiconformally conjugate to Blaschke products
(see Definition 3.45). The limit set of a quasi-Fuchsian group is a quasicircle
but the corresponding statement is not always true in complex iteration, as
shown by the Julia set of Q1{4 pzq “ z2 ` 1{4, where the Fatou set has two
completely invariant components, and the Julia set is a Jordan curve which is
not a quasicircle since it contains cusps.
Another important class of Kleinian groups are those called Schottky
groups. They are analogues of hyperbolic rational maps with a connected
Fatou set.
Definition 3.78 (Schottky group) A Kleinian group is a Schottky group if

it is free, finitely generated and all its non-trivial elements are loxodromic.
It was shown by Maskit [Mas] that for a Schottky group there always exist
2n disjoint oriented Jordan curves C1 , C11 , . . . , Cn , Cn1 with disjoint interiors,
and n Möbius transformations γ1 , . . . , γn such that γj maps the interior of Cj
to the exterior of Cj1 , for all 1 ď j ď n, such that “ă γ1 , . . . , γn ą. In this
case the limit set pq is a Cantor set contained in the union of the interiors
of the 2n Jordan curves. The group in Example 3.73 is a Schottky group for
which the Jordan curves can be chosen as circles. These examples exhaust by
no means all types of Klenian groups.
3.5.4 Ahlfors’ Finiteness Theorem and no wandering domains

Assume is a finitely generated Kleinian group.
Definition 3.79 (Fundamental domain) Suppose 1 Ă pq is invariant. We

say that F Ă 1 is a fundamental domain of in 1 if F is connected and
every orbit pzq Ă 1 intersects F in exactly one point.
In Example 3.73, pq has one invariant component. Set F “ pCzDq p Y

1 1
pCα Y Cβ q where D “ intpCα q Y intpCα q Y intpCβ q Y intpCβ q. Then F is a
fundamental domain. Consider F and identify Cα1 with Cα , Cβ1 with Cβ under
the action of α respectively β. It follows that pq{ is a compact Riemann
surface of genus 2. In greater generality, for any Schottky group with
g ě 2 generators, the ordinary set pq is invariant and pq{ is a compact
Riemann surface of genus g. In fact, every compact Riemann surface of genus
g is quasiconformally equivalent to pq{ where is a Schottky group with
g generators. (Cf. the Retrosection Theorem in [Be2, Sec. 1.6].)
In Example 3.74, pq has two invariant components: 1 “ D and 2 “
CzD.
p Compare with Figure 3.17. Let Q1 denote the quadrilateral bounded by
the four tangent circles. Then 1 X pQ1 zpCα1 Y Cβ1 qq is a fundamental domain
of in 1 . Based on the identifications by α and β on BQ1 it follows that 1 {
is isomorphic to a torus with one puncture. Indeed, the four corners correspond
to just one point which does not belong to pq, and therefore corresponds to a
puncture. By symmetry, 2 { is also isomorphic to a torus with one puncture.
Hence the quotient space pq{ is isomorphic to two punctured tori, with
their punctures at infinity.
The examples above illustrate a general fact known as Ahlfors’ Finiteness
Theorem, which was originally proven in [Ah1].
Q1
Figure 3.17 The fundamental domain Q1 in Example 3.74 and the Riemann surface
which corresponds to 1 pq{.
Theorem 3.80 (Ahlfors’ Finiteness Theorem) For a finitely generated

Kleinian group , the set pq{ is a finite union of compact Riemann
surfaces each with at most finitely many punctures.
In Section 4.3 it is explained how the corollary below follows from Ahlfors’
Finiteness Theorem.
Corollary 3.81 (No wandering domains) If is a finitely generated Kleinian

group and U is a component of pq, then, there exists a non-trivial element
γ P such that γ pU q “ U . In other words, no component of pq is a
wandering component.
Ahlfors’ Finiteness Theorem is the analogue of Sullivan’s Theorem on the

non-existence of wandering domains (Theorem 4.2) for rational maps. In fact
Sullivan gave a unified proof of both results [Su2] (cf. [Be3]) using quasicon-
formal mappings and deformations. We shall explain the ideas and one of the
proofs in detail in Section 4.3.
3.5.5 Kleinian groups and quasiconformal mappings: parameter spaces

To keep the discussion limited we consider a simple class of groups in order to
illustrate the similarities between Kleinian groups and holomorphic dynamics.
Let “ă α, β ą be a free group with two loxodromic generators with the
condition that αβα ´1 β ´1 is parabolic with trace ´2. Examples 3.74 and 3.75
above consider such groups; their quotient space is a pair of punctured tori. For
simplicity, we shall assume that the group is Fuchsian of the first kind and that
the limit set is the unit circle as in Example 3.74.
Let φ be a quasiconformal map of C p such that αφ :“ φ ˝ α ˝ φ ´1 and βφ :“

´ 1
φ ˝ β ˝ φ are Möbius transformations (how to choose φ will be explained
in Section 4.1.3). Then φ “ă αφ , βφ ą is again Kleinian and satisfies that
αφ βφ αφ´1 βφ´1 is parabolic. It follows that pφ q “ φpS1 q is a quasicircle. The
quotient space pφ q{φ is the union of two punctured tori
φ φ
T1 “ φpDq{φ and T2 “ φpCzDq{
p φ.
The group φ is called a quasiconformal deformation of . The space of

quasiconformal deformations of is analogous to the space of rational maps
with two attracting fixed points, a component of Rat2 which depends on two
complex parameters.
The generators αφ and βφ each depend on three complex parameters (the
coefficients of the matrix) and we can normalize so that three points are fixed.
The relation
tracepαφ βφ αφ´1 βφ´1 q “ ´2
(corresponding to the commutator being parabolic) is a relation between the

remaining three parameters so that tφ u is a two-dimensional space. These
two parameters can be thought of as the moduli of the punctured tori.
φ
If the quasiconformal map φ is restricted to be conformal in CzD,
p then T2 “
φpCzDq{
p φ is conformally equivalent to T2 “ pCzDq. The space of quasi-
p
Fuchsian groups
B :“ tφ “ φφ ´1 | φ is conformal in CzD,

p and αφ , βφ P MöbpCqu
p
is called a Bers slice and is analogous to the main hyperbolic component of

the Mandelbrot set. We shall explain these deformations in more detail in
Section 4.1.
If we consider a sequence φn of quasiconformal maps such that φn P B
and so that αφn P φn converges to a parabolic element when n Ñ 8, then φn
tends to a point in the boundary of B. Such a boundary group is considered in
Example 3.76, which is an example of what is known as a geometrically finite
group. To define this concept in the context of Kleinian groups (compare to
Definition 3.44 in holomorphic dynamics) one needs to consider the extension
of an action of a Kleinian group to the hyperbolic 3-ball, which is an orientation
preserving isometry in the hyperbolic metric. The concept of fundamental
domains extends to the 3-ball. We then say that a a Kleinian group is
geometrically finite if it has a fundamental domain which is a polyhedron in
the 3-ball with finitely many sides.
Non-geometrically finite groups are called degenerate. Such groups can

appear as limits of quasi-Fuchsian groups in Bers’ slices. An analogy in
holomorphic dynamics would be Siegel or Cremer polynomials in the cardioid,
the boundary of the main hyperbolic component of the Mandelbrot set.
The Density Conjecture by Bers, Sullivan and Thurston, states that any
finitely generated Kleinian group is a limit of geometrically finite groups.
This has recently been proved as the culmination of extensive and remarkable
work by many authors, in particular the so-called Tameness Conjecture, see
[Mar] for details. The Ahlfors Conjecture, that the limit set of any finitely
generated Kleinian group is either C p or has Lebesgue measure zero, follows
from previous results of Canary and the Tameness Theorem, see [Mar].
As in the case of rational maps, if the limit set of a geometrically finite
Kleinian group is connected, then it is also locally connected [AMa]. (Compare
with comments after Definition 3.44.) Using the above machinery, in a long
Table 3.1 Sullivan’s dictionary
Complex iteration Kleinian groups

f P Rat Y Ent non-elementary
f P Rat with degpf q ě 2 non-elementary and finitely generated
Julia set J pf q Limit set pq
Fatou set F pf q Ordinary set or set of discontinuity pq
J pf q is the smallest, closed, non- pq is the smallest, closed, non-empty,
empty, completely invariant set invariant set.
Repelling periodic points are dense in Loxodromic and parabolic fixed points are
J pf q dense in pq
#tcomp. of F pf qu “ 0, 1, 2 or 8 #tcomp. of pqu “ 0, 1, 2 or 8
F pf q has at most two completely pq has at most two invariant components
invariant components
No Wandering Domains Theorem Ahlfors’ Finiteness Theorem
Blaschke product Fuchsian group
Hyperbolic rational map with two com- Quasi-Fuchsian group
pletely invariant components
Hyperbolic rational map with connected Schottky group
F pf q
f geometrically finite geometrically finite
f hyperbolic rational geometrically finite without parabolic
elements
Main hyperbolic component of the Man- Bers’ slice
delbrot set
series of papers Mahan Mj has claimed an extension of this result to all finitely
generated Kleinian groups.
3.5.6 The Sullivan dictionary

We conclude this section with a summary of the terms that appeared in the pre-
vious discussion, paired with their analogues in iteration theory (see Table 3.1).
The first version of this dictionary appeared in [Su2]. Since then, many empty
spaces have been filled in, and several new lines have been added.
4
Introduction to surgery and first occurrences
In this chapter we describe what quasiconformal surgery is about and explain

the surgeries which were developed as the first occurrences of this technique:
the parametrization of hyperbolic components of the Mandelbrot set and the
proof of the No Wandering Domains Theorem.
Generalities What is known as quasiconformal surgery in holomorphic

dynamics is a technique commonly used to construct holomorphic maps with
prescribed dynamics. The ‘prescribed dynamics’ are given by a map f which
in general is not holomorphic, although it may be. We shall refer to f as the
model map. The word surgery appears because one may need to ‘cut’ and
‘paste’ different spaces and maps together to construct f . This is usually the
first step in the construction and is known as topological surgery.
We are leaving the holomorphic world in order to have a greater choice for
our models, and then checking whether the model map has a ‘holomorphic
dynamical copy’, i.e. whether there exists a holomorphic map conjugate to
f . The main tool for obtaining ‘holomorphic dynamical copies’ is to apply
the Integrability Theorem (Theorem 1.28), which provides a quasiconfor-
mal conjugacy to return to the holomorphic setting (see the Key Lemma
(Lemma 1.39)). It follows that we should look for models in the space of
quasiregular maps (see Proposition 1.37).
Let us be more precise. Let S be a Riemann surface, conformally isomorphic
to D, C or C, p and suppose we have chosen a quasiregular map f : S Ñ S.
We are interested in finding a new map F , which is holomorphic and qua-
siconformally conjugate to f , hence has qualitatively the same dynamics. In
view of the Key Lemma (Lemma 1.39), for this to be possible, we must be
able to construct a Beltrami form μ on S, satisfying }μ}8 :“ k ă 1 and being
f -invariant (i.e. f ˚ μ “ μ), or equivalently to construct an almost complex
147
148 Introduction to surgery and first occurrences
structure σ on S, satisfying Kpσ q ă 8 and being f -invariant. Under these

conditions, the existence of F is assured.
To define an f -invariant almost complex structure with bounded dilatation
is not always possible. It is a natural question to ask under which conditions it
can be done. We address this problem in general in Chapter 5.
We see later, in Chapter 9, that it is also possible to obtain similar results
if the almost complex structure has some areas where the dilatation is not
bounded, as long as these regions are sufficiently small. Such cases involve
the generalized version of the Integrability Theorem (Theorem 9.7) and the
surgery is then called trans-quasiconformal surgery. In this setting, the steps
described above are similar but some of them are more involved.
Let us recall the elementary example of Section 1.5, and see how it
provides – in the simplest case when the model map is holomorphic – an
explicit example of what we just explained. We consider the C-linear map
f :“ M0 : D Ñ D, where M0 pzq “ λ0 z for some 0 ă |λ0 | ă 1. We define a
Beltrami coefficient μν on D, which is the pullback of μ0 , first by an R-
linear map Lν of the plane (hence quasiconformal) and then by a branch of
the logarithm (hence holomorphic). The dilatation of μν is the same as the
dilatation of the map Lν , and therefore constant. By construction, this Beltrami
coefficient μν is M0 -invariant. In this simple example, we are able to write
down explicitly an integrating map φν and see that (after composing with a
logarithm) it conjugates M0 pzq “ λ0 z to F pzq :“ Mν pzq “ eν z.
In general, finding a holomorphic map F that is conjugate to the quasiregular
map f is the last step in the surgery construction. The first step, the construc-
tion of a model map, often requires some creativity. It is at this stage that the
actual ‘surgery’ is involved. There is no general theory on how to do this, since
each problem requires its own particular solution.
One rough classification of the different surgery procedures comes from the
regularity of the model map and its domain.
Soft surgery: the model map is holomorphic The simplest type of surgery
is a change of complex structure without changing the model map (e.g. the
elementary example of Section 1.5). One could argue whether it deserves
the name surgery. Start with a Riemann surface S and a holomorphic map
f : S Ñ S. Suppose we can find an almost complex structure σ on S (different
from σ0 ), which is f -invariant, i.e. f ˚ σ “ σ , and with bounded dilatation.
After applying the Integrability Theorem as in the Key Lemma, we obtain
a holomorphic map F , which is quasiconformally conjugate to the original
map f and is called a quasiconformal deformation of f . One needs to check
whether f and F are different maps, since it might happen that we are back
Introduction to surgery and first occurrences 149
to where we started (as seen in the elementary example, when eν “ λ0 ). If we

always get the original map f back, no matter which invariant almost complex
structure we choose, we say that f is quasiconformally rigid.
Soft surgery in holomorphic dynamics is equivalent to its analogue and pre-
decessor in Kleinian groups. Given a Kleinian group , suppose we can find an
almost complex structure σ on C p which is γ -invariant by all elements γ P ,
and with bounded dilatation. If φ is an integrating map given by the Integra-
bility Theorem, the elements φ ˝ γ ˝ φ ´1 are again Möbius transformations
and φ :“ tφ ˝ γ ˝ φ ´1 | γ P u is again a Kleinian group, quasiconformally
conjugate to . It is called a quasiconformal deformation of .
This type of surgery is sometimes used as a building block for other more
complicated ones, but it is most interesting when performed so that it depends
on one or more parameters. It then allows us, for instance, to parametrize
subsets of the parameter space where the maps (or groups) are structurally
stable. The hyperbolic components of the Mandelbrot set or the Bers’ slices
are examples of this. In this chapter and in Chapter 6 we shall look at several
constructions of this kind.
The space of quasiconformal deformations of a given map f is of great
interest. With some identifications it is known as the Teichmüller space of f ,
see for instance [MS].
Cut and paste surgery: the model map is quasiregular This is the type of
surgery for which most applications have been found. The given model map
f : S Ñ S is quasiregular and often obtained by pasting together different
holomorphic and quasiconformal maps on the Riemann surface S. Sometimes
S itself is constructed by cutting and pasting together different pieces of
the complex plane or other Riemann surfaces. We shall look at examples in
Chapters 7 and 8.
When we obtain results, via surgery, that relate different parameter spaces,
the model maps and the Beltrami forms usually depend on a parameter.
The Riemann surface may be fixed or also depend on the parameter. The
latter case is more involved since it requires the use of the Uniformiza-
tion Theorem. In Chapter 8 we shall consider examples of this kind, both
when the underlying Riemann surface is fixed and when it depends on a
parameter.
First occurrences Inspired by the quasiconformal deformations of Fuchsian

groups (see Section 3.5.5 and the description of soft surgery above), Sullivan
was the first to use quasiconformal mappings in holomorphic dynamics, to
prove the two results which we explain in this section.
The earliest recorded surgery construction in holomorphic dynamics took

place in the fall of 1981. Sullivan in a seminar at Institut des Hautes Études
Scientiques (IHES), and Douady and Hubbard soon after, proved the following
theorem (cf. Section 3.4).
Theorem 4.1 (The multiplier map) Let be a hyperbolic component of

period p of the Mandelbrot set. Then, the multiplier map : Ñ D is a
conformal isomorphism.
The multiplier map is holomorphic, so what remains is to prove its bijectiv-

ity. The proof they gave falls into two parts. Let ˚ “ z´1 p0q consist of
the parameters in corresponding to non-superattracting periodic orbits. First
we prove (following Sullivan) that : ˚ Ñ D˚ is a holomorphic covering
map, and therefore that ´1 p0q is a single point, the centre of . This is
an application of the elementary example in Section 1.5 and a soft surgery
construction, which can be viewed as changing the modulus of a fundamental
domain in the basin of attraction (a torus). The construction is explained
in Section 4.1. We finish the argument by showing that the degree of the
multiplier map is actually one, using a later result of A. Gleason (see [DH2]).
At the end of the section, we sketch the analogous (and earlier) construction
for Fuchsian groups, changing the modulus of a representing punctured torus.
We also show (Theorem 4.7) how the same construction and similar argu-
ments prove the generalization of Theorem 4.1 to Multibrot sets, i.e. the
connectedness loci of unicritical polynomials normalized to be of the form
Qd,c pzq “ zd ` c. More precisely, if is a hyperbolic component of period p
of a Multibrot set Md , then the multiplier map : Ñ D is a holomorphic
branched covering map of degree d ´ 1.
In degree two, there is an alternative proof of Theorem 4.1 due to Douady
and Hubbard, which uses cut and paste surgery on Blaschke products. We
explain this procedure in Section 4.2.
Around the same time, Sullivan proved the celebrated No Wandering
Domains Theorem [Su2].
Theorem 4.2 (No Wandering Domains Theorem) Let f be a rational map

of degree d ě 2. Then, every connected component of the Fatou set of f is
preperiodic.
The proof involves a soft surgery construction which we explain in Sec-

tion 4.3. Using the same arguments Sullivan gave a new proof of Ahlfors’
Finiteness Theorem (Theorem 3.80).
4.1 Changing the multiplier of an attracting cycle

Let be as above, a hyperbolic component of period p of the Mandelbrot
set M.
The technique we shall use to prove that the multiplier map on a hyperbolic
component, : ˚ Ñ D˚ , is a covering map is a soft surgery that can be
applied in many other situations. The essential step, for a given quadratic
polynomial in ˚ , is to change the multiplier of the attracting cycle. Therefore
we shall start by explaining how to change the multiplier in greater generality.
To ease notation we choose to work with rational maps f : C p Ñ C,p although
the technique also applies to entire transcendental maps of C, or transcendental
self-maps of C˚ .
Let fλ0 : C
p ÑC p denote a rational map of degree d ě 2 with an attracting
cycle of period p of multiplier λ0 P D˚ . For any λ P D˚ our first goal is to con-
struct a map fλ quasiconformally conjugate to fλ0 such that the corresponding
p-periodic cycle has multiplier λ. For later purposes we restrict to an arbitrary
simply connected neighbourhood U0 of λ0 in D˚ so that λ P U0 .
First we introduce some notation. Set α “ Opα0 q “ tα0 , α1 , . . . , αp´1 u, an
attracting cycle, where αj `1 “ fλ0 pαj q and p is the smallest integer so that
αp “ α0 . Let Apαq “ Af pαq denote the basin of attraction of α and A˝j the
Ťp´1
connected component of Apαq containing αj , so that A˝ pαq “ j “0 A˝j is
the immediate basin of attraction of α. After composing with a Möbius trans-
formation if necessary, we may assume that Apαq does not contain infinity.
Let 0 be a neighbourhood of α0 in A˝0 , chosen as the preimage of D under a
p
linearizing map ψ0 : 0 Ñ D, that conjugates fλ0 to z ÞÑ λ0 z (see Figure 4.1).
The linearizing map ψ0 : 0 Ñ D is a conformal isomorphism.
A˝1
p
fλ A˝0 α1
0
α0
z ÞÑ λ0 z 0
D ψ0
A˝2
α2
Figure 4.1 The immediate basin of the cycle α and the neighbourhood 0 of α0
p
mapped onto D under the linearizing map ψ0 of fλ . In the figure p “ 3.
0
As explained in Section 3.3.2, the map ψ0 can be extended to the entire basin
p
of attraction Af p pα0 q of α0 as a fixed point of fλ0 , so that ψ0 pAf p pα0 qq “ C
p p
and so that ψ0 conjugates fλ0 to z ÞÑ λ0 z, i.e. ψ0 pfλ0 pzqq “ λ0 ψ0 pzq for all
z P Af p pα0 q. The extended map ψ0 is holomorphic but no longer injective.
Most of the preparation for changing the multiplier was done in the ele-
mentary example of Section 1.5. We adapt the notation in the example to our
situation. Let ν0 be determined so that λ0 “ eν0 and Impν0 q P r0, 2π q, and let
L : U0 Ñ C be the branch of the logarithm satisfying Lpλ0 q “ ν0 . For any
λ P U0 set ν “ Lpλq. In the example we showed how to change the standard
complex structure μ0 ” 0 in C to the Beltrami coefficient μλ pzq “ νν ´ ν0 z
`ν0 z̄
for z P C . (In the example it was denoted μν .) The Beltrami coefficient is
˚
invariant under z ÞÑ λ0 z, and the integrating map φλ (in the example called
φν ), normalized to fix 0 and 1, is a quasiconformal conjugacy between z ÞÑ λ0 z
and z ÞÑ λz. Hence, for all z P C,
φλ˚ μ0 “ μλ and λ φλ pzq “ φλ pλ0 zq.
We now define a Beltrami form on the basin of attraction Af p pα0 q by

pulling back μλ under the holomorphic map ψ0 . By doing so, we obtain a
p
p λ on Af p pα0 q, as the following commutative
fλ0 -invariant Beltrami form μ
diagram indicates:
p
fλ
0
pAf p pα0 q, μ
p λ q ÝÝÝÝ Ñ pAf p pα0 q, μ
pλq
§ §
ψ0 đ
§ §ψ
đ 0
zÞÑλ z
pC, μλ q ÝÝÝÝ0Ñ pC, μλ q
§ §
φλ đ
§ §φ
đ λ
zÞÑλz
pC, μ0 q ÝÝÝÝÑ pC, μ0 q
Observe that the dilatation of μ

p λ is exactly that of μλ , since we pull back by a
holomorphic map.
Remark 4.3 The form μ p λ can be obtained differently: define μp λ initially
just on the neighbourhood 0 of the periodic point α0 . Then, spread μ pλ
p
by the dynamics of fλ0 , to the whole basin of attraction of α0 , similarly to
Section 1.5.1. The main difference is that we have multiple preimages of the
p
annulus 0 zfλ0 p0 q when we pull back.
If p ą 1 we extend μ p λ to the whole basin of attraction Apαq by successive
pullbacks by fλ0 . We shall denote the extended Beltrami coefficient by μ
p λ as
p λ is defined as:
well. More precisely, μ
• on Af p pαp´1 q as the pullback of μ

p λ on Af p pα0 q by fλ0 : Af p pαp´1 q Ñ
Apα0 q, . . . ;
• on Af p pα1 q as the pullback of μ p λ on Af p pα2 q by fλ0 : Af p pα1 q Ñ
Af p pα2 q.
p
Since μp λ on Af p pα0 q is fλ0 -invariant, it follows that the pullback by fλ0 :
Af p pα0 q Ñ Af p pα1 q of μ
p λ on Af p pα1 q coincides with the μ p λ on Af p pα0 q
already defined. We set
p λ “ 0 in CzApαq.
μ p
Since μp λ satisfies the hypothesis of the Integrability Theorem, we know

there exists an integrating map φpλ : C
p Ñ C.
p Therefore, the mapping fλ “
´1 p
φλ ˝ fλ0 ˝ φλ : C Ñ C is holomorphic, hence a rational map of degree d,
p p p
and quasiconformally conjugate to fλ0 . The following diagram commutes:
f λ0
pC,
p μ p λ q ÝÝÝÝ Ñ pC,
p μpλq
§ §
§ §
φpλ đ đφpλ
fλ
pC,
p μ0 q ÝÝÝÝÑ pC,
p μ0 q
Remark 4.4 (1) If in particular fλ0 is a polynomial of degree d to start

with, the almost complex structure μ p λ is equal to μ0 “ 0 in the basin of
attraction of infinity. Normalizing the integrating map φpλ to fix infinity,
it follows that infinity is a superattracting fixed point of fλ , with no other
preimages than itself. Hence fλ is also a polynomial of degree d.
Moreover, if fλ0 is a monic polynomial and φpλ is furthermore chosen
to satisfy φpλ pzq{z Ñ 1 as z Ñ 8 then fλ is also monic. If w “ φpλ pzq
then this follows from
˜ ¸d
fλ pwq φpλ pfλ0 pφpλ´1 pwqq φpλ pfλ0 pzqq fλ0 pzq z
“ “ ÝÑ 1
wd wd fλ0 pzq zd φpλ pzq
as w Ñ 8.
Furthermore, if fλ0 is a monic centred polynomial and if we normalize
φpλ so that the sum of the images by φpλ of the critical points, added with
multiplicity, is also zero, then fλ is again a monic centred polynomial.
With the above three normalization conditions on φpλ we are ready to
apply the Integrability Theorem with holomorphic dependence on the
parameter λ. We shall do so in Section 4.1.2.
(2) Suppose instead that the given map fλ0 is an entire transcendental map,
and normalize again the integrating map to fix infinity. By the topological
properties of essential singularities, the map fλ has an essential singular-

ity at infinity, and hence fλ is an entire transcendental map. Similarly,
if fλ0 : C˚ Ñ C˚ has essential singularities both at the origin and at
infinity, and if the integrating map is normalized to fix the origin as well
as infinity then fλ : C˚ Ñ C˚ has essential singularities at the origin and
infinity.
We need to check that fλ has an attracting p-cycle of multiplier λ. Since
fλ is conjugate by φpλ to fλ0 , the fλ -cycle αλ “ tφpλ pα0 u, . . . , φpλ pαp´1 qu is
attracting, so the first part is clear. The second part follows from the follow-
ing fact.
Claim 4.5 (Linearizing map) The composition
ψλ “ φλ ˝ ψ0 ˝ φpλ´1 : φpλ p0 q ÝÑ D

p
is holomorphic and conjugates fλ to z ÞÑ λz. Therefore ψλ is a linearizing
p
map around the attracting fixed point φpλ pα0 q of fλ .
Proof The map ψλ : φpλ p0 q Ñ D preserves, under pullback, the standard
complex structure, as shown in the following commutative diagram:
p
fλ
pφpλ p0 q, μ0 q / pφpλ p0 q, μ0 q
O O
φpλ φpλ
p
fλ
0
p0 , μ
pλq / p0 , μ
pλq
ψλ ψ0 ψ0 ψλ
zÞÑλ0 z
pD, μλ q / pD, μλ q
φλ φλ
zÞÑλz

pD, μ0 q / pD, μ0 q
Therefore by Weyl’s Lemma, ψλ is holomorphic. Since ψλ is a conformal

p
map conjugating fλ around φpλ pα0 q to the linear map z ÞÑ λz around the
origin, it follows that ψλ is a linearizing map.
This concludes our first goal.

4.1.1 The monic polynomial case: reading the Böttcher coordinates

Suppose fλ0 is a monic polynomial of degree d ě 2. Let φpλ be the integrating
map normalized to fix infinity and satisfying φpλ pzq{z Ñ 1 as z Ñ 8. Then by
Remark 4.4 above, fλ is a monic polynomial. The basin of attraction Apαq of
the cycle α is contained in the filled Julia set Kλ0 :“ Kfλ0 .
If the filled Julia set Kλ0 is connected then the Böttcher map ϕλ0 is defined
on the entire basin of attraction of infinity of fλ0 , i.e. ϕλ0 : CzKλ0 Ñ CzD. For
simplicity we assume that Kλ0 is connected.
Claim 4.6 (Böttcher coordinates) Let ϕλ0 : CzKλ0 Ñ CzD be the Böttcher
mapping of fλ0 on the basin of attraction of infinity. Then, the map
ϕλ “ ϕλ0 ˝ φpλ´1 : Czφpλ pKλ0 q Ñ CzD
is the Böttcher mapping of fλ on the basin of attraction of infinity.
Proof Recall that a mapping ϕ which conjugates a monic polynomial to z ÞÑ

zd in neighbourhoods of infinity is the Böttcher mapping around infinity if and
only if it satisfies ϕpzq{z Ñ 1 as z Ñ 8.
Observe that we defined μ p λ pzq “ 0 for all points z in CzKλ0 . It follows that
the integrating map φpλ is holomorphic on this set, since it transports μ0 “ 0 to
itself. Its inverse on this set is holomorphic as well. This means that ϕλ “ ϕλ0 ˝
φpλ´1 is holomorphic and conjugates fλ to z ÞÑ zd on the basin of attraction of
infinity of fλ as the following commutative diagram indicates:
zÞÑzd
pCzD, μ0 q / pCzD, μ0 q
9 O O e
ϕλ0 ϕλ0
fλ0
ϕλ pCzKλ0 , μ0 q / pCzKλ , μ0 q ϕλ
0
φpλ φpλ

fλ
pCzφpλ pKλ0 q, μ0 q / pCzφpλ pKλ q, μ0 q
0
Finally, we have ϕλ pwq{w Ñ 1 as w “ φpzq

p Ñ 8 since
ϕλ0 pφpλ´1 pwqq ϕλ pzq z

“ 0
w z φpλ pzq
and both factors tend to 1.

If the polynomial has a disconnected Julia set, then the Böttcher map is still
defined by ϕλ “ ϕ0 ˝ φpλ´1 , but only on a neighbourhood of infinity.
4.1.2 Application to families of unicritical polynomials

In this section we prove a generalization of Theorem 4.1 to unicritical poly-
nomials of degree d ě 2. Monic centred unicritical polynomials of degree d
are of the form Qc pzq “ Qd,c pzq “ zd ` c. Let Md denote the connectedness
locus in the c-plane of Qd,c . Figure 4.2 shows the Multibrot set for d equal to
3 and 4.
Figure 4.2 The Multibrot set for cubic and quartic unicritical polynomials. Notice the
central hyperbolic components containing c “ 0 and corresponding to the component
in the Mandelbrot set bounded by a cardioid. The bounding curve of the central
hyperbolic component of the Multibrot set is always an epicycloid with d ´ 1 cusps.
In the following d is fixed. Let be a hyperbolic component of Md of

period p. Then for any c P the polynomial Qc has an attracting p-cycle. Let
: Ñ D denote the multiplier map. We shall prove the following theorem.
Theorem 4.7 (Multiplier map for unicritical polynomials) Let denote a

hyperbolic component of Md . Then the multiplier map “ d : Ñ D is a
holomorphic branched covering of degree d ´ 1, ramified over 0 in D.
Set ˚ “ ´1 pD˚ q. Proving that : ˚ Ñ D˚ is a covering map is due

to Sullivan. The computation of the degree follows from an argument of
Gleason
The steps in the proof are:
(i) construct a local holomorphic inverse to : ˚ Ñ D˚ showing that the

map is a covering map;
(ii) deduce that ´1 p0q is a single point in ;
(iii) determine the degree of the map .
Fix a point c0 P ˚ with pc0 q “ λ0 ‰ 0. Choose an arbitrary simply con-

nected neighbourhood U0 of λ0 , compactly contained in D˚ . For any λ P U0
we shall construct a monic centred polynomial Qcpλq pzq “ zd ` cpλq, which
is quasiconformally conjugate to Qc0 and satisfies
cpλ0 q “ c0 , pcpλqq “ λ, and λ ÞÑ cpλq is holomorphic.
It follows that cpλq P ˚ , and that we have constructed a holomorphic local

inverse of in the neighbourhood U0 of λ0 . Therefore : ˚ Ñ D˚ is a
holomorphic covering map (see Figure 4.3).
˚ D˚
cpλq
c0
U0
λ0

´1 p0q 0
Figure 4.3 The map λ ÞÑ cpλq is a holomorphic local inverse of the multiplier
map .
Notice that the construction of the previous sections, taking fλ0 “ Qc0 , and
using the normalization of the integrating map φpλ explained in Remark 4.4
(1) already provides a monic centred polynomial fλ of degree d with one
critical point at 0 and with a p-cycle of multiplier λ. These properties
imply that fλ pzq “ zd ` cpλq, and moreover cpλq “ φpλ pc0 q, the critical value
of fλ .
It is clear that, if α0 is one of the points in the p-cycle for Qc0 , then
p
pcpλqq “ pQcpλq q1 pφpλ pα0 qq “ λ, as we wanted to show.
The next step in the proof is to draw conclusions from the holomorphic
dependence on parameters.
p λ satisfy
Claim 4.8 There exists k ă 1 such that the Beltrami coefficients μ
μλ ||8 ď k for all λ P U0 . Moreover,
that ||p
p λ pzq is holomorphic in λ for all z P C.

λ ÞÑ μ
The integrating mappings φpλ with λ P U0 are normalized as to fix 0 and have
φpλ pzq
z Ñ 1 as z Ñ 8. It follows that:
(a) λ ÞÑ φpλ pzq is holomorphic in λ for all z P C;

(b) in particular the map λ ÞÑ φpλ pc0 q “ cpλq is holomorphic.
Proof Recall that L : U0 Ñ C is chosen as the branch of the logarithm sat-

isfying Lpλ0 q “ ν0 where Impν0 q P r0, 2π q. Set ν “ Lpλq for λ P U0 . Then ν
depends holomorphically on λ. As it was pointed out in the elementary exam-
ple of Section 1.5, the Beltrami coefficient μλ in C varies holomorphically with
respect to ν, hence also with respect to λ. Now μ p λ on AQpc pα0 q is obtained
0
from μλ on C by pulling back with the extended linearizing map ϕ0 , which is
independent of λ. Therefore for z P AQpc pα0 q
0
p λ pzq “ ϕ0˚ μλ pzq

λ ÞÑ μ
is also holomorphic in λ. If the period p ą 1, this holomorphic dependence is

spread to all z P Apαq by successive pullbacks with Qc0 , also independent of
p λ pzq “ 0 on CzApαq is trivially holomorphi-
p λ by μ
λ. The final extension of μ
cally dependent on λ for all z P CzApαq. Set for all λ P U0
ˇ ˇ
ˇ Lpλq ´ Lpλ q ˇ
0 ˇ
kpλq “ ˇ ˇ.
ˇ
ˇ Lpλq ´ Lpλ0 q ˇ
Then λ ÞÑ kpλq is a continuous real valued function on U0 that takes values in

r0, 1q. Moreover, we have
μλ ||8 “ ||μλ ||8 “ kpλq.

||p
Set
k “ sup kpλq.
λPU0
Since U0 is compact, the supremum is realized as the maximal value of kpλq

for λ P U0 . Hence k ă 1 and ||p
μλ ||8 ď k for all λ P U0 . We are now ready to
apply the Integrability Theorem with parameters. It follows that λ ÞÑ φpλ pzq is
holomorphic in λ for all z P C. Since φpλ is a conjugacy between Qc0 and fλ ,

it maps the critical value c0 of Qc0 , to the critical value φpλ pc0 q of fλ . In other
words,
fλ “ Qcpλq where cpλq “ φpλ pc0 q.
It follows from above, that cpλq depends holomorphically on λ. This finishes

the proof of Claim 4.8. (Compare with Lemma 1.40.)
Claim 4.9 (Uniqueness of centre) The inverse image by the multiplier map
: Ñ D of the centre of D is a single point of , called the centre of .
Proof By the Riemann–Hurwitz formula (see e.g. [Bea, Thm. 2.7.1 or

5.4.1.]), the Euler characteristic of ˚ has to be one, which implies that
´1 p0q is a single point.
Corollary 4.10 (Quasiconformal conjugacy between polynomials in the

same hyperbolic component) Any pair Qc1 , Qc2 of unicritical polynomials
of degree d with c1 , c2 P ˚ are quasiconformally conjugate.
See Figure 3.12. The last step in the proof of Theorem 4.7 is to prove that
the degree of the multiplier map : Ñ D is d ´ 1. Let c˚ denote the centre
p
of the hyperbolic component, and assume has period p. Then Qc˚ p0q “ 0.
The following lemma is due to Gleason. The proof can be found in [BDH,
BPer, DH2].
Lemma 4.11 Consider the polynomials Gn pcq “ Qnc p0q for n ě 1. All roots
of Gn are simple.
Proof The polynomials Gn are defined recursively by
G1 pcq “ c and Gn pcq “ pGn´1 pcqqd ` c.
It follows that Gn is monic with integer coefficients. The roots of Gn are

therefore algebraic integers over Z, since the ring of algebraic integers A over
Z consists of complex numbers which are roots of monic polynomials with
integer coefficients.
Assume there exist an n ą 1 and a root cr of Gn which is not simple. Then cr
must be a root of both Gn and its derivative
G1n pcq “ dpGn´1 pcqqd ´1 G1n´1 pcq ` 1.

It follows that
´1
cqqd ´1 G1n´1 pr
pGn´1 pr .
cq “
d
Since the ring A is closed under addition and multiplication [NZM,
Thm. 9.12], it follows from the equation above that ´d1 P A. But this
is impossible since Q X A “ Z [NZM, Thm. 9.2]. Hence Gn has no
multiple roots.
We are ready to prove that the centre c˚ of is a zero of multiplicity d ´ 1
of the multiplier map . For any c P the points in the attracting p-cycle are
solutions of
p
F pc, zq “ Qc pzq ´ z “ 0.
The map F satisfies the hypothesis of the Implicit Function Theorem at the
point pc˚ , 0q, since Fz pc, zq |pc˚ ,0q “ pc˚ q ´ 1 “ ´1 and Fc pc, zq |pc˚ ,0q “
G1p pc˚ q ‰ 0 by Lemma 4.11. In a neighbourhood of c˚ the points in the attract-
ing p-cycle can therefore be expressed as functions of c. Let α0 pcq denote
the periodic point in the Fatou component containing the critical point 0, so
that α0 pc˚ q “ 0. It follows that α01 pc˚ q ‰ 0, so that α0 pcq has a simple zero at
j
c “ c˚ . Set αj pcq “ Qc pα0 pcqq for j “ 1, . . . , p ´ 1.
The multiplier map can be written as
p
ź ´1
pź
pcq “ Q1c pαj pcqq “ dpαj pcqqd ´1 .
j “1 j “0
The multiplicity of c˚ as a zero of is equal to the sum of the multiplicities of

c˚ as a zero of αj pcq, raised to the power d ´ 1. But this is zero for all j ‰ 0
and one for α0 pcq. We conclude that has a zero of multiplicity d ´ 1 at c˚ .
This finishes the proof of Theorem 4.7 and in particular of Theorem 4.1.
4.1.3 Changing the modulus of a fundamental domain

of a Fuchsian group
In this section we sketch the soft surgery used to obtain quasiconformal
deformations of a Fuchsian group, making the contents of Section 3.5.5 more
precise. We shall use the concepts and notation from Section 3.5. Our goal is
to stress the parallelism with the surgery described above, where we changed
the multiplier of an attracting cycle.
For simplicity, we start with the Fuchsian group “ă α, β ą, where
is free, α and β are two loxodromic elements, αβα ´1 β ´1 is parabolic, and
pq “ S1 . See for instance Example 3.74 in Section 3.5. Let Q denote a
fundamental domain for in D, whose sides are identified in pairs by the

generators α and β (see Figure 3.17, where Q “ Q1 ). With these identi-
fications, D{ is conformally equivalent to a punctured torus with marked
generators. Hence, there exists a unique ν0 P H so that this torus is conformally
equivalent to the torus obtained from the parallelogram T0 in C with vertices at
t0, 1, ν0 , 1 ` ν0 u and opposite sides identified linearly. Let ϕ : Q Ñ T0 denote
the unique conformal map that induces the conformal equivalence among the
tori, the quotient spaces.
The number ν0 is in fact the complex modulus of the torus arising from
Q, its imaginary part is the (real) modulus of the quadrilateral Q without
identifications (see Section 1.3.4). Because the puncture can be at any point,
we may identify the punctured torus arising from Q with the unpunctured
quotient. If is the group in Example 3.74, then ν0 is purely imaginary due to
the symmetry.
Choose an arbitrary ν P H, let Tν denote the parallelogram with vertices at
t0, 1, ν, 1 ` νu, and let Lν : C Ñ C denote the unique R-linear map which
maps T0 to Tν (preserving the order of the vertices). It can be checked that
BLν M BLν ν ´ ν0
μpLν q “ “ .
Bζ Bζ ν0 ´ ν
The Beltrami coefficient μpLν q “ L˚ν μ0 defines a field of infinitesimal ellipses
in T0 which can be pulled back to Q under the conformal map ϕ : Q Ñ T0 . Set
μν “ ϕ ˚ μpLν q and observe that this defines an almost complex structure in Q
of bounded dilatation.
We now use the fact that Q is a fundamental domain to spread μν by the
action of . More precisely, for every point z P D, there exists a unique γ P
such that γ pzq P Q. Indeed if two different elements γ , γ 1 P were to satisfy
this condition, mapping z to two different points in Q, then both points would
be in the orbit of z, violating the condition that Q is a fundamental domain.
On the other hand, if both elements mapped z to the same point, then either
γ 1 γ ´1 would be a relation violating that is free or z would be a fixed point
of γ 1 γ ´1 and hence lie in the limit set. Thus, for every z P D, it makes sense
to define μν pzq “ γ ˚ μν pγ pzqq, where γ is as defined above. This extends μν
to the whole unit disc.
Set μν ” 0 in CzD,
p and observe that μν , by construction, is invariant under
the action of . Its dilatation is bounded given that all pullbacks are done by
Möbius transformations. We may then apply the Integrability Theorem and
obtain a quasiconformal map φν satisfying φν˚ μν “ μ0 . If we normalize φν
to fix the two fixed points of α and the attracting fixed point of β, then φν is
unique and depends holomorphically on the parameter ν, since μν does.
Define a new group

ν :“ φν φν´1 :“ tφν ˝ γ ˝ φν´1 | γ P u.
Because μν is -invariant, it follows that the elements of ν are conformal
(and hence Möbius transformations), so that ν is again a Kleinian group.
Thus for every ν P H, we obtain a quasiconformal deformation of . Since
pν q “ φν ppqq “ φν pS1 q and pν q “ φν ppqq,
the limit set of the new group is a quasicircle, and therefore ν is a quasi-
Fuchsian group. Let ν1 :“ φν pDq and ν2 :“ φν pCzDq p denote the invari-
ant components of pν q. The quadrilateral Qν :“ φν pQq is a fundamental
domain for ν in the component ν1 and ν1 {ν is isomorphic to a punc-
tured torus of complex modulus exactly ν. Indeed, Lν ˝ ϕ ˝ φν´1 is con-
formal and maps Qν to Tν . Also observe that φν is conformal on CzD,
hence no quasiconformal deformation takes place on the invariant component
of pν q.
This provides a family B of quasi-Fuchsian groups parametrized by the
complex parameter in the upper half plane, viewed as the complex modulus
of one of the two punctured tori of pq{. This is analogous to the family of
quadratic polynomials Qc with c belonging to the main hyperbolic component
of the Mandelbrot set. This component is parametrized by a complex parameter
in D, viewed as the multiplier of the attracting fixed point αc of Qc , or
equivalently, by the complex modulus of the torus pAc zαc q{Qc where Ac is
the basin of attraction of αc .
As a final remark, if ν 1 “ ν0 ` for some P Z observe that we obtain the
same group and the same torus, just with a different set of generators, i.e.with a
different marking (recall Definition 3.68). This fact is similar to the parameter
periodicity observed when changing the multiplier of an attracting cycle.
4.2 Changing superattracting cycles to attracting ones

4.2.1 Preliminary: Blaschke products
In this section we present a simple example of cut and paste surgery. It will
serve as a building block for the construction in the next section.
Consider the family of Blaschke products
z`λ
Bλ pzq “ z , parametrized by λ P “ D.
1 ` λz
Each Bλ is a quadratic rational map with fixed points at 0, 8 and 11´
´λ
λ
P
BD. The fixed points at 0 and 8 are attracting with multipliers λ and λ
respectively. (The inversion I : z ÞÑ 1z conjugates Bλ to Bλ .) The third fixed

point is repelling. The unit disc D and its inversion CzD are invariant under Bλ
and form the basins of attraction of 0 and 8 respectively (by Schwarz Lemma).
Hence the unit circle is the Julia set of Bλ . Note that B0 pzq “ z2 .
We restrict our attention to the dynamics of Bλ on D. For any 0 ă r ă 1, if
we set Dr “ t|z| ă ru, then Dλ1 “ Bλ´1 pDr q compactly contains Dr , i.e. Dr Ă
Dλ1 . This is easy to see by observing that if |z| “ r then |Bλ pzq| ă r, since
Bλ pzq{z maps D to itself.
If λ ‰ 0 and r is small, Dλ1 consists of two connected components. Instead if
r ą |λ|, Dλ1 consists of one single connected component and therefore Dλ1 zDr
is an annulus. Indeed, when r ą |λ| the second preimage ´λ of zero lies
inside Dr , which implies that the map Bλ : Dλ1 Ñ Dr has degree two, the total
degree of Bλ . Summarizing, for any parameter λ P D, and for any r ą |λ|,
we have
Bλ pDr q Ă Dr , Bλ pDλ1 q “ Dr , Bλ pDzDλ1 q “ DzDr .
See Figure 4.4.

We use this family of Blaschke products to illustrate the basic construction
of pasting maps together.
D
Bλ
Dλ1
Bλ p D r q
r
0
Figure 4.4 The initial setup for a Blaschke product Bλ , in the case r ą |λ|.
Fix a parameter λ0 P D, and fix |λ0 | ă r ă 1. Then, for all |λ| ă r, we

define a quasiregular mapping gλ : D Ñ D, which is the result of pasting
together three maps.
Denote by Aλ0 the closed annulus Dλ1 0 zDr . We define gλ : D Ñ D as
$
&Bλ0
’
’ on DzDλ1 0 ,
gλ “ Bλ on Dr ,
’
’
%h
λ on Aλ0 ,
where hλ is a quasiregular map, which coincides with Bλ0 on the outer

boundary of Aλ0 and with Bλ on the inner boundary S1r (see Figure 4.5).
Observe that these boundary maps are analytic maps of degree two on analytic
curves. Moreover, the outer boundary map is fixed and the inner boundary map
varies continuously with λ for each fixed z P S1r , i.e. λ ÞÑ Bλ pzq “ z 1z`
`λ
λz
is
continuous in λ. Then by Exercise 2.3.3 at the end of Chapter 2 we conclude
that hλ can be chosen as a covering map of degree two depending continuously
on λ for each z P Aλ0 .
D D
Aλ0
Bλ
Aλ
r 1 r 1
hλ
Bλ0
Figure 4.5 Pasting together different Blaschke products via an interpolating

quasiregular map hλ .
We proceed to define a gλ -invariant Beltrami form μλ . We first define it on

Dr as μλ “ μ0 . Since gλ “ Bλ is holomorphic on Dr and Bλ pDr q Ă Dr , it
follows that μλ is invariant in Dr . Next we define μλ in the region where gλ is
not holomorphic, i.e. on the annulus Aλ0 , by setting μλ “ h˚λ μ0 . Finally, we
spread μλ to the remaining part DzDλ1 0 , by the dynamics of gλ . To do this it
is important to observe that for every point z P DzDr there is a unique n ě 0
such that gλn pzq belongs to the half open annulus Aλ0 zBDr . Moreover, if n ą 0
then gλn pzq “ Bλn0 pzq. Hence μλ is defined recursively on D as
$
μ on Dr ,
& 0
’
’
μλ “ ´h˚λ μ0 on Aλ0 ,
% Bn ˚ μ
’ ¯
on Bλ´0n pAλ0 q.
’
λ0 λ
By construction μλ is gλ -invariant. Moreover, for each λ P Dr 1 , the Beltrami

coefficient μλ has dilatation bounded away from 1 on the compact annulus
Aλ0 . The same is true on all of D, since the dilatation is unchanged when
pulling back by the holomorphic map Bλ0 , i.e. }μλ }8 :“ kpλq ă 1. Further-
more, λ Ñ μλ pzq varies continuously with λ for all z P Aλ0 , hence also for all
z P D. Therefore kpλq varies continuously with λ. For all λ P Dr 1 the Beltrami
coefficient has dilatation bounded by k :“ max|λ|ďr 1 kpλq ă 1.
We are ready to apply the Integrability Theorem. Note that for any λ P D
we have a quasiconformal integrating map φλ : D Ñ D which fixes 0, and
is uniquely determined up to post-composition by a rotation about 0. Any
such map extends to the boundary. Let´ us still ¯ denote the extension by
φλ : D Ñ D and choose it such that φλ 11´ ´λ
λ0
“ 11´
´λ
λ
. Consequently, the
0
composition Gλ “ φλ ˝ gλ ˝ φλ´1 is a holomorphic self-map of D of degree
two, hence a Blaschke product. The multiplier of the fixed point at 0 is λ,
since φλ is holomorphic in Dr . Since 11´
´λ
λ
is a fixed point of Gλ , it means that
Gλ “ Bλ .
The map gλ and the gλ -invariant Beltrami coefficient μλ can be extended to
all of C.
p If this is done we would end by applying the Integrability Theorem
depending on parameters for C,p as suggested in Exercise 4.2.1.
In the previous section we described an alternative change of multiplier. The
resulting map was always conjugate to the original one. As a consequence,
none of the multipliers could be taken to be equal to 0. Note, however, that this
construction works for either λ or λ0 equal to 0.
4.2.2 Gluing the Blaschke product into basins of attraction

By applying part of the construction from Section 4.2.1 we are able to give an
alternative proof of Theorem 4.1, proving in one step that the multiplier map
is a covering and that its degree is exactly one. We recall the precise statement
in the following proposition.
With the technique based on the Blaschke product model we show that, for
any c0 P , the multiplier map has a local inverse around λ0 “ pc0 q P D,
including the case λ0 “ 0.
We start by giving the basic idea in the proof. Let c0 be any point in ,
and let as before α “ tα0 , . . . , αp´1 u be the attracting p-cycle of Qc0 . Let
A˝j be the component of the immediate attracting basin A˝ pαq containing αj ,
and let the numbering be chosen so that A˝0 contains the critical point 0. Let
R0 : A˝0 Ñ D denote the uniquely determined Riemann mapping that maps
α0 to 0, and when extended continuously to the boundary of A˝0 , maps the
p p
fixed point of Qc0 in BA˝0 to 11´ ´λ
λ0
in BD. Then R0 conjugates Qc0 : A˝0 Ñ
0
`λ0
A˝0 to the Blaschke product Bλ0 : D Ñ D where Bλ0 pzq “ z 1z` λ0 z
, since any
holomorphic self-map of D of degree two is a Blaschke product of this form. In
order to construct a local inverse to pc0 q “ λ0 in a neighbourhood of λ0 , we
paste the two Blaschke products Bλ0 and Bλ together via an interpolating map
hλ , for λ in the neighbourhood of λ0 , as explained in Section 4.2.1. We change
the complex structure in D accordingly. Then we pull back the construction
first by the Riemann mapping R0 to A˝0 , then by Qc0 to all of Apαq, and finally
we let the map and the complex structure be left unchanged outside Apαq.
We shall give more details only when c0 P ´1 p0q since this is the case that
was not covered by the earlier construction. In this case the Blaschke product
Bλ0 is equal to B0 pzq “ z2 and the Riemann mapping R0 equals the Böttcher
p
mapping that conjugates Qc0 : A˝0 Ñ A˝0 to B0 : D Ñ D.
Proof of Theorem 4.1 Let c0 belong to ´1 p0q. Choose any 0 ă r 1 ă r ă 1

and |λ| ď r 1 ă r. Then the disc Dr “ t|z| ă ru is compactly contained in
Bλ´1 pDr q and Bλ´1 pDr qzDr is an annulus. Using the construction from
Section 4.2.1 we obtain the following commutative diagram:
fλ
pA˝0 , ηλ q ÝÝÝÝÑ pA˝0 , ηλ q
§ §
R0 đ đR0
§ §
gλ
pD, μλ q ÝÝÝÝÑ pD, μλ q
§ §
φλ đ
§ §φ
đ λ
λB
pD, μ0 q ÝÝÝÝ Ñ pD, μ0 q
1
where fλ “ R´ 0 ˝ gλ ˝ R0 and ηλ “ R0 μλ is an fλ -invariant Beltrami form
˚
on A0 .
˝
Note that D01 “ B0´1 pDr q “ D?r so that gλ “ B0 on DzD?r , hence fλ “

p 1
Qc0 on A˝0 zR´
0 pD r q. We define Fλ : C Ñ C as follows:
?
# ˇ´1
pQc0 ˇ qp´1 ˝ fλ on A˝0 ,
Fλ “
Qc 0 on CzA˝0 ,
ˇ´ 1
where pQc0 ˇ qp´1 is short for
ˇ´1 ˇ´1 ˇ´1
pQc0 ˇ qp´1 “ Qc0 ˇA˝ ˝ ¨ ¨ ¨ ˝ Qc0 ˇA˝ ,
1 p´1
and we use that Qc0 : A˝j Ñ A˝j `1 is a bijection for j “ 1, . . . , p ´ 1.

The global mapping Fλ is holomorphic everywhere except on the annu-
1
lus R´ ?
0 pD r zDr q, where it is quasiconformal. In fact Fλ is equal to Qc0
1
everywhere except on R´ ?
0 pD r q.
We spread ηλ on A0 to an Fλ -invariant Beltrami form on C, which we denote
˝
again by ηλ . Notice that, for any z P ApαqzA˝0 , there is a unique n ą 0 such

that Qnc0 pzq P A˝0 , and define ηλ on C by
$
&μ0
’
’ on CzApαq,
η λ “ ηλ on A˝0 ,
’
%pQn q˚ η
’
on Q´ n ˝ ˝
c0 λ c0 pA0 qzA0 .
By construction ηλ is an Fλ -invariant Beltrami form with dilatation bounded

by the bound for the dilatation of μλ on D?r zDr . We are ready to apply
the Integrability Theorem to obtain a quasiconformal integrating mapping φpλ ,
normalized so that 0 and 8 are fixed and φpλ is tangent to the identity at
8. It follows that the mapping φpλ ˝ Fλ ˝ φpλ´1 is a quadratic polynomial of
the form Qcpλq pzq “ z2 ` cpλq with an attracting p-cycle namely φpλ pαq “
tpφλ pα0 q, . . . , φλ pαp´1 qqu of multiplier λ.
Finally we add parameters to the problem. We know from Section 4.2.1
that λ ÞÑ gλ pzq is continuous for all z P D. It follows that λ ÞÑ fλ pzq and
λ ÞÑ Fλ pzq are continuous for all z in A0 respectively in C. Moreover, μλ pzq
depends continuously on λ for each z P D and has a dilatation that is uniformly
bounded away from 1. It follows that the same is true for the resulting ηλ pzq
for all z P C. Therefore, for each z P C the integrating map φpλ pzq depends
continuously on λ. Hence, the attracting p-cycle, φpλ pαq, varies continuously
with λ. Note that cpλq “ φpλ ˝ Fλ p0q and that Fλ p0q is independent of λ. Hence
cpλq varies continuously with λ and belongs to . We have constructed a local
inverse to :
λ P Dr 1 ÞÑ cpλq P with pcpλqq “ λ.
It follows that ´1 p0q consists of a single point, the centre of .
The dynamics of Qc for c P ˚ is not globally conjugate to the dynamics

of the centre polynomial Qc0 . In the first case the critical orbit has infinitely
many points, while it has only p points in the second. However, it follows that
all polynomials in are quasiconformally conjugate in a neighbourhood of
their Julia set. In particular, they are all J -conjugate (see Definition 3.28).
Corollary 4.12 (Conjugacy to the centre in a neighbourhood of J ) Any

quadratic polynomial Qc with c P is quasiconformally conjugate to the cen-
tre polynomial Qc0 of , in some neighbourhoods of their respective Julia sets.
Remark 4.13 In Section 7.5 we shall see an alternative construction to the one
presented here. There, the gluing is done along the boundary of the immediate
basin of attraction, and the resulting map is conjugate to the original only on
the Julia set.

4.2.1 Observe that τ pzq “ 1{z, the reflection with respect to S1 , conjugates
`λ
Bλ pzq “ z 1z` λz
to itself. As in Section 4.2.1 choose 0 ă r 1 ă r ă 1
and λ, λ0 P Dr 1 . Extend gλ : D Ñ D to gλ : C p ÑC p by defining
#
Bλ0 on BD,
gλ “
τ ˝ gλ ˝ τ on CzD.
p
Define a gλ -invariant Beltrami form μλ on C

p depending continuously
on λ for each z P C. Show that this can be done so that with a proper
p
normalization of the integrating maps φλ : C
p Ñ C,
p the resulting holo-
morphic map Gλ :“ φλ ˝ gλ ˝ pφλ q : C
´ 1 p ÑC p is equal to Bλ .
Remark: The exercises below are special cases of a more general theorem of
McMullen (see Section 7.5).
4.2.2 Suppose P is a polynomial with an attracting fixed point α and assume

that P has degree d when restricted to the immediate basin of α, say
A˝ pαq. Show that there exists a polynomial Q and a quasiconformal
homeomorphism φ : C Ñ C conjugating P and Q on a neighbourhood
of the respective Julia sets, and in all Fatou components except A˝ pαq,
such that φpαq is a superattracting fixed point of Q of multiplicity d,
and is the unique critical point in its basin. In particular, the dynamics
of Q in the immediate basin of φpαq is conjugate to z ÞÑ zd . Finally
observe the analogous result for periodic attracting orbits.
Hint: Under a Riemann map, P |A˝ induces a degree d map from D
onto itself, which therefore is a Blaschke product.
4.2.3 Suppose P is a polynomial with a superattracting fixed point, say α,
whose immediate basin, A˝ pαq, contains no other critical point than
itself. Let U be a Fatou component which is mapped to A˝ pαq with
degree k ě 3 (i.e. containing at least two critical points). Prove that
there exists a polynomial Q and a quasiconformal homeomorphism
φ : C Ñ C conjugating P and Q on a neighbourhood of the respective
Julia sets, and in all Fatou components except U , such that φpU q
contains a unique critical point of multiplicity k ´ 1 which is mapped
to φpαq.
4.2.4 Generalize the procedure in Exercise 4.2.2 to other components of the
basin of attraction, to show that the same statement would hold if we
asked Q to have one single critical grand orbit in the whole basin of
attraction φpαq.
4.3 No wandering domains for rational maps

In this section we prove the celebrated theorem of Sullivan about the
non-existence of wandering domains for rational maps (Theorem 4.2). The
original proof is in [Su2], using soft surgery. In the same paper he used similar
arguments to also prove the non-existence of wandering domains for Kleinian
groups as well as Ahlfors’ Finiteness Theorem (Theorem 3.80).
We start by giving an idea of the common strategy to prove the absence of
wandering components in both settings. Afterwards we give the details in the
rational case.
Let R denote a rational map and a finitely generated Kleinian group.
We recall the definition of a wandering domain: a Fatou component U is
wandering for a rational map R if R n pU q ‰ R m pU q for all m ‰ m. Likewise,
a component U of pq is a wandering component if γ pU q ‰ U for all γ P
different from the identity.
The idea is as follows. Start by recalling that Ratd , the space of rational
maps of degree d, can be identified with C2d `1 , a complex vector space of
finite dimension. Analogously, the space of quasiconformal deformations of a
finitely generated Kleinian group is finite dimensional. This is due to the fact
that every deformation induces a representation of into PSLp2, Cq, and the
space of such representations Homp, PSLp2, Cq is finite dimensional, when
is finitely generated (see e.g. [An]).
Suppose U is a wandering component of FR (or pq). Then, Sullivan
shows that there exists a space of ‘non-equivalent’ Beltrami forms supported
on U , of arbitrarily high dimension. One may then use R (or ) to spread these
Beltrami forms and obtain an arbitrarily high dimensional space of Beltrami
forms supported on FR (or pq). After having applied the Integrability
Theorem we obtain an arbitrarily high dimensional space of non-equivalent
quasiconformal deformations of R (or ), and hence an arbitrarily high dimen-
sional subspace of Ratd (or Homp, PSLp2, Cq). This is a contradiction.
To see the full proof of Ahlfors’ Finiteness Theorem we refer to [Su2, Ah1,
Be3] or [MT].
We now proceed to explain in detail the rational case. The proof can be split
into three steps:
(1) Show that if there were a wandering domain U , all its iterates Un “
f n pU q for large enough n would be simply connected.
(2) Construct, for arbitrarily large N , an N -dimensional real analytic family
of Beltrami forms μλ , for λ in a subset of RN , invariant by f and with
bounded dilatation. After integrating, these Beltrami forms induce a real
analytic family fλ of quasiconformal deformations of f , all rational maps

of the same degree as f .
(3) Since the complex dimension of the space Ratd of rational maps of degree
d is 2d ` 1, there must be – for N ą 4d ` 2 – a path of Beltrami forms
μλpt q , t P r0, 1s, inducing the same rational map g, that is fλpt q “ g. This
will contradict the properties of the Beltrami forms we constructed in the
second step.
There exist several rewritings and simplifications of Sullivan’s proof,

although all of them essentially follow the same three steps. We choose to
present Baker’s approach to the first step (Proposition 4.14), which is a signif-
icant improvement. We present Lemma 4.15 following Steinmetz [St] as part
of the second step. The idea is similar to Sullivan’s, but more constructive. For
the rest, we choose to follow Sullivan’s original arguments, being consistent
with our goal to give some ‘historical flavour’ to this section.
Let us start by addressing the first step.
Proposition 4.14 (Eventual properties of wandering domains) Let f be a

rational map of degree d ě 2. Suppose U “ U0 is a wandering domain and let
Un :“ f n pU q for n ě 0. Then, by renaming U0 “ Um if necessary for m large
enough, we have that, for all n ě 0,
(1) Un contains no critical points;

(2) Un is simply connected;
(3) f is univalent on Un .
Proof Since f has a finite number of critical points, and the Un ’s are pair-
wise disjoint, we may assume, by renaming U “ f m pU q for some m ą 0 if
necessary, that neither U nor its forwards iterates contain critical points.
Let us also suppose (conjugating by a Möbius transformation if necessary)
that the point at 8 belongs to a component of f ´1 pU q, so that Yně0 Un is
bounded in the complex plane.
With this set up we shall first prove that any limit function of the normal
family tf n |U uně0 is a finite constant; next that this implies that the diameter
of the sets Un must tend to 0 as n tends to 8; and finally that all the Un are
simply connected and that f |Un : Un Ñ Un`1 is univalent for all n.
Claim: A limit function of the normal family tf n |U uně0 is constant.
Indeed, suppose this were not the case, and let g be a non-constant limit
function of a subsequence, which we denote again by tf n u. Then g is holo-
morphic in U and g 1 is not identically equal to 0. Choose z0 P U such that
g 1 pz0 q ‰ 0, and let D Ă U be a compact set around z0 on which g is injective.

In particular gpz0 q R gpBDq. Set :“ infzPBD |gpzq ´ gpz0 q| ą 0. Since f n
converges uniformly to g in D there exists N so that |f n pzq ´ gpzq| ă for all
n ą N and all z P D. Set Fn pzq :“ f n pzq ´ gpzq and Gpzq :“ gpzq ´ gpz0 q.
Then for n ą N and for z P BD we have
|Fn pzq ´ Gpzq| “ |f n pzq ´ gpzq| ă ď |Gpzq|.
By Rouché’s Theorem, it follows that G and Fn have the same number of zeros
in D for n ą N . Since G has at least one zero, at z “ z0 , then Fn has at least
one zero as well. That is, there exist for n ą N , points z˚ pnq P D such that
f n pz˚ pnqq “ gpz0 q. Hence gpz0 q P Un for all n ą N , which contradicts that
U is a wandering domain.
Claim: The diameter of the sets Un tends to 0 as n tends to 8.
Indeed, suppose this were not the case. Let L be a compact subset of U ,
nj a subsequence tending to 8 such that diampf nj pLqq ě δ for some δ ą 0
and all nj ą N for some N . Take a convergent subsequence of f nj , which
we denote again by f nj , converging to the constant limit function c. Then,
given any 0 ă ă δ, it follows from uniform convergence that f nj pLq Ă
Dc pq for nj sufficiently large. This contradicts that the diameter is bounded
from below.
Claim: The Un are all simply connected and f |Un : Un Ñ Un`1 are all
univalent.
Take an arbitrary closed curve γ Ă U . We will show that γ is homotopic
in U to a constant curve. Set γn :“ f n pγ q Ă Un . It follows from above that
the diameter of the curves γn tends to 0 as n tends to 8. Hence, if Dn
denotes a bounded connected component of Czγ p n , then diampDn q tends to
0 as n tends to 8. Therefore, for n large enough, there can be no poles in
any of the components Dn . Indeed, f has a finite number of poles, and hence
there exists δ ą 0 so that f pDp pδqq Ă f ´1 pU q for any pole p of f , since
8 P f ´1 pU q.
Let γnfilled denote the curve γn union all the bounded components of Czγ p n.
It follows that all iterates are holomorphic on a neighbourhood of γn filled for n
sufficiently large and, since |f n | is bounded on the curves, by the Maximum
Modulus Principle it is also bounded on γnfilled . We conclude from Montel’s
Theorem that the iterates form a normal family in a neighbourhood of γnfilled
and therefore that γnfilled Ă Un . This implies that the curves γn are homotopic
in Un to a constant curve for n sufficiently large. Since f n |U : U Ñ Un is
a covering map, the homotopy can be lifted to a homotopy in U of γ to a
constant curve, which shows that U and all its forwards iterates are simply
connected.
Finally, it follows from the Riemann–Hurwitz formula that all f |Un : Un Ñ
Un`1 must be univalent.
The fact that f is univalent on all Un and that the Un ’s are pairwise
disjoint makes it possible to push forward any almost complex structure that
we may define on U . After that, one can pull it back along the grand orbit
of U .
We proceed to define a real analytic family of almost complex structures on
U of real dimension N for some N ą 4d ` 2. These structures are induced
by a real analytic family of K-quasiconformal mappings whose properties we
state in the following lemma (cf. [St]). The proof of the lemma is given at the
end of the section.
Lemma 4.15 (Family of diffeomorphisms) Let B “ tλ P RN | ||λ|| ď 1u.

Given K ą 1, there exists a family of maps
ψ : B ˆ D ÝÑ D,
pλ, zq ÞÝÑ ψλ pzq,
satisfying:
(1) ψλ is K-quasiconformal on D for all λ P B;

(2) λ ÞÑ ψλ pzq is real analytic for all z P D;
(3) ψ0 “ Id |D ;
(4) for all λ P B and all z “ reiθ P Dzt0u with θ P rπ, 2π s: ψλ pzq “ z;
(5) ψλ1 ı ψλ2 if λ1 ‰ λ2 (injectivity property).
Each of these homeomorphisms induces a Beltrami coefficient on D

defined by
Bψλ
r λ :“ ψλ˚ μ0 “
μ ,
Bψλ
with dilatation bounded by K. Because of the properties of ψλ , it follows that
λ ÞÑ μr λ pzq is real analytic for any z P D.
We now transport μ r λ to the wandering domain U via some Riemann map
R : D Ñ U . Let θ1 , θ2 , θ3 P rπ, 2π s be three distinct arguments, for which
the radial limit of Rpreiθj q exists as r tends to 1 and are all distinct (existence
follows from a theorem by F. and M. Riesz [CL]). Let a1 , a2 , a3 P BU denote
the corresponding limiting values. We define the Beltrami coefficient μλ on
U as μλ :“ R˚ μ r λ , the push forward by the Riemann map of μ r λ on D. The

dilatation of μλ is again bounded by K.
Next, we spread μλ to C. p As mentioned before, since f is univalent on Un ,
we first push μλ on U forward to all Un . Afterwards we pull μλ back to all
iterated preimages of Un for any n, in the usual way. For the remaining points
on Cp we set μλ ” μ0 . More precisely, we recursively define
$
n on Un , for n ě 0,
&pf q˚ μλ
’
’
μλ :“ pf m q˚ μλ on any component of f ´m pUn q, for any m, n ě 0,
’
on Cz
p Ť f ´m pUn q.
’
%μ
0 n,mě0
Then μλ pzq is a Beltrami form which depends real analytically on λ. Clearly

μλ has dilatation bounded by K, since all pullbacks and pushforwards
were done by the holomorphic function f . Moreover, μλ is f -invariant by
construction.
We then apply the Theorem of Integrability with dependence on parameters
to obtain a family of quasiconformal maps φλ : C p ÑC p which integrate the
˚
almost complex structures μλ , i.e. such that φλ pμ0 q “ μλ . We define
gλ :“ φλ ˝ f ˝ φλ´1 .
Since μλ is f -invariant, it follows that μ0 is gλ -invariant, and gλ is holomor-

phic on C,p i.e. a rational map of degree d. If we normalize the integrating
maps so that they fix the three points a1 , a2 , a3 , then gλ can be written as
a rational map whose 2d ` 1 coefficients depend real analytically on λ (see
Section 1.4).
This process defines a map π : B ÝÑ Ratd . Since B has larger real dimen-
sion than 2p2d ` 1q, it follows that there must exist a fibre of π containing a
non-trivial arc. That is, there exist a rational map of degree d, say g, and a path
tλptq, t P r0, 1su Ă B, such that gλpt q “ g for all t P r0, 1s. By defining
ϕt :“ φλ´p10q ˝ φλpt q ,
we obtain a path of K 2 -quasiconformal maps of the sphere fixing a1 , a2 , a3

and such that they commute with f , as shown in the following commutative
diagram:
f
pC,
p μλp0q q / pC,
p μλp0q q
E Y
φλp0q φλp0q

g
ϕt pC, μ0 q
p / pC,
p μ0 q ϕt
O O
φλptq φλptq
f
pC,
p μλpt q q / pC,
p μλpt q q
Note that ϕ0 “ Id. We shall use the following two lemmas which we prove at
the end of the section.
Lemma 4.16 (Families of topological self-conjugacies) If tϕt ut Pr0,1s is a

continuous family of topological conjugacies between a rational map f and
itself, such that ϕ0 “ Id, then ϕt fixes all points in the Julia set Jf for all t,
and maps each Fatou component onto itself.
Lemma 4.17 (Extensions to the unit circle) Suppose U Ă C is an open

simply connected set and tϕt : U Ñ U u0ďt ď1 is a continuous family of quasi-
conformal homeomorphisms which extend to the identity map on the boundary
of U . Let R : U Ñ D be a Riemann map and set ϕpt “ R ˝ ϕt ˝ R´1 : D Ñ D.
Then, for all t P r0, 1s, the map ϕpt extends to BD as some circle map ϕp
independent of t.
Hence by Lemma 4.16, the conjugacy ϕt fixes all points in the Julia set and
maps U onto itself. In particular, it fixes all points on the boundary of U . Let
us now consider the map
ϕpt :“ R ˝ ϕt ˝ R´1 : pD, μ

r λpt q q ÝÑ pD, μ
r λp0q q.
By Lemma 4.17 ϕpt extends to a map ϕp on the unit circle, independent of t.

Finally set
pt :“ ψλp0q ˝ ϕpt : pD, μ
ψ r λpt q q ÝÑ pD, μ0 q,
which extends to S1 as ψλp0q ˝ ϕp, and compare these maps with the original
maps in Lemma 4.15, which are
ψλpt q : pD, μ
r λpt q q ÝÑ pD, μ0 q.
Observe that, for any t, both maps are quasiconformal with the same Beltrami
coefficient. Moreover, they both fix three points on the unit circle (in fact they
fix the lower half of the unit circle). It follows that they must be identical,
since integrating maps on D are unique up to post-compositions with Möbius
transformations of the disc. More precisely we have the following commutative
diagram:
ϕt
pU, μλpt q q / pU, μλpt q q
R R
ϕpt
pD, μ r λpt q q / pD, μ r λp0q q
JJ
JJ ψpt
JJ
JJ
ψλptq
JJ ψλp0q
JJ
J$
Id
pD, μ0 q / pD, μ0 q
But this contradicts the injectivity property of the maps ψλpt q . Indeed, since ϕpt
extends as ϕp to the unit circle, and ψλp0q does not depend on t, it follows that ψ
pt
extends to the boundary as ψλp0q ˝ ϕp for all t. But we just concluded that ψt “
p
ψλpt q , and the injectivity property assures that the maps ψλpt q cannot coincide
for different values of t (and hence different values of λ). This concludes the
proof of the No Wandering Domains Theorem, up to the three lemmas we left
behind.
Proof of Lemma 4.15 For any δ ą 0 we consider the C 8 bump function of R

# 2
δ 2 expp x 2δ´δ 2 q for |x| ă δ,
ωpxq :“
0 for |x| ě δ,
which satisfies
0 ď ωpxq ď δ 2 {e and |ω1 pxq| ď 8δ{e2 .
Choose a partition 0 “ x0 ă x1 ă ¨ ¨ ¨ ă xN ă xN `1 “ π of the interval

r0, π s and let δ ă 12 min1ďj ďN `1 pxj ´ xj ´1 q. We define the following N
linearly independent functions
ωj pxq :“ ωpx ´ xj q, for j “ 1, . . . , N .

Observe that the supports of the functions ωj are pairwise disjoint, namely the
compact symmetric intervals around xj of size 2δ. We then define for pλ, θ q P
RN ˆ r0, 2π q the real function (see Figure 4.6)
N
ÿ N
ÿ
pλ, θ q :“ λj ωj pθ q “ λj ωpθ ´ xj q where λ “ pλ1 , . . . , λN q.
j “1 j “1
0.03
0
0 π π 3π π 2π
4 2 4
Figure 4.6 The function pλ, θ q where δ “ 0.3, N “ 3, xj “ j4π for j “ 1, 2, 3 and
λ “ p0.5, 0.7, 0.3q.
Observe that this is a linear map in λ and C 8 in θ , which equals 0 for θ P

rπ, 2π s. We now use this map to define the family of C 8 diffeomorphisms (for
δ small enough) of the punctured disc
ψ : B ˆ Dzt0u ÝÑ Dzt0u where ψλ prei θ q :“ rei pθ `pλ,θ qq ,
for θ P r0, 2π s and 0 ă r ď 1. We extend ψλ to D by setting ψλ p0q “ 0 for all

λ P B.
It remains to check that ψλ is K-quasiconformal in D. This can be seen
either by observing that this map is K-bilipschitz and therefore quasiconformal
(see Section 1.3.6) or by computing its Beltrami coefficient directly, obtaining
Bz ψλ pzq e´2i θ θ pλ, θ q B
“ where z “ rei θ , λ P B and θ “ .
Bz ψλ pzq 1 ` θ pλ, θ q Bθ
It is then clear that the norm of the Beltrami coefficient can be made as small
as we want, by taking δ small enough. The rest of the properties are obvious
from the construction.
Proof of Lemma 4.16 Since ϕt conjugates f to itself it maps the set Pn “ tz P

C
p | f n pzq “ zu onto itself for any n ą 0. Observe that Pn is a discrete set, that
ϕ0 “ Id, and that ϕt varies continuously with respect to t. As a consequence,

ϕt must be the identity on the set Pn , for any t P r0, 1s. Since this holds for all
n ą 0, and the set of periodic points is dense in the Julia set, by continuity we
get that ϕt is the identity on the Julia set for every t P r0, 1s.
To prove Lemma 4.17, we first give some preliminary notions on the bound-
ary behaviour of Riemann maps. For details we refer the reader to [CL, Go, Pi]
or [Pom].
Let U Ă C be an open, simply connected, bounded set. Let R : D Ñ U be
a Riemann map. Carathéodory’s Theorem states that R extends continuously
to the boundary of U if and only if BU is locally connected. This means that
there is a bijection between points in S1 and points in BU and, even more, the
boundary can be continuously parametrized by the unit circle.
If the boundary of U is not locally connected this is no longer true. Never-
theless, one can still study how the Riemann maps behaves when we approach
the boundary of the unit disc. A theorem by Fatou states that the radial limit at
ζ “ ei θ , i.e. limr Ñ1 Rprei θ q, exists for almost every ζ P S1 . The exceptional
set E of points for which the radial limit does not exist has measure zero. Even
more, a theorem of M. and F. Riesz states that the limit map R : S1 zE Ñ BU
cannot be constant on any set of positive measure.
A point a P BU is called accessible from U if there exists a curve γ :
r0, 1q Ñ U which lands at a, i.e. γ ptq tends to a as t tends to 1. If the boundary
of U is locally connected, then all its points are accessible. Following [Go] we
say that two such curves γ1 and γ2 are in the same access to a, if for every
neighbourhood V Ă C p of a there exists a curve α : r0, 1s Ñ U X V , such that
αp0q P γ1 and αp1q P γ2 . Equivalently, an access is a homotopy class within
the family of curves γ : r0, 1q Ñ U , such that limt Ñ1 γ ptq “ a.
It turns out there is a bijective relation between accesses and points in S1 zE.
Clearly every point in S1 for which the radial limit exists, corresponds to an
access to an accessible point (the limit point). The following Lindelöf-type
theorem says that this correspondence is bijective.
Theorem 4.18 (Accesses and radial limits) Let γ : r0, 1q Ñ U be a curve
which lands at a point a in the boundary of U . Then the curve R´1 ˝ γ in
D lands at some point ζ of BD, and R has the radial limit at ζ equal to a.
Moreover, if γ1 , γ2 : r0, 1q Ñ U are curves landing at a common point a P BU ,
then γ1 and γ2 are in the same access to a if and only if R´1 pγ1 q and R´1 pγ2 q
land at the same point in BD.
Given an accessible point a P BU we define the fibre of a as the set of
accesses corresponding to the point a or, equivalently, the set of points ζ P S1
such that the radial limit exists and equals a. Observe that the theorem of F. and
M. Riesz, mentioned above, implies that each fibre forms a totally disconnected
set in S1 . Indeed, in between two points ζ1 and ζ2 in S1 having both radial limits
at a, there must exist a point ζ with a different radial limit.
Proof of Lemma 4.17 First observe that the induced maps on the unit disc
ϕpt : D Ñ D are quasiconformal and hence they extend to the unit circle.
Since ϕt is the identity of the boundary of U it can at most permute the
accesses to a given point. This means that the induced maps on the disc ϕpt can
at most permute points which belong to the same fibre, which form a totally
disconnected set. But at the same time, the maps ϕt depend continuously on
t, so the permutation of the accesses cannot vary with t. It then follows that
they have to permute these points in the same way for all t. Hence, the maps
ϕpt extend to the set S1 zE in the same way for all t. Since this is a dense set,
the extension to S1 is also the same map for all t.
5
General principles of surgery
The previous chapters contain examples of quasiregular maps for which an

invariant almost complex structure with bounded dilatation exists. It is a
natural question to ask under which conditions this can be accomplished, so
that a holomorphic dynamical copy is obtained by means of the Integrability
Theorem.
We first present two statements, both due to Shishikura, describing typical
scenarios in surgery constructions. The first one was called the Fundamental
Lemma of Quasiconformal Surgery in [Sh1]. It applies to most procedures of
cut and paste surgery, where we paste together holomorphic and quasiregular
mappings. This is the case, for example, in Section 4.2, where we glue a
Blaschke product into a basin of attraction. Shishikura stated his principle for
rational maps. We include more general types and slightly modify one of the
hypotheses. The second principle may be viewed, in some cases but not all, as
a particular case of the first.
Finally, we present Sullivan’s Straightening Theorem, also called the Gener-
alized Shishikura Principle, which is the strongest of the three. It gives a nec-
essary and sufficient condition for a quasiregular map f to admit an invariant
almost complex structure. Namely, it requires the iterates f n to be uniformly
K-quasiregular for some K ă 8. Although the Shishikura principles follow
from this theorem, we choose to prove them independently, since the proof
illustrates a procedure we shall use in many of the surgeries to come.
The principles of this section apply to maps in Rat, Ent and Ent˚ . In fact,
appropriately formulated they also apply to maps in Mer, or to holomorphic
maps defined on subsets of the Riemann sphere. These cases present no extra
difficulties but allowing for them all makes formulation heavier. It is for this
reason that we present the statements for rational maps and comment on the
differences in the other cases.
179
180 General principles of surgery
To simplify the exposition we use the following terminology.
Definition 5.1 (Quasirational map) A quasiregular map f : C p ÑC p is called

quasirational if it is quasiconformally conjugate to a rational map.
This is equivalent to the existence of an f -invariant almost complex struc-

ture of bounded dilatation on C.
p
Likewise, we shall use the terminology quasientire and quasimeromorphic
defined in the obvious way.
5.1 Shishikura principles

Proposition 5.2 (First Shishikura Principle) Let f : C
p ÑC
p be quasiregu-
lar. Let p ě 1 be given, and suppose there exist:
• U “ U1 Y ¨ ¨ ¨ Y Up consisting of p disjoint open subsets Uj of C

p such that,
for 1 ď j ă p,
f pUj q “ Uj `1 , f pUp q Ď U1 ;
• ψ :U ÑU r Ă C,
r , where U p and ψ is a quasiconformal homeomorphism
(the gluing map);
• H :U rÑU r is quasiregular with H p holomorphic;
satisfying:
(i) f |U “ ψ ´1 ˝ H ˝ ψ; and
(ii) Bz f “ 0 a.e. in Czf
p Ń pU q, for some N ě 0.
Then f is quasirational.
Comment: In Shihikura’s original formulation p “ 1.

When p “ 1 and N “ 0 we can think of this scenario as pasting the holo-
morphic map H via the gluing map ψ : U Ñ U r (see Figure 5.1) onto U .
Allowing N ą 0 adds the possibility of having some regions in CzU p where
f is not holomorphic, but only quasiregular, as long as orbits of points in these
regions after at most N iterates fall into the set U which is invariant under f .
Finally, the case p ą 1 allows for some more complicated situations, like maps
H that are not necessarily holomorphic, although a certain iterate is (see e.g.
[BF2] and Figure 5.2).
f U
r
U
ψ
H
ψ ´1 H ψ
Figure 5.1 The hypothesis in the First Shishikura Principle, for N “ 0 and p “ 1.
The gluing map ψ is quasiconformal while H is holomorphic.
ψ
U1 U
r1
ψ ´1 H ψ ψ ´1 H ψ H
H
U2 U
r2
ψ
Figure 5.2 The hypothesis in the First Shishikura Principle, for N “ 0 and p “ 2. In
this case H might not be holomorphic but H 2 is.
Proof Suppose ψ is K1 -quasiconformal, H is K2 -quasiregular and f is K3 -

quasiregular.
We first define an H -invariant Beltrami form μ r almost everywhere (a.e.)
on U . Set Uj :“ ψpUj q for 1 ď j ď p. These sets are disjoint, since ψ is
r r
a homeomorphism, and they satisfy H pU rj q “ Urj `1 and H pU r1 by
rp q Ď U
construction. Now define a.e. on U ,
r
$
’
’ μ0 on U
rp ,
’
&H ˚ pμ q
’
on U
rp´1 ,
0
μ
r“
’
’
’ ¨¨¨
% p ´1 ˚
’
pH q pμ0 q on U r1 .
By construction, μ r is H -invariant since H p is holomorphic, and the dilatation

p ´1
of μr satisfies Kprμq ď K2 a.e. We pull back μ r by ψ to a Beltrami form
μ“ψ μ ˚ r defined a.e. on U . Given that μ r is H -invariant we have that μ is
p ´1
f -invariant a.e. on U . Furthermore Kpμq ď K1 K2 . Finally, we spread μ
recursively by the dynamics of f as

$
’f ˚ pμq
’
’ on f ´1 pU q,
’
’¨ ¨ ¨
’
’
&
μ “ pf n q˚ pμq on f ń pU q,
’
’
’¨ ¨ ¨
’
’
’
on Cz
p Ť ´j pU q .
’
%μ
0 j ě0 f
All together the Beltrami form μ is well defined a.e. on C, p because the
preimages of U form an increasing sequence of sets
ď
U Ď f ´1 pU q Ď f ´2 pU q Ď ¨ ¨ ¨ Ď f ń pU q Ď ¨ ¨ ¨ Ď f ń pU q,
ně0
where μ is f -invariant by construction, and the union ně0 f ń pU q is

Ť
´1
forward and backward invariant. We have }μ}8 ă k “ K K `1 , where K “
p ´1
K1 K2 K3N , since f satisfies Bfz “ 0 a.e. outside f Ń pU q, and f is K3 -
quasiregular inside f Ń pU q.
Finally, the Integrability Theorem provides a quasiconformal map
φ:Cp ÑC p integrating μ, and therefore F :“ φ ˝ f ˝ φ ´1 : C
p ÑC p is holo-
morphic and hence a rational map quasiconformally conjugate to f (see the
Key Lemma (Lemma 1.39)).
Remark 5.3 It follows from the proof that Bφ “ 0 on the set

Cz
p Ť
j ě0 f
´j pU q. In particular, F and f are conformally conjugate on
the interior of this set, if any.

Corollary 5.4 (Special case of the First Shishikura Principle) Let f : C
pÑ
C
p be quasiregular. Let X Ă C
p be open and satisfy:
(a) Bz f “ 0 a.e. on CzX

p and f is quasiconformal on X;
(b) there exists N ě 1 such that maxzPX 7tOpzq X Xu ď N (i.e. orbits of f
pass through X at most N times); and
(c) there exists M ą 0 such that f n pXq X X “ H for all n ě M.
Then f is quasirational.
Proof Set
ď
U“ f n pXq,
něM
and observe that U is forward invariant. From the assumption it follows that
f is holomorphic on U , since U and X are disjoint. Moreover, it follows that
Bz f “ 0 a.e. on Czf
p ´M pU q, since X Ď f ´M pU q. The hypotheses of the first
principle are now satisfied with p “ 1, ψ “ Id |U and H “ f on U .
The second scenario we describe has a similar setting to the Corollary above,
but the proof is essentially different.
Proposition 5.5 (Second Shishikura Principle) Let f, X and N be as in

Corollary 5.4 satisfying paq and pbq but not necessarily pcq. Then, the same
conclusion holds.
Proof For a point x P X let mx ě 1 be the smallest number of iterates so from

then on its orbit does not come back to X, that is f n pxq R X for all n ě mx .
This number is always finite by hypothesis, but it could be arbitrarily large,
depending on x (this is the difference between this case and Corollary 5.4).
Our goal is to define an f -invariant Beltrami form μ on Cp with }μ}8 ă 1.
We do this by pulling back along grand orbits in such a way that the Beltrami
form is well defined a.e.
The set of grand orbits induces a partition of the phase space C.
p We may
think of a grand orbit as an infinite tree where each node is mapped to one
single other node, but several nodes (as many as the degree of the map) can be
mapped to the same one (see Figure 5.3).
z x z0
Figure 5.3 The infinite tree representing the grand orbit of z, where mz “ 7. Points
marked with a star are points of X. In this sketch N “ 2, so every biinfinite path in the
tree can contain up to two stars. From z0 forward, all points belong to CzX.
p
There are two kinds of grand orbits: either GOpzq and X are disjoint sets
or the intersection is non-empty but it consists of at most N points along each
path of the tree. In the first case we let μ “ μ0 along GOpzq. In the second case
for some x P X X GOpzq set z0 “ f mx pxq. Then f n pz0 q R X for all n ě 0. We
define μ on GOpxq “ GOpz0 q as follows:
#
μ0 if w “ f n pz0 q for some n ě 0,
μpwq “
pf p q˚ pμ0 q where f p pwq “ f q pz0 q for some q ě 0.
Note that μ is well defined a.e. By construction μ is f -invariant with dilatation
bounded by K N a.e. if f is K-quasiconformal on S. Thus we obtain a Beltrami
form on Cp with bounded dilatation, and the proposition follows.
5.2 Sullivan’s Straightening Theorem

We now state the third and last result, which is a generalization of the two
principles above.
Theorem 5.6 (Sullivan’s Straightening Theorem) Let f : C p ÑC p be

quasiregular. Suppose there exists a constant K such that the iterates f n are
K-quasiregular for all n ě 1. Then f is quasirational.
Notice that this theorem provides a necessary and sufficient condition.

The necessity is clear: if f is conjugate to a holomorphic map F by a K 1 -
quasiconformal homeomorphism φ, then for every n ě 1, f n “ φ ´1 ˝ F n ˝ φ
is pK 1 q2 -quasiregular.
For rational maps the proof is outlined in [Su1]. We present P. Tukia’s
approach [Tu1], 1 and afterwards adapt it to more general maps. See also [Ge1].
We use the following lemma, which we prove at the end of the section
(see also [BrHa, Prop. 2.7]). Recall that the Haudorff metric on the space of
non-empty bounded subsets of D is not a metric but a pseudo-metric.
Lemma 5.7 Let B Ă D be a non-empty bounded set in the Poincaré metric.

Then there exists a unique closed disc of minimal radius which contains B. Its
centre cB is called the circumcentre (or centre of Young). The map B ÞÑ cB is
continuous with respect to the Hausdorff pseudo-metric.
The unit disc D, endowed with the Poincaré metric, comes into play via the
Beltrami forms in the following way: let M be the set of Beltrami forms μ on
p with }μ}8 ă 1; for any z P C,
C p let Mpzq denote the set of the corresponding
Beltrami coefficients μpzq at z (expressed in some chosen chart around z). We
identify the values μpzq with points in D.
Let Dpf q denote the set of points z P C p for which either f is not
R-differentiable at z or f is R-differentiable but Dz f is singular. We recall
that if f is quasiregular, then Dpf q has measure zero. For z P CzDpf p q set
μf pzq “ Bz f pzq{Bz f pzq and λ “ Bz f pzq{Bz f pzq. Notice that |λ| “ 1 and that
1 We are grateful to Peter Haı̈ssinsky for pointing this out and providing the proof.
5.2 Sullivan’s Straightening Theorem 185
μf pzq is determined once z is given. The formula (1.9) induces a map f ˚ :

M Ñ M, which for every z P CzDpf
p q restricts to f ˚ |Mpf pzqq : Mpf pzqq Ñ
Mpzq defined by μpf pzqq ÞÑ f μpzq as follows:
˚
1 λμf pzq ` μpf pzqq

f ˚ μpzq “ ¨ a.e.
λ 1 ` λμf pzqμpf pzqq
The mapping f ˚ |Mpf pzqq : Mpf pzqq Ñ Mpzq is a Möbius transformation of
the unit disc onto itself and hence a hyperbolic isometry. We use this fact later
in the proof.
Proof of Sullivan’s Theorem Set N pf q “ nPZ f n pDpf qq. Then N pf q is
Ť
a measurable set of measure zero, which is forward invariant, and its comple-
ment Y “ CzNp pf q has full measure and is backwards invariant. Moreover, for
any z P Y , the map f n is R-differentiable at z with an invertible differential.
In Y we define the sets of Beltrami forms
B :“ tpf n q˚ μ0 unPZ Ă M.
We note that, by definition and since f is uniformly quasiregular, Bz :“
B X Mpzq is a uniformly bounded set in Mpzq, and f ˚ Bf pzq “ Bz .
For all z P Y , let μpzq be the circumcentre of the bounded set Bz in Mpzq.
Since f ˚ |Mpf pzqq acts as an isometry, it follows from the uniqueness of the
centre and the invariance of B described above that f ˚ μpf pzqq “ μpzq. For
z P N pf q, set μpzq “ μ0 pzq. By construction μ is f -invariant and }μ}8 ă 1.
It remains to prove that μ is measurable. To do so, let us define for n ě 0,
Bpnq “ tpf k q˚ μ0 , |k| ď nu and Bz pnq “ Bpnq X Mpzq, which is a finite set.
Let μnz be the circumcentre of Bz pnq. The map z P Y ÞÑ μnz is measurable,
since N pf q is measurable and so are the pullbacks pf k q˚ . Also, by continuity
of the centre, μnz tends to μpzq, hence the measurability of μ. But N pf q has
measure zero, so it follows that f ˚ μ “ μ almost everywhere.
Finally, as in the other cases, the Integrability Theorem applied to μ
implies the existence of a quasiconformal map φ such that φ ˝ f ˝ φ ´1 is
holomorphic.
Proof of Lemma 5.7 We may as well assume that B contains at least
two points. Let rB ą 0 be the infimum over the radius of any open disc
containing B, and let us consider a sequence of open discs Dpxn , rn q such that
B Ă Dpxn , rn q, and rn tends to rB . We will prove that txn unPN is a Cauchy
sequence. This will suffice to prove the existence and uniqueness of the centre.
Let ą 0; one may fix R 1 ă rB and R ą rB such that no geodesic in
Dp0, RqzDp0, R 1 q has length larger than {2. This implies that, for every
z P D, no geodesic in Dpz, RqzDpz, R 1 q has length larger than {2 since
Möbius transformations that preserve D are isometries with respect to the

Poincaré metric.
We fix m, n large enough so that rn , rm ă R. Let y be the middle point of
the geodesic segment γ joining xn to xm .
Since R 1 ă rB , there is at least one point z P B such that dpy, zq ą R 1 . By
definition of R, one has xn , xm P Dpz, Rq. It follows from the convexity of
the distance function that at least half of γ is contained in Dpz, RqzDpz, R 1 q.
Therefore, dpxm , xn q ď , and txn unPN is a Cauchy sequence. This proves the
existence and uniqueness of the circumcentre.
To prove the continuity of the map B ÞÑ cB let us consider a sequence
tBn unPN of uniformly bounded sets tending to a bounded set B in
the Hausdorff pseudo-metric. This means that given ą 0, for n large
enough we have Bn Ă B and B Ă Bn , where the superscript denotes an
-neighbourhood of the set.
Denote by cn the circumcentre of Bn with its associated radius rn . Define
r “ lim inf rn . Extracting a subsequence if necessary, we might as well assume
that tcn unPN tends to some c P D. It follows that B Ă Dpc, rq so that rB ď r
by minimality. If we had rB ă r, then we could find r 1 P prB , rq such that, for
n large enough, rn ą r 1 ; moreover, from the convergence of tBn unPN , it would
follow that Bn Ă Dpc, r 1 q for n large enough, contradicting the minimality
of rn . Therefore, r “ rB , and by the uniqueness of the circumcentre, c is the
circumcentre of B.
5.3 Non-rational maps

Both the Shishikura principles and Sullivan’s Straightening Theorem, appro-
priately formulated, hold for the following cases.
Locally defined maps Let f : U 1 Ñ U be quasiregular of degree d ě 2,

where U, U 1 » D and U 1 Ă U . The type of maps we have in mind are
polynomial-like maps, which will be treated in Section 7.1. Observe that points
can be iterated infinitely many times in the set
č
f ń pU 1 q.
ně0
All proofs go through replacing C

p by the domain of definition. After integrat-
ing we obtain a holomorphic map
F : D 1 Ñ D, with D 1 Ă D,
quasiconformally conjugate to f on the appropriate domains.
5.3 Non-rational maps 187
Transcendental maps with no poles Let f : C Ñ C (or f : C˚ Ñ C˚ ) be

quasiregular maps with an essential singularity at infinity (or at infinity and
zero). The critical points (understood as points where f fails to be a local
homeomorphism) of a map of this type form a discrete set in the plane, so they
do not interfere in any pullback operation.
However, transcendental maps may have asymptotic values, and even sets of
positive measure of asymptotic values. But these do not create any problems
either. In the presence of asymptotic values, we pull back along those branches
of the inverse, which are well defined, if any. The non-regular branches map
the values to infinity, and therefore they do not create new finite points where
the pullbacks are not well defined.
Additionally, observe that any Beltrami form in C (or C˚ ) can be considered
a Beltrami form in C p and integrated as such, where the integrating map is
normalized to fix 8 (or 8 and 0).
In conclusion, the proofs are transferred to the transcendental case with
almost no changes. After integrating we obtain a holomorphic map F P Ent
(or F P Ent ˚ ) quasiconformally conjugate to the original one.
Transcendental maps with poles Let f : C Ñ C p be quasiregular, with an

essential singularity at infinity.
On the poles and prepoles, the forward orbit gets truncated after finitely
many iterations. On the poles, the differential is not well defined. However,
the set of prepoles, say Ppf q, is a countable set and therefore of measure
zero. Since the preimage of a set of measure zero has measure zero, repetitive
pullbacks still make sense for this type of maps.
The proof of the First Shishikura Principle works for transcendental mero-
morphic maps satisfying the hypothesis of the principle. The f -invariant
Beltrami form μ, is defined as μ0 on the set of poles and their iterated
preimages, since their orbits cannot eventually fall into the invariant set U .
The resulting map F : C Ñ C p is a transcendental meromorphic map.
For the Second Principle, observe that poles are points where the infinite
tree gets truncated in the forward direction. We define μ “ μ0 on the set
of poles, and pull backwards from there. The resulting map F : C Ñ C p is a
transcendental meromorphic map.
Sullivan’s Straightening Theorem also goes through, after reformulating the
hypotheses to require that the nth iterates of f are uniformly quasiregular
wherever defined. The set of poles and prepoles is of measure zero and should
be included in N pf q.
6
Soft surgeries
In this chapter we treat two different applications of soft surgery. Recall that
a soft surgery starts from one given holomorphic map and consists of altering
the almost complex structure in an invariant way, in order to obtain a family of
invariant Beltrami forms.
In the first application, Section 6.1, a contribution by Xavier Buff and
Christian Henriksen, we start with one map, which possesses a rotation ring
(for example, a Herman ring), and vary the complex structure to produce a
one-parameter family of holomorphic maps, parametrized by D, all having
a rotation ring. Different parameters result in rings of different modulus and
different twist number, a quantity which measures the relative rotation between
the two ring boundaries.
In the second application, Section 6.2, we start with a polynomial with
one or more critical points escaping to infinity and change the complex
structure to obtain a one-parameter family of polynomials with the same
property but where the Böttcher coordinate of the fastest escaping critical
point has been changed. This allows, for example, to parametrize (by D˚ ) a
certain one-parameter family of polynomials in the complement of the con-
nectedness locus. In the simplest case, the one-parameter family of quadratic
polynomials tQc ucPC , this gives a parametrization of the complement of the
Mandelbrot set.
Other applications of soft surgery have already been explained in
Sections 4.1 and 4.3. We remark that the elementary example in
Section 1.5 lies behind all these applications except the one for no wandering
domains.
188
6.1 Deformation of rotation rings 189
6.1 Deformation of rotation rings

Xavier Buff and Christian Henriksen
In this section, we show how one can deform the complex structure of a
rational map having a rotation ring, in order to get a family of rational maps
that are quasiconformally conjugate but not conformally conjugate to the
initial one.
Recall that the postcritical set Pf of a rational map f : C
p ÑC
p is
ď ď
Pf “ f n pvq,
v PVf ně0
where Vf is the set of critical values of f .
Definition 6.1 (Rotation ring) A rotation ring for a rational map f is a

periodic connected component of CzP
p f that is an annulus with finite modulus.
Note that if U is a rotation ring of period p, then for any integer k ě 1

and any component Uk of f ´k pU q, the map f k : Uk Ñ U is a covering map.
Since U is a component of f ´p pU q, we have that f p : U Ñ U is a covering
map. The modulus of the annulus U is preserved, so f p : U Ñ U is an
isomorphism. This isomorphism cannot have finite order, since otherwise an
iterate of f would be the identity on U , thus on all C
p by analytic continuation.
As a consequence, f : U Ñ U is conjugate to an aperiodic rotation and U is
p
contained in a Siegel disc or a Herman ring. The result we are interested in is

the following.
Theorem 6.2 (Main Theorem) Let f P Ratd have a rotation ring H of

modulus m. Then there is an analytic family of rational maps Fτ P Ratd ,
parametrized by the left half plane, such that:
• f is conformally conjugate to Fτ0 with τ0 “ ´2π m;

Repτ q
• Fτ has a rotation ring of modulus ´ ;
2π
• f and Fτ are quasiconformally conjugate, the conjugacy being conformal
outside the grand orbit of H .
Remark 6.3 The family Fτ is non-trivial since the modulus of the rotation
ring is non-constant.
There are several cases where this result may be applied. We will illustrate
this with two concrete examples: the case of a rational map with a Herman ring
and the case of a polynomial with a Siegel disc capturing a critical orbit.
190 Soft surgeries
6.1.1 Examples of rotation rings

Herman rings Consider the family of cubic Blaschke products
1 ´ az
ft,a pzq :“ e2π i t z with t P R and a P p0, 1{3q.
1 ´ a{z
When a P p0, 1{3q, the restriction of ft,a to the unit circle S1 is an orientation
preserving analytic circle diffeomorphism. When t P Z, there is a fixed point
at z “ 1 and the rotation number is 0. According to Theorem 3.20, for each
irrational θ P p0, 1q and each a P p0, 1{3q, there is a unique t P p0, 1q such
that the rotation number of f “ ft,a is θ . According to Denjoy, in that case,
there is a circle homeomorphism ψ : S1 Ñ S1 conjugating Rθ , the rotation of
angle θ to f , that is
for all z P S1 , f ˝ ψpzq “ ψpe2π i θ zq. (6.1)
If ψ : S1 Ñ S1 is analytic (according to Theorem 3.16 this is true as soon as

θ P H), then it has a holomorphic extension to some annulus A :“ A1{R,R “
t1{R ă |z| ă Ru with R sufficiently close to 1. Equation (6.1) holds in A by
analytic continuation. In particular, the open set ψpAq is invariant and thus con-
tained in the Fatou set of f . It cannot be contained in a Siegel disc since each
component of CzψpAq
p contains superattracting fixed points of f (one at 0 and
the other at 8). So, ψpAq is contained in a Herman ring Ha (see Figure 3.8).
Lemma 6.4 (Herman rings are rotation rings) Assume f “ ft,a has a
Herman ring H containing S1 . Then H is a rotation ring for f .
Proof Among the four critical points of f , two are fixed (one at zero and the
other at infinity). The boundary of H is contained in Pf . In particular, at least
one critical point of f is contained in the Julia set. Since f commutes with the
map z ÞÑ 1{z, two critical points of f must be contained in the Julia set. So,
the Herman ring H does not intersect Pf and H is a rotation ring for f .
Siegel captures Let θ P B be a Bryuno number. Consider the family F “

tfa ua PC of cubic polynomials given by
fa pzq :“ ρz ` az2 ` z3 with ρ “ e2π i θ .
Such families have been studied by Zakeri [Z1] (see also [BHe]).
First, fa p´zq “ ´fá pzq, so that the affine map z ÞÑ ´z conjugates fa to
fá . If g is a cubic polynomial with a fixed point with multiplier ρ, there exists
a unique b P C such that g is conjugate to f˘?b . So, the a 2 -plane is the moduli
space of cubic polynomials having a fixed point of multiplier ρ.
Since θ is a Bryuno number, for all a P C, the polynomial fa has a Siegel

disc a around the fixed point 0. The orbit of at least one critical point must
accumulate on Ba . As a consequence, at most one critical point may have an
unbounded orbit. The Julia set is connected if and only if both critical points
have bounded orbits. Figure 6.1 shows the connectedness locus of the family
F in the a 2 -plane.
Figure 6.1 The connectedness locus of the family F in the a 2 -plane. Points are
coloured according to the behaviour of the critical points of fa : white if fa has an
unbounded critical orbit; black if fa has a critical orbit that eventually lands on the
fixed point at 0; and yellow otherwise.
The following result provides an example of rotation ring (see Figure 6.2).
Figure 6.2 The Julia set of a cubic polynomial fa having a rotation ring
(light yellow).
Lemma 6.5 (a contains a rotation ring) If a 2 is sufficiently close, but not
equal, to 4ρ, then a zt0u contains a critical value of fa . Its orbit is dense in
192 Soft surgeries
an invariant analytic Jordan curve Ca . The connected component of a zCa

which does not contain 0 is a rotation ring for fa .
Proof On the one hand, if a 2 “ 4ρ, the cubic polynomial fa has a critical
point at á{3 and fa pá{3q “ 0. If a 2 is close but not equal to 4ρ, then fa
has a critical value close, but not equal, to 0.
On the other hand, according to Bryuno, there is a constant c ą 0 (which
only depends on θ ) such that if fa is univalent in Dr , then a contains Dcr .
If a 2 is sufficiently close to 4ρ, this disc contains a critical value of fa whose
orbit is dense in an invariant analytic Jordan curve Ca .
At least one critical point of fa is contained in Jfa since its orbit must
accumulate on the boundary of the Siegel disc. It follows that a intersects at
most one critical orbit. So, when a 2 is sufficiently close to 4ρ, then a X Pfa
coincides with Ca .
In that case, the connected component of a zCa which does not contain 0 is
an annulus Ha whose boundary components are Ca and Ba . Both boundary
components are contained in the postcritical set. And, moreover, Pfa does not
intersect Ha . Thus, Ha is a rotation ring for fa .
Define the capture component C1 by
C1 :“ tb P C | f˘?b maps a critical point into ˘?b u.
We shall conclude the section by proving that C1 is a simply connected domain

of C by exhibiting a natural conformal isomorphism from D to C1 .
6.1.2 Changing the modulus of the rotation ring

Let f P Ratd be a rational map with a rotation ring H . In this paragraph,
we show that we can deform the complex structure of C p so that f remains
holomorphic with respect to the new structure, but the modulus of the rotation
ring H is changed (cf. [FG]).
Let m be the modulus of the rotation ring H and set r “ e´2π m . There exists
an isomorphism ϕ : H Ñ Ar , where Ar “ tr ă |z| ă 1u.
Given α ą 0, we can define an R-differentiable map φα : Ar Ñ Ar α by
φα pzq :“ z |z|α ´1 .
Note that φα : Ar Ñ Ar α conjugates Rθ : Ar Ñ Ar to Rθ : Ar α Ñ Ar α . In

addition,
α ´ 1 α`1 α´1 ´1 α ` 1 α`1 ´1 α´1
Bz φα “ z 2 z 2 and Bz φα “ z 2 z 2 .
2 2
It follows that
α´1 z
Bz φα “ μα Bz φα with μα “ .
α`1 z
In particular,
ˇ ˇ
ˇα ´ 1ˇ
}μα }8 “ ˇˇ ˇă1
α ` 1ˇ
and φα is quasiconformal.
Thus, we can pull back the standard structure σ0 on Ar α to define an almost
complex structure σα “ φα˚ σ0 on Ar . Since φα conjugates Rθ : Ar Ñ Ar to
Rθ : Ar α Ñ Ar α and σ0 is invariant under Rθ : Ar α Ñ Ar α , we get that σα is
invariant under Rθ : Ar Ñ Ar .
Pulling back the ellipse field σα by ϕ defines an ellipse field ςα on H . Since
σα is invariant under Rθ : Ar Ñ Ar , and ϕ conjugates f p “ H Ñ H to Rθ :
Ar Ñ Ar , the ellipse field ςα is invariant under f p : H Ñ H . Pulling back
the ellipse field ςα by f along the cycle of H yields an ellipse field which is
invariant under f on
p
ď
H :“ f k pH q.
k “1
Of course, pulling back p times along the cycle of H , we come back to H

(since H is periodic of period p for f ). Moreover, pulling back p times ςα by
f , we obtain ςα again since ςα is invariant by f p : H Ñ H .
Next, we extend ςα to an f -invariant ellipse field defined on all of C. p
First we extend ςα to f pHq Ą H by letting ςα “ f ςα . This does not
´ 1 ˚
redefine ςα on H. Then we spread ςα by the dynamics of f to to the grand

orbit of H . On the complement of this grand orbit, we let ςα “ σ0 . Since we
pull back by holomorphic maps, the ellipse field ςα has uniformly bounded
dilatation.
By the Integrability Theorem (Theorem 1.28) there exists a quasiconformal
map Qα : C p ÑC p with ςα “ Q˚ σ0 . The ellipse field ςα depends analytically
α
on α (see Theorem 1.30). Therefore, if Qα is normalized suitably (the choice
of normalization depends on the context), it depends analytically on α. The
construction is illustrated in Figure 6.3.
194 Soft surgeries
Qα
ϕ φα ˝ ϕ ˝ Q´ 1
α
φα
Figure 6.3 The construction used to change the modulus of the rotation ring. The
1
vertical maps ϕ and φα ˝ ϕ ˝ Q´
α are conformal, whereas the horizontal maps Qα
and φα are quasiconformal. The circle/ellipse fields shown at the bottom are drawn by
hand, whereas those at the top are computed.
Now consider the map Fα :“ Qα ˝ f ˝ Q´ 1

α as shown in the following
diagram:
Qα
pC,
p ςα q / pC,
p σ0 q
f Fα

pC,
p ςα q / pC,
p σ0 q.
Qα
Since the ellipse field ςα is f -invariant and since Qα sends it to the field of
circles σ0 , we see that Fα is a quasiregular map that preserves σ0 . According to
Weyl’s Lemma, the map Fα : C pÑC p is holomorphic, thus a rational map. We
constructed a family of rational maps Fα : C p ÑC p which are quasiconformally
conjugate to f : Cp Ñ C,p the conjugacy being conformal outside the grand

orbit of H . The annulus Hα :“ Qα pH q is a rotation ring for Fα . The map
α : Hα Ñ Ar α
1
φα ˝ ϕ ˝ Q´
is quasiconformal (it is a composition of quasiconformal maps) and preserves

the standard complex structure σ0 . Again, according to Weyl’s Lemma, it is
holomorphic, thus an isomorphism. This shows that the modulus of the rotation
ring Hα is
1 1
modpHα q “ log α “ α modpH q.
2π r
In particular, the rational map Fα is not conformally conjugate to f (except
when α “ 1) since the modulus of the rotation ring is changed. Figure 6.4
shows an example where the modulus is doubled.
Figure 6.4 We illustrate how one can fatten a rotation ring. Given a cubic polynomial
f with a fixed rotation ring, we can construct a quasiconformal map Q2 that
conjugates f to a polynomial F2 with a rotation ring with twice the modulus. A circle
field on the grand orbit of the rotation ring is illustrated to the right, and the pullback
of the circle field by Q2 to the left.
We shall apply this construction to the family of cubic Blaschke products

1 ´ az
ft,a pzq “ e2π i t z with t P R and a P p0, 1{3q.
1 ´ a{z
As mentioned above, for each irrational number θ and each a P p0, 1{3q, there
is a unique t “ Tθ paq P p0, 1q such that the rotation number of ft,a : S1 Ñ S1
is equal to θ . The following proposition establishes regularity properties of the
curve Tθ : p0, 1{3q Ñ p0, 1q.
196 Soft surgeries
Proposition 6.6 (Regularity properties of Tθ ) For each irrational θ , there is

an aθ P r0, 1{3s such that:
• the curve Tθ is R-analytic on p0, aθ q; and

• for pt, aq P p0, 1q ˆ p0, 1{3q, the map ft,a has a Herman ring containing
S1 with rotation number θ if and only if a P p0, aθ q and t “ Tθ paq.
Proof Fix an irrational number θ . If no map ft,a has a Herman ring contain-
ing S1 with rotation number θ , the proposition trivially holds with aθ “ 0.
So, let us assume that there is a pair pt0 , a0 q P p0, 1q ˆ p0, 1{3q such that
f “ ft0 ,a0 has a Herman ring H containing S1 with rotation number θ . The
rotation number of f : S1 Ñ S1 is θ . Thus, t0 “ Tθ pa0 q. It is therefore enough
to prove that the curve Tθ is R-analytic in a neighbourhood of a0 and that
fTθ pa q,a has a Herman ring containing S1 for all a P p0, a0 s.
As mentioned earlier, H is a rotation ring for f . The previous surgery
provides a family of ellipse fields ςα depending analytically on α P p0, `8q.
Let Qα : C p ÑC p be the unique quasiconformal map satisfying ςα “ Q˚ σ0 ,
α
normalized by the conditions
Qα p0q “ 0, Qα p8q “ 8 and Qα p1{aq “ 1.
Then, the rational map
Fα :“ Qα ˝ f ˝ Q´
α
1
is a cubic rational map with a superattracting fixed point at 0, a superattracting

fixed point at 8 and a Herman ring Hα “ Qα pH q of modulus
modpHα q “ α modpH q.
Since f has a zero at 1{a, the rational map Fα has a zero at 1. Since f has a
pole at a, the rational map Fα has a pole at bα “ Qα paq. It follows that
1´w
Fα pwq “ ρα w
1 ´ bα {w
for some complex number ρα P Czt0u depending analytically on α.
Lemma 6.7 We have that
´ á ¯¯
bα P p0, 1{9q and ρα “ exp 2π i Tθ bα .
Proof We shall first prove that bα ą 0. The rational map f commutes

with the antiholomorphic involution τ : z ÞÑ 1{z. It follows that the iso-
morphism ϕ : H Ñ Ar satisfies ϕ ˝ τ ˝ ϕ ´1 “ τr , where τr : Ar Ñ Ar is
the antiholomorphic involution of Ar given by τr pzq “ τ pz{rq. Furthermore,
φα ˝ τr “ τr α ˝ φα . It follows that the ellipse field σα is invariant by τr and the

ellipse field ςα is invariant by τ . Hence the map Qα ˝ τ ˝ Q´ 1
α is antiholomor-
phic. It exchanges 0 and 8 and also 1 and bα . It follows that
1 bα bα
Qα ˝ τ ˝ Q´
α pwq “ and “ 1.
w bα
So, bα is real. In addition, ς1 is a field of circles and Q1 is the holomorphic
map z ÞÑ w “ az. In particular, b1 “ a 2 ą 0. Since bα ‰ 0 for all α ą 0, we
see that bα ą 0 for all α ą 0.
The unit circle is the set of fixed points of τ . Its image by Qα is the set
?
of fixed points of w ÞÑ bα {w. This is the circle of radius bα . The map f :
S1 Ñ S1 is a circle diffeomorphism with rotation number θ . It follows that the
restriction of Fα to S1?b is a circle homeomorphism with rotation number θ
α
and without critical points, thus a circle diffeomorphism. In particular,
?
á ¯ a 1 ´ bα a
Fα bα “ ρα bα ? “ ρα bα
1 ´ bα { bα
?
has modulus bα . This shows that ρα has modulus 1.
?
Now, making the change of coordinates z “ w{ bα , the rational map Fα is
conjugate to
?
1 ´ bα z
z ÞÑ ρα z ? .
1 ´ bα {z
The restriction to S1 is a circle diffeomorphism with rotation number θ . So,
a ´ á ¯¯
bα P p0, 1{3q and ρα “ exp 2π i Tθ bα
as required.
Lemma 6.8 We have that bα tends to 0 as α tends to `8.

Proof The Herman ring Hα of Fα separates the zero at 1 from the pole at
bα . As α tends to `8, the modulus modpHα q “ α modpH q tends to `8. It
follows that bα tends to 0 as α tends to `8.
?
The map Fα depends analytically on α. In particular, bα , bα and
?
Fα p bα q
ρα “ ?
bα
all depend analytically on α. It follows that Tθ is analytic on the image of the
? ?
curve α ÞÑ bα . For α “ 1, we have bα “ a0 and as α Ñ `8, we have
?
bα Ñ 0. So, Tθ is analytic in a neighbourhood of a0 as required. In addition,
198 Soft surgeries
every a P p0, a0 s is equal to bα for some α P p0, `8q and fTθ pa q,a is conjugate
?
to Fα via the scaling map z ÞÑ w “ bα z. Since Fα has a Herman ring around
?
the circle of radius bα , the map fTθ pa q,a has a Herman ring around S1 .
a
Remark 6.9 The previous argument shows that aθ “ lim bα . According to
α Ñ0
Sections 7.2.3 and 7.3, we have aθ ą 0 if and only if θ P B is a Bryuno number.
In addition, when θ P H, we have aθ “ 1{3. According to a deep result by
Herman (see Theorem 3.21), there exist θ P BzH such that 0 ă aθ ă 1{3. It is
unknown whether this holds for all θ P BzH.
6.1.3 Twisting in the rotation ring

We have seen that given a map f with a rotation ring, we can vary its modulus
by modifying the complex structure. However, this is not the only way we can
deform f quasiconformally. To motivate the construction return to Figure 6.1.
Some points in the figure are coloured in red. These are points which, at least
1
numerically, all possess fixed rotation rings with modulus equal to 2π log 43 . In
Figure 6.5 the dynamics of four of these polynomials are illustrated. We see
very similar pictures; the largest difference from one picture to another seems
to be that the two boundary components are rotated with respect to each other.
In the following we show how this can be accomplished with a soft surgery.
The construction is as follows. As before, let ϕ : H Ñ Ar denote a con-
formal isomorphism. Passing to logarithmic coordinates, Ar corresponds to
the strip
SX :“ tx ` i y | X ă x ă 0u with X “ logprq.
Let Lt : SX Ñ SX be the R-linear map that fixes i and maps X to X ` i t
Lt px ` i yq :“ x ` i py ` tx{Xq,
or equivalently
it it
Lt pzq “ az ` bz with a “ 1 ` and b“ .
2X 2X
Then Lt maps vertical lines isometrically to vertical lines, and it restricts to the
identity on the imaginary axis. In particular, it descends as a quasiconformal
mapping φt : Ar Ñ Ar mapping circles (centred at the origin) to themselves
by rigid rotations. On the unit circle, φt coincides with the identity, whereas on
the circle of radius r, φt coincides with the rigid rotation Rt .
Figure 6.5 The dynamics of four polynomials corresponding to the four marked
points in Figure 6.1. The modulus of the rotation rings are all the same (approximately
1 4
2π log 3 ). Inspecting the two boundary components of the rotation rings, the principal
difference between the pictures seems to be that the two boundary components are
rotated with varying degree with respect to each other. It turns out that this is not far
from the truth and that the polynomials can be obtained from each other by
quasiconformally twisting the rotation ring.
Proceeding as we did when we modified the modulus of H , we define an

almost complex structureσt on Ar by pulling back the standard one by φt .
Since φt commutes with Rθ and this preserves σt and σ0 , it follows that ςt “
ϕ ˚ σt is invariant under the action of f p : H Ñ H . We now spread ςt by the
dynamics to H and by the circle field on CzH,
p so it is defined everywhere and
invariant by f .
The map Lt is quasiconformal with constant dilatation, hence the dilatation
of σt is uniformly bounded, and so is the dilatation of ςt because we pull back
by holomorphic mappings. Therefore there exists a quasiconformal homeo-
morphism Qt : C p ÑC p such that ςt “ Q˚ σ0 .
t
200 Soft surgeries
As in the previous paragraph, ςt depends analytically on t and so, if suitably

normalized, Qt depends analytically on t (see Theorem 1.30). Again, the map
1 p
Ft :“ Qt ˝ f ˝ Q´
t :CÑC
p
is a rational map. It is quasiconformally conjugate to f , and even conformally

outside the grand orbit of H .
The construction is illustrated in Figure 6.6 for a cubic polynomial
˜ ? ¸
2 3 5´1
f pzq “ ρz ` az ` z with ρ “ exp 2π i ,
2
Figure 6.6 We illustrate how one can twist a rotation ring. Given a cubic polynomial
f with a fixed rotation ring, we can construct a quasiconformal map Qt that
conjugates f to a polynomial Ft with a rotation ring with the same modulus. A circle
field on the grand orbit of the rotation ring is illustrated to the right, and the pullback
of the circle field by Qt to the left.
where a 2 is sufficiently close to 4ρ so that f has a rotation ring. In that case,

the grand orbit of H is bounded (it is contained in the filled Julia set of f which
is compact in C). It follows that the support of ςα avoids a neighbourhood of
infinity, and so we may normalize Qt by the following conditions:
Qt p0q “ 0, Qt pzq “ z ` Op1q as z Ñ 8.
Note that ς0 “ σ0 . As a consequence, Q0 “ Id and F0 “ f . In addition, for

all t P R, the cubic rational map Ft fixes 0 and 8 and the only preimage of 8
is 8 itself. So, Ft is a cubic polynomial. It is holomorphically conjugate to f
in a neighbourhood of 8 by the map Qt which is tangent to the identity at this
point. It follows that Ft is a cubic polynomial. The support of ςt also avoids a
neighbourhood of 0, and so the multiplier of 0 as a fixed point of Ft is equal to

ρ. This proves that
Ft pzq “ ρz ` at z2 ` z3
for some complex number at which depends analytically on t P R.
In the next paragraph, we show that the two previous constructions, varying
the modulus of the rotation ring and twisting, may be combined in a single
operation that depends holomorphically on the parameter. We shall also see
that, in some cases, the curve t ÞÑ Ft is periodic of period 2π .
6.1.4 Holomorphic deformation in the rotation ring

Fix τ0 in the left half plane H and for τ P H denote by Lτ : C Ñ C the
unique R-linear map that fixes i and maps τ0 to τ :
τ ` τ0 τ ´ τ0
Lτ pzq :“ z` z.
τ0 ` τ 0 τ0 ` τ 0
Observe that this is the same map we considered in the elementary example of
Section 1.5. Note that the complex dilatation of Lτ is constant and equal to
τ ´ τ0
μτ :“ .
τ ` τ0
It is important to note that μτ depends holomorphically on τ . In addition, Lτ
maps vertical lines to vertical lines isometrically and, in particular, commutes
with the translation z ÞÑ z ` 2π i , and Lτ restricts to the identity on the imag-
inary axis. For τ P H , denote by Sτ the strip
Sτ :“ x ` i y | Repτ q ă x ă 0 .
(
Then, Lτ restricts to a quasiconformal homeomorphism Lτ : Sτ0 Ñ Sτ . Since

Lτ commutes with the translation by 2π i , it induces a quasiconformal homeo-
morphism φτ : A|eτ0 | Ñ A|eτ | satisfying exp ˝Lτ “ φτ ˝ exp
Lτ
Sτ0 / Sτ
exp exp

A|eτ0 | / A|eτ | .
φτ
Now, the surgery goes as follows. As previously, let ϕ : H Ñ Ar be a

conformal isomorphism. Set τ0 “ log r and for τ in the left half plane denote
by ςτ the ellipse field on H obtained by pulling back the standard complex
structure on A|eτ | by φτ ˝ ϕ
202 Soft surgeries
ςτ “ pφτ ˝ φq˚ σ0 .
As previously, ςτ is invariant by f p and may therefore be promoted to a

globally f -invariant ellipse field by pulling back ςτ via f on the grand orbit
of H and extending by σ0 outside. We may then invoke the Integrability
Theorem to obtain a quasiconformal homeomorphism Qτ : C p ÑC p such that
˚
pQτ q σ0 “ ςτ . Since ςτ is invariant by f , the map
Fτ :“ Qτ ˝ f ˝ Q´
τ :CÑC
1 p p
is holomorphic – we have built a family of rational maps quasiconformally

conjugate to f .
As observed previously, μτ depends holomorphically on τ , and so the family
of ellipse fields φτ˚ σ0 depends holomorphically on τ . The maps we then use to
define ςτ via pullback, i.e. ϕ and f , are holomorphic and do not depend on τ . It
follows that ςτ depends holomorphically on τ . So, when suitably normalized,
Qτ depends holomorphically on τ , and so does Fτ by Lemma 1.40.
Note that Hτ “ Qτ pH q is a rotation ring for Fτ . In addition, the map
τ : H Ñ A|eτ |
1
φτ ˝ ϕ ˝ Q ´
is a conformal isomorphism. In particular,
1 1 Repτ q
mod pHτ q “ log τ “ ´ .
2π |e | 2π
Since Lτ0 “ Id, the complex structure φτ˚0 σ0 coincides with σ0 , and so does
ςτ0 . As a consequence, Qτ0 is a Möbius transformation and Fτ0 is conformally
conjugate to f .
6.1.5 Conformal conjugacies

We shall now see that in the two examples we considered (a cubic polynomial
with a capture and a cubic rational map with a Herman ring), the rational maps
Fτ `2π i and Fτ are conformally conjugate. Note that by assumption, for all
k ě 1, the restriction f k : f ´k pH q Ñ H is a covering map since H does not
intersect the postcritical set Pf .
Lemma 6.10 (Conformal conjugacy) Assume that for all k ě 1 each compo-
nent of f ´k pH q maps with degree 1 to H by f k . Then, Fτ `2π i is conformally
conjugate to Fτ for every τ P H .
Proof The quasiconformal homeomorphism
g :“ φτ´1 ˝ φτ `2π i : Ar Ñ Ar
commutes with the rigid rotation Rθ . It follows that the quasiconformal homeo-
morphism
h :“ ϕ ´1 ˝ g ˝ ϕ : H Ñ H
commutes with f . In addition,
h˚ ςτ “ ςτ `2π i .
Furthermore, there is a uniform bound on the hyperbolic distance in Ar ,

between a point and its image under g. So, this also holds for h in H . Compar-
ing hyperbolic and euclidean distances as we approach the boundary, shows
that h extends as the identity to BH . According to Rickman’s Lemma 1.20,
the homeomorphism h1 : C p Ñ C, p which coincides with h on H and is
equal to the identity otherwise, is quasiconformal. Since f : f ´1 pH q Ñ H
is a covering map, we may lift h1 : H Ñ H to obtain a quasiconformal
automorphism h2 : f ´1 pH q Ñ f ´1 pH q such that the following diagram
commutes:
h2
f ´1 pH q / f ´1 pH q .
f f

H / H
h1
Since each component of f ´1 pH q maps to H with degree 1 by f , we automat-

ically have h2 “ Id on the boundary of f ´1 pH q. Extending h2 by the identity
outside f ´1 pH q, we obtain a quasiconformal homeomorphism h2 : C p ÑC p
for which the above diagram commutes on C. Continuing like this, we may
p
construct a sequence of quasiconformal automorphisms hk : C p ÑC p such that
´ k
hk “ Id outside f pH q and the following diagram commutes:
hk
C
p / C
p
f f

C
p / C.
p
hk´1
204 Soft surgeries
A point z P C:
p
• either belongs to the grand orbit of H , in which case there is a unique k

such that hk `1 pzq ‰ hk pzq;
• or belongs to the complement of the grand orbit of H , in which case hk pzq “
z for all k ě 0.
It follows that the sequence phk : C

p Ñ Cq
p converges to some map h8 : C p Ñ C.
p
Since f is holomorphic, the quasiconformal homeomorphisms hk : C Ñ C p p
all have the same bounded dilatation K. By compactness of the set of
K-quasiconformal homeomorphisms fixing three points (see Theorem 1.26),
the limit map h8 : C p ÑC p is a K-quasiconformal homeomorphism. This limit
map coincides with the identity outside the grand orbit of H and conjugates f
to itself.
Now, since h˚ ςτ “ ςτ `2π i on H , we deduce that h˚8 ςτ “ ςτ `2π i on C.
p As
a consequence, the quasiconformal homeomorphism
ψ :“ Qτ `2π i ˝ h8 ˝ Q´1 : C
τ
p ÑC p
satisfies ψ ˚ σ0 “ σ0 , thus is a Möbius transformation. In addition, ψ conju-

gates Fτ `2π i to Fτ , which proves the Lemma.
We can always normalize Qτ so that it depends holomorphically on τ and

τ fixes three points in C. For example, this may be
so that Qτ `2π i ˝ h8 ˝ Q´ 1 p
achieved by choosing three repelling periodic points z0 , z1 and z2 of f (those
belong to the complement of H ), and then normalizing Qτ by the condition
Qτ pz0 q “ 0, Qτ pz1 q “ 1 and Qτ pz2 q “ 8.
For such a normalization, the Möbius transformation Qτ `2π i ˝ h8 ˝ Q´ 1
τ is
the identity, and so Fτ `2π i “ Fτ . Setting
λ :“ eτ and gλ :“ Fτ ,
we obtain a well-defined holomorphic family of rational maps pgλ : C p Ñ Cq p
parameterized by Dzt0u.
On the one hand, the assumptions of Lemma 6.10 are satisfied in the case of
the family of cubic Blaschke products presented in Section 6.1.1
1 ´ az
ft,a pzq “ e2π i t z , t P p0, 1q and a P p0, 1{3q.
1 ´ a{z
Indeed, each point in the unit circle has three preimages, one in the unit circle,
one in the unit disc, and one outside the unit disc. It follows that if f “ ft,a
has a Herman ring H around the unit circle, then H separates two connected
components of f ´1 pH q, one contained in the unit disc and one contained in
its complement. Each component maps with degree 1 to H . Normalizing Qτ

as in Section 6.1.2, i.e.
Qτ p0q “ 0, Qτ p8q “ 8 and Qτ p1{aq “ 1,
the rational map gλ “ Fτ takes the form
1´w
gλ pwq “ ρλ w
1 ´ bλ {w
for some complex numbers ρλ P Czt0u and bλ P Czt0, 1u which depend
holomorphically on λ P Dzt0u (see [BFGH] for further applications of this
construction).
On the other hand, the assumptions of Lemma 6.10 are not satisfied in the
case of the family of cubic polynomials presented in Section 6.1.1
fa pzq “ e2π i θ z ` az2 ` z3 , a P C and θ P B.
Here, the component of fa´1 pH q which surrounds the critical point close to
á{2 (the one mapping to the Siegel disc), maps with degree two to H . Even
in that case, however, it is still true that the polynomial Fτ `2π i is conformally
conjugate to the polynomial Fτ . Indeed, let a be the Siegel disc of fa
which contains 0. Extend h1 : H Ñ H to a by the identity. Since h1 fixes
the critical value v P a , both h1 ˝ fa : fa´1 pa ztvuq Ñ a ´ tvu and fa :
fa´1 pa ztvuq Ñ a ´ tvu are two to one coverings, and we can lift h1 ˝ fa
by f to get a homeomorphism h2 : fa´1 pa q Ñ a . Of the two possible lifts,
we choose the one that extends as the identity to Bfa´1 pa q. The restriction of
f to fań pa qzf ń`1 pa q is a covering map for all n ą 1, and h2 commutes
with f . Therefore, we may lift h2 to a homeomorphism hk : fa´k `1 pa q Ñ
fa´k `1 pa q which coincides with the identity on Bfa´k `1 pa q, and extend hk
by the identity to the complement of fa´k `1 pa q, for k “ 3, 4, . . . .
As in the proof of Lemma 6.10, we obtain a sequence of quasiconformal
homeomorphisms hk : C p ÑC p such that the following diagram commutes:
hk
C
p / C
p
f f

C
p / C
p
hk´1
which is what we used to conclude that Fτ `2π i is conformally conjugate to
Fτ . We therefore get a family of cubic polynomials gλ “ Feτ parametrized
by the punctured unit disc Dzt0u. We normalize Qτ as in Section 6.1.3 by
the condition
206 Soft surgeries
Qτ p0q “ 0, Qτ pzq “ z ` Op1q as z Ñ 8.
Then, gλ takes the form
gλ pzq “ ρz ` aλ z2 ` z3 with ρ “ e2π i θ ,
for some complex number aλ depending holomorphically on λ P Dzt0u. The

conformal conjugacy class of gλ is uniquely determined by bλ “ aλ2? .
Recall that C1 is the set of parameters b P C such that z ÞÑ ρz ˘ bz2 ` z3
has a critical value contained in the Siegel disc ˘?b . We shall conclude this
section by proving the following result.
Proposition 6.11 (The capture component is simply connected) For each

Bryuno number θ , the set C1 is a simply connected domain of C containing 4ρ.
Proof As mentioned in Section 6.1.1, 4ρ is contained in the interior of C1 . Let

b be a parameter in C1 zt4ρu and let a be a square root of b. Let v be the critical
value of fa contained in the Siegel disc a and let H Ă a be the rotation ring
of fa that separates the orbit of the critical value v from the boundary of the
Siegel disc. Let m be the modulus of the rotation ring. The previous surgery
provides a holomorphic family of cubic polynomials
gλ pzq “ ρz ` aλ z2 ` z3 ,
parametrized by the unit disc, with ge´2π m “ fa .

The image of the map B : λ ÞÑ aλ2 P C defined for λ P Dzt0u is open and
contained in C1 . This shows that C1 is open.
The critical value Qτ pvq of gλ and the fixed point 0 are both contained in
the bounded component of CzH . The modulus of the rotation ring Qτ pH q
1
is ´ 2π log |λ|, which tends to 8 as λ tends to 0. It follows that as λ tends to
0, Qτ pvq tends to 0. There is a unique parameter b P C such that f˘?b has
a critical value at 0: the parameter b “ 4ρ. It follows that aλ2 tends to 4ρ as
λ tends to 0. In particular, each connected component of C1 contains 4ρ, and
so C1 is connected. At the same time, we see that the map B has a removable
singularity at 0 with b0 “ 4ρ.
We now prove that B : D Ñ C1 is injective. Suppose bλ “ bλ1 . Clearly,
bλ “ 4ρ if and only if λ “ 0, so we can suppose λ and λ1 both differ from
0. The quasiconformal map Q´ 1
τ ˝ Qτ 1 commutes with f . Since one critical
point is captured by the Siegel disc , and the other is not, Q´ 1
τ ˝ Qτ 1 fixes
each critical point, and thus each point in the postcritical set. In particular,
1
Qτ pvq “ Qτ 1 pvq, and Qτ ˝ Q´ τ 1 coincides with the identity on B. Consider
the holomorphic map c : A|eτ 1 | Ñ A|eτ | given by the composition
c :“ φτ ˝ ϕ ˝ Q´ 1
τ ˝ Qτ 1 ˝ ϕ
´1
˝ φτ´1 1 .
This map extends to the unit circle as the identity, because Q´ 1

τ ˝ Qτ 1 is the
´ 1 ´1
identity on the unit circle, and both ϕ ˝ ϕ and φτ ˝ φτ 1 extend to the unit
circle as the identity. Hence c is the identity. Using that Qτ pvq “ Qτ 1 pvq,
we get cpφτ ˝ ϕpvqq “ φτ 1 ˝ ϕpvq. Therefore φτ pϕpvqq “ φτ 1 pϕpvqq, and it
follows that τ “ τ 1 ` 2ki π for some integer k. In particular, λ “ λ1 .
We finally prove that B is surjective. Indeed, if b1 belongs to C1 zt4ρu, the
surgery provides an injective map B1 : D Ñ C1 . The map B ´1 ˝ B1 : D Ñ D
1
sends 0 to 0. In addition, since ´ 2π log |λ| is the modulus of the rotation ring
corresponding to Bpλq and to B1 pλq, we have that B ´1 ˝ B1 preserves the
modulus of λ. According to the Schwarz Lemma, B ´1 ˝ B1 is a rotation. In
particular, b1 , which belongs to the image of B1 by construction, also belongs
to the image of B.
6.2 Branner–Hubbard motion

In this section we construct, by soft surgery, special holomorphic motions,
called wring-operations, which are also group actions. They are used to obtain
families of holomorphic maps by varying the complex structure in a systematic
way in the dynamical space. For polynomials of a fixed degree d ě 2 with at
least one critical point escaping to 8, or for mappings with an attracting (not
superattracting) cycle, the complex structure is varied in the basin of attraction
of 8 and the attracting cycle respectively.
In this section we shall treat the first case and restrict to cubic polynomials.
The wring-operation was originally introduced in [BH, Chapt. II] in order to
deal with polynomials of this kind. Without referring to the wring-operation, a
model for changing the multiplier in the attracting linear case is described in
Section 1.5, and also applied in Sections 4.1 and 6.1.
The wring-operation – also called the Branner–Hubbard motion – has been
treated thoroughly in [PeT] in a more general form.
We begin the section by the definition of a holomorphic motion respect-
ing a group structure, then we introduce the group underlying the Branner–
Hubbard motion. We end the introduction by presenting the model for
changing the complex structure in the announced systematic way; in general
by wringing, in particular by stretching or turning when restricting to certain
subgroups.
208 Soft surgeries
6.2.1 Holomorphic motion respecting a group structure

Recall the definition of holomorphic motion (Definition 3.54). In this section
the holomorphic motions will be over a parameter space which is a group
Definition 6.12 (Holomorphic motion respecting a group structure) Let

X, be two complex analytic manifolds and A a subset of X. Suppose
p, ˚q is a group and choose the basepoint λ0 P as the identity element.
Let H : ˆ A Ñ X be a holomorphic motion of A over . We say that
H is a holomorphic motion respecting the group structure if H satisfies the
requirements (1)–(3) for holomorphic motion as well as (4):
(1) For any fixed z P A the mapping λ ÞÑ H pλ, zq is holomorphic.

(2) For any fixed λ P the mapping z ÞÑ Hλ pzq :“ H pλ, zq is injective.
(3) The mapping Hλ0 “ IdA .
(4) H pλ1 , H pλ, zqq “ H pλ1 ˚ λ, zq for all z P A and all λ1 , λ P .
Below we define a group structure on the right half plane Hr as well as the
wring-operation, which introduces holomorphic motions on C respecting the
group structure on the right half plane Hr .
6.2.2 Group structure on the right half plane

Let Hr denote the right half plane with elements ω “ s ` i t, where s ą 0 and
t P R. The map
„ j
s0
ω “ s ` i t P Hr ÞÑ
t 1
maps the right half plane bijectively onto the subgroup of the general linear
group GL2 pRq of 2 ˆ 2 matrices of the given form under matrix multiplica-
tion. The right half plane inherits via the bijection, the group-multiplication:
ω ˚ ω1 “ ps ` i tq ˚ ps 1 ` i t 1 q “ ss 1 ` i pts 1 ` t 1 q.
We refer to this group as pG, ˚q. The group is non-commutative but has two
commutative subgroups that generate pG, ˚q:
pS, ˚q :“ pR` , ˚q and pT , ˚q :“ p1 ` i R, ˚q.
6.2.3 Wringing, stretching and turning the complex structure

Let ω “ s ` i t P Hr and define rω : C Ñ C as the R-linear map
rω pzq :“ 12 pω ` 1qz ` 12 pω ´ 1qz . (6.2)

Then rω is the unique R-linear map that maps the ordered triple pi , 0, 1q onto
the ordered triple pi , 0, ωq. It equals the identity on the imaginary axis i R and
preserves the right half plane Hr as well as the left half plane H . Note that
rω pz ` 2π i q “ rω pzq ` 2π i , and that rω1 ˝ rω “ rω1 ˚ω , so that the mapping
G ˆ C Ñ C defined by pω, zq ÞÑ rω pzq is a group action of pG, ˚q on the left.
It follows from (6.2) that rω is a quasiconformal homeomorphism with
constant Beltrami coefficient
ω´1
r ω pzq :“ Bz rω {Bz rω pzq “
μ .
ω`1
Furthermore, define ω : C Ñ C as
ω pzq :“ z ¨ |z|ω´1 , (6.3)
so that the following diagram commutes:

r
C ÝÝÝωÝÑ C
§ §
expđ
§ §exp
đ
C˚ ÝÝÝÝÑ C˚
ω
where C˚ “ Czt0u. Then ω is equal to the identity on S1 and preserves the

unit disc D as well as the set outside of the unit circle CzD. Moreover, it follows
that ω1 ˝ ω “ ω1 ˚ω , hence that G ˆ C Ñ C defined by pω, zq ÞÑ ω pzq is a
group action of pG, ˚q on the left.
The Beltrami coefficient of ω for z P C˚ is
ω´1z
μω pzq :“ Bz ω {Bz ω pzq “ ,
ω`1z
see also (1.22) in Section 1.5. It follows that ω : C˚ Ñ C˚ is a quasiconfor-
mal diffeomorphism, and that
ω ÞÑ ω pzq and ω ÞÑ μω pzq
depend holomorphically on ω P G. (For fixed z ‰ 0 the Beltrami coefficient

μω pzq is a Möbius transformation in ω.)
Since r1 and 1 equal IdC we conclude that
r : G ˆ C Ñ C, pω, zq ÞÑ rω pzq and

: G ˆ C Ñ C, pω, zq ÞÑ ω pzq
are holomorphic motions of C over G, respecting the group structure.

210 Soft surgeries
We say that the complex structure is:

• wringed if ω P G “ Hr is arbitrary;
• stretched if ω “ s P S “ R` ;
• turned if ω “ 1 ` i t P T “ 1 ` i R.
Note that ω is mapping the circle |z| “ r onto the circle |z| “ r s , where s “
Re ω, and the radial line z “ rei θ onto the logarithmic spiral z “ r s ei pθ `t q ,
where r ą 0 and ω “ s ` i t. For a pure stretching any radial line is mapped
onto itself. For a pure turning any circle is mapped onto itself.
The Beltrami coefficient μω has constant dilatation equal to Kpω pzqq :“
1`|μω |
1´|μω |
since |μω | “ |μω pzq| is independent of z.
The ellipse field corresponding to the almost complex structure determined
by the Beltrami coefficient μω satisfies that along any radial line ´the argument
¯
of the minor axis to the radial line is equal to the constant 12 arg ωω´ 1
`1 . This
is similar to the ellipse field shown in Figure 1.18 for D˚ (which is easily
extended to C˚ ).
Since rω : C Ñ C is R-linear it conjugates z ÞÑ kz to itself for any k P R,
and hence ω conjugates z ÞÑ zk to itself for any k P R.
In particular, we have the following property in any degree d P N.
Lemma 6.13 For any ω P G the homeomorphism ω : C Ñ C conjugates z ÞÑ
zd to itself.
Remark 6.14 The homeomorphisms ω are, up to multiplication by ‘trivial’
maps, the only ones that conjugate z ÞÑ zd to itself. More precisely, suppose
: C Ñ C is such a homeomorphism. Then there exists a ω P G and continu-
ous functions ρ ÞÑ αpρq and ρ ÞÑ βpρq with ρ, αpρq, βpρq non-negative and
satisfying αpd ¨ ρq “ d ¨ αpρq and βpd ¨ ρq “ d ¨ βpρq so that
peρ `2π i θ q “ ω peρ `2π i θ qepα pρ q`2π i β pρ qq .
See Exercise 6.2.1. When restricting to the family tω uωPG we have chosen
αpρq ” 0 ” βpρq.
Lemma 6.15 Let ω P G be given.
(1) The R-linear map rω : C Ñ C conjugates the translation z ÞÑ z ` τ with
τ P C to the translation z ÞÑ z ` rω pτ q.
(2) The homeomorphism ω : C Ñ C conjugates the linear map z ÞÑ λz with
λ P C˚ to the linear map z ÞÑ ω pλqz.
The proof is straightforward. Note that the model can be used to change
the multiplier λ P D˚ of the fixed point at 0 of the linear map z ÞÑ λz to
any multiplier ω pλq in D˚ by varying ω P G, similar to how it is done in

Section 1.5.
6.2.4 The wring-operation on the family of monic centred marked

cubic polynomials
In [BH, Sec. 8] the wring-operation is defined on any family of monic centred
polynomials of degree d ě 2. We restrict to the case d “ 3.
Let P3 denote the family of monic centred cubic polynomials with marked
critical points, parametrized by pa, bq P C2 :
Ppa,bq pzq “ z3 ´ 3a 2 z ` b,
and let Kpa,bq denote the filled Julia set of Ppa,bq . The polynomial has critical
points at ˘a. Set Pp0,0q “ P0 .
The parameter space C2 splits into the connectedness locus C3 , consisting
of parameters for which the corresponding polynomial has a connected filled
Julia set, and its complement, the escape locus E3 , consisting of parameters for
which at least one critical point escape to infinity, i.e.
C3 “ tpa, bq P C2 | ˘ a P Kpa,bq u and
E3 “ tpa, bq P C2 | t˘au X pCzKpa,bq q ‰ Hu .
Let gpa,bq : C Ñ r0, 8q denote the Green’s function associated to the filled
Julia set Kpa,bq of Ppa,bq . It is harmonic on CzKpa,bq and identically zero on
Kpa,bq (see Section 3.3.4). Set Gpa, bq “ max gpa,bq p˘aq, i.e. the potential
of the fastest escaping critical point, and set rpa,bq “ eGpa,bq . The Böttcher
coordinates ϕpa,bq are defined on the neighbourhood Npa,bq of 8 consisting
of points of potential larger than Gpa, bq, making the following diagram
commutative:
Ppa,bq
Npa,bq ÝÝÝÝÑ Npa,bq
§ §
ϕpa,bq §
đ
§ϕpa,bq
đ
CzDrpa,bq ÝÝÝÝÑ CzDrpa,bq
P0
Let μω,pa,bq denote the Ppa,bq -invariant Beltrami coefficient satisfying

#
ϕp˚a,bq μω on Npa,bq ,
μω,pa,bq :“
μ0 on Kpa,bq ,
212 Soft surgeries
and spread to the remaining part of C by the dynamics of Ppa,bq . The Beltrami
coefficient is well defined on Cz ně0 Ppá,b n
Ť
q
p˘aq with bounded dilatation
Kpμω,pa,bq q “ Kpμω q, since it is constructed through successive pullbacks by
holomorphic maps. Note that ω ÞÑ μω,pa,bq pzq depends holomorphically on ω
for each z P C.
Proposition 6.16 (Integration and normalization) For any pa, bq P C2 and

any ω P G there exists a unique integrating map
φω,pa,bq : pC,
p μω,pa,bq q Ñ pC,
p μ0 q,
normalized so that:
(i) φω,pa,bq p8q “ 8;

´1
(ii) φω,pa,bq ˝ Ppa,bq ˝ φω,pa,bq
P P3 ;
´1
(iii) ω ˝ ϕpa,bq ˝ φω,pa,bq
is tangent to the identity at infinity.
The map ω ÞÑ φω,pa,bq pzq depends holomorphically on ω for any fixed z P C

and pa, bq P C2 .
Remark 6.17 (Tangency at infinity) Compare with Remark 4.4 where an

integrating map is conformal in a neighbourhood of 8. In that case, we may
normalize the integrating map to be tangent to the identity at 8. In the case
we consider here the integrating map is quasiconformal in a neighbourhood of
8 so the same requirement would not work. However, for any integrating map
the composition in (iii) is conformal in a neighbourhood of 8, hence it makes
sense to require this composition to be tangent to the identity at 8.
Proof An integrating map is determined up to post-composition by a Möbius

transformation. Let φr “ φrω,pa,bq denote an integrating map fixing 8, and φr ˝
Ppa,bq ˝ φr´1 the resulting cubic polynomial. By post-composing φr with an
affine map z ÞÑ αz ` β we can obtain a monic centred cubic polynomial. Note
that α in such affine maps is determined up to multiplication by a pd ´ 1q-root
of unity, hence for d “ 3 multiplication by ˘1, while β is well defined for each
choice of α.
Abusing notation let φr denote the new integrating map so that φr ˝ Ppa,bq ˝
φ 1 is a monic centred cubic polynomial. Consider the following commutative
r ´
diagram:
φr˝Ppa,bq ˝φr´1
pφpN
r pa,bq q, μ0 q / pφpN
r pa,bq q, μ0 q
O O
φr φr
Ppa,bq
pNpa,bq , μω,pa,bq q / pNpa,bq , μω,pa,bq q
ϕr ϕpa,bq ϕpa,bq ϕr

P0
pCzDrpa,bq , μω q / pCzDr , μω q
pa,bq
ω ω
~
P0
pCzDrpa,bq
s , μ0 q / pCzDr s , μ0 q
pa,bq
The composition ϕr :“ ω ˝ ϕpa,bq ˝ φr´1 conjugates the monic polynomial φr ˝

Ppa,bq ˝ φr´1 to P0 in the neighbourhood φpNr pa,bq q. The conjugating map is
therefore tangent to multiplication by δ “ ˘1 at infinity. Set φω,pa,bq pzq :“
φpδzq.
r Then φω,pa,bq is the unique integrating map satisfying the requirements
in Proposition 6.16.
Since ω ÞÑ μω,pa,bq pzq depends holomorphically on ω for any fixed z P C
and pa, bq P C2 , it follows from the Integrability Theorem 1.30 with depen-
dence on parameters that ω ÞÑ φω,pa,bq pzq also depends holomorphically on ω
for any fixed z P C and pa, bq P C2 .
We define the wring-operation
W : G ˆ C2 Ñ C2 by Wpω, pa, bqq “ paω , bω q,

´1
where Ppaω ,bω q :“ φω,pa,bq ˝ Ppa,bq ˝ φω,pa,bq
.
Note that the Böttcher coordinate of Ppaω ,bω q equals
´1
ϕpaω ,bω q :“ ω ˝ ϕpa,bq ˝ φω,pa,bq
,
and is mapping φω,pa,bq pNpa,bq q onto CzDrpa,bq

s , where s “ Re ω.
The wring-operation W can be applied to any polynomial Ppa,bq P P3 . For
any choice of pa, bq P C2 , the map
Hpa,bq : G ˆ C
pÑC
p defined by Hpa,bq pω, zq :“ φω,pa,bq pzq
is a holomorphic motion of C
p over G, respecting the group structure.
214 Soft surgeries
Moreover, it follows from Theorem 6.18, Corollary 6.19 and Remark 6.20
below that
H : G ˆ E3 Ñ E3 defined by pω, pa, bqq ÞÑ Wpω, pa, bqq “ paω , bω q
is locally injective but in general not globally injective. Hence, the wring-
operation on E3 over G is almost a holomorphic motion, respecting the group
structure; almost in the sense that we have local but not global injectivity. If
we restrict to the subgroup S of real elements of G we do have injectivity.
Theorem 6.18 (Properties of the wring) The wring-operation W : G ˆ

C2 Ñ C2 has the following properties:
(1) the map W is a group action, i.e. Wpω1 , pWpω, pa, bqqq “ Wpω1 ˚ ω,
pa, bqq;
(2) the map ω ÞÑ paω , bω q is holomorphic in ω for each fixed pa, bq;
(3) if pa, bq P C3 then paω , bω q “ pa, bq for all ω P G;
(4) if pa, bq P E3 then ω ÞÑ paω , bω q is locally injective.
Proof
(1) For any pa, bq P C2 and ω P G the quasiregular map
Ppa,bq : pC, μω,pa,bq q Ñ pC, μω,pa,bq q
with the Ppa,bq -invariant Beltrami coefficient μω,pa,bq is the result of the
wring-operation by ω before integrating. The ‘Böttcher coordinate’ of the
quasipolynomial Ppa,bq is then
ω ˝ ϕpa,bq : pNpa,bq q, μω,pa,bq q Ñ pCzDrpa,bq

s , μ0 q.
Composing with the wring-operation by ω1 and applying ω1 ˝ ω “ ω1 ˚ω

we obtain
ω1 ˝ ω ˝ ϕpa,bq “ ω1 ˚ω ˝ ϕpa,bq : pNpa,bq q, μω1 ˚ω,pa,bq q
Ñ pCzDr s 1 s , μ0 q
pa,bq
and the result follows.

(2) For any pa, bq P C2 and z P C the map ω ÞÑ φω,pa,bq pzq is holomor-
phic in ω. The critical points and critical values of Ppa,bq are mapped
onto the critical points and critical values of Ppaω ,bω q respectively. We
find that φω,pa,bq paq “ aω varies holomorphically with ω. If a “ 0, then
aω “ 0 and φω,p0,bq pbq “ bω varies holomorphically with ω. If a ‰ á
then φω,pa,bq pPpa,bq paqq “ bω ´ 2aω3 varies holomorphically with ω and

so does bω .
(3) Choose pa, bq P C3 . For any ω P G the quasiconformal mapping φω,pa,bq
conjugates Ppa,bq to Ppaω ,bω q and Bφω,pa,bq “ 0 a.e. on Kpa,bq . The poly-
nomials are therefore hybrid equivalent. Since the filled Julia set Kpa,bq
is connected they are uniquely determined up to affine conjugation. The
polynomial Ppa,bq is affine conjugate to Pp˘a,˘bq . Using that φ1,pa,bq “
IdCp we obtain paω , bω q ” pa, bq for all ω P G.
(4) Since the map pω, pa, bqq ÞÑ paω , bω q respects the group action is suffices
to prove local injectivity around pa, bq. Note that if paω , bω q “ pa, bq then
the critical values must coincide, i.e. bω ˘ 2aω “ b ˘ 2a. We obtain the
local injectivity by applying the Böttcher coordinate to the critical value of
the fastest escaping critical point.
Choose pa, bq P E3 and z P Npa,bq . If ϕpa,bq pzq “ eρ ì θ , then
ϕpaω ,bω q pφω,pa,bq pzqq “ ω peρ ì θ q “ esρ ì ptρ `θ q where ω “ s ` i t.
It follows that
gpaω ,bω q pφω,pa,bq pzqq “ sgpa,bq pzq for all z P C.
Applying this to the fastest escaping critical value of P pa, bq, say
Ppa,bq paq, with Böttcher coordinate equal to
ϕpa,bq pPpa,bq paqq “ rp3a,bq e2π i α ,
we obtain that Ppaω ,bω q paω q is the fastest escaping critical value of Ppaω .bω q
with Böttcher coordinate
3ps ì t q 2π i α
ϕpaω ,bω q pPpaω ,bω q paω qq “ rpa,bq e .
If bω ´ 2aω “ b ´ 2a then s ` 2π i 3t P 1 ` 2π i Z. It follows that the

wring-operation is locally injective.
Corollary 6.19 (Stretching rays and turning curves in E3 ) If pa, bq P E3

then the action of the subgroups S and T on pa, bq defines the stretching ray
and the turning curve through pa, bq respectively. The action s ÞÑ pas , bs q is
injective on S “ R` . The action of 1 ` i t ÞÑ pa1ì t , b1ì t q may be injective
on 1 ` i R or periodic depending on pa, bq.
Proof Assume that a is the fastest escaping critical point of Ppa,bq . The
function s ÞÑ sgpa,bq paq maps R` bijectively onto R` showing the global
216 Soft surgeries
injectivity of stretching. Examples of turning curves that are globally injective

and others that are periodic are given in [Bra2].
Remark 6.20 (Stretching rays and turning curves for quadratic poly-
nomials) If we apply the stretching and turning to c P CzM, the complement
of the Mandelbrot set, then the stretching ray through c equals the external ray
through c and the turning curve through c equals the equipotential through c.
In this case the turning-operation is periodic.
For completeness we quote some of the results from [BH] about the topo-
logical structure of the parameter space of the cubic escape locus E3 . See [BH,
Thm. 11.1].
Theorem 6.21 (Foliation of the escape locus) The mapping G : E3 Ñ R`

is a trivial fibration with fibre Sr “ tpa, bq P C2 | Gpa, bq “ log ru for any
r ą 1 homeomorphic to S3 .
The proof uses that a stretching ray has one point in common with each
Sr , and that Sr is homeomorphic to S3 for r sufficiently large. It suffices
to understand the structure of Sr for a fixed r in order to understand the
topological structure of E3 .
The parametrization Ppa,bq is chosen so that the critical points are marked as
˘a. Each Sr splits into the disjoint sets Sr` “ tpa, bq P Sr | gpàq “ log r ą
gpáqu, Sr´ “ tpa, bq P Sr | gpáq “ log r ą gpàqu, and their common
boundary BSr` “ BSr´ “ tpa, bq P Sr | gpàq “ log r “ gpáqu.
The critical value P pàq has three preimages when counted with multi-
plicity, the critical point à counts for two and its co-critical point ´2a for
one. Analogously, the preimages of the critical value P páq are the critical
pointá and its co-critical point `2a.
Let ψr˘ : Sr˘ Ñ S1 denote the map that determines the argument of the co-
critical point of the fastest escaping critical point, i.e.
ϕpa,bq p¯2aqq
ψr˘ pa, bq :“ .
|ϕpa,bq p¯2aqq|
The next result follows with some modification from [BH, Prop. 11.3 and
Sec. 12].
Proposition 6.22 (Fibres are discs) The maps ψr˘ are locally trivial topolog-
ical fibrations with fibres homeomorphic to D.
That the fibrations are locally trivial follows from the fact that a turning
curve locally has one point in common with each fibre. It is a deeper result to
prove that fibres are homeomorphic to discs.

6.2.1 Write z P C˚ as epρ `2π i θ q , where ρ ą 0 is uniquely determined but
θ is determined only up to addition by an integer. Suppose
peρ `2π i θ q “ eR pρ,θ q`2π i pθ `T pρ,θ qq
is a homeomorphism that conjugates z ÞÑ zd to itself. Show that the
functions R and T cannot depend on θ so that must be of the form
stated in Remark 6.14.
7
Cut and paste surgeries
In this chapter we treat several different applications of cut and paste surgery.
The general principle is to paste together dynamics of different maps, under
the condition that they agree on the boundaries of their distinct domains of
definition.
Section 7.1 explains one of the earliest uses of quasiconformal mappings in
holomorphic dynamics, due to Douady and Hubbard. It is especially relevant as
a building block in many other applications. This celebrated result justifies why
polynomial Julia sets appear in the dynamical plane of many non-polynomial
families. The parameter version of the Straightening Theorem explains why
one can also find copies of the Mandelbrot set in parameter spaces other than
that of the quadratic family.
Section 7.2 groups several classical results about Siegel discs under one
common surgery, due to Ghys. The construction consists in pasting a rigid
irrational rotation on a topological disc bounded by an invariant quasicircle: the
gluing curve. With this construction one shows, for example, that a Siegel disc
of a quadratic polynomial with rotation number of bounded type is a Jordan
domain containing the critical point on the boundary.
Section 7.3 is dedicated to a construction due to Shishikura, where two
rational maps with Siegel discs of a given rotation number are pasted together
along an invariant curve, resulting in a map with a Herman ring of the
same rotation number. This construction explains, for example, the possible
classes of irrational numbers that can be realized as rotation numbers of actual
Herman rings.
Two cut and paste surgeries due to McMullen are explained in Sections 7.4
and 7.5. The first is inspired by the famous theorem of Bers on Kleinian
groups, known as the Simultaneous Uniformization Theorem. In the actual
construction two different expanding Blaschke products are pasted together
218
7.1 Polynomial-like mappings and the Straightening Theorem 219
along the unit circle to produce a rational map whose Julia set is a quasicircle.
In Section 7.5.2, the dynamics of a rational map is replaced in certain domains
by dynamics of other maps, under milder conditions than those in Section 7.2.
They are pasted along parts of the given Julia set, even when it is not locally
connected. Among the many possible applications, this surgery proves that
any given hyperbolic rational map is J -conjugate to a postcritically finite
rational map.
Section 7.6 is contributed by Kevin M. Pilgrim and Tan Lei. Their surgery
replaces the dynamics of a given rational map in certain domains of the
Fatou set, increasing the connectivity of some Fatou components of the result-
ing map.
Section 7.7 deals with the celebrated surgery by Shishikura, which provides
a bound on the number of non-repelling cycles of any given rational map, in
terms of the number of critical points. The surgery consists in constructing an
appropriate perturbation of the starting map so that all the non-repelling cycles
are turned into attracting ones.
The last section in this chapter, Section 7.8, a contribution by Shaun Bullett,
goes beyond the realm of holomorphic dynamics and enters the world of holo-
morphic correspondences, a natural generalization. This surgery construction
is special in the sense that we are pasting (two copies of) a limit set of a
certain Kleinian group with connected ordinary set together with (two copies
of) a Julia set of an appropriate quadratic polynomial with a connected filled
Julia set.
7.1 Polynomial-like mappings and the Straightening Theorem

The notion of polynomial-like mapping was given in the paper ‘On the dynam-
ics of polynomial-like mappings’ by Douady and Hubbard [DH3], motivated
by the observation that rational maps may locally behave qualitatively like a
polynomial. Figure 7.1 shows an example of a copy of a polynomial Julia set
in the dynamical plane of a certain Newton’s method.
How polynomial-like mappings (defined below) are related to actual polyno-
mials is explained by a cut and paste surgery which results in the Straightening
Theorem (Theorem 7.4).
Definition 7.1 (Polynomial-like mapping) Let U and V be bounded, simply

connected subsets of the plane, bounded by analytic curves and such that U Ă
V Ă C. The triple pf ; U, V q is called a polynomial-like mapping of degree d
if f : U Ñ V is holomorphic and proper of degree d.
220 Cut and paste surgeries
Figure 7.1 Left: Dynamical plane of the rational map NP , which is the Newton’s
method of a cubic polynomial P . Initial conditions whose orbits do not converge to
any attracting fixed point of NP (roots of P ) are shown in black. There is a clear
analogy with the filled Julia set K´1 (the basilica) of the polynomial
Q´1 pzq “ z2 ´ 1 (right).
We are only interested in the case d ě 2. Note that f : U Ñ V is a branched

covering map with d ´ 1 critical points in U when counted with multiplicity.
Remark 7.2 The original definition of polynomial-like mappings did not
include assumptions on niceness of the boundaries. However, if U 1 and V 1
are bounded, simply connected subsets of the plane, such that U 1 Ă V 1 Ă C
and f : U 1 Ñ V 1 is holomorphic and proper of degree d, then one can consider
an open simply connected set V with analytic boundary such that V Ă V 1 and
U 1 Ă V . Defining U to be the preimage of V in U 1 , it follows that f : U Ñ V
is a polynomial-like mapping of degree d.
Note that for any polynomial P of degree d we can choose V “ DR “
t|z| ă Ru for R sufficiently large so that pP |U ; U, V q with U “ P ´1 pV q is
polynomial-like of degree d.
Definition 7.3 (Filled Julia set and Julia set of a polynomial-like mapping)
The filled Julia set of a polynomial-like mapping pf ; U, V q is defined as
č
Kf :“ f ń pV q,
ną0
i.e. the set of points z P U for which f n pzq P U for all n ě 0, in other words
the points in U that never escape. The Julia set Jf is the boundary of Kf .
We are ready to state the Straightening Theorem (see [DH3, Chapt. 1,

Thm. 1]).
Theorem 7.4 (The Straightening Theorem)
(1) Every polynomial-like mapping pf ; U, V q of degree d is hybrid equivalent

to a polynomial P of degree d.
(2) If Kf is connected then P is unique up to affine conjugation.
The proof of the first part of the theorem is by surgery and is presented
below.
Proof
(1) Choose any ρ ą 1. Let R : CzV

p Ñ CzD
p
ρ d be a Riemann map, fixing
8. It follows from Theorem 2.9(c) that R extends continuously to the
boundaries as an analytic map, say ψ1 : BV Ñ S1ρ d . Choose a lift of this
map ψ2 : BU Ñ S1ρ so that ψ1 pf pzqq “ ψ2 pzqd for all z P BU .
Let A0 and Aρ,ρ d denote the annuli A0 “ V zU and Aρ,ρ d “ tρ ă
|ζ | ă ρ d u. The boundary maps ψ1 and ψ2 are analytic. By Proposi-
tion 2.30(b), they extend continuously to a map ψ : A0 Ñ Aρ,ρ d , which
is quasiconformal in intpA0 q.
We define a new map F : C Ñ C by gluing the map z ÞÑ zd into V zU ,
via the gluing map ψ and into CzV via R. More precisely,
$
&f pzq `
’
’ if z P U ,
F pzq :“ R´1 ψpzqd
˘
if z P V zU ,
’
%R´1 `Rpzqd ˘ if z P CzV .
’
The map F : C Ñ C is quasiregular, since it is the composition of quasi-

conformal and holomorphic mappings.
We define a Beltrami coefficient μ on C as follows. On the annulus A0
we set μ “ ψ ˚ μ0 . On CzV , we set μ “ μ0 “ R˚ pμ0 q. Observe that, up
to where it is defined, μ is F -invariant (see Figure 7.2). We now spread μ
to the domain U by the dynamics of f . To do so, note that A0 is a funda-
mental domain of F , since orbits pass through A0 at most once. Indeed,
all points in A0 are mapped into CzV by F , and then tend to infinity under
iteration by F . As a consequence, the sets An “ tz P U | f n pzq P A0 u
are disjoint for different values of n. If Kf is connected these sets are
V
zd
f
ψ
U
ρ ρd
A0
Figure 7.2 Sketch of the proof of the Straightening Theorem.
half-closed annuli; if Kf is disconnected they eventually consist of several

connected components. Therefore, it makes sense to define
$
˚ if z P A0 ,
&ψ μ0 pzq
’
’
μpzq :“ pf q μpzq if z P An ,
n ˚
’
’
%μ pzq
0 elsewhere.
By construction μ is F -invariant. The dilatation of μ is equal to the

dilatation of ψ ˚ μ0 on A0 , since all other pullbacks are by holomorphic
maps. By the Integrability Theorem there exists a quasiconformal map
φ : C Ñ C so that φ ˚ μ0 “ μ. Observe that μ “ μ0 on Kf and therefore
Bφ “ 0 on Kf .
It follows that the map
P :“ φ ˝ F ˝ φ ´1
is a holomorphic map of C of degree d and hence a polynomial. Moreover,

it follows that φ is a hybrid equivalence between f and P .
(2) Suppose f is hybrid equivalent to two polynomials P1 and P2 with a
connected Julia set. Then P1 and P2 are hybrid equivalent to each other
and hence, by Theorem 3.53, affine conjugate.
Remark 7.5 (Applying the First Shishikura Principle) After defining F we

could have skipped the rest of the proof had we applied the First Shishikura
Principle (Proposition 5.2). Indeed, let U 1 “ CzU
p and be a gluing map given
by ψ on V zU and R on CzV p . Then, F is quasiregular on C p and holomorphic
on CzU
p 1 , with U 1 invariant under F . Moreover, F “ ´1 ˝ pz ÞÑ zd q ˝ on
1
U . From the principle we conclude that F is quasiconformally conjugate to a
polynomial. From Remark 5.3 we obtain that the conjugacy is conformal on U
and hence a hybrid equivalence.
Remark 7.6 (Applying Sullivan’s Straightening Theorem) Likewise, we
could have concluded that F is quasiconformally conjugate to a polynomial by
directly applying Sullivan’s Straightening Theorem (Theorem 5.6). Upgrading
to hybrid equivalence is not so obvious in this case.
The Straightening Theorem explains why it is common to find copies of
polynomial Julia sets in dynamical spaces of general holomorphic maps (see
Figure 7.1).
In one-dimensional parameter spaces of holomorphic families of maps we
may encounter copies of the Mandelbrot set. In order to explain why it is so
we introduce the notion of holomorphic families.
Definition 7.7 (Holomorphic family of polynomial-like mappings) Let

be a complex analytic manifold and F “ tpfλ ; Uλ , Vλ qu a family of
polynomial-like mappings. Set
V :“ tpλ, zq | z P Vλ u,
U :“ tpλ, zq | z P Uλ u,
f pλ, zq :“ pλ, fλ pzqq.
F is a holomorphic family of polynomial-like mappings if it satisfies the

following properties:
(1) U and V are homeomorphic over to ˆ D.

(2) The projection from the closure of U in V to is proper.
(3) The map f : U Ñ V is holomorphic and proper.
If these properties are satisfied, the degree of the polynomial-like mappings

is constant and is called the degree of F. By the Straightening Theorem, for
each λ the map fλ is hybrid equivalent to a polynomial of the degree of F
We define the connectedness locus F as
CF :“ tλ P | Kfλ is connected u.
Restrict to degree two. Let be a Riemann surface isomorphic to D

and F “ tpfλ ; Uλ , Vλ qu be a family of quadratic-like mappings. We then
denote the connectedness locus by MF . For each λ P MF , the quadratic-like
mapping fλ is hybrid equivalent to a unique polynomial of the form Qc pzq “

z2 ` c. Hence the map
χ : MF ÝÑ M,
λ ÞÝÑ c “ χ pλq
is well defined.
The family F is called Mandelbrot-like if it satisfies the requirement in the
following theorem, see [DH3, Chapt. IV].
Theorem 7.8 (Homeomorphic copies of the Mandelbrot set) Let K P
be a closed set of parameters homeomorphic to a closed disc and containing
MF . Let ωλ denote the critical point of fλ , and suppose for each λ P zK,
that the critical value fλ pωλ q P Vλ zUλ . Furthermore, assume that the vector
fλ pωλ q ´ ωλ turns once around 0 as λ turns once around BK. Then, the map
χ is a homeomorphism, and it is holomorphic in the interior of MF .
Figure 7.3 Deformed copy of M in the parameter plane of an entire transcendental

family of maps.
Remarks 7.9
(1) The assumption ‘fλ pωλ q P Vλ zUλ if λ P zA’ is equivalent to MF being
compact.
(2) If the winding number of fλ pωλ q ´ ωλ around 0 is δ ą 1, then MF is a
branched covering of degree δ.
7.2 Gluing Siegel discs along invariant curves

The surgery we present in this section is due to Ghys [Gh] from 1984. Roughly
speaking, starting from an analytic circle diffeomorphism with irrational rota-
tion number, Ghys pasted the corresponding rigid rotation into the unit disc.
At the end of the construction he obtained a holomorphic map in a neigh-

bourhood of the origin with a Siegel disc, whose boundary is the image of
the unit circle under a quasiconformal map. His motivation was to explain that
it is a priori possible to have Siegel discs whose boundary is a Jordan curve
not containing any critical point of the map, provided a special class of circle
maps exists. Such circle maps were proven to exist by Herman in [He2] (see
Theorem 3.21). Therefore the foreseen examples by Ghys are actually realized.
We shall explain this in Section 7.2.1.
In a Bourbaki seminar in 1987, Douady [Do3] explained that the Ghys
surgery also applies to certain real analytic circle homeomorphisms with a
critical point, provided they are quasisymmetrically conjugate to an irrational
rigid rotation. Douady posed the following question: do there exist irrational
rotation numbers, such that these analytic critical circle homeomorphisms are
quasisymmetrically conjugate to rigid rotations? A few days after the Bourbaki
seminar, Herman gave a positive answer to the question. Using inequalities of
Świa̧tek [Sw1], Herman [He3] obtained the result known today as the Herman–
Świa̧tek Theorem (Theorem 3.17), which states that any real analytic circle
homeomorphism with rotation number of bounded type is quasisymmetrically
conjugate to the corresponding rigid rotation. This led to the celebrated result
that Siegel discs of quadratic polynomials with a bounded type rotation number
have Jordan curve boundaries containing the critical point (see Section 7.2.2).
The same surgery was used later by Shishikura [Sh1] to construct Siegel discs
starting from Herman rings (see Section 7.2.3).
We begin this section by explaining the Ghys construction in a very general
form for invariant quasicircles. Then we will show how to use it to prove the
different results mentioned above, as well as others.
Let us remark that all constructions in this section may also be viewed as
special cases of the McMullen surgery described in Section 7.5. However,
as presented here, they form a coherent group with applications of a similar
nature.
The Ghys surgery The general set up is as follows:

Let γ denote an oriented Jordan curve in C, p and f : γ Ñ γ an orientation
preserving homeomorphism. Let θ “ rotpf, γ q denote the rotation number of
f on γ (see Section 3.2.3).
We refer to the interior (respectively the exterior) of γ , denoted by intpγ q
(respectively extpγ q), as the connected component of the complement of γ in
p to the left (respectively the right) of γ . Set Gγ :“ intpγ q.
C
Note that the rotation number of f on γ with the reversed orientation

is 1 ´ θ .
The Ghys surgery shows how it is possible, under certain conditions, to paste
the rigid rotation Rθ of rotation number θ into Gγ .
Theorem 7.10 (The Ghys surgery: gluing rigid rotations along quasicir-
cles) Let γ denote an oriented Jordan curve in C,p and f : U Ñ V a holomor-
phic map, where U and V are open neighbourhoods of γ , so that f |γ : γ Ñ γ
is an orientation preserving homeomorphism. Suppose θ “ rotpf, γ q is irra-
tional, and f |γ is quasisymmetrically conjugate to the rigid rotation Rθ on S1 .
Then there exists a quasiconformal map φ : Gγ Y U Ñ D and a holomorphic
map F : D Ñ C p such that F has a Siegel disc of rotation number θ containing
φpGγ q, and F “ φ ˝ f ˝ φ ´1 on DzφpGγ q .
Remarks 7.11 (Classification of the resulting map)
(a) If f in Theorem 7.10 is globally defined, then so is F . In fact, if f P Rat,

then so is F . If f P Ent, then F is in Ent if 8 R Gγ and in Rat if 8 P Gγ .
If f P Ent˚ , then F is either of the same kind or in Ent or in Rat depending
on the position of the singularities with respect to γ .
(b) By reversing the orientation of γ , we can paste the rigid rotation Rp1´θ q
in CzG
p γ.
Proof By hypothesis, there exists an orientation preserving quasisymmetric

map ψ : S1 Ñ γ conjugating Rθ to f |γ . It follows from Ahlfors–Beurling or
Douady–Earle and Proposition 2.30 that the quasisymmetric map ψ : S1 Ñ γ
extends to a map : D Ñ Gγ , which is quasiconformal in D. We use as
the gluing map.
Define fr : Gγ Y U Ñ Gγ Y V as follows:
#
p ˝ Rθ ˝ ´1 qpzq if z P Gγ ,
frpzq :“
f pzq if z P U zGγ .
The map fr is continuous, since ´1 |γ “ ψ ´1 and f |γ “ ψ ˝ Rθ ˝ ψ ´1 |γ . It

is quasiregular, because it is the result of pasting together a holomorphic and a
quasiconformal map along a quasicircle (see Theorem 1.19).
A slight modification of the First Shishikura Principle on locally defined
maps applies to this setup. However, we choose an explicit approach so as to
deduce the properties of the resulting map.
We define a Beltrami form μ on Gγ as the pushforward by of μo ” 0
in D. The dilatation of μ is bounded, because is quasiconformal. Then
we spread μ by the dynamics of fr to the rest of U . Observe that this is
possible since Gγ is invariant under the map fr. For the recursively defined
corresponding Beltrami form μ, we have
$
´1 ˚
&p q μ0 on Gγ ,
’
’
μ :“ pfrn q˚ μ on frń pGγ q, whenever defined
’
’ Ť ń
%μ
0 on U z fr pG q.
n γ
The Beltrami form μ has bounded dilatation in U , since all the pullbacks
outside Gγ are by the holomorphic map fr “ f . Moreover, μ is fr-invariant
by construction. It follows from the Integrability Theorem (Theorem 1.27) that
there exists a quasiconformal map φ : Gγ Y U Ñ D such that φ ˚ μ0 “ μ. We
normalize φ so that p0q is mapped to 0. Finally, define F :“ φ ˝ fr ˝ φ ´1 :
U 1 Ñ φpGγ Y V q. It is holomorphic by Weyl’s Lemma (see Figure 7.4).
Figure 7.4 The new map F on D, showing the Siegel disc in the case when it is
bounded by the curve φpγ q.
Note that F is quasiconformally conjugate to f , via φ, on the complement

of φpGγ q. On φpGγ q the map ´1 ˝ φ ´1 preserves μ0 and is therefore
conformal. By construction, it conjugates F to the rigid rotation Rθ . Hence
φpGγ q is contained in a Siegel disc of F . This concludes the proof.
Remarks 7.12 (Properties of the resulting map)

(a) Note that F is conformally conjugate to f on any domain that never enters
Gγ under iteration of f .
(b) Further properties of the map F will depend on the original map f and the
curve γ . For example, if f has a critical point on γ , then φpGγ q is maximal
and equal to a Siegel disc of F . In other cases, the Siegel disc may extend
beyond φpGγ q. This is the case, for instance, if γ is an invariant curve
inside a Herman ring.
7.2.1 Application to Siegel discs with rotation numbers θ R H

Recall from Chapter 3 that if a holomorphic map has a Siegel disc with rotation
number in the Herman class H and if the map is defined in a neighbourhood of
the Siegel disc, and if the boundary of the Siegel disc is a Jordan curve, then it
contains a critical point (see Theorem 3.42).
As explained in the introduction to Section 7.2, the proposed application by
Ghys, combined with a theorem of Herman, namely Theorem 3.21, gives the
following existence theorem (cf. [Gh, He2]]).
Theorem 7.13 (Existence of Jordan boundaries without critical points)

There exist holomorphic functions with a Siegel disc of rotation number not in
H, whose boundary is a Jordan curve, which does not contain a critical point.
In particular, there exist such examples within the family of quadratic
polynomials Pθ pzq “ e2π iθ z ` z2 and mappings in the standard family
Eθ pzq “ e2π i θ zez , for certain θ R H.
Proof For certain values of θ we shall prove the existence of Pθ and Eθ

satisfying the statement in the theorem. In both cases we apply the Ghys
surgery to certain globally defined maps f and obtain as a result a map F
within the respective families.
To obtain a quadratic polynomial Pθ , we start with the Blaschke products
zá
Ba pzq :“ z2 with a ą 3.
1 ´ az
These cubic rational maps are symmetric with respect to the unit circle, have
superattracting fixed points at 0 and 8, and two critical points on the real
line, not on the unit circle. They have a pole at z “ 1{a P D and a zero at
z “ a P CzD. The unit circle is invariant and, by the Argument Principle, the
map has degree one on S1 . In fact, Ba |S1 is a diffeomorphism.
By Herman’s Theorem (Theorem 3.21), for each a there exists t P r0, 1q
and θ P r0, 1q such that Ba,t :“ e2π i t Ba “ e2π i t z2 1z´
á
az restricted to S is
1
quasisymmetrically conjugate to the irrational rigid rotation Rθ , but not C 2

conjugate. Let us fix a ą 3 and t “ tpaq, and define B :“ Ba,t .
We apply the Ghys surgery to f “ B : C p ÑC p and γ “ S1 . We paste the
rigid rotation Rθ inside D, and obtain a quasiconformal map φ : C p ÑC p and
a rational map F of degree two, which has a Siegel disc around the origin
containing φpDq. In fact, φpDq is the maximal domain of linearization. If this
were not the case, there would exist a neighbourhood U of φpS1 q invariant
under F . Since φ is a homeomorphism, and a conjugacy outside φpDq, we
have that φ ´1 pU q X pCzDq is invariant under B. But since B is symmetric with
respect to S1 , it implies that an annular neighbourhood of S1 is invariant under
B. Therefore, B|S1 is analytically conjugate to the rigid rotation, contradicting
the choice of tpaq.
It follows that the Siegel disc of F is φpDq, that the boundary φpS1 q is a
Jordan curve, in fact a quasicircle, and that the unique critical point of F in
C is φpcq, where c is the critical point of B in CzD. Since c is at a positive
distance from S1 , we have that φpcq is at a positive distance from φpS1 q.
Finally note that if we normalize the quasiconformal map φ so that φp0q “
0, φp8q “ 8 and φpcq “ é2π i θ {2, then the rational map F has a superat-
tracting fixed point at infinity with no finite preimages (hence F is a quadratic
polynomial), a fixed point at 0 with derivative e2π i θ , and a critical point at
é2π i θ {2. It then follows that F “ Pθ .
To obtain a map Eθ P Ent as stated in the theorem, we start with maps in
Ent˚ in the Arnol’d family:
Ha,t pzq :“ e2π i t zea {2pz´1{zq “: e2π i t Ha with 0 ă a, t ă 1.
The unit circle is invariant under Ha,t and the map is symmetric with respect
to S1 . Actually, Ha,t |S1 is the orientation preserving diffeomorphism
a
x ÞÑ x ` t ` sinp2π xq pmod 1q.
2π
For any fixed a, t P p0, 1q, the two critical points of Ha,t are both on the real
line, bounded away from the unit circle. By Herman’s Theorem, for each a
there exists t “ tpaq and θ P p0, 1q such that Ha,t |S1 is quasisymmetrically
conjugate to the irrational rigid rotation Rθ but not C 2 conjugate. Let us fix a
and t “ tpaq.
We apply the Ghys surgery to f “ Ha,t with γ “ S1 , and obtain a quasi-
conformal map φ : C Ñ C and a map F P Ent. As before, the boundary of
the Siegel disc of F is the Jordan curve φpS1 q at a positive distance from
the unique critical point of F . If we set φp0q “ 0 and φpcq “ ´1, where c
is the critical point of Ha,t outside the unit disc, one can check that F must
be of order one since we have not modified the behaviour of Ha,t near infinity,
having a fixed point at 0 with derivative e2π i θ , and a critical point at ´1. The
origin is also an asymptotic value and has no other finite preimages than itself,
since Ha,t has no zeros. One can check that such a map must be of the form
Eθ pzq “ e2π i θ zez .
7.2.2 Application to Siegel discs with rotation number of bounded type

As explained in the introduction to Section 7.2, Douady observed that the Ghys
surgery applies to real analytic circle homeomorphisms with a critical point.
The following theorem uses results of Douady, Ghys, Herman and Świa̧tek.
Theorem 7.14 (Quadratic bounded type Siegel discs) Let Pθ pzq “ λθ z `
z2 , where λθ “ e2π i θ and θ is an irrational number of bounded type. Then Pθ
has a Siegel disc of rotation number θ around z “ 0 whose boundary is a
quasicircle containing the critical point.
See Figure 7.5.
?
Figure 7.5 The Julia set of Pθ pzq “ e2π i θ z ` z2 for θ “ 52´1 “ r1, 1, 1, . . .s, the
golden mean, showing the Siegel disc around the fixed point at 0.
Proof To prove the theorem we start with a special model map in the family
of cubic Blaschke products in the previous section, the one with a “ 3. The
´3
rational map B3 pzq “ z2 1z´ 3z has superattracting fixed points at 0 and 8, a
double critical point at z “ 1, a zero at z “ 3 and a pole at z “ 1{3. The
preimage under B3 of the unit circle consists of three loops meeting at the
critical point z “ 1: the unit circle, one loop inside D and one loop outside D.
It follows that B3 |S1 : S1 Ñ S1 is a homeomorphism (see Figure 7.6).
Consider the Blaschke products
z´3
Bθ pzq :“ e2π i t pθ q z2 with 0 ă tpθ q ă 1,
1 ´ 3z
1:1
2:1
Figure 7.6 The map B3 and the maps Bθ pzq “ e2π i t pθ q 1z´ ´3 obey the following:
3z
each point in the unit disc (the white region to the right), has three preimages in the
white regions to the left, one outside the unit disc and two inside the unit disc.
Similarly, each point outside the unit disc (the shaded region to the right) has three
preimages in the shaded regions to the left, one inside the unit disc and two outside
the unit disc.
where tpθ q P p0, 1q is the uniquely determined value so that the rotation
number of the circle map Bθ |S1 is precisely θ , the given irrational number
of bounded type (see Theorem 3.20). Since θ is of bounded type, it follows
from the Herman–Świa̧tek Theorem (Theorem 3.17) that Bθ is quasisymmet-
rically conjugate to the rigid rotation Rθ . Hence we apply the Ghys surgery
to f “ Bθ : C p ÑC p and γ “ S1 . We obtain as explained in the proof of
Theorem 7.10 a quasipolynomial fr “ B rθ , a B
rθ -invariant Beltrami form μθ
and an integrating quasiconformal map φθ : C Ñ Cp p normalized to fix 0 and 8
and to map the critical point z “ 1 of Bθ to z “ ´λθ {2, and a holomorphic
map Fθ “ φθ ˝ B rθ ˝ φ ´1 which must be a quadratic rational map. Indeed,
θ
after gluing the Siegel disc inside S1 , each point in the unit disc has only one
preimage in D, and each point outside D has no longer a preimage in D. The
point at 8 is a superattracting fixed point of Fθ with no preimage in C.
This implies that Fθ is a quadratic polynomial with a fixed point at z “ 0 and
the critical point at z “ ´λθ {2. Hence Fθ is of the form Fθ pzq “ apλθ z ` z2 q.
Since “ φθ pDq is a Siegel disc of rotation number θ around the fixed point
0, the multiplier of the fixed point must be Fθ1 p0q “ λθ , and we conclude that
Fθ “ Pθ . The maximal Siegel disc around 0 is equal to , since its boundary
B “ φθ pS1 q contains the critical point ´λθ {2. Moreover, the boundary of
is a quasicircle, being the image of S1 under the quasiconformal map
φθ . Compare Figure 7.5 to the one to the right in Figure 7.7, and note that
rń pS1 qq “ JP .
Ť
φθ p ně0 Bθ θ
Remark 7.15 (Generalization) The generalization of this result to higher

degree polynomials has some extra difficulties. Provided one has the
Figure 7.7 Left: The set ně0 Bθń pS1 q. Right: The set ně0 B rń pS1 q. The right
Ť Ť
θ
figure is quasiconformally homeomorphic to Figure 7.5, via the integrating map φθ .
appropriate Blaschke products in hand, the surgery can be performed with

no problem. But in degrees higher than 2, there are whole slices in parameter
space of polynomials with a Siegel disc of a given rotation number. This means
that issues of surjectivity of the surgery procedure need to be addressed. Zakeri
[Z1] generalized the result to cubic polynomials. Further generalizations are
due to Shishikura, Zakeri [Z2] and Zhang [Zh].
As we saw in the section above, the transcendental analogue of the family of
quadratic polynomials is the family of semi-standard maps z ÞÑ λzez , λ P C˚ ,
for which z “ ´1 is the unique critical point, and z “ 0 is both an asymptotic
value and a fixed point of multiplier λ. For any Bryuno number θ the map
Sθ pzq “ λθ zez , where λθ “ e2π i θ , has a Siegel disc around the origin.
Analogously to the case of quadratic polynomials, Geyer [Ge2] proved the
following theorem.
Theorem 7.16 (Bounded type Siegel discs for λzez ) Let Sθ pzq “ λθ zez ,
where λθ “ e2π i θ and θ is an irrational number of bounded type. Then Sθ
has a Siegel disc of rotation number θ around z “ 0 whose boundary is a
quasicircle containing the critical point.
See Figure 7.8 and Exercise 7.2.3.
7.2.3 Turning Herman rings into Siegel discs

In the examples so far, we have glued a Siegel disc inside the unit circle. How-
ever, in general we may use any oriented quasicircle, as long as its dynamics is
quasisymmetrically conjugate to an irrational rotation. The most regular case
is given by an invariant curve inside a Herman ring of a holomorphic map.
Such a curve is always analytic and the map is actually analytically conjugate
to a rigid rotation, so there is no restriction on the rotation number.
Figure 7.8 The dynamical plane of Sθ pzq “ e2π i θ zez , showing the Siegel disc
around the fixed point at 0.
Versions of the following theorem can be found in [Sh1] for rational maps,
and in [DF] for transcendental ones. The converse construction is explained in
Section 7.3.
Theorem 7.17 (Converting Herman rings into Siegel discs) Suppose f P
Ent˚ Y Rat has a Herman ring A of rotation number θ P p0, 1q. Choose γ to
be any of the invariant curves in A and let Gγ be one of the two components
of Czγ
p . Then there exists a quasiconformal homeomorphism φ : C p ÑC p and
a map F P Ent Y Rat such that F has a Siegel disc of rotation number θ or
1 ´ θ containing φpGγ q, and F is equal to φ ˝ f ˝ φ ´1 on the complement of
φpGγ q. It follows from the construction that if f P Ent˚ then F P Ent, and if
f P Rat so is F .
The theorem is an immediate consequence of the Ghys surgery in
Theorem 7.10. The boundary of the resulting Siegel disc of F is equal to the
image under φ of the boundary component of A, which is not in Gγ .
We remark that the theorem also holds for meromorphic functions. None of
the poles or essential singularities inside Gγ will exist for F .
Let us revisit the two cases we discussed above. If we consider the Blaschke
family
zá
Ba,t pzq “ e2π i t z2 , where a P p3, 8q, t P p0, 1q,
1 ´ az
then Ba,t |S1 is a circle diffeomorphism. We can choose a and t so that the
rotation number of the circle map is any predetermined irrational value θ
(see Theorem 3.20), so that there is a Herman ring around S1 (for example,
for any θ P H). We choose γ to be any of the invariant curves in the Herman
ring, with the orientation compatible to the ordinary orientation of S1 , and
paste a rigid rotation inside γ . As in the constructions above, it is easy to
check that the resulting map is the quadratic polynomial Pθ . If γ were given
the opposite orientation, and a rigid rotation were pasted inside γ , then the
resulting map would be the quadratic polynomial Pp1´θ q . For any pa, tq for
which the circle map Ba,t |S1 is analytically linearizable (i.e. there is a Herman
ring), with rotation number θ , it follows that Pθ is analytically linearizable.
A similar phenomenon holds for the Arnol’d family
Ha,t pzq “ e2π i t zea {2pz´1{zq , for a, t P p0, 1q,
in relation to the semi-standard map Sθ pzq “ e2π i θ zez (see Exercise 7.2.4
below).

7.2.1 Suppose f P Rat has a Siegel disc centred at 0 of rotation number θ .
Check that conjugating by 1{z, we obtain a map fr P Rat with a Siegel
disc centred at 8 with the same rotation number. However, conjugating
by 1{z, we obtain a rational map with a Siegel disc centred at 8 with
rotation number 1 ´ θ .
7.2.2 Suppose f P Rat Y Ent Y Ent˚ has a Siegel disc of rotation number
θ . Show that there exists fr of the same type as f having a Siegel disc
of rotation number 1 ´ θ .
Hint: Consider complex conjugation.
7.2.3 Prove Theorem 7.16, applying the Ghys surgery to the auxiliary model
family given by the Arnol’d maps
Hθ pzq “ e2π i t pθ q ze1{2pz´1{zq with 0 ă tpθ q ă 1.
Hint: Proceed analogously to the proof of Theorem 7.14.

7.2.4 Prove that if the Arnol’d circle map is analytically linearizable for some
pa, tq and has rotation number θ , then Sθ is analytically linearizable.
Deduce also that if θ is of bounded type then each connected component
of the boundary of the Herman ring is a quasicircle containing a critical
point.
7.2.5 Suppose f P Rat Y Mer. For any given Herman ring A of f with clo-
sure A in C, let GA denote the bounded component of CzA. Let A1
and A2 be two nested invariant Herman rings of f with closure in
C (i.e. A2 Ă GA1 ). Show that there is a pole in GA2 and another one
in GA1 zpA2 Y GA2 q. Generalize this fact to a larger number of nested

rings.
Hint: Glue a Siegel disc in GA2 and apply the Maximum Modulus
Principle.
7.2.6 Use surgery to prove that if f is in Rat (or in Mer) with a Herman ring
A, then there has to be a critical point in each of the two components
of the complement of A. More precisely, let G1 and G2 denote the two
complementary components and γ1 and γ2 the two components of BA
belonging to G1 and G2 respectively. Then, there exists a critical point
in Gj whose orbit accumulates in γj , for j “ 1, 2.
7.3 Turning Siegel discs into Herman rings

The cut and paste surgery we explain in this section can be thought as the
converse to the one in Theorem 7.17 in the previous section. We start with two
holomorphic maps f and fr with invariant Siegel discs of opposite rotation
numbers and end up with a map F possessing an invariant Herman ring A. The
map F combines the dynamics of f and fr, in the sense that F corresponds to f
in one of the connected components of the complement of A, and corresponds
to fr in the other connected component.
This idea is originally due to Shishikura [Sh1]. He started with rational maps
with Siegel discs and constructed rational maps with Herman rings. We explain
his construction in the simplest case, i.e. when the Siegel discs are invariant.
First we address the case of rational maps and then explain how it adapts to
more general maps (see [DF]).
7.3.1 The rational case

Let f and f be rational maps of degree d and d̃, respectively. Suppose both
r
maps have an invariant Siegel disc, and r of rotation number θ and 1 ´ θ
respectively, for some θ P p0, 1q. Furthermore, assume that both Siegel discs
are centred at the origin and that neither of them contains 8. Let ϕ : Ñ D
and ϕr :
r Ñ D be linearizing coordinates conjugating f and fr to Rθ and
R1´θ respectively, i.e. f “ ϕ ´1 ˝ Rθ ˝ ϕ on , and fr “ ϕr´1 ˝ R1´θ ˝ ϕr
on .
r
One might view the surgery in an abstract way as choosing some arbitrary
invariant curves γ and γr in the Siegel discs, and obtaining a new Riemann
sphere by gluing the two unbounded half-spheres along the chosen curves, via
a map that conjugates their dynamics (see Figure 7.9).
Figure 7.9 The geometric idea of the main construction.
The special role assigned to 0 and 8 is not essential and could be substituted
by any pair of distinct points for f as well as for fr. However, this normaliza-
tion simplifies the explanation of the surgery.
The first step is to obtain a gluing map. Let h : γ Ñ γr be a conjugacy
between f |γ to fr|γr , defined as follows. Let r and r r denote the radii of the
circles Sr “ ϕpγ q and Srr “ ϕrpr r
γ q. Since the inversion Lpzq “ rz
r
conjugates
Rθ to R1´θ , the map h :“ ϕr ˝ L ˝ ϕ is an analytic map conjugating f |γ to
´ 1
fr|γr as shown in the following commutative diagram (see also Figure 7.10):
f
γ / γ
ϕ ϕ
Rθ
Sr / Sr
h L L h
R1´θ
Srr / Srr
O O
ϕr ϕr
fr
γr / γr
The second step in the construction consists in extending h to a globally

defined homeomorphism : C p Ñ C, p the gluing map. Let Gγ and Gγr denote
the bounded components of the complement of γ and γr respectively. The map
is constructed so that it satisfies:
(a) |γ “ h;
(b) pGγ q “ CzG
p γr and pCzG
p γ q “ Gγr ;
f fr
r
h
Gγ Gγr
ϕ ϕr
Rθ R1´θ
r L r
1
r
1
0 0
Figure 7.10 The definition of h : γ Ñ γr.
(c) p0q “ 8 and p8q “ 0;

(d) is conformal everywhere outside a closed annular neighbourhood Aγ
of γ .
To construct the we paste four different maps together. Consider an annular

neighbourhood Aγ of γ obtained by choosing two f -invariant curves in .
Let γ o and γ i denote the outer and inner boundary of Aγ respectively. Note
that Aγ splits into two f -invariant subannuli, denoted by Ai and Ao , bounded
by γ and respectively γ i and γ o (see Figure 7.11). Likewise, choose γri
and γro in
r and define A rγr “ Ari Y A
ro so that Arγr is an fr-invariant annular
neighbourhood of γr .
Figure 7.11 The global definition of the gluing map , as the result of pasting four
maps together.
Let
Ro : CzG
p γo Ñ G i
γr and Ri : Gγ i Ñ CzG
p γro
be chosen as Riemann mappings satisfying
Ro p8q “ 0 and Ri p0q “ 8.
Since the curves are analytic, Ro and Ri extend to analytic maps R

p o on γ o
and R on γ respectively (see Theorem 2.9). On the closed annulus Aγ we
p i i
define
ho : Ao Ñ A
ri and hi : Ai Ñ A
ro
as quasiconformal interpolating maps such that ho agrees with R

p o on γ o and
with h on γ ; and h agrees with h on γ and R on γ (see Proposition 2.30).
i p i i
The map defined as the combination of the four mappings above:

$
’
’ Ro on CzG
p γo,
’
&ho
’
on Ao ,
:“ i
’
’
’ h on Ai ,
R
% i
’
on Gγ i ,
has all the desired properties (see Figure 7.11).
The third step in the construction is to define a quasiregular map g : C
pÑC
p
with the dynamical properties we have announced for F . We shall define g on
the dynamical Riemann sphere of f , by modifying the dynamics in Gγ so it
reflects the dynamics of fr outside Gγr . Set
#
f on CzG
p γ,
g :“
´1 ˝ fr ˝ on Gγ .
First we observe that g is continuous on γ , because |γ “ h, and h is

defined so that h´1 ˝ fr ˝ h “ f on γ . It follows that g is quasiregular. Indeed,
g is holomorphic outside Gγ and is quasiregular on Gγ .
Moreover, note that the topological annulus X ´1 pq r is g-invariant,
and that g is conjugate to the rotation Rθ on this annulus. Indeed, g is equal
to f on the ‘outer part’ zGγ , and conjugate to fr on the ‘inner part’ Gγ X
´1 pq
r of the annulus. This annulus will become the Herman ring of F after
straightening g.
Finally, observe that the poles of g are the same as the poles of f , together
with the former zeros of fr. Likewise, the zeros of g are the zeros of f together
with the former poles of fr.
We end the proof by applying Sullivan’s Straightening Theorem (The-

orem 5.6). To do that we check that the iterates of g are uniformly
K-quasiregular for some K ă 8. Observe that g is holomorphic everywhere
except on Ai Y pg ´1 pAi q X Gγ q. On Ai we have that g n “ phi q´1 ˝ frn ˝ phi q
and hence pK 1 q2 -quasiconformal, where K 1 is the quasiconformal constant of
. On pg ´1 pAi q X Gγ qzAi we have that g “ phi q´1 ˝ fr ˝ Ri and hence it
is K 1 -quasiregular. After being mapped to Ai orbits remain there. Hence the
iterates of g are uniformly K-quasiregular where K “ pK 1 q3 . We conclude
that there exists a quasiconformal homeomorphism φ : C pÑC p conjugating g
to the rational F :“ φ ˝ g ˝ φ . Observe that the annulus
´ 1
´ ¯
A :“ φ X ´1 pqr
is F -invariant and topologically conjugate to a rotation of angle θ . The bound-

ary of A belongs to the Julia set of F , since it corresponds to boundaries of
the Siegel discs of f and fr respectively. Therefore A is maximal and hence a
Herman ring.
Alternatively to the use of Sullivan’s Straightening Theorem we could have
constructed explicitly a g-invariant Beltrami form (see Exercise 7.3.1).
7.3.2 The transcendental case

We extend the construction above, to maps that are not necessarily rational (cf.
[DF]). More precisely, we consider f, fr P Rat Y Ent Y Mer, not necessarily
in the same subclass.
As in the section above we assume that f and fr have invariant Siegel discs
of opposite rotation numbers centred at the origin. Since the construction is
symmetric, and we already considered the case where both maps are rational,
we assume that f P Ent Y Mer and allow fr to be in any of the three classes
above. We observe that the surgery construction goes through with only minor
modifications. Let us check this step by step.
The first and second steps, consisting of the definition of the gluing map ,
are independent of the maps f and fr, and therefore no modification is needed.
The third step, the definition of the quasiregular map g, needs to be slightly
modified, since f p8q is not defined, and frp8q might not be defined either.
Hence, first we set g : C˚ Ñ C p to be
#
f on CzGγ ,
g :“
´1 ˝ fr ˝ on Gγ zt0u.
If fr is defined at 8, i.e. if fr P Rat, we extend the definition of g to C by setting

gp0q “ ´1 pfrp8qq. The domain of g is therefore C˚ if fr P Ent Y Mer and
C if fr P Rat. Poles of f , if any, are still poles of g, and also any former zero
of fr, except for z “ 0 itself, is now a pole of g. Notice that z “ 0 might be
an essential singularity, hence all former poles of fr or zeros of f are now pre-
essential points for g, i.e. preimages of the essential singularity at the origin.
It follows that g is a quasimeromorphic map.
After applying Sullivan’s Straightening Theorem we obtain a new map F
with at least one essential singularity at infinity.
The class to which F belongs depends on both f and fr. In order to classify
the different cases, we need to introduce the class of maps with a small set of
essential singularities:
Mer8 :“ tf : CzB
p ÑC
p | B a compact countable set and f meromorphicu.
Then one can check that we have the following table:
f zfr Pol Rat Ent Mer

Pol Rat
Rat Rat Rat
Ent Mer Mer Ent˚ , Mer8
Mer Mer8 Mer8 Mer8 Mer8
7.3.3 Applications and extensions

The first application in the rational case is in Shishikura’s original paper [Sh1].
His construction is more elaborate and applies to cycles of Siegel discs, which
are in turn transformed into cycles of non-nested Herman rings. Using this
surgery, Shishikura proves the following theorem.
Theorem 7.18 (Existence of periodic Herman rings for rational maps)

Suppose there exists a rational map with a p-cycle of Siegel discs of rota-
tion number θ . Then, there exists a rational map with a p-cycle of Herman
rings of rotation number θ . In particular, for any Bryuno number θ , and any
p ě 1, there exists a rational map with a p-cycle of Herman rings of rotation
number θ .
The simplest version (the invariant case) is enough for many applica-
tions. We show one for the transcendental case and leave others as exercises
(cf. [DF]).
Proposition 7.19 (Existence of Herman rings for meromorphic maps with

exactly one pole) Given any Bryuno number θ , there exists F P Mer with
exactly one (non-omitted) pole having a Herman ring of rotation number θ .
Moreover,
ebz
F pzq “ az2 ,
z`1
for some parameters a, b P C˚ .
Proof We start by
f pzq :“ e2π i θ zez and frpzq :“ e2π i p1´θ q zp1 ` zq,
where θ is an arbitrary Bryuno number.

Observe that f belongs to Ent and has a Siegel disc centred at the origin of
rotation number θ . Notice that z “ 0 is also an asymptotic value and has no
other preimage than itself. The point z “ ´1 is the only critical point of f .
The map fr is a quadratic polynomial with a fixed point at z “ 0 of multiplier
e i p1´θ q . Since 1 ´ θ P B (see Exercise 3.2.1), fr has a Siegel disc centred at
2π
the origin of rotation number 1 ´ θ . The critical point of fr is at z “ ´1{2, and

the other zero is at z “ ´1.
Gluing the two maps together as in the construction above, we obtain a new
map F with a Herman ring of rotation number θ and with:
(a) one essential singularity at 8;

(b) one simple pole, since f has no poles and fr has one simple zero, different
from 0. The pole is not omitted, because z “ ´1 is not omitted by fr.
Thus F P Mer with exactly one pole, which is not omitted

To find an expression for F , we choose to normalize the integrating map
φ to fix 0, 8 and ´1. Then the point z “ ´1 is the non-omitted pole of F .
Observe that the origin is a superattracting fixed point, and also an asymptotic
value. Hence, factorizing out the simple pole and the double zero at zero, the
remaining map is entire without zeros. Hence F must be of the form
ek pzq
F pzq “ az2
z`1
for some value a P C˚ and some entire map kpzq. Since we have not modi-
fied f around infinity, the order of F is one and hence kpzq “ bz for some
b P C˚ .
We end this section by considering the special case where we reflect with
respect to an invariant curve in the Siegel disc. This is used in [Z2, Zh] to
study the properties of boundaries of Siegel discs.
Theorem 7.20 (Reflection of f with respect to an invariant curve) Suppose

f P Rat (resp. f P Ent) has a Siegel disc, centred at the origin, of rotation
number θ . Then there exists F P Rat (resp. F P Ent˚ Y Mer8 ) with a Herman
ring of rotation number θ such that F is symmetric with respect to the unit
circle, which is therefore invariant under F . Moreover, f and F are quasi-
conformally conjugate on the complement of the unit disc and conformally
conjugate in any open set which does not eventually map to the Herman ring.
In spirit this is a special case of the Shishikura construction, gluing f to

‘itself’ (with opposite rotation number). This procedure is used to construct
‘models’ (see Exercises 7.3.4 and 7.3.5).
We do the construction in the plane of linearizing coordinates, so that the
symmetry is clear.
Proof We prove the theorem in the rational case. As before let Sr denote
the circle of radius r centred at the origin. Let be the Siegel disc of f
and let ϕ : Ñ D3 be a linearizing map, which conjugates f to Rθ . Set
γ “ ϕ ´1 pS1 q and γ 1 “ ϕ ´1 pS2 q, let Gγ (resp. Gγ 1 ) denote the bounded com-
ponent of the complement of γ (resp. γ 1 ), and let A be the annulus bounded by
γ and γ 1 .
The topological part of the surgery consists in defining a quasiregular sym-
metric map in the linearizing plane.
We first construct the gluing map. Choose a Riemann mapping
p γ 1 Ñ CzD
R : CzG p 2 fixing infinity, which then extends to an analytic map
R on γ . Let h : A Ñ A1,2 be a quasiconformal interpolation between the
p 1
boundary mappings R| p γ 1 and ϕ|γ (see Figure 7.12). Define the gluing map
: C Ñ C by
p p
R
γ1
γ
A
r
Gγ ϕ
1 2 3
A
Figure 7.12 The gluing map equals a Riemann map R outside of γ 1 and a
quasiconformal interpolation on the annulus A between the linearizing map ϕ|γ on γ
and R|
p 1 on γ 1 .
γ
$
&R on CzG
’
’ p γ1,
:“ h on A,
’
’
%ϕ on Gγ 1 .
Observe that is quasiconformal on the annulus A but conformal

everywhere else. Moreover, conjugates f to Rθ on γ .
By using the reflection τ : Cp ÑC p in the unit circle, i.e. τ pzq “ 1{z, we
define a map g : C
p ÑC p on the linearizing plane by
#
˝ f ˝ ´1 on CzD,
p
g :“
τ ˝ p ˝ f ˝ ´1 q ˝ τ on D.
The map is continuous and quasiregular. It is symmetric with respect to S1 by

r :“ A1{2,2 . We
construction. It is actually holomorphic outside the annulus A
leave to the reader to check that the map
#
ϕ ˝ ´1 on A1,2 ,
ϕr :“ ´1
τ ˝ ϕ ˝ ˝ τ on A1{2,1
is a quasiconformal conjugacy between g and Rθ on the given annulus.

We define a g-invariant Beltrami form μ on C,
p recursively, by
$
˚ on A,
&ϕr μ0
’
’ r
μ :“ pg n q˚ μ on g ń pAqr for n ě 1,
’
g ń pAq.
’ Ť
%μ
0 on Xz ně1
r
Since g is holomorphic everywhere except on A, r it follows that μ has bounded

dilatation. It is g-invariant by construction. Since ϕr and g are symmetric with
respect to S1 , so is μ (see Exercise 1.2.5 applied recursively).
Let φ : Cp ÑC p be the integrating map fixing 0, 1 and 8. Because of the
symmetry of μ it follows that φ is also symmetric with respect to S1 , i.e. τ ˝
φ ˝ τ “ φ (see Exercise 1.4.1). The holomorphic map F : C p Ñ C,
p defined as
F “ φ ˝ g ˝ φ, is also symmetric with respect S . It is easy to check that ϕ̃ ˝
´ 1 1
φ ´1 is conformal and conjugates F to Rθ on the annulus φpAq. r The rotation

domain for F containing φpAq r must be symmetric with respect to S1 and be a
Herman ring.

7.3.1 In the main construction of Section 7.3.1, define a g-invariant Beltrami
form whith bounded dilatation. Also express the linearizing map for
the newly obtained Herman ring in terms of the maps involved in the
surgery.
7.3.2 Given any set of Bryuno numbers tθ1 , . . . , θN u, with N ą 0, show that
there exists f P Mer or f P Rat with exactly N poles and N Herman
rings that are not nested.
Hint: Construct a polynomial with N invariant Siegel discs of rotation
numbers θ1 , . . . , θN . Then apply the construction above N times.
7.3.3 Let Fα,a,b pzq :“ e2π i α z 1z´
á z ´b
az 1´bz . Prove that for any Bryuno number
θ , the parameters α, a and b can be adjusted so that F has two Siegel
discs and one Herman ring, all of rotation number θ . Hint: Let fθ pzq “
1
e2π i θ z zz`
´1 and consider fθ and f1´θ .
7.3.4 Let Pθ pzq “ e2π i θ z ` z2 with θ a Bryuno number. Let γ be any invari-
ant curve in the Siegel disc of Pθ . Reflect Pθ with respect to γ as in
Theorem 7.20, and show that the resulting map belongs to the Blaschke
á
family Ba,t pzq “ e2π i t 1z´ āz .
7.3.5 Let fθ pzq “ e 2π i θ z
ze with θ a Bryuno number. Let γ be any invariant
curve in the Siegel disc of fθ . Reflect fθ with respect to γ as in
Theorem 7.20, and show that the resulting map belongs to the Arnol’d
family Ha,t pzq “ e2π i t zea {2pz´1{zq .
7.3.6 Let Ba,t be the Blaschke family of rational maps as in Section 7.2.1.
Fix a ą 3 and let t “ tpaq be given by Theorem 3.21, so that Ba,t pa q
on the unit circle has rotation number, say θ , and is quasisymmetrically
conjugate to Rθ but not C 2 -conjugate. Show that there exist a 1 and t 1 ,
such that Ba 1 ,t 1 has a Herman ring of rotation number θ , with both
boundaries being Jordan curves containing no critical point.
Hint: First apply the surgery in Section 7.2.1 and then the one in
Section 7.2.3.
7.3.7 In the original Shishikura construction in Section 7.3.1, we paste
together two rational maps f and fr of degrees d and d̃ respectively.
Calculate the degree of the resulting map F .
7.4 Simultaneous uniformization of Blaschke products

In this section we explain a cut and paste surgery due to Curtis T. McMullen.
The construction is inspired by Bers’ famous Simultaneous Uniformization
Theorem, which states that for any two Fuchsian groups 1 and 2 uniformiz-
ing two compact Riemann surfaces of the same genus there exists a quasi-
Fuchsian group whose limit set pq is a quasicircle, such that the two
Riemann surfaces arising from pq{ are conformally equivalent to the
given ones. The quasi-Fuchsian group is a quasiconformal deformation of

1 and 2 (in the respective invariant components of pq). For an extended
statement of Bers’ Simultaneous Uniformization Theorem see [Be1] and in
particular the exposition in [Be2]. The notion of Klenian group, limit set and
ordinary set is introduced in Section 3.5. It is also mentioned that the analogue
in holomorphic dynamics to the action of a Fuchsian group is the iteration of a
Blaschke product.
McMullen’s construction consists of pasting two expanding Blaschke prod-
ucts along the unit circle respecting the dynamics. The result is a rational
map whose Julia set is a quasicircle, separating the two completely invariant
Fatou components in which the dynamics is conjugate to the original Blaschke
products. More precisely, the theorem is as follows.
Theorem 7.21 (Simultaneous uniformization of Blaschke products) Let

Bj , j “ 1, 2, be two expanding Blaschke products of degree d ą 1. Then there
exists a rational map F of degree d whose Julia set is a quasicircle, and such
that F is conjugate to Bj on the uniformizations of the two components of
CzJ
p F respectively.
The manuscript [McM] containing the construction is unpublished; we

thank McMullen for providing it to us.
The key construction relies on a result about quasicircles, which is the
content of Theorem 7.22 below. This is known in the literature as the Sewing
Theorem (cf. [LV, p. 92]) or conformal welding (cf. [As, Thm. 5.10.1]).
The proof we give here is an application of the Integrability Theorem and is
interesting in itself. However, it may also be proved by other means (see e.g.
[LV, II.7.5]). The setup is as follows.
Let γ be an oriented Jordan curve in C, p and let U, V denote the inte-
rior (respectively exterior) component of the complement of γ . Choose two
Riemann mappings RU : U Ñ D and RV : V Ñ CzD. p They extend continu-
ously to homeomorphisms of the boundaries R p U : BU Ñ S1 and R p V : BV Ñ
S1 . The composition ψ :“ R pV ˝ Rp ´1 : S1 Ñ S1 is an orientation preserving
U
homeomorphism that we call a boundary correspondence associated to γ or a
conformal welding (see Figure 7.13).
A Riemann mapping is determined up to post-composition with a Möbius
transformation that preserves the unit disc. It follows that ψ is determined up
to equivalence by Möbius transformations that preserve the unit disc. Indeed,
let Rp 1 “ Mi ˝ R
U
p U and R
p 1 “ Mo ˝ R
V
p V , where Mi , Mo : D Ñ D (and there-
fore they also preserve CzD). Then ψ 1 “ Mo ˝ ψ ˝ Mi´1 is also a boundary
correspondence associated to γ .
D
R
pU
γ
ψ
U
D
V
R
pV
Figure 7.13 A boundary correspondance ψ or conformal welding.
Theorem 7.22 (Quasicircles and welding) A function ψ : S1 Ñ S1 is qua-

sisymmetric if and only if there exists an oriented quasicircle γ inducing the
boundary correspondence ψ.
Proof One direction is immediate, for if γ is an oriented quasicircle, then any

associated boundary correspondence is quasisymmetric.
Now assume ψ : S1 Ñ S1 is a quasisymmetric map, and let ψ p : D Ñ D be
a quasiconformal extension of ψ. Set
#
pψp ´1 q˚ μ0 on D,
μ :“
μ0 on CzD,
p
and choose an integrating quasiconformal map φ : C pÑC p satisfying φ ˚ μ0 “

1
μ a.e. Then γ “ φpS q is an oriented quasicircle. Moreover, the composi-
p : pD, μ0 q Ñ pφpDq, μ0 q is conformal, and so is φ : pCzD,
tion φ ˝ ψ p μ0 q Ñ
pφpCzDq, μ0 q. Set U :“ φpDq and V :“ φpCzDq, RU “ ψ ˝ φ 1 and
p p p ´ 1 ´
RV “ φ ´1 . It follows that RpV ˝ Rp ´1 “ ψ. Hence ψ is a boundary corre-

U
spondence associated to the quasicircle γ “ φpS1 q. Since the integrating map
φ is only determined up to post-composition with a Möbius transformation
M:C p Ñ C,p so is γ .

It follows from the definition of a boundary correspondence associated to

an oriented quasicircle and Theorem 7.22 that we have established a bijection
between the quotient spaces
QS{ „ ÐÑ QC ` { «,
where QS denotes the set of quasisymmetric maps ψ : S1 Ñ S1 with the

equivalence relation „ defined as ψ1 „ ψ2 if and only if there exist Möbius
transformations M 1 , M preserving D such that ψ2 “ M 1 ˝ ψ1 ˝ M; and QC `
denote the set of oriented quasicircles with the equivalence relation « defined
as γ1 « γ2 if and only if there exists a Möbius transformation M such that
γ2 “ Mpγ1 q.
Proof of Theorem 7.21 Given two expanding Blaschke products Bj , with j “
1, 2, of degree d, choose an orientation preserving homeomorphism ψ : S1 Ñ
S1 which conjugates B1 to B2 on S1 . There are d ´ 1 such homeomorphisms,
corresponding to the number of fixed points of Bj |S1 and all are quasisymmet-
ric. Then by Theorem 7.22 there exists an oriented quasicircle γ inducing the
boundary correspondence ψ, with Riemann maps RU and RV , where U, V
denote the interior and exterior components of the complement of γ .
We glue the dynamics of B1 , B2 into U, V respectively via the extension of
the conformal mappings R p U : U Ñ D and R p V : V Ñ CzD.
p That is, we define
#
Rp ´ 1 ˝ B1 ˝ R
pU on U ,
F :“ U
´1
R ˝ B2 ˝ R V
p
V
p on V ,
which is well defined on γ (i.e. the two definitions agree) as shown in the
following commutative diagram:
B1
S1 / S1
_?? ?
?
p U?
R
?? RU
p

ψ γ ψ
???

RV p V?
R
? ?
p
S1 / S1
B2
Hence the new map is a continuous self-map of C p which is holomorphic

outside γ , and locally quasiconformal on γ . Since γ has zero measure it
follows from Weyl’s Lemma that the map is globally holomorphic, hence a
rational map F with Julia set JF “ γ .
Example 7.23 We construct an explicit quadratic rational map from two

expanding Blaschke products of degree two as in Theorem 7.21. Consider the
one-parameter family
z`λ
Bλ pzq :“ z with λ P D,
1 ` λz
corresponding to those Blaschke products of degree two which have attracting
fixed points at 0 and 8 (of multiplier λ and λ respectively) and the third fixed
point at 1 P S1 . For any pair Bλ1 , Bλ2 we obtain a quadratic rational map whose
Julia set is a quasicircle and with the dynamics of Bλ1 , Bλ2 glued into the two
Fatou components as explained above. If we normalize the integrating map to
fix 0, 1 and 8, then the rational map takes the form
z ` λ1
Fλ1 ,λ2 pzq :“ z .
1 ` λ2 z
Applications An appropriate version of Theorem 7.21 depending on parame-

ters could have interesting applications. As an example, let us fix one of the two
Blaschke products to be B0 pzq “ zd . By gluing expanding Blaschke products
B of degree d (with a marking, namely the quasisymmetric conjugacy to zd
on the unit circle) to B0 , we obtain polynomials of degree d belonging to the
main hyperbolic component in the parameter space of complex polynomials
of degree d, i.e. the polynomials having one attracting fixed point in C whose
immediate basin of attraction contains all critical points in the plane of the
polynomial.
Defining the appropriate space of ‘marked Blaschke products’ and adding
the appropriate parameters and normalizations to the construction, this would
lead to a parametrization of the main hyperbolic component of the connected-
ness locus of Pold .
7.5 Gluing along continua in the Julia set

The surgery in this section is a cut and paste construction due to Curtis
T. McMullen and is contained in his 1988 paper ‘Automorphisms of rational
maps’ [McM1]. It is similar in spirit to some of those in the preceding sections,
where we replace the dynamics of a map on a given domain, by the dynamics of
a different map, under the condition that the two agree along the gluing curve.
In the previous constructions the gluing curve was always a quasicircle. In
the surgery in this section the agreement condition of the two maps is milder:
the maps only need to agree along the ideal boundary of the domain. The
boundary is therefore allowed to be a non-locally connected continuum. In the
applications it will be a subset of a Julia set.
The original motivation for this surgery was to study the finite group Autpf q
of Möbius transformations commuting with a given rational map f . In this
section we shall not comment on this, but explain the surgery involved which
is interesting in itself.
Before making the construction precise, we briefly introduce the concept of
ideal boundary and some useful related results.
Prime ends and the ideal boundary In this section a disc is an open simply
connected subset of C p whose complement contains at least two points. If U
is a disc, a Riemann map RU maps U conformally onto D. Carathéodory’s
Theorem states that RU extends continuously to BU if and only if BU is locally
connected. In that case, BU can be parametrized by the unit circle. If BU is not
locally connected there is still a relation with the unit circle, via prime ends of
U . (For an introduction to prime ends see [Mi2, Sect. 17], [Co, Sect. 14.3] or
[Pom, Sect. 2.4].)
Definition 7.24 (Ideal boundary of a disc) Let U be a disc. The ideal

boundary of U , denoted by I pU q, is the set of prime ends of U .
The ideal boundary I pU q is identified with S1 , via a Riemann map RU .

This identification is well defined up to post-composition with a Möbius
transformation that preserves D.
Induced maps on ideal boundaries
Case 1 Any proper holomorphic map between two discs U and V , say
f : U Ñ V , induces naturally a real analytic map between their respective
ideal boundaries. Indeed, if we consider Riemann maps RU : U Ñ D and
1
RV : V Ñ D, the composition fp :“ RV ˝ f ˝ R´ U : D Ñ D is a proper holo-
morphic map, which extends to a neighbourhood of S1 via the Schwarz Reflec-
tion Principle (see Chapter 2, Remark 2.11). Therefore it induces a real analytic
map from the unit circle to itself, i.e. from the ideal boundary I pU q to the ideal
boundary I pV q. We shall refer to this map as fp : I pU q Ñ I pV q.
Case 2 In fact it is enough to have f defined on a one-sided neighbourhood of

BU , which is mapped properly onto a one-sided neighbourhood of BV . We
can still uniformize U and V as before, and obtain a map fp from a one-
sided neighbourhood of S1 to another one-sided neighbourhood of S1 . By the
Schwarz Reflection Principle, we again obtain a real analytic map from S1 to
itself, hence a real analytic map fp : I pU q Ñ I pV q.
Remark 7.25 If the map f in any of the two cases above is quasiconformal,
the same construction goes through, and f induces a map fp : I pU q Ñ I pV q,
which is quasisymmetric (being the extension to the unit circle of a quasicon-
formal homeomorphism of D (see Theorem 2.9) or between annular subsets
of D (see Corollary 2.13)). In both cases the induced map fp is unique up to
equivalence by Möbius transformations that preserve D.
The following proposition contains a useful relation between the ideal
boundary I pU q and the topological boundary BU . For a proof see [McM1,
Prop. 5.2].
Proposition 7.26 (Ideal boundary versus topological boundary) Let U be

a disc. If φ : U Ñ U is a quasiconformal homeomorphism, which induces the
identity on I pU q, then φ extended by the identity on CzU
p is a quasiconformal
homeomorphism on C. In particular, it equals the identity on BU .
p
In greater generality, we can define the ideal boundary of an invariant con-

tinuum. Recall, that C Ă C p is a non-degenerate continuum if C is a compact,
connected set that contains more that one point.
Definition 7.27 (Ideal boundary of an invariant continuum) If C Ă C p is a

non-degenerate continuum, then every connected component of CzC is a disc.
p
We define the ideal boundary of C, denoted by I pCq, as the union of the ideal
boundary of each of the connected components of its complement.
In the applications in this section, C will be a forward invariant connected

component of the Julia set of a given rational map f . If so, the connected
components of the complement of C might be Fatou components, but in
general they are unions of Fatou components and other components of the
Julia set.
Think for instance of the well-known example of a cubic Blaschke product
with an invariant Herman ring (see Figure 3.8). If C is the exterior component
of the Julia set, i.e. the boundary of the immediate basin of the superattracting
fixed point at infinity, then f pCq “ C and only the unbounded component of
the complement is a Fatou component. All the bounded components contain
infinitely many Fatou components (annuli) together with their boundaries and
accumulations of such.
It may also happen that the continuum C is disjoint from the boundary of
any Fatou component. This is the case in the McMullen examples of a buried
quasicircle, which is an invariant component of the Julia set of a rational
map in a special family of rational maps (see Figure 7.14). In this case, C
is accumulated by infinitely many Fatou components.
Figure 7.14 The Julia set of the maps fλ pzq “ z3 ` λ{z3 consists, for λ ‰ 0 and
sufficiently small, of a Cantor set of quasicircles. Uncountably many of them are
buried, i.e. they do not intersect the boundary of any Fatou component. There exists a
unique buried quasicircle Cλ that is invariant under fλ . These examples were
discussed for the first time in [McM1]. The Julia set is drawn for λ “ 0.006 ` 0.002i .
Another possibility is shown in Figure 3.7, where taking C “ S1 we see

that both connected components of CzC p are infinitely connected Fatou com-
ponents, with connected components of the Julia set accumulating on C.
If f is a polynomial with connected Julia set, then C can only be the
whole Julia set. In this case, every component of the complement is a Fatou
components of f .
If f is a rational map and C is a forward invariant component of the Julia
set, then the preimage f ´1 pCq has finitely many components. The induced
map fp : I pCq Ñ I pCq is mapping each component of I pCq onto a component
of I pCq. On each of them the map fp is defined as described above. Indeed, let
U be a component of CzC.p If U contains no components of f ´1 pCq, then f
maps U properly onto a component V of CzC. p In this case fp : I pU q Ñ I pV q
is defined as in Case 1 above. If, instead, U does contain at least one component
of f ´1 pCq, then let U 1 be the component whose closure contains BU . Then f
maps U 1 properly onto a subset V 1 of a component V of CzC. p In particular,
f maps a one-sided neighbourhood of BU onto a one-sided neighbourhood of
BV as a covering so that fp : I pU q Ñ I pV q is defined as in Case 2 above.
In the applications we perform a sequence of surgeries. Therefore, it is
convenient to work in the wider class of maps, the quasirational maps, to avoid
having to integrate at each step.
Recall that a map f : C p ÑC p is quasirational if it is quasiconformally
conjugate to a rational map. By the Julia set of f we mean the set which under
conjugation is mapped to the Julia set of the rational map. Similarly for the
Fatou components of f . Recall also that by Sullivan’s Straightening Theorem
(Theorem 5.6), being quasirational is equivalent to the iterates f n for n ě 1

being uniformly K-quasiregular for some K ě 1.
Under the same assumptions as above, observe that if the original map f
is only assumed to be quasirational, it still makes sense to define the induced
map fp : I pCq Ñ I pCq. In this case, however, on each component the map
is only quasisymmetrically equivalent to a real analytic map (actually qua-
sisymmetrically conjugate, if the component were mapped to itself). More
precisely, let φ : C
p ÑC p be an integrating quasiconformal homeomorphism,
so that g :“ φ ˝ f ˝ φ 1 is a rational map. Then g induces a real analytic
´
map gp : I pφpU qq Ñ I pφpV qq via Riemann mappings Rφ pU q and Rφ pV q . On

the other hand, the maps φ : U Ñ φpU q and φ : V Ñ φpV q induce quasisym-
metric maps φpU : I pU q Ñ I pφpU qq and φpV : I pV q Ñ I pφpV qq. The induced
map fp : I pU q Ñ I pV q can be written as
fp :“ φpV´1 ˝ gp ˝ φpU : I pU q Ñ I pV q.
Hence, fp is quasisymmetrically equivalent to the real analytic map gp. Notice

that if V “ U then φpV “ φpU and therefore fp : I pU q Ñ I pU q is conjugate to gp.
7.5.1 The surgery

The surgery consists in replacing a map f on some disc U by a new map,
which is quasiconformally conjugate to a Blaschke product.
Setup Let f be quasirational and C a non-degenerate invariant continuum. Its

preimage f ´1 pCq consists of C and finitely many other disjoint components.
Let U and V be components of CzC p so that, as above, fp : I pU q Ñ I pV q
denotes the induced map on the ideal boundaries. Moreover, let B denote a
Blaschke product, mapping D onto D.
Definition 7.28 (Compatibility of maps on the ideal boundary) In the setup

above we say that f and B are compatible on the ideal boundary of U via
quasiconformal maps φU : U Ñ D and φV : V Ñ D, if the induced maps
φpU : I pU q Ñ S1 and φpV : I pV q Ñ S1 define a quasisymmetric equivalence
between fp : I pU q Ñ I pV q and the restriction of B to the unit circle, as shown
in the following commutative diagram:
fp
I pU q ÝÝÝÝÑ I pV q
§ §
§ §p
φpU đ đφV
B
S1 ÝÝÝÝÑ S1
If V “ U (and hence I pU q is fixed), then we are able to choose φU and φV

such that φp :“ φpU “ φpV , and then fp and B are quasisymmetrically conjugate
under φp : I pU q Ñ S1 (see Figure 7.15).
U V
f
RU RV
I pU q I pV q
φU fp φV
φpU φpV
S1 S1
Figure 7.15 f and B are compatible on the ideal boundary of U via quasiconformal
maps φU : U Ñ D and φV : V Ñ D.
We are ready to state and prove the key proposition in the surgery, denoted
the Gluing Lemma by McMullen.
Proposition 7.29 (Gluing Lemma) In the setup above suppose that:
• either I pU q is a fixed component of I pCq under the map fp : I pCq Ñ I pCq;

• or I pU q is strictly preperiodic under fp, and f n pV q X U “ H for all n ą 0.
Assume f and B are compatible on the ideal boundary of U , via quasicon-

formal maps φU : U Ñ D and φV : V Ñ D, which in the fixed case satisfy
φU “ φV :“ φ. Then the map
#
f pzq if z R U ,
gpzq :“ ´1
φV ˝ B ˝ φU pzq if z P U ,
is quasirational. If I pU q is fixed, then B and g are quasiconformally conjugate

on U .
Proof Consider f ´1 ˝ g locally near BU , using the branch of f ´1 which

fixes the points not in U . This map is the identity outside U , and quasiconfor-
mal inside U . It also induces the identity map on the ideal boundary since
fp “ φpV´1 ˝ B ˝ φpU “ gp : I pU q Ñ I pV q. Using Proposition 7.26 we deduce

that f ´1 ˝ g is locally quasiconformal. By composing with f we obtain that
g is locally quasiconformal except on neighbourhoods of the branch points.
Hence, g is quasiregular. The dilatation of g is bounded by the maximum of
Kpf q and KpφU q ¨ KpφV q.
If I pU q is a fixed component of fp, then gpU q “ U and g n pU q “ φ ´1 ˝
B ˝ φpU q. Hence, we have Kpg n q “ Kpφq2 on U . If z R U , it may have sev-
n
eral iterates in CzU

p before falling into U . Hence, Kpg n q ď Kpf n q ¨ Kpφq2 ď
K ¨ Kpφq2 , where we used that f is quasirational so that Kpf n q ď K for all n
and some K ě 1. Hence, g is quasirational (see Figure 7.16).
g
g“f
U g “ φV´1 B φU V
φU φV
Figure 7.16 The Gluing Lemma.
In the case where fppI pU qq “ I pV q and f n pV q X U “ H for all n ą 0, the

dilatation is Kpg n q ď K ¨ KpφU q ¨ KpφV q since no orbit can pass through U
more than once.
Corollary 7.30 (The Gluing Lemma in the periodic case) In the setup
above suppose U0 “ Up , U1 , . . . , Up´1 are components of CzCp such that
f pI pU0 qq “ I pUj q for 0 ď j ď p ´ 1. For j “ 0, 1, . . . , p ´ 1 let Bj :
pj
D Ñ D be Blaschke products so that f |Uj and Bj are compatible on the

ideal boundary via the quasiconformal homeomorphisms φj : Uj Ñ D, where
φp “ φ0 . Then the map
# Ťp ´ 1
f pzq if z R Uj ,
gpzq :“ j “0
φj´`11 ˝ Bj ˝ φj pzq if z P Uj for j “ 0, 1, . . . , p ´ 1,
is quasirational.
Proof Applying the same arguments as in the proof of the Gluing Lemma
above we conclude that g is quasiregular. To see that Kpg n q is uniformly
bounded, observe that g p is quasiconformally conjugate to a Blaschke
řp´1
product of degree d on any of the Uj ’s, where d “ j “0 degpBj q (see
Figure 7.17).
g
g“f g
g3
U0 g U1 g U2
φ0 φ1 φ2
B
B0 B1
B2
Figure 7.17 The Gluing Lemma in the periodic case for p “ 3. Notice that g 3 |Uj is
conjugate to the Blaschke product Bj `2 ˝ Bj `1 ˝ Bj , where indices are taken mod 3.
Remark 7.31
(1) The degree of g may be lower than the degree of f .

(2) Observe that, even if f pU q is not equal to V , then after the surgery gpU q “
V . If I pU q is periodic of period p, we obtain g p pU q “ U so that U is a
periodic Fatou component of the quasirational map g.
7.5.2 Application: changing any rational map to one with

connected Julia set
One application of the surgery above is to simplify the dynamics of a given
rational map by pasting in rigid models, defined as follows.
Definition 7.32 (Rigid models) We define the three rigid models for proper
self-maps of the unit disk D as:
(a) the elliptic model z ÞÑ e2π i θ z, for θ P RzQ;

(b) the hyperbolic model z ÞÑ zd for some d ą 1; and
(c) the parabolic model z ÞÑ Pd pzq for d ą 1, where
zd ` a d ´1
Pd pzq :“ d
for a “ ,
1 ` az d `1
is the unique map of degree d with a single critical point at z “ 0 and a
parabolic fixed point of multiplicity three at z “ 1.
Remark 7.33
(1) If B is any of the above models, and φ is a quasisymmetric map of S1 ,

which conjugates B to itself on S1 , then φ must be a rigid rotation z ÞÑ αz
with |α| “ 1. For the elliptic model, α can be arbitrary; for the hyperbolic
model, α must satisfy α d “ 1, and in the parabolic model, α “ 1. This is
the reason for calling the models rigid. Note that different types of rigid
models are not quasisymmetrically conjugate on S1 .
(2) However, the models are rigid only when we require conjugacy. If we
consider equivalence, then the hyperbolic and parabolic models are in
fact conformally equivalent on D. Set ϕ1 “ IdD and ϕ2 pzq “ 1z` à
az , then
d
ϕ2 pz q “ Pd pϕ1 pzqq. Analogously, we see that the rigid rotation Rθ is
conformally equivalent to IdD .
The main application is stated in the following theorem ([McM1, Prop. 6.9]).
Theorem 7.34 (Main Theorem) Let f be a rational map and C Ă Jf a for-

ward invariant connected component different from a single point. Then there
exists a rational map g and a quasiconformal homeomorphism φ : C pÑC p
such that φ conjugates g|Jg to f|C . Moreover, g has the following properties.
The Julia set Jg is connected and the periodic Fatou components of g are
Siegel discs, basins of superattraction or parabolic basins. On each periodic
Fatou component, the first return map is conjugate to one of the three rigid
models. Moreover, if c is a critical point in a strictly preperiodic Fatou compo-
nent, and n is the smallest number for which g n pcq is in a periodic component
U , then:
(a) if U is a Siegel disc, g n pcq is the indifferent periodic point in U , i.e. the
centre of the Siegel disc;
(b) if U is a basin of superattraction, g n pcq is the superattracting periodic
point;
(c) if U is a parabolic basin, g n`k pcq is a critical point, where k is the smallest
number for which g k pU q contains a critical point.
Remark 7.35
(1) The theorem above states that any forward invariant connected component
of the Julia set of a rational map is conjugate to the Julia set of a different
rational map.
(2) If f has a connected Julia set to start with, then we are replacing Fatou
components by Fatou components with rigid dynamics. If Jf is discon-
nected, we are erasing all of the Julia set of f , which is not part of C, and
replacing all components of CzCp by Fatou components of g.
(3) The replacements could be done with Blaschke products other than
the rigid models as long as the induced maps on the ideal boundaries
coincide.
Corollary 7.36 (Special case of hyperbolic rational maps) If f is a hyper-

bolic rational map with connected Julia set, then there exists, up to Möbius
conjugacies, a unique postcritically finite rational map satisfying the conclu-
sions of the Main Theorem. In fact the same is true if f and C are as in the
Main Theorem, and f is expanding on C.
To prove the Main Theorem we need to understand what kind of maps f

induces on the different components of the ideal boundary of C. The results
are summarized in Propositions 7.37 and 7.39 below. We postpone the proofs
of the propositions to the end of the section and first finish the proof of the
Main Theorem.
Dynamics induced by f on the ideal boundary I(C) Suppose f is quasirational

and C Ă Jf is a forward invariant connected component different from a
single point. Then the preimage of C has finitely many components. The two
propositions below are in [McM1, Theorem 6.1].
Proposition 7.37 (Global dynamics of fp on I(C))
(a) All but finitely many components of CzC

p map to their images by a quasi-
conformal homeomorphism.
(b) Every component of I pCq is preperiodic and there are only finitely many
periodic ones.
(c) For any component I pU q of I pCq, there exist quasisymmetrical mappings
φp1 and φp2 mapping I pU q and fppI pU qq respectively to S1 providing an
equivalence between fp and z ÞÑ zd for some d ě 1. If I pU q is fixed, fp
and z ÞÑ zd may not be quasisymmetrically conjugate.
Remark 7.38
(1) Note that the equivalence in Proposition 7.37 between fp and z ÞÑ zd for
some d ě 1 can be changed to an equivalence between fp and any Blaschke
product of degree d on S1 .
(2) There is also a lot of freedom in the choices of the quasisymmetric maps φpj
for j “ 1, 2 providing the equivalence. If one of them, say φp1 , is changed
to an arbitrary quasisymmetric map ψ p1 , then there exists a quasisymmetric
map ψ p2 so that the pair ψ p2 of maps provide the equivalence.
p1 , ψ
Proposition 7.39 (Dynamics of fp on periodic components of I(C)) On

each periodic component of I pCq, the first return map is quasisymmetrically
conjugate to one of the rigid models.
We shall apply Propositions 7.37 and 7.39 in the sequence of surgeries which
prove the Main Theorem.
Sequence of surgeries and proof of the Main Theorem The proof of the Main
Theorem is divided into a sequence of surgeries. Since we work in the class of
quasirational maps we may as well start from a quasirational map f and a
forward invariant connected component C Ă Jf different from a single point.
After each surgery construction the resulting map g is quasirational, the subset
C of Cp is unchanged, C is a connected component of Jg , and gp “ fp on I pCq.
Hence, after each step we are allowed to denote the resulting map again by f .
We start by a survey of the different steps in the proof. Afterwards we make
the constructions more precise. We first consider the cycles of components of
I pCq under fp, one after the other. It follows from Proposition 7.37(b) that
there are only finitely many of them. Suppose I pU q is periodic of period
p ě 1. Then from Proposition 7.39 we know that fpp is quasisymmetrically
conjugate to exactly one of the rigid models. We then change the dynamics on
the p components of CzCp whose ideal boundaries form the p-cycle of I pU q
under f , in such a way that the composition is quasisymmetrically conjugate
p
to the correct rigid model. After applying the Gluing Lemma we obtain a new
quasirational map. After a finite number of such surgeries we have dealt with
all the cycles of components of I pCq.
We are left with the strictly preperiodic components of I pCq under fp. We
work backwards along successive preimages of the cycles of I pCq. It follows
from Proposition 7.37(a) that there are only finitely many components of CzCp
on which the dynamics is not a quasiconformal homeomorphism and has to be
replaced. We perform the replacements one after the other, each time obtaining
a new quasirational map. The changes are chosen such that the conditions in
the Main Theorem are satisfied for the rational map at the end.
We now go though the proof with more details.
The periodic case Suppose I pU q is periodic of period p ě 1. Following

Proposition 7.39 let φp : I pU q Ñ S1 be a quasisymmetric map so that φp ˝
fpp ˝ φp´1 equals exactly one of the rigid models, say BU . Then extend φp to
a quasiconformal homeomorphism of D, denote the extension by φ, p choose
a Riemann mapping RU : U Ñ D and set φ :“ φ ˝ RU : U Ñ D. It follows
p
that f p and BU are compatible on the ideal boundary via φ.
If p “ 1 we are done: replacing f on U by the map φ ´1 ˝ BU ˝ φ and
keeping f outside U , it follows from the Gluing Lemma that we obtain a new
quasirational map.
If p ą 1 then we have to replace f on the disc components Uj where
I pUj q :“ fpj pI pU qq for 0 ď j ď p ´ 1 with maps that are compatible with fp
along the ideal boundaries and whose composition agrees with φ ´1 ˝ BU ˝ φ
on U . Let dj ě 1 denote the degree of fp : I pUj q Ñ I pUj `1 q for 0 ď j ď
řp´1
p ´ 1. Then the sum j “0 dj “ d where d “ 1 in the elliptic case. From
Proposition 7.37(c) we know that there exist a pair of quasisymmetric maps
mapping I pUj q and I pUj `1 q respectively to S1 and providing an equivalence
between fp and Bj pzq “ zdj , for 0 ď j ď p ´ 2. From Remark 7.38(2) we
know that we may choose one of the quasisymmetric maps arbitrarily. Let
φp0 “ φp : I pU0 q Ñ S1 from above. For j “ 1, . . . , p ´ 1 establish recursively
φpj : I pUj q Ñ S1 such that φpj ˝ fp “ Bj ´1 ˝ φj ´1 . Depending on the case we
choose Bp´1 to be the Blaschke product defined as
$
’
’ e2π i θ z in the elliptic case,
&
d
Bp´1 pzq :“ z p´1 in the hyperbolic case,
’ dp´1
z ` a
in the parabolic case, for a “ d ´1 .
’
%
dp´1 d `1
1àz
The conditions in Corollary 7.30 are therefore satisfied. Hence, by replacing
f on Uj by the map φj´`11 ˝ Bj ˝ φj for j “ 0, 1, . . . , p ´ 1 and keeping f
Ťp´1
outside j “0 Uj we obtain a new quasirational map.
See Figure 7.17, observing that the choice of U “ U0 is arbitrary. More
precisely g p “ φj´1 ˝ B pj q ˝ φj on Uj for any j “ 0, . . . , p ´ 1 where B pj q
is the appropriate cyclic composition of B0 , . . . , Bp´1 .
The strictly preperiodic case After having dealt with all periodic components
of I pCq we work backwards along the preimages of each of them. If I pV q is
one of them, and fppI pV qq “ I pW q, then fp is quasisymmetrically equivalent to

z ÞÑ zd for some d ě 1. If d “ 1 we keep the quasiconformal homeomorphism
f |V : V Ñ W . Only if d ą 1 we shall replace the map. Suppose this has been
done up to V . We construct as before gluing maps φ1 and φ2 using Riemann
mappings RV and RW and define g : V Ñ W as g :“ φ2´1 ˝ pφ1 qd . The new
critical point c equals φ1´1 p0q P V with critical value gpcq equal to φ2´1 p0q.
The orbit of c satisfies one of the conditions in the Main Theorem if the orbit
of gpcq does. We choose the Riemann mapping RW so that φ2´1 p0q has the
correct orbit behaviour.
At each step we obtain a new quasirational map. The next surgery starts
from the resulting quasirational map.
Conclusion of the proof of the Main Theorem After a finite number of steps
we have constructed a quasirational map, satisfying the conditions in the Main
Theorem. After integration we get the rational map g.
Proof of Propositions 7.37 and 7.39

Proof of Proposition 7.37 There are only finitely many components
V1 , . . . , Vk of CzC,
p which intersect f ´1 pCq or contain a critical point. Set
V “ tVj u1ďj ďk .
Proof of (a). If U is a component of CzC p not in V, then f maps U
onto another component of CzC as a covering. Since U is a disc, f |U is a
p
quasiconformal homeomorphism onto its image.
Proof of (b). Let U be a component of CzC p not in V. Suppose f n pU q is
not in V for any n ě 0. Then f is normal on U and therefore f n pU q must
n
eventually cycle, otherwise U would be a wandering domain and that is not

possible for a rational map. Hence, U is eventually mapped onto a Fatou
component whose first return map is a quasiconformal homeomorphism, and
therefore a Siegel disc. It follows that I pU q is preperiodic. If, on the other
hand, f n pU q is eventually mapped onto one of the components in V then
I pU q must also be preperiodic. Indeed, if some iterate is mapped onto I pW q
for W R V, we can repeat the argument from before, the component W is
eventually mapped onto either a Siegel disc or onto one of the components
in V. Since there are only finitely many components in V the ideal boundary
I pU q will eventually cycle. There is a finite number of periodic components
in I pCq, since there is at most finitely many cycles of Siegel discs and finitely
many components in V.
Proof of (c). If U is not in V then f |U is a quasiconformal homeomorphism
and fp : I pU q Ñ f pI pU qq is quasisymmetrically equivalent to the identity. If
U is one of the components in V, then we can find an annulus U 1 Ă U such that
fp
RU RV
U V
φ1 φ2
U1
f
V1
ϕ1 ϕ2
ϕ2 ˝ f ˝ ϕ1´1
ϕ ϕ
z ÞÑ zd
Figure 7.18 Sketch of the proof of Proposition 7.37(c).
BU Ă BU 1 and f maps U 1 onto f pU 1 q Ă V by a covering. (Since U contains

only finitely many critical points and finitely many components of f ´1 pCq
then U 1 can be chosen to avoid them all.) If we uniformize U 1 and f pU 1 q
separately to standard annuli of the form Ar “ tr ă |z| ă 1u, using maps ϕ1
and ϕ2 respectively, then the induced covering map on the annulus must be
conjugate to z ÞÑ zd , with the same extension to S1 as before. Let ϕ denote the
conjugacy.
1
Choose Riemann mappings RU and RV . Then φ1 :“ ϕ ˝ ϕ1 ˝ R´ U and
´1
φ2 :“ ϕ ˝ ϕ2 ˝ RV extend to S1 providing an equivalence between fp :
I pU q Ñ I pV q and z ÞÑ zd for some d ě 1 (see Figure 7.18).
Observe that if I pU q is fixed then the proof provides an equivalence between
fp and z ÞÑ zd but not necessarily a conjugacy, since φp1 and φp2 are in general
different maps.
The remainder of this section is dedicated to proving Proposition 7.39, for

which we shall study each type of periodic component of I pCq separately. Let
U be a component of CzC
p for which I pU q is periodic under fp. Replacing f by
an iterate, we may assume that I pU q is fixed. As discussed above, there is an
annulus U 1 Ă U with BU Ă BU 1 and such that f is a covering onto its image.
Lemma 7.40 Suppose fppI pU qq “ I pU q. If U 1 can be chosen to be a proper

subset of f pU 1 q then fp is quasisymmetrically conjugate to z ÞÑ zd for some
d ą 1.
In the proof of the above lemma we shall apply the following theorem of
Shub (cf. [Shu], [dMvS, Chapt. II, Sect. 2]).
Theorem 7.41 (Quasisymmetric conjugacy of expanding circle maps) Any

expanding C 1`α orientation preserving endomorphism of S1 is quasisymmet-
rically conjugate to z ÞÑ zd for some d ą 1.
Proof of Lemma 7.40 Choose a Riemann mapping RU : U Ñ D. Then

RU pU 1 q is a one-sided neighbourhood of S1 , which is mapped properly over
1
itself by the map RU ˝ f ˝ R´ U . By the Schwarz Reflection Principle, we
obtain a holomorphic covering F : A1 Ñ A, where A1 and A are annuli sur-
rounding S1 such that A1 is a proper subset of A. It follows that F |S1 is
expanding relative to the Poincaré metric on A. Hence by Theorem 7.41 the
lemma follows.
The next proposition deals with the buried case where BU is not part of the
boundary of any Fatou component.
Proposition 7.42 (The buried case) Suppose fppI pU qq “ I pU q. If BU is not

contained in BV for any Fatou component V Ă U , then fp : I pU q Ñ I pU q is
quasisymmetrically conjugate to z ÞÑ zd , for some d ě 1.
To prove this proposition we shall apply the following theorem from topol-
ogy (cf. [Why, p. 35]).
Theorem 7.43 (Zoretti’s Theorem) Let C be a component of a compact set

J Ă R2 . For all ε ą 0, there exists a simple closed curve γ which encloses C
and such that γ X J “ H and distpC, xq ă ε for all x P γ .
Proof of Proposition 7.42 Since C is a component of Jf , we can apply

Zoretti’s Theorem. Choose a decreasing sequence tn uną0 of positive numbers
tending to 0 as n tends to 8. Let γn denote a simple closed curve in the
Fatou set, lying in an n -neighbourhood of BU . Since BU is disjoint from the
boundary of any Fatou component, these curves lies in infinitely many different
Fatou components. For n sufficiently large these components are not periodic,
and the open annuli bounded by γn and BU contain no critical values of f .
Let V 1 denote such an annulus. Let U 1 denote the unique component of
f 1 pV 1 q which lies in U and is bounded by BU and some preimage of γn .
´
Since γn is not contained in a periodic Fatou component, it must be disjoint

from its preimages, and therefore either U 1 Ă V 1 or V 1 Ă U 1 . But f maps U 1
properly onto V 1 with some degree. Since they are both annuli, the modulus
of V 1 must be larger than or equal to the modulus of U 1 . Hence U 1 must be a
proper subset of V 1 . The conclusion then follows from Lemma 7.40.
In the remaining cases BU Ă BV , where V Ă U is a Fatou component. By

the classification of periodic Fatou components for rational maps V can only
be a Siegel disc, a Herman ring, a basin of (super)attraction or a parabolic
basin. We deal with these cases separately.
Proposition 7.44 (Siegel discs and Herman rings) If V is a fixed Siegel disc
or a fixed Herman ring of rotation number θ P RzQ, then fp : I pU q Ñ I pU q
is real analytically conjugate to the rigid rotation Rθ pzq “ e2π i θ z.
Proof If V is a Siegel disc then U “ V and a Riemann mapping RU , which

maps the indifferent fixed point to 0, conjugates f to the rigid rotation on D.
Hence fp on S1 is actually equal to Rθ . If V is a Herman ring, then there is
a conformal map ϕ : V Ñ Ar “ tr ă |z| ă 1u conjugating f to Rθ . In this
case, arguing as we did in Proposition 7.37, we obtain that fp on I pU q is real-
analytically conjugate to Rθ .
Proposition 7.45 (Basins of (super)attraction and certain parabolic basins)

If V is an immediate basin of (super)attraction, or a parabolic basin whose
parabolic fixed point is not on BU , then fp on I pU q is quasisymmetrically
conjugate to z ÞÑ zd for some d ą 1.
Proof Let W Ă V be a Jordan domain, for which f pW q is a proper subset

of W and W does not intersect BU . In the (super)attracting case W may be
chosen as a disc surrounding the (super)attracting fixed point and mapping
inside itself. In the parabolic case W may be chosen as an attracting petal in
the parabolic basin so that W intersects BV at the parabolic fixed point.
Let Wn be the component of f ń pW q containing W , then
W Ă W1 Ă W2 ¨ ¨ ¨
Ť
and n Wn “ V . In the parabolic case, BU cannot contain any point in the
grand orbit of the parabolic point since f pBU q “ BU , so Wn X BU “ H for
all n ą 0. Observe that, if V is not simply connected, Wn will become multiply
connected after a certain n0 , since the union of all the Wn ’s must account for
all of V .
Wn
Wn ´ 1
An
f
Figure 7.19 Sketch of the annuli An in the superattracting case.
For every n, there is a component of BWn which together with BU bounds

an annulus An disjoint from Wn (see Figure 7.19). In general, An may contain
infinitely many different components of the Julia set. Each annulus lies in
an n -neighbourhood of BU , where n Ñ 0 as n Ñ 8. Otherwise there is a
continuum in BV which lies in U and meets BU , contradicting the fact that C
is a component of the Julia set.
Choose n sufficiently large so that An contains no critical point. Then,
An`1 is an annulus properly contained in An and f pAn`1 q “ An . Hence
Lemma 7.40 applies and the conclusion follows.
The only case left is the general parabolic.
Proposition 7.46 (General parabolic basins) If V belongs to a parabolic

basin, with a parabolic fixed point on BU , then fp on I pU q is quasisymmetri-
cally conjugate to Pd for some d ą 1.
Proof Let RU : U Ñ D be a Riemann map sending the parabolic fixed point

to z “ 1 (Riemann maps have radial limits at parabolic fixed points since these
are accessible). Then f and V carry over to D where we also denote them by
f and V . Observe that f is not defined on the whole disc, and that its domain
contains V and a one-sided neighbourhood of S1 . By the Schwarz Reflection
Principle, f can be extended to a holomorphic map in a neighbourhood of
S1 having a parabolic fixed point at z “ 1. The restriction to the unit circle is
precisely fp. We construct by hand a quasisymmetric conjugacy between fp and
Pd , where d is the degree of fp on S1 .
Consider an attracting petal W , containing z “ 1 on its boundary. For
simplicity choose W so that BW is a C 8 curve avoiding the critical orbits.
As before, let Wn denote the component of f ń pW q containing W . For n
sufficiently large, Wn contains all critical points of V , and hence there is a
unique component γn of BWn , lying between BU and Wn , and touching S1 at

the n-th preimages of z “ 1. This curve γn is contained in the annulus where
f is a covering (see Figure 7.20).
f P2
γn`1 1
γn`1
An A1n A1n
An γn1
γn
Wn Wn1
1 1
An
An A1n A1n
Figure 7.20 The construction of Wn , γn and An in the parabolic case, drawn

for d “ 2.
The curves γn and γn`1 bound a pinched annulus An with several compo-
nents. This pinched annulus is, in a sense, a fundamental domain, since all
forward orbits of points in between S1 and γn`1 must pass once and only once
through An Y γn .
We may carry out the same discussion on another copy of the unit disc for
the map Pd , obtaining analogous sets Wn1 , γn1 , A1n , etc.
To construct the conjugacy, we start by fixing n sufficiently large and by
choosing a C 1 -diffeomorphism φ from the curve γn onto γn1 , fixing z “ 1, and
sending preimages of z “ 1 to their analogues in an appropriate fashion. Using
the map f , we define φ on γn`1 so that Pd ˝ φ “ φ ˝ f . We then extend φ
quasiconformally to the different components of An sending them to their
analogues A1n . Using the dynamics of f and Pd we extend φ recursively
to the whole pinched annulus in between γn and S1 , so that it conjugates
f and Pd . Such map extends quasiconformally to the whole unit disc, and
quasisymmetrically to S1 , conjugating fp and Pd as required.
7.5.1 Prove as stated in Remark 7.33(1) that if B is any of the rigid mod-
els defined in Definition 7.32 and φ is a quasisymmetric map of S1
conjugating B to itself on S1 , then φ must be a rigid rotation z ÞÑ αz
with |α| “ 1. Deduce conditions on α for each type of rigid model.
7.5.2 Prove that every Blaschke product of degree d ě 1 mapping D to

itself is quasisymmetrically equivalent to z ÞÑ zd on S1 . (This is
Remark 7.38(1).)
7.5.3 Suppose f is quasisymmetrically equivalent to z ÞÑ zd on the unit cir-
cle. Show that for any quasisymmetric map ψ1 : S1 Ñ S1 , there exists
a quasisymmetric map ψ2 : S1 Ñ S 1 such that ψ1 and ψ2 also provide
an equivalence between f and z ÞÑ zd . (This is Remark 7.38(2).)
7.5.4 Consider the cubic Blaschke product fλ,a given in the Herman ring
example illustrated in Figure 3.8. Let C denote the exterior connected
component of the Julia set, i.e. the immediate basin of the superattract-
ing fixed point at infinity. Go through the proof of the Main Theorem
(Theorem 7.34) in this case to see that fλ,a only needs to be modified
on one of the disc components of CzC. p Observe that the resulting
map is quasiconformally conjugate to a quadratic polynomial with a
Siegel disc.
7.5.5 Let P be a polynomial of degree d0 ě 2, with an attracting fixed point
α. Let A˝ “ A˝ pαq denote the immediate basin of attraction and A “
Apαq the whole basin. Suppose P |A˝ has degree d ě 2. Set C :“ BA˝ ;
d b
it is a forward invariant continuum in JP . Let Bpzq :“ 1z`` bzd
where
1
0 ă b ă dd ´
`1 is arbitrary.
(a) Show that B has degree d on D, has an attracting fixed point in
p0, 1q whose immediate basin of attraction is D, and has a unique
critical point at z “ 0.
(b) Prove that P and B are compatible on the ideal boundary of A˝ .
(c) Prove that there exists a quasiconformal map φ : C Ñ C such that
it conjugates P to a new polynomial Pr for which φpαq is an
attracting fixed point, with immediate basin of attraction φpA˝ q,
containing a unique critical point.
7.5.6 Let P , A, A˝ , . . . be as in Exercise 7.5.5. Set C :“ BA; it is a forward

invariant subset of the Julia set, perhaps disconnected.
(a) Observe that all bounded components of CzC p are topological discs
and CzA is totally invariant. Therefore, we may define I pCq as the
union of the ideal boundaries of all components of A and redefine
fp appropriately.
(b) Using Exercise 7.5.5 and replacing P on the appropriate preperi-
odic components, show that there exists a quasiconformal homeo-
morphism φ : C Ñ C conjugating P to a polynomial Pr, outside A,
such that φpAq is a basin of attraction with a single critical grand
orbit, having at most one critical point in each component.
7.6 Disc-annulus surgery on rational maps 267

Kevin M. Pilgrim and Tan Lei
Let f : Cp ÑC p be a rational map. In this section, we introduce a flexible cut
and paste surgery supported on the Fatou set of f . The resulting new map has
dynamics closely related to that of f , although the connectivity of some of the
Fatou components may increase.
The construction is in many ways similar to a related construction in the
context of Kleinian groups. The ‘Klein–Maskit combination of type I using
trivial discs’ (cf. [Mar]) takes as input two Kleinian groups 1 , 2 and two
round discs D r 2 with trivial stabilizer contained in the ordinary sets 1 , 2 .
r1 , D
The output is a Kleinian group “ 1 ˚ 2 containing Möbius conjugate
copies of 1 and 2 . That is, the dynamics of contains that of 1 and
2 . Moreover, it can easily be shown that the connected components of the
limit set of are either translates of connected components of the limit sets
j , or points. This operation also has a natural combinatorial analogue in the
setting of three-manifolds, namely boundary connect sum. Let Mj “ pH3 Y
j q{j , j “ 1, 2, be the quotient three-manifolds. Since j is assumed non-
empty, Mj has non-empty boundary, and the discs D r j descend to discs Dj on
3
BMj . The manifold M “ pH Y q{ is obtained from the manifolds Mj by
identifying D1 and D2 via an orientation-reversing homeomorphism.
In the setting of rational maps, however, the fact that maps with interesting
dynamics have degree strictly greater than one makes the formulation of
combination theorems taking as input two (or several) rational maps rather
subtle. One difficulty is the lack of a natural, simple combinatorial analogue
to describe the gluing data, which can be quite complicated to describe. We
therefore content ourselves with the more modest program of beginning with
a single map f and modifying it (in a more-or-less arbitrary fashion) inside a
smoothly bounded disc V Ă Ff to obtain a new map F .
Contents In Section 7.6.1 we define precisely the first surgery, which we call
disc-annulus surgery. In Section 7.6.2 we give applications and relate the Julia
set of F to that of f . In Section 7.6.3 we discuss the dependence of the map
F on the various choices made in the construction and prove a uniqueness
theorem (Theorem 7.60). In Section 7.6.4 we discuss the inverse operation to
disc-annulus surgery. Finally, in Section 7.6.5 we give a related construction in
which the modification procedure is slightly more involved. Nevertheless it is
useful for the construction of examples such as those found in [PT2]. One such
1 This section is a revised version of the paper [PT1] and is reproduced with the permission of
World Scientific.
example is a hyperbolic rational map whose Julia set contains a wandering

component which is a Jordan curve but not a quasicircle.
7.6.1 The disc-annulus surgery

We say that V Ă C is a smooth disc if BV is a real-analytic Jordan curve. By a
p
branched covering we mean a proper C 1 -map between smooth, oriented (real)
2-manifolds, possibly with boundary, such that the boundary map is a covering
map of (real) 1-manifolds, and such that, on the interior, the map is given in
appropriate local (complex) coordinates by z ÞÑ zd for some d.
The following two lemmas are the technical ingredients for the definition of
the disc-annulus surgery.
Lemma 7.47 (Key Lemma) Let A Ă C p be an open annulus bounded by two
C Jordan curves γ , and let W be an open disc bounded by a C 1 Jordan curve
1 ˘
η. Give orientations to the curves such that A and W lie to the left of their
boundaries. Let f ˘ : γ ˘ Ñ η be two orientation preserving C 1 -coverings of
degree d ˘ ě 1. Then there exists a branched covering g : A Ñ W satisfying
the following properties:
(1) g|γ ˘ “ f ˘ ;
(2) gpAq “ W and the degree of g is d ` ` d ´ ;
(3) g can be chosen to be C 1 in A and holomorphic and proper in a union of
any collection of finitely many smooth discs compactly contained in A with
pairwise disjoint closures.
See Figure 7.21.
We call the map g a covering extension of the boundary maps f ˘ . In
practice, we will take g to be holomorphic in a neighbourhood of its critical
points.
We recall the next lemma which was already mentioned in Remark 2.11 of
Chapter 2, as an application of the Schwarz Reflection Principle.
Lemma 7.48 (Extension Lemma) Let D, D 1 be two smooth discs in C. p Then
a holomorphic proper mapping F : D Ñ D extends to a holomorphic map in
1
a neighbourhood of D. In particular, F : BD Ñ BD 1 is a C 1 -covering.

A branched covering F is quasirational if it is quasiconformally conjugate
to a rational map. The Julia set of a quasirational map is thus well defined, and
has the same qualitative metric and measure theoretic properties as the Julia set
of a rational map. The following lemma is standard in the construction of new
rational maps via surgeries. It is a restatement of a special case of the Second
Shishikura Principle (Proposition 5.5).
A γ`
γ´
f` f´ g
d´ : 1
d` : 1 d` ` d´ : 1
W η
Figure 7.21 Left: The setup for Lemma 7.47. Right: Example of a branched covering
of degree two from an annulus onto a disc, such that each of the boundaries of the
annulus is mapped onto the boundary of the disc by degree one. The blue curve is
mapped to the blue segment two to one. The outer and inner boundaries of the annulus
are mapped to the boundary of the disc as indicated in the figure. The red and green
segments are respectively mapped to the red and green slits in the disc two to one. The
dotted line is mapped bijectively to the dotted line. Each of the connected domains in
the annulus minus the segments are mapped bijectively to the open disc minus the two
slits. See the exercises at the end of the section for generalizations.
Lemma 7.49 (Shishikura Principle) Let F : C p ÑC p be a C 1 branched cov-

ering which is holomorphic a.e. in CzB,
p holomorphic in a neighbourhood of
the critical points in B, and for some integer k, F j pBq X B “ H for all j ě k.
Then F is quasirational.
We now describe disc-annulus surgery.
Theorem 7.50 (Disc-annulus surgery) Let pf, V , H, h, gq satisfy the follow-
ing conditions:
• f is a rational map;
• V is a smooth disc such that:
– BV contains no critical points;
– f : V Ñ f pV q is proper;
– there exists 1 ď p ď 8 such that
f j pV q X V “ H for 0 ă j ă p,
and, in case p ă 8,
f p pV q Ă V and f p : V Ñ f p pV q is proper;
‚ H is a smooth disc with H Ă V zf p pV q;

‚ h : H Ñ Czf
p pV q is a holomorphic proper map, and
‚ g : V zH Ñ f pV q is a covering extension of the boundary maps such that
g is holomorphic and proper on f p pV q and near the critical points.
Then the map

$
’f on CzV
’
&
p
F :“ h on H
’
’
%g on V zH
is quasirational.
Remark 7.51 (V in Ff ) Observe that the hypotheses imply that V is a subset

of the Fatou set of f . If p ă 8, by the Schwarz Lemma there is an attracting
periodic point in f p pV q, and therefore V is part of a basin of attraction of
f (and of F ). If p “ 8 then either (i) V belongs to a strictly preperiodic
component of f or (ii) since V is non-recurrent, V is part of a basin of
attraction of an attracting or parabolic cycle. In none of the cases V can be
part of a periodic rotation domain.
Summarizing (see Figure 7.22), f is the original map; V is the disc with con-
trolled recurrence on which the modification is supported; h : H Ñ Czf p pV q
is the new, added dynamics; and g is an interpolating map gluing f outside
of V to h on H . We refer to F as a disc-annulus extension of f supported on
V . If h is univalent we call the extension F univalent or degree one. Note that
F pV q “ C p and degpF q “ degpf q ` degphq.
V F V
H H
h
f p pV q f p pV q
f pV q f pV q
Figure 7.22 The domains V , f p pV q and H in the case 2 ď p ă 8. The new map F
sends H to Czf
p pV q via h, and V zH to f pV q via g, using Lemma 7.47. On the white
region, F “ f .
Proof We first extend h to H by Lemma 7.48. If p “ 8, set f p pV q “ H.

Then F satisfies the conditions of Lemma 7.49 for B “ V zpH Y f p pV qq and
k “ 1. So F is quasirational.
7.6.2 Applications and properties

Here is an application of Theorem 7.50 with p “ 8.
Corollary 7.52 Let f be a rational map. Let z0 be a point in the Fatou set
such that z0 is neither a periodic point nor contained in a rotation domain.
Then there is a disc-annulus extension F of f supported on a neighbourhood
V of z0 .
Proof Choose V a smooth disc containing z0 such that V ztz0 u contains no
critical points, f : V Ñ f pV q is proper, and f j pV q X V “ H for all j ą 0.
Then any choice of pH, h, gq satisfying the conditions of Theorem 7.50 (for
p “ 8) will do.
Thus, for any map f and any non-recurrent point z0 in its Fatou set, one can
modify f via a disc-annulus surgery supported in a small neighbourhood V of
z0 so that on a disc H in this neighbourhood, the new dynamics is more or less
completely arbitrary.
Theorem 7.50 also applies to recurrent points in the Fatou set.
Corollary 7.53 Let f be a rational map. Let z0 be a (super)attracting periodic
point of period p. Then there is a disc-annulus extension F of f supported on
a neighbourhood V of z0 .
Proof Choose V to be a smooth disc containing z0 such that V ztz0 u contains
no critical points, f p : V Ñ f p pV q is proper with f p pV q Ă V and f j pV q X
V “ H for 0 ă j ă p. Then any choice of pH, h, gq satisfying the conditions
of Theorem 7.50 (for p ă 8) will do.
We now examine how the dynamics of f and F in the above two Corollaries
are related. Given a rational map or branched covering f , recall that the
postcritical set of f is defined as
ď
Pf “ f n pCf q,
ną0
where Cf is the set of critical points of f . We say that two compact sets K1
and K2 are conformally homeomorphic if there is a conformal map from a
neighbourhood of K1 to a neighbourhood of K2 which sends K1 onto K2 .
Corollary 7.54 (Properties of disc-annulus extensions) Let f and F be as

in Corollaries 7.52 or 7.53. Then:
(1) Jf Ă JF , JF is disconnected, and every connected component of Jf is a
connected component of JF .
(2) If H X Pf “ H and h is univalent, then H X PF “ H.
(3) If H X PF “ H, then:
• every Julia component of F passing through H infinitely many times is
a point, and
• every other Julia component of F is conformally homeomorphic to a
Julia component of f .
Proof (1) The set Jf is equal to the closure of the set of repelling periodic
points of f . Since V is contained in the Fatou set of f and F “ f outside V ,
repelling periodic points of f stay repelling periodic points of F . So Jf Ă
JF . Since also F pH q “ hpH q “ Czf p pV q Ą Jf and JF is totally invariant,
JF X H ‰ H. Moreover, since f “ F on BV and V is contained in the Fatou
set of f , it is easy to show that BV and hence BH are in fact also in the Fatou
set of F . So JF X H ‰ H, JF X pCzH p q ‰ H, and BH Ă FF . Hence JF is
disconnected. That each component of Jf is also a component of JF follows
from the fact that a rational map sends connected components of the Julia set
onto such components; see e.g. [Bea] for details.
(2) The critical points of F are the union of the critical points of f |Cp zV ,
g and h. The F -orbit of the g-critical points do not intersect H . So if h is
conformal and H X Pf “ H, H X PF “ H.
(3) Assume now H X PF “ H. Take a closed disc H 1 Ă H such that pJF X
H q Ă H 1 . By a lemma of Fatou, the diameters of the components of F ń pH 1 q
tend to zero as n Ñ 8 (see Lemma 3.40). A Julia component J 1 such that
F n pJ 1 q Ă H for some n is contained in a component of F ń pH 1 q. Therefore
if F n pJ 1 q Ă H for infinitely many n, J 1 is a point.
For example, if h is conformal, it has a unique repelling fixed point in H ,
which is a point component of JF .
Now let J 1 be a component of JF such that F pJ 1 q Ă H and F n pJ 1 q X
H “ H for n ą . Then F `1 pJ 1 q is a component of Jf , and F `1 is con-
formal in the component of F ´ pH q containing J 1 .
Corollary 7.55 (Connectivity of Fatou components) Denote by Wf , WF the

Fatou components of f and F containing BV . If Wf is periodic, then WF
is periodic and infinitely connected. If Wf is strictly preperiodic, and h is
univalent, then the connectivity of WF is equal to m0 ` m1 , where m0 and m1
are the connectivities of Wf and f pWf q.
Proof By assumption V Ă Wf and V zH Ă WF . If Wf is periodic with

p ă 8 then WF is also periodic since F p pV zH q “ f p pV q Ă V . If p “ 8
the periodicity is not changed either since F “ f outside V . Since BV is in the
Fatou set, but H contains points of the Julia set, it follows that WF is not simply
connected. But any invariant component of the Fatou set which is not a Herman
ring is either simply connected or infinitely connected ([Bea] Section 7.5). By
Remark 7.51, WF is not a Herman ring, hence WF is infinitely connected.
Now suppose Wf is strictly preperiodic and h is univalent, and let B “
f pWf qzf pV q, which has m1 ` 1 boundary components. Observe that points
in h´1 pBq are in H and belong to WF , since they never come back to Wf
under f . In particular, every boundary component of B except Bf pV q, is
a boundary component of WF , and H contains no other. All those original
boundary components of Wf are also so in WF , and the counting follows.
The following corollary has previously been obtained by Baker (cf. [Bea,
§11.7]).
Corollary 7.56 There exists rational maps with a non-periodic Fatou compo-
nent of any given number of connectivity.
These corollaries provide many examples of maps with disconnected Julia
sets. Explicit examples of rational maps with similar properties are given in
[Bea, Chapter 11]. The following examples use the freedom of g|f p pV q and
h|H to assign interesting dynamics to F .
Example 7.57 (Surgery of a polynomial to obtain a higher degree
polynomial) Let f be a polynomial of degree greater than 1, so that 8 is
a superattracting fixed point. Let V 1 “ t|z| ą Ru, where R is large enough so
that V :“ f ´1 pV 1 q Ą V 1 and V zt8u contains no critical point of f . Choose
any smooth disc H with H Ă V zf pV q and any holomorphic proper map
h : H Ñ Czf p pV q. Finally, choose a covering extension g on V zH so that
g is holomorphic and proper from f pV q onto gpf pV qq with 8 as a criti-
cal fixed point of local degree degpf q ` degphq; the preimage of the map
zdegpf q pz ´ 1qdegphq outside a small neighbourhood of 0 provides a model,
showing that such a covering extension exists. This provides a quasirational
map F with F ´1 p8q “ 8. So F is quasiconformally conjugate to a polyno-
mial of degree degpf q ` degphq.
Example 7.58 (Capturing a critical point) In the setting of Example 7.57,
assume that a critical point c of f satisfies f pcq P V zf pV q for some
ą 0. We take H to be a smooth disc containing f pcq and h : H Ñ Czf
p pV q

conformal such that hpf pcqq “ c. Therefore the critical point c, escaping to
8 under f , is ‘captured’ back and becomes periodic (hence superattracting)

under F .
Similarly, one can exploit the same idea to send a critical point c to, for
example, a point x P Jf whose orbit is dense in Jf , thereby obtaining a map F
whose postcritical set has complicated topology. Note, however, that in such an
example the altered critical point lies in the Julia set of F and is non-recurrent.
Example 7.59 (Case 1 ă p ă 8) We describe a surgery on the quadratic

polynomial f pzq “ z2 ´ 1 to obtain a cubic rational map with disconnected
Julia set, and with 0 a double critical point. Let V be a smooth disc con-
taining 0 whose closure is contained in the basin of attraction of 0 so that
the second iterate f 2 pV q is relatively compact in V , and f : V Ñ f pV q is a
degree two covering ramified at 0. Now take H to be a smooth disc whose
closure is contained in V zf 2 pV q. Define h : H Ñ Czf pV q to be conformal,
g : V zH Ñ f pV q to be a covering extension which is a holomorphic branched
covering of degree three in f 2 pV q, with 0 as a double critical point and ´1 as
the critical value. Now pasting together f |Cp zV , h and g gives a quasirational
map F . This map has a simple critical point at 8, a double critical point at 0
with orbit 0 ÞÑ ´1 ÞÑ 0, and another simple critical point in V zpH Y f 2 pV qq
whose orbit is attracted to the cycle t0, ´1u.
The map F is of degree three and the Julia set of F contains a fixed copy J0
of the Julia set Jf , a countable collection of homeomorphic preimages of J0 ,
and a Cantor set worth of point components.
7.6.3 Uniqueness of disc-annulus extensions

The construction of a disc-annulus extension of f depends on the
non-canonical choices of V , g, H and h. Clearly, the flexibility of choice in
the map g implies that one cannot expect the quasiconformal conjugacy class
on the whole sphere to be independent of such choices. With this in mind,
we say that two maps F, F 1 are J -conjugate if there is a quasiconformal
homeomorphism φ : C pÑC p which conjugates F on a neighbourhood of JF
to F on a neighbourhood of JF 1 . Note that this notion of J -conjugacy is
1
stronger than the one used elsewhere in the book.

In this section, we establish a prototype uniqueness theorem.
Theorem 7.60 (Uniqueness up to J -conjugacy) Let f be a rational map

and U a Fatou component of Ff . Then the J -conjugacy class of a univalent
disc-annulus extension F supported on V Ă pU zPf q depends only on U and
not on V , g, h, H .
This fails without the assumption of the univalence of h, or the fact that
H X Pf “ H (which is a consequence of the assumption V X Pf “ H).
Assume on the contrary that there is a point z P H which is either a point
of Pf or a critical point of h. Let h1 : H Ñ Czf
p pV q be another holomorphic
map satisfying h pzq ‰ hpzq and h pzq P Jf . Then V and H are identical but
1 1
the two resulting extensions F, F 1 are distinct on their Julia sets.
Analogy with Kleinian groups The univalent disc-annulus extensions con-

structed in Theorem 7.60 are analogous to adding a handle to a three-manifold
with boundary. Let M1 “ pH3 Y 1 q{1 be the three-manifold with bound-
ary associated to a Kleinian group 1 . Let D, D 1 be disjoint round discs in
BM1 , and consider the three-manifold M obtained by gluing D to D 1 via an
orientation-reversing homeomorphism (equivalently, join D to D 1 with a solid
tube). The resulting manifold admits a hyperbolic structure inherited from the
quotient of H3 by a new group “ 1 ˚Z which is an HNN-extension of .
The discs D, D 1 yield a compressing disc in M, i.e. its boundary is an essential
curve on BM. Indeed, if is a basin of attraction, then V descends to a closed
disc on the quotient punctured torus associated to U , which is like the boundary
of the quotient three-manifold.
The following lemma is the key step in the proof of Theorem 7.60.
Lemma 7.61 Let f be a rational map, U a Fatou component of Ff . Suppose
(see Figure 7.23):
(1) W Ă U is an open subset such that there is 1 ď p ď 8 with
(a) f : W Ñ f pW q proper,
(b) f j pW q X W “ H for 0 ă j ă p, and
in case p ă 8, f p pW q Ă W and f p : W Ñ f p pW q is proper;
(2) pV , H q and pV 1 , H 1 q are two pairs of smooth discs in W satisfying the
conditions of Theorem 7.50 and the additional condition 1
´ ¯ that H Y H is
contained in a disc or annulus Y Ă W z f p pW q Y Pf .
Then any two univalent extensions F “ F pV , g, h, H q, F 1 “ F 1 pV 1 , g 1 ,

h1 , H 1 q are J -conjugate.
Proof of Lemma 7.61 We will construct a combinatorial equivalence (in

the sense of McMullen [McM4], Appendix) between the holomorphic cover-
ings F : X1 Ñ X0 , F 1 : X11 Ñ X01 , where X0 , X01 are neighbourhoods of the
respective Julia sets and X1 and X11 are respective subsets of X0 and X01 .
Such an equivalence consists of a pair of quasiconformal homeomorphisms
φj : Xj Ñ Xj1 , j “ 0, 1 such that the diagram
Y Pf
H V
W
f p pW q f pW q
H1
V1
Figure 7.23 Setup of Lemma 7.61. Two univalent extensions with the modified
domains close enough to each other and avoiding the postcritical set, give rise to two
J -conjugate rational maps.
F
X1 ÝÝÝÝÑ X0
§ §
φ1 đ
§ §φ
đ 0
F1
X11 ÝÝÝÝÑ X01
commutes and such that φ0 extends to a quasiconformal homeomorphism from
X0 to X01 and φ0 is isotopic to φ1 rel. the ideal boundary of X0 .
From this, a standard pullback argument ([McM4], Thm. A.1) implies the
existence of a conjugacy φ : X0 Ñ X01 , yielding the result.
We now proceed to define the combinatorial equivalence. In case p ă 8,
let L1 “ f pW q Y ¨ ¨ ¨ Y f p pW q. In case p “ 8, choose L1 to be a union of
finitely many closed discs or annuli contained in Ff such that L1 X W “ H,
L1 Ą ną0 f n pW q and f pL1 q Ă L1 . Denote by P 1 the union of finitely many
Ť
disjoint smooth open discs containing Pf such that f pP 1 q Ă P 1 and BP 1 X

BL1 “ H.
Let X0 “ X01 “ CzpL
p 1 Y P 1 q; note that X Ą J , J 1 , because f pL1 q Ă
0 F F
L so f pL q X pH Y H 1 q “ H. Moreover, X0 has finitely many boundary
1 n 1
components.
Next we lift φ0 to obtain φ1 . We call this the lifting step.
Lifting step Define X1 “ F ´1 X0 , X11 “ pF 1 q´1 X0 and φ0 |X0 “ Id.

Observe that pL1 Y P 1 q Ă F ´1 pL1 Y P 1 q and therefore X1 Ă X0 . Also
X11 Ă X0 . Extend φ0 to a quasiconformal homeomorphism from Czf p pV q
to Czf
p pV 1 q (this is possible since both f pV q and f pV 1 q are compactly
contained in f pW q). Define φ1 “ Id on X0 zW , and on H define φ1 as
the lift of φ0 under h, h1 ; this is possible since h, h1 are univalent and
indeed we have φ1 |H “ ph1 q´1 ˝ φ0 ˝ h. Observe that X1 Ă pX0 zW q Y H ,
X11 Ă pX0 zW q Y H 1 and the following diagram commutes, with the top and
the bottom maps being holomorphic coverings:
F
X1 ÝÝÝÝÑ X0
§ §
φ1 đ
§ §Id
đ X0
F1
X11 ÝÝÝÝÑ X0
Finally we show that φ0 is isotopic to φ1 rel. the ideal boundary of X0 . We
call this the isotopy step.
Isotopy step There is a C 1 -extension of φ1 on X0 such that φ1 is isotopic to
φ0 rel. BX0 . To see this, extend φ1 to W such that φ1 “ Id on W zY and φ1 |Y
is isotopic to the identity rel. BY . Then φ1 is isotopic to Id rel. CzY
p Ą BX0 .
The pair φ0 , φ1 gives the desired combinatorial equivalence, and so we
obtain a quasiconformal mapping φ : X0 Ñ X0 such that the following dia-
gram commutes:
F
X1 ÝÝÝÝÑ X0
§ §
φđ
§ §φ
đ
F1
X11 ÝÝÝÝÑ X0
Extending φ arbitrarily to the whole sphere shows that F and F 1 are
J -conjugate.
Remark 7.62 In Theorem 7.60, the global quasiconformal conjugacy class
depends on the critical orbit relations of F , which are difficult to control since
the gluing map g necessarily introduces new critical values in f pV q Ă FF .
Also, in the non-univalent case, it may not be possible to carry out the lifting
step. Or the lifting step may be possible but the isotopy step impossible.
Proof of Theorem 7.60 Let V , V 1 Ă U zPf be any two smooth discs. Set
V0 “ V , V1 “ V 1 , and let Vt be a path of smooth discs joining V0 to V1 through
U zPf . Any two univalent extensions supported on sufficiently close Vs , Vt
are J -conjugate, by Lemma 7.61; the theorem follows by compactness of the
interval r0, 1s.
7.6.4 Inverse of disc-annulus extensions: simplification

The next result offers an inverse procedure.
Theorem 7.63 (Inverse disc-annulus surgery) Let F be a rational map. Let
A be a smooth annulus satisfying the following conditions:
• F maps A properly onto a disc;

• BA contains no critical points;
• there is 1 ď p ď 8 such that
F j pAq X A “ H for 0 ă j ă p,
and, in case p ă 8,
F p pAq Ă A and F p : A Ñ F p pAq is proper;
• there is a component H of CzA

p such that H X Ť j
1ăj ăp F pAq “ H.
Then there is a quasirational map f which coincides with F on CzpA

p Y Hq
and f pA Y H q “ F pAq.
Note that degpf q ă degpF q. In the analogy with three-manifolds, one may
think of this operation as a special case of cutting along a compressing disc.
Proof Define f to be F on CzpA
p Y H q and extend f so that f : A Y H Ñ
F pAq is a C proper map which is holomorphic in F p pAq. By assumption,
1
f n pBq X B “ H for B “ pA Y H qzF p pAq and n ą 0. So f is quasirational

by Lemma 7.49.
7.6.5 Surgery with two gluing regions

Here, we describe another, similar surgery construction. In the disc-annulus
surgery, the new dynamics h is holomorphic on a disc. Here, we allow for
the more flexible setting of adding the new dynamics h which is holomorphic
on an annulus. Since the annulus has two boundary components, the gluing
map g will be defined on two disjoint pieces. Controlling the recurrence of the
additional region where conformal distortion occurs requires our introduction
of an additional smooth disc G (see Figure 7.24).
V h V
H G F H G
f p pV q f p pV q
g
f pV q f pV q
Figure 7.24 The domains V , f p pV q, H and G in the setup of Theorem 7.64.

Theorem 7.64 (Surgery with two gluing regions) Let pf, V , H, G, h, gq

satisfy the following conditions:
• f is a rational map;
• V is a smooth disc contained in a Fatou component U such that:
– BV contains no critical points of f ;
– there is p ă 8 such that f j pV q X V “ H for 0 ă j ă p,
f p pV q Ă V , and f p : V Ñ f p pV q is proper;
• H is a smooth annulus such that H Ă V zf p pV q and Bf p pV q is a boundary

component of H ;
• G is a smooth disc contained iná Fatou component 1
¯ U such that U and
U 1 have disjoint orbit and mod CzpG
p Y f pV qq “ d 1 mod H for some
integer d 1 ą 0;
• h : H Ñ CzpG
p Y f pV qq a holomorphic covering of degree d 1 with
hpf pBV qq “ BG and g : V zpH Y f p pV qq Ñ f pV q, g : f p pV q Ñ G a
p
covering extension holomorphic in a neighbourhood of the critical points.
Then
$
&f on CzV
’
’ p
F :“ h on H
’
’
%g on V zH
is quasirational. Moreover, if H X Pf ‰ H and f is hyperbolic, the

J -conjugacy class of F depends only on U , U 1 , d 1 and the homotopy class
of BH relative to Pf (see Figure 7.24).
Proof Set B “ V zH . We have F j pBq X B “ H for j ą p. So by

Lemma 7.49 F is quasirational.
To prove the unicity, we first establish a local result. Assume
pf, V , H, G, h, gq and pf, V 1 , H 1 , G1 , h1 , g 1 q are two sets of choices of the
theorem such that:
´ ¯ ´ ¯
(1) mod CzG p Y f pV q { mod H “ mod CzG p 1 Y f pV 1 q { mod H 1 “
d 1 is an integer, and there is a set U 1 which is the union of finitely
many disjoint closed discs such that G Y G1 Ă intpU 1 q Ă U 1 Ă Ur and
1
f pU q Ă U ;1
(2) there is a smooth annulus Hp containing H Y H 1 such that f ´p pB Hp q X

Hp “ H.
Then the two maps F and F 1 are J -conjugate.

To prove this, set W :“ Hp Y f p pV q. It is a disc containing H Y H 1 Y

f p pV q Y f p pV 1 q. By assumption 2, f p pW q is contained in f p pV q X f p pV 1 q.
Let X0 “ CzpU
p 1 Y f pW q Y ¨ ¨ ¨ Y f p pW q Y P 1 q, where P 1 is the union of
finitely many disjoint smooth open discs containing Pf such that f pP 1 q Ă
P 1 and BP 1 X BpU 1 Y f pW q Y ¨ ¨ ¨ Y f p pW qq “ H. Note that X0 Ą JF , JF 1
and that X0 has finitely many boundary components.
Lifting step Define X1 “ F ´1 pX0 q, X11 “ F 1´1 pX0 q, and a C 1 -
diffeomorphism φ0 |X0 isotopic to the identity rel. BX0 , mapping Bf pV q
to Bf pV 1 q.
Now extend φ0 to a C 1 -diffeomorphism from CzpG p Y f pV qq to
CzpG
p 1 Y f pV 1 qq.
Define φ1 |X0 zW “ Id and H Ñ H 1 to be a lifting of φ0 : Czpf
p pV q Y Gq Ñ
Czpf
p 1 1
pV q Y G q. Extend φ1 to Hp so that it is isotopic to the identity rel. B Hp .
Then φ1 is isotopic to Id rel. BX0 . The only problem is that it is not exactly a
lift of φ0 (near f ´1 pBf pV qq.
The pair φ0 , φ1 gives the desired combinatorial equivalence and hence a
quasiconformal-conjugacy of F : X1 Ñ X0 to F 1 : X11 Ñ X0 . Therefore F
and F 1 are J -conjugate.
From this one can easily obtain the global unicity result.
Lemma 7.65 Let pf, V q be as in Theorem 7.64. If there is a Fatou component

U whose orbit under f is disjoint from V , then the disc G and the annulus
H exist.
Proof One can first choose either G or H first. Assume that H and hence
modH are given. Choose z0 P U . Since the modulus of the annulus between
z0 and Bf pV q is infinite, one can find a smooth disc G containing z0 such
that modpCzpG
p Y f pV qqq “ d 1 mod H for some (probably very large) integer
1
d ą 0. Of course, we can also start from G. Let G be any smooth disc
whose closure is contained in U . There is a minimal integer d 1 such that
modpCzpG
p Y f pV qqq ă d 1 mod H . Then one can find H in V zf p pV q such
that modpCzpG Y f pV qqq “ d 1 mod H .
p
In practice we may want d 1 to be as small as possible. We will show that in

case p “ 1, one can choose V , H, G so that d 1 “ 2 if degpf q ě 3 and d 1 “ 3
if degpf q “ 2.
Surgery on the basin of infinity of a polynomial Let f pzq “ z2 ` c and let

Kc “ Kf be its filled Julia set, which we assume to have non-empty interior
(for example c “ ´1). Fix a smooth disc G with closure in intpKc q. Denote
by ϕ the Böttcher coordinate of f in the basin of infinity. Let R be chosen

sufficiently large such that
´ ¯
log R ą 13 mod CzpG
p Y ϕ ´1 pt|z| ě R 2 uqq ;
this is possible since the right-hand side is comparable to 23 log R. Now take
V “ ϕ ´1 pt|z| ą Ruq. One can choose H as in Theorem 7.64 with d 1 “ 3.
See [PT2] for a computer-generated picture of the Julia set of such a map with
c “ ´1 and d 1 “ 3.
A similar surgery can be done for any degree d polynomial f with
intpKf q ‰ H, or with rational maps with multiple superattracting cycles.
Moreover, if d ě 3 one can choose V , G, H so that d 1 “ 2. Finally, note that if
d 1 ě 2 then JF may contain preimages of components of Jf under covering
maps of positive (indeed, arbitrarily large) degree.
We conclude with an easy consequence of Theorem 7.64.
Corollary 7.66 (Properties of the extensions) Let F be the quasirational
map given by Theorem 7.64. Then JF is not connected, Jf Ă JF , degpF q “
degpf q ` degphq “ degpf q ` d 1 and JF zJf contains uncountably many
wandering components which are not points.
7.6.1 It is shown in Figure 7.21 how one can construct a branched covering
of degree two of an annulus onto a disc such that each of the bound-
aries of the annulus is mapped onto the boundary of the disc by degree
one. Generalize this idea to construct a degree 2d branched covering,
mapping each of the boundaries of the annulus onto the boundary of the
disc by degree d.
Hint: Draw 2d segments in the annulus similar to the two drawn in
the figure. Map each connected component of the annulus minus the
segments bijectively onto the disc minus the two slits, and map each seg-
ment onto one of the two critical slits, alternately. Alternatively, change
coordinates so that the annulus separates 0 and 8 and precompose the
degree 2 map by zd .
7.6.2 Give a (topological) example of a degree three branched covering map
from an annulus onto a disc such that the outer boundary of the annulus
is mapped onto the boundary of the disc by degree two while the inner
boundary of the annulus is mapped by degree one.
Hint: Construct first a branched covering of degree two as in Fig-
ure 7.21. Afterwards, make a slit in the annulus from the outer boundary
to an interior point. Open up along the slit and glue in a copy of a slitted
Figure 7.25 Sketch of the construction in Exercise 7.6.2.
disc, as shown in Figure 7.25. The resulting space is still an annulus.

Explain how to map the open annulus 3 : 1 onto the open disc.
7.6.3 For any natural numbers d ` ą 1 and d ´ ą 1, generalize the construc-
tions in the exercises above to give an example of a branched covering
of degree d ` ` d ´ from an annulus onto a disc, such that the outer
respectively inner boundary of the annulus is mapped onto the boundary
of the disc by degree d ` respectively d ´ .
7.7 Perturbation and counting of non-repelling cycles

In the paper [Sh1], published in 1987, Shishikura used a cut and paste surgery
construction to prove the following celebrated theorem. The result originally
appeared in his Masters thesis (in Japanese) in 1985.
Theorem 7.67 (Counting non-repelling cycles) Suppose f P Ratd with

d ě 2. Let nnr pf q be the number of non-repelling cycles of f . Then
nnr pf q ď 2pd ´ 1q.
Shishikura proved that every non-repelling cycle of a rational map has a

critical point associated to it and that the critical point can not be shared by
any other cycle.
It is well known that every immediate basin of an attracting or parabolic
cycle must contain a critical point in its interior (see Theorem 3.39). This
proves that the number of attracting and parabolic cycles of a rational map
must be less than or equal to the number of critical points, i.e. 2pd ´ 1q. In
fact, Fatou conjectured that one could add Siegel cycles and Cremer cycles to
the equation. His idea was to perturb the rational map to convert the irrationally
indifferent cycles into attracting ones, without destroying the attracting nature
of the others. But he only succeeded in doing so for at least half of them at the
same time. Hence, the proof was not complete. Later on, Douady and Hubbard
[Do2] proved the statement for polynomials, using polynomial-like mappings
(see Exercise 7.7.1). Although their proof could be used for some rational
maps, it did not work in general. Shishikura proved the following theorem,
from which Theorem 7.67 follows as a corollary.
Theorem 7.68 (Perturbation Theorem) Let f P Ratd with d ě 2. Denote by

z0 , . . . , zN all non-repelling periodic points of f . For 0 ă ε ă ε0 , there exist
ε in C
fε P Ratd and points z0ε , . . . , zN p such that:
(a) if ε Ñ 0, then fε Ñ f uniformly and ziε Ñ zi in the spherical metric;

(b) if f pzi q “ zj then fε pziε q “ zjε , and each ziε is an attracting periodic point.
Let nattr pfε q denote the number of attracting cycles of fε . Since fε P Ratd
we have nattr pfε q ď 2pd ´ 1q. By construction we have nnr pf q ď nattr pfε q,
and Theorem 7.67 follows.
We dedicate this section to the proof of Theorem 7.68. We start by giving the
idea of the construction and afterwards fill in the details. We essentially follow
the steps presented by Shishikura in [Sh1, Sect. 4], and parts from Beardon in
[Ber, Sect. 9.6].
Without loss of generality we may assume that zi ‰ 8 for all i, and there-
fore the set of non-repelling cycles is contained in a compact set in C.
We start by choosing a polynomial h satisfying
hpzi q “ 0 and h1 pzi q “ ´1 for all i “ 0, . . . , N . (7.1)
Let k be the degree of h. For ε P C˚ define a new map gε (the model) as

follows:
! )
• gε pzq :“ f pz ` ε hpzqq in the disc Dε “ |z| ď |ε|´1{k ;
! )
• gε pzq :“ f pzq in the set |z| ě 2|ε|´1{k ;
• gε is a quasiregular! map (to be specified) ) in the annulus Aε :“
A|ε|´1{k ,2|ε|´1{k “ |ε|´ 1 {k ď |z| ď 2|ε|´ 1 {k , interpolating between the
two maps above.
By defining gε appropriately we shall prove that, for ε small enough, gε

is a quasiregular map with the same topological degree as f . In fact, we set
gε “ f ˝ Hε , where Hε : C p ÑC p is a quasiconformal homeomorphism, see
Lemma 7.69 (compare with Figure 7.26).
Observe that if ε is sufficiently small then Dε contains all the non-repelling
cycles of f . Since hpzi q “ 0, it turns out that if zi is p-periodic for f , then it
is also p-periodic for gε with multiplier
pgεp q1 pzi q “ p1 ´ εqp pf p q1 pzi q.

8
gε “ f .
Vε A gε “ f ˝ Hε .
ε
gε pzq “ f pz ` εhpzqq
gε z4 z5
z3
z0
z6
z1
Eε
Eε
Figure 7.26 Definition of the quasiregular model gε by pasting three different maps
together. On the annulus Aε , the map Hε is a quasiconformal auxiliary map so that
gε “ f ˝ Hε on Aε interpolates between its two boundary maps. The shaded area Vε
is mapped to the invariant set Eε .
If ε is real, positive and small, and the multipliers of the non-repelling cycles of
f have moduli less than or equal to one, it follows that z0 , . . . , zN are attracting
periodic points of gε .
Hence, the model gε satisfies the desired properties, namely having (at least)
as many attracting cycles as non-repelling cycles of f , and possessing the same
topological degree as f . However, the model is not holomorphic. Therefore,
to deduce the existence of one critical point per cycle, we need to find a
holomorphic map conjugate to gε . This is where the smoothing of the surgery
comes in.
Note that gε is holomorphic everywhere except in the annulus Aε , where
it is quasiregular. We shall show that, if ε is small enough, orbits of gε pass
through Aε at most once (Lemma 7.70).
Once this is proven, we may apply the Shishikura Principle, which says that
an invariant almost complex structure with bounded dilatation can be defined,
and hence a quasiconformal integrating map φε exists, so that fε :“ φε ˝ gε ˝
φε´1 is holomorphic, and hence a rational map of degree d. We finally show
that the new map fε preserves the attracting nature of the cycles ziε “ φε pzi q.
The proof is then complete.
We proceed to make the details of the proof more precise. We start by
showing that, for any arbitrary polynomial h of degree k ě 1 we can define a
family of quasiconformal maps Hε : C p ÑC p with certain properties, specified
in Lemma 7.69 below.
Lemma 7.69 (Definition and properties of special perturbations of the

identity) Let h be any polynomial of degree k ě 1, and let ρ be a C 8 -function
on R satisfying ρpRq “ r0, 1s, ρ “ 1 on p´8, 1s and ρ “ 0 on r2, 8q. Define
Hε : C
p ÑC p for ε P C by
´ ¯
Hε pzq “ z ` ε hpzq ρ |ε|1{k |z| , for z P C,
and Hε p8q :“ 8. Then, for ε small enough, Hε is a quasiconformal home-

omorphism. Moreover, Hε Ñ IdCp uniformly w.r.t. the spherical metric and
||μHε ||8 Ñ 0 when ε Ñ 0.
For any rational map f set gε :“ f ˝ Hε . It follows that gε is quasiregular
and of the same degree as f and, when ε Ñ 0, that gε Ñ f uniformly in the
spherical metric.
Proof Note that for ε P C˚ :

! )
• Hε pzq “ z ` ε hpzq in the disc Dε “ |z| ď |ε|´1{k ;
! )
• Hε pzq “ z in the set |z| ě 2|ε|´1{k ;
! )
• Hε is C 8 on the annulus Aε “ |ε|´1{k ď |z| ď 2|ε|´1{k , interpolating
between the two maps above.
It is therefore enough to prove uniform convergence on the set t|z| ď 2|ε|´1{k u

since Hε pzq “ z everywhere else.
Choose a constant M ą 1 so that |hpzq| ď M if |z| ď 1 and
|hpzq| ď M|z|k and |h1 pzq| ď M|z|k ´1 if |z| ě 1. (7.2)
Then, using these estimates and the expression for the chordal distance σ
between two points (3.1), and its relation to the spherical distance (3.3)
we obtain
2|ε||hpzq|ρ p|ε|1{k |z|q
dCp pHε pzq, zq ď σ pHε pzq, zq “ p1`|z|2 q1{2 p1`|H pzq|2 q1{2
ε
2|ε||hpzq| 2M |ε| maxt1,|z|k u (7.3)

ď p1`|z|2 q1{2 ď maxt1,|z|u
ď 2M|ε| maxt1, |z|k ´1 u ď 2k M|ε|1{k ,
which shows that dCp pHε pzq, zq Ñ 0 uniformly. Since any rational map satis-
fies a Lipschitz condition with respect to the chordal metric (see e.g. [Bea,
Thm. 2.3.1]), there exists L ą 1 such that
dCp pgε pzq, f pzqq ď σ pf pHε ppzqq, f pzqq ď Lσ pHε pzq, zq ď C|ε|1{k , (7.4)
for C “ 2k ML and |z| ď 2|ε|´1{k , which shows that gε Ñ f uniformly in the

spherical metric.
Observe that if |z| ď 2|ε|´1{k we have |ε||z|k ď 2k and |ε||z|k ´1 ď
2k ´1 |ε|1{k . Putting both observations together, and using (7.2), we obtain that
|εhpzq| ď 2k M and |εh1 pzq| ď 2k ´1 M|ε|1{k ,
for all z such that 1 ď |z| ď 2|ε|´1{k . In particular, this holds for all points in
the annulus Aε if ε is small enough.
With these estimates we proceed to show that |μHε | Ñ 0 as ε Ñ 0. It
suffices to consider the annulus Aε since Hε is holomorphic in the complement.
Differentiating at any point z P C with respect to z and z we obtain
z
Bz Hε “ εhpzq|ε|1{k ρ 1 p|z||ε|1{k q
2|z|
and
z
Bz Hε “ 1 ` εh1 pzqρp|z||ε|1{k q ` εhpzqρ 1 p|z||ε|1{k q|ε|1{k .
2|z|
Using the estimates above and taking into account that ρ and ρ 1 are bounded,
we conclude that for all z such that 1 ď |z| ď 2|ε|´1{k ,
|Bz Hε | “ Op|ε1{k |q Ñ 0 as ε Ñ 0,
and
|Bz Hε | “ 1 ` Op|ε1{k |q Ñ 1 as ε Ñ 0.
It follows that |μHε | Ñ 0 as ε Ñ 0 as we wanted to show.

We now prove that Hε is injective. We know from Sections 1.1 and 1.2 that
the Jacobian of Hε is
J “ |Bz Hε |2 ´ |Bz Hε |2 .
From the estimates above on the partial derivatives, we deduce that the
Jacobian is strictly positive for all z for which 1 ď |z| ď 2|ε|´1{k , and hence
Hε is a local homeomorphism in this domain. The same is true inside D.
Indeed, on this compact set Hε is holomorphic and |h1 pzq| is bounded, so for
ε small enough Hε1 pzq “ 1 ` εh1 pzq ‰ 0, and hence Hε has no critical points.
On the remaining part of the sphere Hε is the identity. Hence Hε is a C 8 -

covering map of the sphere. Since the sphere is simply connected, the identity
map is a universal covering and therefore any covering map must be of degree
one. It then follows that Hε is a global quasiconformal homeomorphism.
The previous lemma works for any polynomial, and for any ε P C˚ small
enough. From now on we restrict to ε real and positive and we require condi-
tions on the polynomial as in (7.1), namely that hpzi q “ 0 and h1 pzi q “ ´1 for
every i “ 0, . . . , N . Recall that such conditions ensure that the non-repelling
cycles for f turn into attracting cycles for gε , with the same period.
The next step is to show that orbits of gε pass through the annulus Aε at
most once. To that end, we construct an invariant set containing the image of
the annulus. Let Vε :“ t|z| ą ε´1{k u “ CzD p ε.
Lemma 7.70 (Orbits pass through Aε at most once) In the notation above,
by further adjusting the polynomial h and the point a :“ f p8q, orbits under
gε pass through Aε at most once.
More precisely, for ε P p0, ε0 s, there exist open sets Eε uniformly bounded
in C, such that Eε is invariant under gε and satisfies:
• Eε X Vε “ ∅;
• gε pVε q Ă Eε .
Compare with Figure 7.26. The proof splits into three cases, depending on
the type of non-repelling cycles that f possesses – attracting, parabolic or
irrationally indifferent.
Case 1: f has an attracting cycle This is the simplest case. Suppose

z0 , . . . , zp´1 is an attracting cycle of f . Choose E0 an open disc around
z0 such that f p pE 0 q Ă E0 . Let Ei “ f i pE0 q for i “ 1, ¨ ¨ ¨ , p ´ 1, and set
E :“ i Ei . We may assume E Ă Dε . Observe that f p pEq Ă E. This is a
Ť
stable condition under small perturbations. The condition hpzi q “ 0, makes

the perturbation arbitrarily small around the periodic cycle, so one can check
that for ε small enough, these discs still map compactly inside themselves.
Hence define Eε :“ E and add no extra conditions on h (see Figure 7.27)
In this case we could require from the start, applying a coordinate transfor-
mation if necessary, that f p8q P E and 8 P f ´1 pEqzE. By making ε small
enough, we can assure that Vε is mapped compactly inside E by f , and since
E does not depend on ε, this is also true for gε . Then the lemma follows.
Case 2: f has a parabolic cycle If f has no attracting cycle but there exists
a parabolic cycle then we need to work harder to construct Eε . Suppose
f gε
z0 Eε “ E
Figure 7.27 Definition of Eε :“ E, in the special case of an attracting fixed point. For
ε small enough, this open set is mapped inside itself by gε .
z0 , . . . , zp´1 is a parabolic cycle. Let E0 be an attracting petal of f at z0 (see

Definition 3.32). Let Ei “ f i pE0 q for i “ 1, . . . , p ´ 1, and set E :“ i Ei .
Ť
Observe that f p pEq Ă E Y Opz0 q. Away from z0 , this is again a situation

that is stable under small perturbations of the map. Since the points zi for
i “ 0, . . . , p ´ 1, are attracting periodic points for gε , it follows that there
exists a small neighbourhood of z0 , say Nε,0 , which maps strictly inside itself
p
under gε . Hence we define,
Eε,0 :“ Eo Y Nε,0 .
p
Observe that, for ε small enough, we have gε pE ε,0 q Ă Eε,0 as required (see
Figure 7.28). We define Eε,i :“ gε pEε,i ´1 q for i “ 1, . . . , p ´ 1, and Eε :“
p ´1
Yi “0 Eε,i . We choose Eε,0 so that Eε Ă Dε .
f f
E
gε
Eε
z0 Nε z0
gε
Figure 7.28 Definition of Eε , in the case of a parabolic fixed point, as the union of an
attracting petal E0 for f and a neighbourhood Nε around the fixed point which maps
compactly inside itself under gε . Observe that Eε does not shrink to a point when
ε Ñ 0 since it contains the set E0 for all ε.
In this case, we could as before require from the start, that f p8q P E and
8 P f ´1 pEqzE. In this way we can ensure that for ε small enough Eε is
uniformly bounded in C, disjoint from Vε , and that gε maps Vε compactly
inside Eε , since E does not shrink when ε Ñ 0.
Case 3: f has an irrationally indifferent cycle If f has no attracting or

parabolic cycles but only irrationally indifferent ones, then we need to work
even harder to construct Eε . Suppose z0 , . . . , zp´1 is an irrationally indifferent
cycle, and adjust f from the start so that f p8q “ z0 and 8 ‰ zp´1 . By the
theory of normal forms (see e.g. [Ar2] or [Bea, Thm. 6.10.5]), there exists a
holomorphic local diffeomorphism ψ around 0 such that ψp0q “ z0 and
pψ ´1 ˝ f p ˝ ψqpuq “ λu ` Opuk `3 q,
where λ is the multiplier of the cycle, and k is the degree of h. A direct

calculation of the power series expansion (see [Bea, Lem. 9.6.4]) shows that,
close to u “ 0,
´ ¯
pψ ´1 ˝ gεp ˝ ψqpuq “ pψ ´1 ˝ f p ˝ ψqpuq ´ λu ` λup1 ´ εqp ` Opεu2 q.
(7.5)
Putting together the two expressions we obtain

´ ¯
pψ ´1 ˝ gεp ˝ ψqpuq “ λu p1 ´ εqp ` Opεuq ` Opuk `2 q .
Set ε :“ t|u| ă ε1{pk `1q u. For ε sufficiently small, ψ is defined on ε , and

p ´1
we may define Eε1 “ ψpε q. Set also Eε :“ Eε1 Y gε pEε1 q Y ¨ ¨ ¨ Y gε pEε1 q.
p 1 1
We want to show that if ε is small enough, then gε pE ε q Ă Eε , and hence Eε
is invariant under gε . Equivalently, we need to check that
pψ ´1 ˝ gεp ˝ ψqpε q Ă ε .
1 and using that p1 ´ εq “ 1 ´ εp ` Opε q, we

1 p 2
From (7.5), setting t “ k `
obtain, for u P ε ,
´ ¯
pψ ´1 ˝ gεp ˝ ψqpuq “ λu 1 ´ pε ` Opε1`t q .
It follows that
ˇ ˇ
ˇ ´1
ˇpψ ˝ gεp ˝ ψqpuqˇ ă |λu| “ |u|
ˇ
p
and therefore ε is mapped into itself under ψ ´1 ˝ gε ˝ ψ. It follows that
p
gε pE 1 ε q Ă Eε1 as required.
Compared to the other cases, the problems start now, since the set Eε1 shrinks
to a point when ε Ñ 0. Even if f p8q “ z0 , we cannot be sure that gε pVε q Ă
Eε1 as we want. To guarantee that it happens, we need to estimate the size of
the two sets Eε1 and gε pVε q in terms of ε.
The size of Eε1 is easily estimated using that ψp0q “ z0 and ψ 1 p0q “ 1. It
follows that ψpuq “ z0 ` Opuq for small |u| and, since Dε has radius ε1{pk `1q ,
we have |z ´ z0 | “ Opε1{pk `1q q for z P Eε1 .
Estimating the size of gε pVε q is somewhat more involved. First observe that
Vε is a neighbourhood of 8 whose chordal radius is
2
σ p8, ε´1{k q “ ´ ¯1{2 “ Opε q
1{k
1 ` ε´2{k
(see (3.2)). Since any rational map satisfies a Lipschitz condition with respect
to the chordal metric (see e.g. [Bea, Thm. 2.3.1]), we have that f pVε q lies in
a disc centred at z0 of chordal size Opε1{k q. Using (7.4) we obtain that gε pVε q
lies in a disc around z0 of chordal size Opε1{k q. But in a compact set the chordal
metric and the Euclidian one are comparable. Hence, the set gε pVε q lies in a
disc with centre z0 and Euclidian radius Opε1{k q. Since ε1{k ă ε1{pk `1q , it
follows that for sufficiently small ε, the inclusion gε pVε q Ă Eε1 is satisfied,
concluding the proof of Case 3, without adding extra conditions to h.
This concludes the proof of Lemma 7.70.
We now summarize and conclude the proof of the Perturbation Theorem
(Theorem 7.68). Suppose z0 , . . . , zN are all the non-repelling periodic points
of f . Choose one specific cycle, say z0 , . . . , zp´1 , and proceed as in one of the
described cases, depending on the type of cycle. This defines the polynomial
h, fixes the image of infinity, and defines a set Eε , satisfying the properties in
Lemma 7.70.
At this point everything is arranged so that the annulus Aε , the only region
where gε is not holomorphic, is mapped into the invariant set Eε , disjoint
from Aε . Choose the standard complex structure on Eε and spread it by
the dynamics. Because of the Shishikura Principle this is well defined. The
dilatation is bounded, and the structure is gε -invariant. Hence, there exists an
integrating map φε , and the map fε :“ φε ˝ gε ˝ φε´1 is a rational map of the
same degree as f . We observe the following:
• The remaining periodic points zp , . . . , zN are attracting under gε . This

means that there exists a collection of neighbourhoods of these points, say
Eε1 , disjoint from Eε and Vε , which are also invariant under gε . This ensures
that φε is conformal not only on Eε but also on Eε1 . Hence, the new periodic
points tziε “ φε pzi quN
i “0 are attracting periodic points of fε for every ε.
• By the Integrability Theorem depending on parameters (Theorem 1.30), the
integrating maps φε appropriately normalized satisfy φε Ñ IdCp and fε Ñ
f uniformly when ε Ñ 0.
7.8 Mating a group with a polynomial 291

7.7.1 Use polynomial-like mappings to prove that a polynomial P of degree
d has at most d ´ 1 non-repelling cycles in C. (A proof can be found in
[Do2].)
Hint: Suppose N R is the set of non-repelling cycles of P . Consider
a perturbation of the form fε “ P ` εQ, choosing ε small enough and
an appropriate polynomial Q, so that fε is polynomial-like of degree d
and the cycles in N R are attracting cycles of fε .
7.7.2 (Existence of polynomials with certain conditions) Suppose v1 , . . . ,
vm are distinct points in C. Assume that for some of these points, say the
first n, with n ď m, there exist neighbourhoods B1 , . . . , Bn which are
closed pairwise disjoint topological discs, and holomorphic functions
hj : Bj Ñ C, satisfying hj pvj q “ 0 and h1j pvj q “ ´1. Prove that, for
any δ ą 0, there exists a polynomial hpzq such that
hpvi q “ 0 h1 pvi q “ ´1 and |h ´ hj | ă δ on Bj ,
for i “ 1, . . . , m and j “ 1, . . . , n.
h ´p
Hint: Apply Runge’s Theorem to the map j p2 1 , where p1 is a
polynomial satisfying p1 pvi q “ 0 and p11 pvi q “ ´1, and let p2 pzq “
śm 2
i “1 pz ´ vi q . Check that h “ p1 ` p2 q satisfies the requirements.
7.8 Mating a group with a polynomial

Shaun Bullett
The projective special linear group PSLp2, Cq is the group of 2 ˆ 2 matrices
with complex entries and determinant 1, under the equivalence relation
ˆ ˙ ˆ ˙
ab á ´b
„ .
cd ć ´d
The elements of PSLp2, Cq are the conformal automorphisms
az ` b
z ÞÑ
cz ` d
of the Riemann sphere C. p
Recall from Section 3.5 that a Kleinian group is a discrete subgroup of
PSLp2, Cq; a point z P C p is ordinary if the action of on some neighbourhood
of z forms a normal family, otherwise z is a limit point. The ordinary set is
denoted pq, while the limit set is denoted pq. Both sets are invariant
under the action of . The sets are the analogues of the Fatou set and the Julia
set respectively of a rational map
7.8.1 The modular group, rational maps and

holomorphic correspondences
The modular group PSLp2, Zq, which is defined in just the same way as
PSLp2, Cq, except that the matrix entries are required to be integers, is an
example of a Kleinian group. The modular group is finitely generated: as an
abstract group it is the free product of C2 ˚ C3 of a cyclic group C2 of order
two, and a cyclic group C3 of order three, generated by
ˆ ˙ ˆ ˙
0 1 ´1 ´1
σ “ and ρ “
´1 0 1 0
respectively (see Figure 7.29). Equivalently, we may take as generators for

PSLp2, Zq the elements τ1 “ σρ ´1 and τ2 “ σρ, that is to say
ˆ ˙ ˆ ˙
11 10
τ1 “ and τ2 “ .
01 11
D
P
Q
−2 −1 0 1 2
Figure 7.29 The action of the modular group PSLp2, Zq on the upper half plane. P
and Q are the fixed points of σ and ρ respectively. The region D is a fundamental
domain. It is also a fundamental domain for the action of PGLp2, Zq “ă σ, ρ, χ ą on
p where χ is the involution χ pzq “ 1{z. The region D Y ρpDq Y ρ ´1 pDq is
C,
outlined in bold.
Regarded as conformal automorphisms of C

p (Möbius transformations), τ1 and
τ2 are the maps
z
τ1 pzq “ z ` 1 and τ2 pzq “ .
z`1
To define the notion of a mating between a (finitely generated) Kleinian
group and a polynomial map, we shall need to work in a class of holomorphic
dynamical systems on the Riemann sphere that is sufficiently general to include
finitely generated groups and polynomial maps as special cases.
Definition 7.71 (Holomorphic correspondence) A holomorphic correspon-

dence on the Riemann sphere C p is a multi-valued map z Ñ w defined by a
relation of the form P pz, wq “ 0, where P pz, wq is a polynomial in z and
w (with coefficients in C) with the property that every non-constant factor
involves both the variables z and w.
Every rational map z Ñ ppzq{qpzq of degree n ě 1 can be regarded as

an n-to-1 correspondence in the obvious way, namely take P pz, wq “ 0 to
be the relation wqpzq ´ ppzq “ 0. In particular, a polynomial map z Ñ ppzq
becomes a correspondence w ´ ppzq “ 0. Moreover, given any finitely gener-
ated Kleinian group , together with a specific choice of generators
aj z ` bj
γj pzq “ , 1 ď j ď n,
cj z ` dj
we may regard as the holomorphic correspondence G : z Ñ w defined by

the polynomial relation P pz, wq “ 0, where P pz, wq is the polynomial
n
ź
pwpcj z ` dj q ´ paj z ` bj qq .
j “1
Observe that the orbits of the action of the group on Cp are the grand orbits
of the correspondence G, that is to say the orbits under mixed forwards and
backwards iteration, z Ñ w and w Ñ z.
For example, since the matrices τ1 and τ2 (defined above) generate the
modular group PSLp2, Zq, the orbits of this group on the Riemann sphere C p
are exactly the same as the grand orbits of the correspondence
pw ´ pz ` 1qqpwpz ` 1q ´ zq “ 0.
An alternative (but equivalent) definition of the notion of a holomorphic

´1
correspondence is that it is a multi-valued map z Ñ w of the form π` ˝ π´ ,
where π` and π´ are holomorphic surjections (branched coverings) from a
compact Riemann surface S onto C. p It is clear that given a polynomial P pz, wq
satisfying the condition in our first definition above, we may satisfy the con-
ditions of this second definition by taking as the Riemann surface S the graph
tpz, wq | P pz, wq “ 0u Ă C p ˆ C,
p and as π´ and π` the coordinate projections
of this graph. The converse is considerably deeper: given a Riemann surface
S and projections π` and π´ onto C, p it can be shown that the graph of the
´1
relation π` ˝ π´ is defined by a polynomial equation P pz, wq “ 0 (see [BP2,
Thm. 2], and the references cited there).
7.8.2 Matings between the modular group and quadratic polynomials

Let Kc denote the filled Julia set of a quadratic polynomial Qc pzq “ z2 ` c.
Recall that if Kc is connected, that is to say if c lies in the Mandelbrot set,
then there is a uniquely defined conformal homeomorphism, the inverse of the
Böttcher coordinate
ψc : CzD
p Ñ CzK
p c,
which is tangent to the identity at infinity and conjugates the map z Ñ z2 on

the complement of the closed unit disc D to the map Qc on the complement of
Kc . When Kc is locally connected, ψc extends to a continuous surjection from
the unit circle S1 onto the boundary Jc of Kc (by the Carathéodory Theorem
(Theorem 2.8), semi-conjugating the argument-doubling map z Ñ z2 on S1 to
the action of Qc on Jc .
Given a quadratic polynomial Qc which has a connected and locally con-
nected filled Julia set Kc , let K´ _ K` denote the union of two copies K´
and K` of Kc , glued together at the boundary point of external argument 0.
Let Hc on K´ _ K` denote the p2 : 2q correspondence defined by sending:
• z P K´ to Qc pzq P K´ and to ´z P K` ;
c pzq P K` .
• z P K` to the pair of points Q´ 1
For example, in the case c “ 0 the filled Julia set K0 is the closed unit disc
D, and H0 on D´ _ D` is the p2 : 2q correspondence defined by sending:
• z P D´ to z2 P D´ and to ´z P D` ;
?
• z P D` to the pair of points ˘ z P D` .
The boundary of K´ _ K` is a wedge J´ _ J` of two copies of the Julia

set Jc , glued together at the boundary point of external argument 0. The p2 : 2q
correspondence Hc on K´ _ K` restricts in the obvious way to J´ _ J` ,
which in the case c “ 0 is a wedge of two copies of the unit circle. Whenever
Kc is connected and locally connected, Jc is a quotient of the unit circle and the
action of the correspondence Hc on J´ _ J` is the corresponding quotient of
the action of H0 on the wedge of two circles.
Given a quadratic polynomial Qc , a mating between Qc and the modu-
lar group PSLp2, Zq will be defined to be a p2 : 2q correspondence on C p
obtained by gluing together the p2 : 2q correspondence Hc on pKc q´ _ pKc q`
constructed above, and the p2 : 2q correspondence G on the complex upper
half-plane defined by the generators τ1 and τ2 of PSLp2, Zq. Before we make
this idea precise, we check that the action of the pair of homeomorphisms
pτ1 , τ2 q on the boundary Rp “ R Y t8u of the complex upper half plane

matches the action of Hc on pJc q´ _ pJc q` .
Let h` denote the homeomorphism from r0, 8s to r0, 1s which sends
x P r0, 8s, represented by the continued fraction rx0 ; x1 , x2 , . . .s, to t P r0, 1s,
the binary number represented by the binary point followed by x0 copies of 0,
then by x1 copies of 1, then by x2 copies of 0 and so on. This homeomorphism
is sometimes referred to in the literature as the Minkowski question mark
map. We shall also need a companion homeomorphism h´ : r´8, 0s Ñ r0, 1s
defined by h´ pxq “ 1 ´ h` p´xq. Note that h` and h´ are both orientation-
reversing.
Consider τ1 and τ2 restricted to the completed real axis r´8, 8s. These are
defined by:
x
τ1 : x Ñ x ` 1 and τ2 : x Ñ .
x`1
Note that on the positive real axis r0, 8s
τ1 : rx0 ; x1 , x2 , . . .s ÞÑ rx0 ` 1; x1 , x2 , . . .s,

τ2 : rx0 ; x1 , x2 , . . .s ÞÑ r0; 1, x0 , x1 , x2 , . . .s.
Let θ1 : r0, 1s Ñ r0, 1{2s and θ2 : r0, 1s Ñ r1{2, 1s denote the halving maps
θ1 : t Ñ t{2 and θ2 : t Ñ t{2 ` 1{2 respectively. It is a straightforward exer-
cise to check that the following diagrams commute, and also that the hor-
izontal maps h` are homeomorphisms between the intervals indicated:
h` h`
r0, 8s ÝÝÝÝÑ r0, 1s r0, 8s ÝÝÝÝÑ r0, 1s
§ § § §
τ1 đ τ2 đ
§ §θ § §θ
đ1 đ2
h` h`
r1, 8s ÝÝÝÝÑ r0, 1{2s r0, 1s ÝÝÝÝÑ r1{2, 1s
Another straightforward computation shows that the following diagrams

also commute. Here the horizontal maps h´ are again homeomorphisms,
and the maps θ1´1 and θ2´1 are the doubling maps t ÞÑ 2t and t ÞÑ 2t ´ 1
respectively:
h´ h´
r´8, ´1s ÝÝÝÝÑ r1{2, 1s r´1, 0s ÝÝÝÝÑ r0, 1{2s
§ § § §
τ1 đ
§ § ´1 τ2 đ
§ § ´1
đθ2 đθ1
h´ h´
r´8, 0s ÝÝÝÝÑ r0, 1s r´8, 0s ÝÝÝÝÑ r0, 1s
The remaining restrictions of τ1 and τ2 , namely τ1 : r´1, 0s Ñ r0, 1s and

τ2 : r´8, ´1s Ñ r1, 8s, are carried by the pair h´ , h` to the maps t ÞÑ
t ` 1{2, from r0, 1{2s to r1{2, 1s, and t ÞÑ t ´ 1{2, from r1{2, 1s to r0, 1{2s,
respectively:
h´ h´
r´1, 0s ÝÝÝÝÑ r0, 1{2s r´8, ´1s ÝÝÝÝÑ r1{2, 1s
§ § § §
τ1 đ τ2 đ
§ §
Þ t `1{2
§ §
đt Ñ Þ t ´1{2
đt Ñ
h` h`
r0, 1s ÝÝÝÝÑ r1{2, 1s r1, 8s ÝÝÝÝÑ r0, 1{2s
Identifying 0 with 8, the homeomorphism h` induces an orientation-

reversing homeomorphism h` : r0, 8s{„ Ñ T` , where T` denotes the circle
obtained from r0, 1s by identifying its end points. Similarly h´ induces an
orientation-reversing homeomrphism h´ : r´8, 0s{„ Ñ T´ , where T´ is a
second copy of the circle obtained from r0, 1s by identifying its end points.
Together h` and h´ induce a homeomorphism
h` _ h´ : R{
p 0„8 Ñ T` _ T´ ,
which conjugates the action of τ1 , τ2 on R{

p 0„8 to that of the pair of argument-
halving maps t Ñ t{2, t Ñ pt ` 1q{2 on the circle T` , the argument-doubling
map t Ñ 2t on the circle T´ , and the antipodal homeomorphism t Ñ t ` 1{2
from T´ to T` .
If we glue together the points 0 and 8 on the boundary of the upper half
plane, then on this boundary R{p 0„8 the action of τ1 and τ2 matches (combi-
natorially, at least) the boundary action of the p2 : 2q correspondence Hc on
the wedge K´ _ K` , constructed earlier in this section. Thus we may glue
K´ _ K` to the upper half plane and obtain the action of a p2 : 2q topological
correspondence on a topological sphere. The question now is whether this
topological mating can be realized by a holomorphic correspondence on the
Riemann sphere C. p
Definition 7.72 (Mating of a quadratic polynomial and the modular group)

Let Qc be a quadratic polynomial with connected filled Julia set Kc , i.e. with c
in the Mandelbrot set M. A holomorphic correspondence F : z Ñ w, defined
by a polynomial relation P pz, wq “ 0 of bidegree p2 : 2q, is called a mating
between Qc and the modular group PSLp2, Zq if:
(a) there exists a completely invariant open simply connected region Ă C p
and a conformal bijection h from to the upper halfplane conjugating the
two branches of F| to the pair of generators τ1 : z ÞÑ z ` 1, τ2 : z ÞÑ
z{pz ` 1q of PSLp2, Zq;
(b) the complement of is the union of two closed sets ´ and ` , which
intersect in a single point and are equipped with homeomorphisms h˘ :
˘ Ñ Kc , conformal on interiors, respectively conjugating F, restricted
to ´ as domain and codomain, to Qc on Kc , and conjugating F,
c on Kc .
restricted to ` as domain and codomain, to Q´ 1
Figure 7.30 illustrates an example of a mating of a quadratic polynomial

and the modular group.
Figure 7.30 A mating between the modular group and a quadratic polynomial with an
attracting cycle of period 2. The plot is of a correspondence in the family (7.6) referred
to in the text, with a “ 4.4. The shaded regions ´ and ` , meeting at the origin,
are copies of the filled Julia set of the polynomial. Also shown is a tiling of the
complementary region by copies of a fundamental domain for PSLp2, Zq.
In this definition we have not specified the identification between the bound-
aries of the regions and ´ Y ` . This is to allow us maximum flexibility
in dealing with the values of c for which the boundaries of the regions ˘
are not locally connected. However, there are two boundary identification
conditions which it can sometimes be useful to require:
(I) The image of the positive imaginary axis under the conformal bijection
h (of condition (a) of the definition) is an arc in C
p both ends of which
converge to the wedge point ´ X ` .
(II) The involution σ : z Ñ ´1{z on and the involution on K´ _ K` which
exchanges z P K´ with z P K` , fit together to yield a conformal involu-
tion (a Möbius transformation) j : C
p Ñ C.p
Notice that if we have a mating which satisfies the boundary condition

(I) then by continuity of the action of the correspondence, the homeomorphism
h from the complex upper half plane to , extends continuously to all ratio-
nal points of the boundary Rp and maps these to a dense set of points in
Bp´ _ ` q. In particular, if Kc is connected and locally connected, then h
extends to a continuous surjection Rp Ñ Bp´ _ ` q realizing the boundary-
matching surjection h´ _ h` : R{ p 0„8 Ñ J´ _ J` constructed earlier in
this section.
The convenience of condition (II) is that every p2 : 2q correspondence
F : z Ñ w which realizes a mating satisfying this condition has an equation
that can be expressed in a canonical form. Condition (II), in conjunction with
conditions (a) and (b) in the definition of a mating, guarantees that the compo-
sition j ˝ F together with the identity map on C p defines a p3 : 3q equivalence
relation on C, and hence that j ˝ F has equation
p
Rpwq ´ Rpzq
“0
w´z
for some degree three rational map R. Conditions (a) and (b) also ensure that
R has one double and two single critical points, and hence that up to pre- and
post-multiplication by Möbius transformations, R is the map Rpzq “ z3 ´ 3z.
Thus up to conjugation the correspondence F has equation of the form
w 2 ` wj pzq ` pj pzqq2 “ 3,
where j is an involution which has as one of its fixed points the wedge
point ´ X ` . Normalizing our coordinate system so that this wedge point
becomes the origin, and j becomes the map z ÞÑ ´z, we deduce that our p2 : 2q
correspondence F lies in the one (complex) parameter family:
aw ´ 1 2 az ` 1 2
ˆ ˙ ˆ ˙ˆ ˙ ˆ ˙
aw ´ 1 az ` 1
` ` “ 3. (7.6)
w´1 w´1 z`1 z`1
Bullett and Penrose formulated the following conjecture [BP1].
Conjecture 7.73 The family (7.6) of p2 : 2q correspondences contains a mat-

ing between PSLp2, Zq and every quadratic polynomial having a connected
Julia set, that is to say every z Ñ z2 ` c with c P M, the Mandelbrot set.
Supporting evidence for Conjecture 7.73 was provided by numerical experi-

ments suggesting a resemblance between the space of matings and the Mandel-
brot set [BP2]. To show that all these matings really do exist, it is natural to try
to use some form of surgery, modifying the complex structures on the pieces
in such a way that they fit together to give an overall complex structure on the
topological mating. However, there are technical problems fitting together the
complex structures at the ‘wedge point’ ` X ´ and at points on the grand
orbit of this point. To circumvent these difficulties we change the question to

that of constructing a mating between a generic faithful discrete representa-
tion of C2 ˚ C3 and a quadratic polynomial Qc . Using surgery we prove the
existence of a mating between every generic discrete faithful representation
of C2 ˚ C3 and every Qc with c P M. One can then deduce the existence
of a mating between PSLp2, Zq and Qc for a large class of values of c,
including all those values for which Qc is hyperbolic, by applying pinching
techniques [BHai, BHar] developed by Bullett, Haı̈ssinsky and Harvey. We
shall explain the surgery construction here, and then summarize the results
concerning matings between PSLp2, Zq and Qc obtainable by pinching.
7.8.3 Using surgery to construct a mating between z Ñ z2 ` c and a

generic faithful discrete representation of C2 ˚ C3 in PSLp2, Cq
For simplicity of exposition, we concentrate our attention on representations of
C2 ˚ C3 in PSLp2, Rq. Recall our generators σ (of order two) and ρ (of order
? of PSLp2, Zq. The fixed points of σ are ˘i and those of ρ are p´1 ˘
three)
i 3q{2. Keeping σ : z Ñ ´1{z unchanged, we can perturb ρ, keeping it of
order three, by moving its two fixed points to another pair of conjugate points
cos θ ˘ i sin θ on the unit circle. For ´1 ă cos θ ă ´1{2 the group generated
by σ and ρ remains discrete and a free product of C2 with C3 . The action
of such a perturbed representation is illustrated in Figure 7.31. The difference
from the action of PSLp2, Zq (Figure 7.29) is that the parabolic fixed points
of PSLp2, Zq (at every point of the rationals Q Ă R) have been opened up to
become intervals, with hyperbolic fixed points at either end. The limit set of
the perturbed group is a Cantor set on the real axis, with these intervals as the
gaps. As in the unperturbed case, the involution χ : z ÞÑ 1{z commutes with
P
Q
R –1S U T 0 +1
Figure 7.31 The region D Y ρpD q Y ρ ´1 pD q for a perturbed representation of

“ă σ, ρ, χ ą.
σ and anti-commutes with ρ, and a fundamental domain for the action of the
group ă σ, ρ ą on the upper halfplane is a fundamental domain D for the
action of the group “ă σ, ρ, χ ą on C.
p
More generally, one can perturb PSLp2, Zq as a representation of C2 ˚ C3
in PSLp2, Cq and not just in PSLp2, Rq. There is a complex one-dimensional
moduli space of such perturbations, parametrized by the cross-ratio of the
fixed points of σ with the fixed points of ρ. Provided that the perturbed
representation remains faithful and discrete, generically it will have connected
ordinary set, and the limit set will be a Cantor set. In fact it will be topologically
(indeed quasiconformally) conjugate to the real examples described above.
However, at each of the boundary points of the moduli space the limit Cantor
set is pinched together and the ordinary set can become disconnected. The
simplest such pinching is that at PSLp2, Zq, where the Cantor set joins up to
become the whole of the completed real axis Rp and the ordinary set becomes
two open discs (the upper and lower half planes).
Every discrete faithful representation of C2 ˚ C3 in PSLp2, Cq, not just those
in PSLp2, Rq, is equipped with an involution χ which commutes with σ and
anti-commutes with ρ. In the Poincaré disc model of hyperbolic 3-space χ
is the involution defined by rotation through angle π around the common
orthogonal to the axis of ρ and the axis of σ .
Recall Definition 3.52 of the term hybrid equivalence and Definition 7.1 of
a polynomial-like map, which in degree 2 is called a quadratic-like map.
Definition 7.74 (Mating of a quadratic polynomial and certain discrete

subgroups of PSLp2, Cq) A p2 : 2q holomorphic correspondence f : z Ñ w
is called a mating between a faithful discrete representation r of C2 ˚ C3 in
PSLp2, Cq having connected ordinary set pq, and a polynomial Qc having
connected filled Julia set Kc , if the Riemann sphere C
p is the disjoint union of a
connected open set pf q and a closed set pf q made up of two components,
` pf q and ´ pf q, such that each of pf q and pf q is completely invariant
under f and:
(a) the action of f on pf q is discontinuous and there is a conformal bijection

between the grand orbit spaces pf q{f and pq{ (where is the group
generated by σ, ρ and χ );
(b) there is a neighbourhood U of ´ pf q such that f restricted to V “
f ´1 pU q X U is a quadratic-like map V Ñ U and there is a quasicon-
formal homeomorphism from U onto a neighbourhood U 1 of Kc in C
which is a hybrid equivalence conjugating the quadratic-like map f to the
quadratic-like map Qc ; similarly there is a hybrid equivalence between
quadratic-like maps f ´1 on a neighbourhood of ` pf q and Qc on a

neighbourhood U 1 of Kc .
Notice that the fact that ` and ´ have disjoint neighbourhoods has
allowed us to be more demanding in our definition of a mating than in the
earlier case of PSLp2, Zq: we ask for hybrid equivalences between f ˘1 on
neighbourhoods of ˘ and Qc on neighbourhoods of Kc , whereas earlier we
only asked for conjugacies which are conformal on interiors between f ˘1 on
˘ and Qc on Kc . Examples of correspondences satisfying Definition 7.74
are illustrated in Figures 7.32 and 7.33.
Figure 7.32 A mating between a generic faithful discrete representation of C2 ˚ C3 in

PSLp2, Rq and a quadratic polynomial with an attracting cycle of period 2. This
example belongs to the family (7.7) (referred to later in the text), and is given by the
parameter values a “ 3.9, k “ 0.9. Pinching to a point the curve illustrated linking
´ to ` , and pinching its images (‘hairs’ on ` and ´ ) to points, transforms this
example into a mating between the modular group and the same quadratic polynomial.
Figure 7.33 Another example of a mating between a generic faithful discrete

representation of C2 ˚ C3 , this time in PSLp2, Cq, and a quadratic polynomial. The
correspondence again belongs to the family (7.7), and is given by the parameter values
a “ 4.6 ` 0.18i, k “ 0.9 ´ 0.05i.
The following theorem is identical to [BHar, Thm. 1].

Theorem 7.75 For every quadratic map Qc with c P M and for every faithful
discrete representation r of C2 ˚ C3 in PSLp2, Cq for which the ordinary set
is connected, there exists a p2 : 2q holomorphic correspondence which is a
mating between Qc and r.
We prove this theorem by constructing the mating using surgery. For sim-
plicity of the description we shall assume the representation of C2 ˚ C3 is in
PSLp2, Rq rather than the more general case of PSLp2, Cq, but there is no
essential extra complication in the complex case.
We first associate an annulus A to Qc . An equipotential for Qc is the image
of a circle tRe2π i t | 0 ď t ă 1u under the inverse of the Böttcher coordinate
ψc . It is an analytic Jordan curve parameterized by external argument t.
The region bounded by such an equipotential is a simply connected domain
V , mapped two-to-one by Qc onto a larger domain U Ą V which also has
boundary an equipotential parameterized by external argument. Let A denote
the annulus U zV , and denote its inner and outer boundaries by B1 A and B2 A
respectively. The map Qc sends B1 A analytically, two-to-one, onto B2 A. The
map j from B2 A to itself given by t Ñ 1 ´ t on external arguments is a
analytic orientation-reversing involution, and there is a conformal involution
ι : z Ñ ´z on V sending each z P V to the other point which has the same
image in U under Qc .
Our next ingredient is an annulus B associated to the group representa-
tion r. Recall the fundamental domain D constructed above for the group
G “ă σ, ρ, χ ą and illustrated in Figure 7.31. Let B denote the annulus
in C{χ
p consisting of the three copies D Y ρpD q Y ρ ´1 pD q of D , with
the boundary identifications induced by χ (Figure 7.34). The rotations ρ and
0 T U S −1 Q P
∼ ∞ ∼ R
Figure 7.34 The annulus B, the quotient of D Y ρpD q Y ρ ´1 pD q under the
boundary identification χ : z Ñ 1{z. The labelling of points follows that of
Figure 7.31.
ρ ´1 , mapping D Y ρpD q Y ρ ´1 pD q to itself, descend to a p2 : 2q corre-

spondence G on B, mapping each z P B to the pair tρz, ρ ´1 zu (or rather to
their equivalence classes under the action of χ ). The set D descends to a
‘fundamental domain’ for the action of G on B. The boundary of B consists
of two analytic Jordan curves, an outer boundary B2 B and an inner boundary
B1 B. The correspondence G maps each of the three sets, B2 B and two halves
of B1 B, to the other two. Thus when its domain is restricted to B1 B, and its
range is restricted to B2 B, the correspondence G defines an analytic two-to-one
orientation-reversing surjection g. The involution σ descends to an (analytic)
involution, which we also denote σ , on the outer boundary B2 B of B, having
fixed points Q and S.
Any two real analytic orientation-reversing involutions of the circle are real
analytically conjugate, so there is an analytic diffeomorphism h : B2 A Ñ B2 B
which conjugates the involution j on B2 A to the involution σ on B2 B. By
lifting via Qc on one side and σ ˝ g on the other we may extend h to
an analytic diffeomorphism BA Ñ BB which conjugates Qc : B1 A Ñ B2 A to
σ ˝ g : B1 B Ñ B2 B.
Lemma 7.76 The analytic diffeomorphism h : BA Ñ BB extends to a homeo-

morphism of annuli h : A Ñ B which is quasiconformal on interiors.
Proof This is immediate from Lemma 2.30(b).
We use h to transport our p2 : 2q correspondence G on B to a p2 : 2q

correspondence G 1 “ h´1 Gh on A. To construct a mating between Qc and
r we first observe that the quotient of U by the boundary identification j
is a Riemann surface of genus zero, and thus there is a conformal bijection
between this quotient U {j and the Riemann sphere. Taking a double cover
of U {j , branched over the fixed points of j , we obtain a Riemann sphere
U Y U 1 , equipped with an involution (also denoted j ) exchanging U with a
second copy, U 1 , and restricting to the original j on the common boundary.
Inside U 1 is a simply connected subdomain V 1 corresponding to V Ă U . Let
Q1c “ j ˝ Qc ˝ j : V 1 Ñ U 1 denote the quadratic map corresponding to Qc :
V Ñ U and let A1 denote the annulus U 1 zV 1 . To define a p2 : 2q topological
correspondence F on U Y U 1 we fit together:
• Qc : V Ñ U , a p2 : 1q correspondence;
• pQ1c q´1 “ j ˝ Q´ c ˝ j : U Ñ V , a p1 : 2q correspondence;
1 1 1
• j ˝ ι : V Ñ V , a p1 : 1q correspondence; and
1
• j ˝ G 1 : A Ñ A1 , a p2 : 2q correspondence.
To construct a complex structure preserved by this topological correspon-

dence we proceed as follows. We define an ellipse field on A by using the
quasiconformal homeomorphism h constructed in Lemma 7.76 to transport
the standard complex structure (a field of infinitesimal round circles) from the
annulus B. Using j we extend this ellipse field to A1 and by pulling back via
1´1 we extend it to an ellipse field on the whole of CzpK
c Y Kc q.
Q´ 1 1
c and Qc
p
Finally, we use the standard round circle field on Kc Y Kc1 to extend the ellipse
field to the whole of C,p and observe that this field transforms equivariantly
under the action of the p2 : 2q correspondence F. By applying the Integrability
Theorem we deduce that there exists a complex structure on C p respected by F.
This completes the proof of the theorem, since by construction our complex
structure is the standard one on Kc and Kc1 , and on their complement pFq it
is the structure corresponding to the representation r of C2 ˚ C3 .
It is apparent from the proof above that all the correspondences F con-
structed in this way satisfy a condition we discussed earlier (in Section 7.8.2)
in the context of matings between quadratic maps and PSLp2, Zq, namely:
(II) The involution σ : z Ñ ´1{z on and the involution on K´ _ K` which
exchanges z P K´ with z P K` , fit together to yield a conformal involu-
tion (a Möbius transformation) j : C
p Ñ C.
p
It follows by the same reasoning as in Section 7.8.2 that up to Möbius

conjugacy we may write F in the form
zÑw ô pj wq2 ` pj wqz ` z2 “ 3.
Applying a further conjugacy to transform j to the involution j pzq “ ´z, the
equation defining the correspondence F becomes a member of the following
family (parameterized by complex a and k):
az ` 1 2 aw ´ 1 2
ˆ ˙ ˆ ˙ˆ ˙ ˆ ˙
az ` 1 aw ´ 1
` ` “ 3k. (7.7)
z`1 z`1 w´1 w´1
This is very satisfactory, since we have exactly two complex degrees of
freedom in the dynamical systems to be mated – one for the variable c in the
quadratic polynomial, and one for the conjugacy class of the representation r
of C2 ˚ C3 .
The technical difficulty that arises if we try to construct a mating between a
quadratic polynomial map and the modular group PSLp2, Zq in an analogous
way to the proof of Theorem 7.75, is that the annuli A and B are now pinched,
their boundaries contain cusps, and Lemma 7.76 is no longer applicable.
Replacing PSLp2, Zq by a generic representation of C2 ˚ C3 allowed us to
avoid this problem. But now we can return to the original question and apply
pinching techniques to the matings provided by Theorem 7.75 to resolve

Conjecture 7.73 in a large class of cases. Two technical restrictions on the
quadratic maps Qc are imposed in [BHai] for the pinching techniques to be
applicable:
(i) if the critical point 0 of Qc is recurrent, the β-fixed point of Qc is not in

the ω-limit set of 0;
(ii) Qc is weakly hyperbolic, that is, there are constants r ą 0 and δ ă 8
such that, for all z P Jc tpreparabolic pointsu, there is a subsequence of
iterates pQnc k qk such that
n
ˆ ˙
Qc k
deg Wk pzq ÝÑ DpQnc k pzq, rq ď δ,
where Wk pzq is the connected component of Q´

c
nk
pDpQnc k pzq, rqq con-
taining z.
The class of weakly hyperbolic quadratic maps is quite large: in addition

to all hyperbolic quadratic maps it contains, for example, all quadratic maps
which satisfy the Collet–Eckmann condition, and all those which contain
parabolic points.
We end by quoting the following [BHai, Thm. 1.6]. See details in [BHai].
Theorem 7.77 For every quadratic map Qc satisfying conditions (i) and (ii)
above, there exists a mating between Qc and PSLp2, Zq in the family (7.6) of
holomorphic correspondences.
7.8.1 For the p2 : 2q correspondence F : z Ñ w defined by (7.6) verify that

when F ´1 pwq consists of a single point then w is one of the three points
1, 3{pa ` 2q or ´1{pa ´ 2q, and the corresponding point z “ F ´1 pwq
is ´1, 0 or ´2{pa ` 1q respectively. When F is a mating between
PSLp2, Zq and Qc , the restriction F|´ is a two-to-one map with
critical point z “ ´2{pa ` 1q and critical value ´1{pa ´ 2q; under the
conjugacy h´ : ´ Ñ Kc these correspond to the critical point 0 and
critical value c of Qc . By considering the orbit of ´2{pa ` 1q show that
if F is a mating between PSLp2, Zq and:
(i) Q0 : z Ñ z2 then a “ 5;
(ii) Q´2 : z Ñ z2 ´ 2 then a “ 4; ?
(iii) Q´1 : z Ñ z2 ´ 1 then a “ p3 ` 33q{2.
As Qc is hyperbolic in all three examples, we know by Theorem 7.77

that these matings occur in the family (7.6), so we may deduce that they
occur at these parameter values.
7.8.2 Construct a mating between PSLp2, Zq and Q´2 directly by taking the
p4 : 4q correspondence G defined on C p by
z Ñ z ` 1, 1{pz ` 1q, z{pz ` 1q, pz ` 1q{z,

and pushing it down to a p2 : 2q correspondence on the quotient sphere
obtained by identifying z with 1{z. To do this explicitly, write down
the polynomial equation P pz, wq “ 0 of G and after dividing through
by w2 manipulate it into the form QpZ, W q “ 0 where Z “ z ` 1{z
and W “ w ` 1{w. Since G has the same grand orbits as PGLp2, Zq
on C,
p the p2 : 2q correspondence QpZ, W q “ 0 has limit set the closed
line interval from 2 through 8 to ´2 (the image of the real axis under
z Ñ Z). It is easy to verify that the action of the correspondence on this
limit set and on its complement is that of a mating between Q´2 and
PSLp2, Zq. Observe also that a further change of coordinates Z Ñ 1{Z,
W Ñ 1{W converts the equation QpZ, W q “ 0 into the form (7.6) with
a “ 4.
8
Cut and paste surgeries with sectors
In this chapter we give three different applications of cut and paste surgeries,
including their dependence on parameters. We plough in the dynamical planes
and harvest in parameter spaces. In all three cases we start by a polynomial in a
limb of the connectedness locus of a one-parameter family, or by a polynomial-
like map hybrid equivalent to such one.
In Section 8.2 we show how it is possible, through surgery, to change a
certain given map and its phase space in such a way that the new map on its new
phase space has an extra critical point. Adding parameters to the construction,
we obtain homeomorphisms between parameter spaces of different families.
The first construction of this kind occurred in [BD] and was later generalized
in [BF1].
In Section 8.3 we show by a similar construction, depending on parameters,
how it is possible to embed a limb of the Mandelbrot set M into other limbs
of M.
In both sections, the topological surgery part involves a change of phase
space for each parameter, either by truncating or extending the given complex
plane, hence constructing an abstract Riemann surface for each parameter. The
given map is kept in certain regions and replaced by a quasiconformal map in
certain sectors outside the filled Julia set. Therefore the complex structure is
changed in these sectors, and subsequently spread by the dynamics recursively
through iterated pullbacks.
In Section 8.4 Adam Epstein and Michael Yampolsky show how to inter-
twine two quadratic polynomials to obtain a cubic polynomial. Sectors also
play an important role in this construction.
In the surgeries in Sections 8.3 and 8.4, it is important to control the opening
modulus of certain sectors and understand how this guarantees that we can
interpolate quasiconformally between certain boundary maps. The opening
307
308 Cut and paste surgeries with sectors
modulus was introduced in [BD]. However, the argument for the quasiconfor-
mal interpolation was not correct in that paper. It was corrected by Bielefeld in
his PhD thesis [Bi2]. He introduced the notion of near translations and showed
how to apply it.
In the paper by Epstein and Yampolsky [EY], reviewed in Section 8.4,
they show that one can control the opening modulus by choosing a suitable
representative of a polynomial in its hybrid class.
In Section 8.1 we first collect the tools: sectors outside the filled Julia
set of a given slope, opening modulus, near translations and quasiconformal
interpolations.
8.1 Preliminaries: sectors and opening modulus

Let P be a monic polynomial of degree d ě 2 with connected Julia set J pP q.
Let ζ be a repelling k-periodic point. There is a finite number of external
rays landing at ζ . They are mapped among themselves by P k with a certain
combinatorial rotation number, say p{q (see Section 3.3.4). If the q arguments
of one cycle of rays are cyclicly ordered as 0 ď θ1 ă θ2 ă ¨ ¨ ¨ ă θq ă 1,
then d ¨ θi mod 1 “ θpi `pq mod q . We may assume ζ to be a fixed point of
P . (If not, consider P k , which is a monic polynomial of degree d k .) Then
for i “ 1, 2, . . . , q, the arguments above are of the form d qj´1 for certain
j P t0, 1, . . . , d q ´ 2u.
8.1.1 Definition of sectors and opening modulus

With P , ζ , and p{q as above, we define a bounded sector S with vertex at ζ as
a simply connected bounded domain satisfying S Ă P q pSq, bounded by two
arcs γ1 and γ2 satisfying γj Ă P q pγj q, for j “ 1, 2, and a third arc connecting
the two corners not equal to ζ . The numbering of γ1 and γ2 are chosen so that
S is on the left of γ1 when oriented from ζ out.
Considering the quotient S{P q , we obtain a fundamental domain confor-
mally equivalent to an annulus AS (see Figure 8.1). We then define the opening
modulus of the sector S as the modulus of the annulus AS , i.e.
modζ pS, P q q “ modpAS q.
The opening modulus of a sector is in general not easy to compute. In what

follows we introduce three special types of sectors. In the first case (the one
we use most often) the opening modulus is computable.
S
Pq
γ2 γ1
γ2 {P q AS γ1 {P q
ζ
Figure 8.1 The definition of a sector with vertex at ζ , invariant under P q , and its
opening modulus.
Log-Böttcher sectors Let ψP : CzD Ñ CzKP denote the inverse of the

Böttcher coordinates conjugating P0 pzq :“ zd to P on the basin of infinity.
Moreover, let Hr denote the right half plane and consider ψP ˝ exp : Hr Ñ
CzKP , which conjugates the map Md (multiplication by d), to P . An external
ray RP pθ q in CzKP corresponds to a horizontal half line Rpθ q of imaginary
part 2π θ and all its vertical translates by 2π . An equipotential of potential ρ
corresponds in Hr to the vertical line of real part ρ.
In Hr we define the unbounded log-Böttcher sector of slope s centred at
Rpθ q as
Spθ q “ S s pθ q “ tρ ` 2π i t P Hr | |t ´ θ | ď sρu.
In the rest of the discussion we omit the word log-Böttcher. The sector Spθ q is
mapped homeomorphically onto the sector Spd ¨ θ q by Md .
The unbounded sector SP pθ q in the dynamical plane of P (see Figure 8.2)
is defined as
SP pθ q “ pψP ˝ expq pSpθ qq.
Note that any such sector overlaps itself, no matter how small the slope is.
To avoid this and to avoid overlap of different sectors (to be specified below)
we consider only bounded sectors.
As before let ζ be a repelling fixed point in J pP q. Consider its backwards
orbit ně0 P ń pζ q and all rays landing at these points. Their arguments
Ť
belong to the set

" *
j
| j P Z, n P N Y t0u .
d n pd q ´ 1q
t
RP p θ q
t “ 2π pθ ` sρ q
SP pθ q
R pθ q
2π i θ Spθ q ψP ˝ exp
t “ 2π pθ ´ sρ q
ρ
0 ρ0
Figure 8.2 The definition of an unbounded sector in the right half plane and its
corresponding unbounded sector in the dynamical plane of P . The shaded region
is the piece of sector up to equipotential ρ.
Fix an arbitrary η ą 0. For any n P N Y t0u define the bounded n-sector of

slope s centred at Rpθ q as
! η)
Sns pθ q :“ Sns pθ, ηq :“ ρ ` 2π i t P Hr | |t ´ θ | ď sρ ; ρ ď n .
d
It is not difficult to check (see [BF1, Prop. 4.1]) that if the slope s is chosen so
that
1
s ă smax :“ ,
2ηpd q ´ 1q
´ ¯
then for any n P N Y t0u the bounded n-sectors Sn d n pdjq ´1q , for j P Z, are
pairwise disjoint. Moreover, the bounded sectors
ˆ ˙ ˆ ˙
j k
S0s , j P Z, and Sn
s
, k P Z, k d, n P N,
dq ´ 1 d n pd q ´ 1q
are all pairwise disjoint, and so are their dynamical counterparts (see
Figure 8.3).
It is easy to calculate the opening modulus of sectors outside the filled Julia
set.
Lemma 8.1 (Opening modulus of log-Böttcher sectors of slope s) Let P be

a monic polynomial of degree d with connected filled Julia set, ζ a repelling
fixed point of combinatorial rotation number p{q, and SPs pθ q a log-Böttcher
sector of slope s with vertex at ζ . Then the opening modulus relative to P q is
2
modζ pSPs pθ q, P q q “ Arctan p2π sq, (8.1)
q Log d
0 ρ
η{d η
Figure 8.3 Disjoint bounded sectors of slope s “ smax in Hr .
hence an increasing function of s. If η tends to 0 and s tends to `8, under the

π
condition that s η ă 1{2pd q ´ 1q, then modζ pSPs pθ q, P q q tends to q Log d.
Proof Since the opening modulus is invariant under holomorphic conjuga-

2
tion, it suffices to show that mod0 pS s p0q, pMd qq q “ q Log d Arctan p2π sq (see
Exercise 8.1.1).
Sectors for polynomial-like maps In greater generality, let f : W 1 Ñ W be a

polynomial-like map with a repelling fixed point ζ of combinatorial rotation
number p{q. Then an invariant sector S Ă W zKf with vertex at ζ is a simply
connected domain bounded by a piece of BW and two arcs γ1 and γ2 (named as
above) with common endpoint at ζ and satisfying γj Ă f q pγj q for j “ 1, 2.
We write S “ zγ1 , γ2 { for the sector between γ1 and γ2 or for simplicity we
set BL S :“ γ1 and BR S :“ γ2 , the left and the right boundary curves of S (see
Figure 8.4).
Sectors in linearizing domains Let denote a neighbourhood of ζ and ϕ :

Ñ D a linearizing coordinate, conjugating f : 1 Ñ to the linear map
z ÞÑ λz, where 1 is the connected component of f ´1 pq containing ζ and
λ “ f 1 pζ q. A sector in is a simply connected domain S Ă with vertex at
ζ , which is invariant under f q , i.e. S X “ f q pSq X . In particular, a log-
Böttcher sector invariant under P q intersects the domain in a sector of this
type. However, in general invariant sectors in may contain part of the filled
Julia set. If S is a sector for a polynomial-like map as above, S X is a sector
in (see Figure 8.4).
f
S B L S “ γ1
B R S “ γ2
W
W1

ζ
Figure 8.4 A general sector for a polynomial-like map. The neighbourhood

denotes a linearizing domain, to be used later.
In this setting we consider the associated quotient torus Tζ “ pztζ uq{f q .

The projection of the sector S in Tζ is the annulus AS , whose modulus is
modζ pS, f q q, the opening modulus of S (see Figures 8.5 and 8.6).
Let S be an invariant sector in under the action of f q . Choose a branch
log of the logarithm defined on ϕpS X q. Then the normalized log-linearizing
coordinate is the log-linearizing coordinate normalized by dividing by log λq ,
1
z ÞÑ w “ plog ˝ϕqpzq.
log λq
It conjugates f q on S to translation T1 by 1 on the horizontal half strip ă ,

the image of S X under the normalized log-linearizing map. Note that the
opening modulus modζ pS, f q q of S is equal to the modulus of the quadrilateral
ă zT1´1 p ă q.
Let denote the double infinite strip that is invariant under T1 and the
extension of the half strip ă . This strip is conformally equivalent to the
double infinite straight horizontal strip m with R as its lower boundary and
height equal to m :“ modζ pS, f q q and such that the conformal equivalence
conjugates T1 on to T1 on m (see Figure 8.5).
Observe that if S is the restriction of a log-Böttcher sector in the linearizing
domain of a polynomial P , i.e. S “ SPs pθ q X then the two quotient annuli
are conformally equivalent and hence, by Lemma 8.1,
2
modζ pS, P q q “ Arctanp2π sq.
q log d
Tζ
AS
D
S
ϕ
ζ 0
q
fq Mλ
log
ă z ÞÑ z
log λq
z ÞÑ z ` 1 log λq
0
z ÞÑ z ` log λq
m m
Figure 8.5 A sector in the linearization domains and its conformal equivalents.
S Rc p0q
Tβc
Jc
βc
Sp
Figure 8.6 An invariant sector S “ Scs p0q with vertex βc , the landing point
of the zero ray Rc p0q.
8.1.2 Quasiconformal interpolation on sectors

In the surgeries that follow we need to construct quasiconformal mappings
from one invariant sector to another, as an interpolation between given bound-
ary maps. In this subsection we collect the different cases.
For j “ 1, 2, let fj be a holomorphic map with a repelling fixed point ζj ,
and let Sj be a sector that is invariant under fj with vertex at ζj . The boundary
of Sj consists of three pieces BL{R Sj , the left and right boundaries of Sj , and
Bout Sj , the boundary opposite to the vertex. We may assume that both BL{R S
are real analytic and that Bout S is C 2 .
Lemma 8.2 (Simple quasiconformal interpolation) Let fj and Sj be as

above. Let gL{R : BL{R S1 Ñ BL{R S2 and gout : Bout S1 Ñ Bout S2 be given dif-
feomorphisms, which are assumed to be analytic on the sides and C 2 on the
outer boundary. Furthermore, assume that gL{R ˝ f1 “ f2 ˝ gL{R whenever
defined. Then there exists an extension g : S 1 Ñ S 2 , which is quasiconformal
in the interior.
Proof The map g is defined inductively using the dynamics on the sec-
tors. For n P N let Qj,n denote the closed quadrilateral defined as the clo-
´pn´1q
sure of fj pSj qzfjń pSj q. Define g1 : Q1,1 X Q1,2 Ñ Q2,1 X Q2,2 from
the inner boundary of Q1,1 to the inner boundary of Q2,1 so that gout ˝
f1 “ f2 ˝ g1 is satisfied. Choose a quasiconformal extension g1 : Q1,1 Ñ
Q2,1 of the given boundary map which is a piecewise C 2 -diffeomorphism
(see Lemma 2.24). Define inductively gn : Q1,n Ñ Q2,n so that f2 ˝ gn´1 “
gn ˝ f1 . Compare with Figure 8.7. The resulting map g : S1 Ñ S2 is
ζ1 ζ2
gn
Q1,n Q2,n
BL S1
BR S1 BL S2 BR S2
Q1,2 Q2,2
g1
Q1,1 Q2,1
Bout S1 Bout S2
Figure 8.7 A simple quasiconformal extension of boundary maps.
quasiconformal with distortion equal to that of g1 since the distortion is not

changed when pulling back by holomorphic mappings.
For j “ 1, 2, let ϕj : j Ñ D be a linearizing coordinate around ζj conju-

gating fj to z ÞÑ λj z, where λj is the multiplier of fj at the fixed point. Then
Sj X j “ fj pSj q X j so that Sj X j is an invariant sector in j . Choose
branches of the logarithm defined on Sj X j and consider the corresponding
normalized log-linearized coordinates
1
z ÞÑ w “ plog ˝ϕj qpzq for z P Sj X j .
log λj
We shall use certain results on extensions, which were explained in Chap-
ter 2. However, we review here the basic concepts we need. Recall that a
map between two backwards periodic Jordan arcs (one-sided infinite) is a near
translation if, roughly speaking, the map is not too far from the identity map
when tending to infinity and with bounded derivative. Periodic maps are always
near translations for example. An example of this kind to keep in mind is given
by the following proposition.
Lemma 8.3 (Example of near translation) Let f : W 1 Ñ W be a quadratic-

like map with connected Julia set, ζ a repelling fixed point with combinatorial
rotation number p{q and S “ zr, l{ Ă W an invariant sector with vertex ζ
bounded by smooth arcs. Let fr : W r1 Ñ Wr be another quadratic-like map hav-
ing a repelling fixed point ζr with the same combinatorial rotation number, and
Sr “ zr l{ similarly defined. Consider a smooth map ψ : r Ñ r
r, r r conjugating
the dynamics:
ψ ˝ f q “ frq ˝ ψ.
In the log-linearizing coordinates the map ψ becomes a near translation.
Indeed, the map ψ expressed in log-linearizing coordinates commutes with

translation by 1 and is therefore a near translation.
More generally, any invariant sector is conformally equivalent to a standard
half strip, invariant under translation T1 by 1. Hence, the extension problem is
reduced to finding maps between standard half strips. We shall use two results.
If two sectors have the same opening modulus – or equivalently if two half
strips have the same height – then the unique Riemann map between them,
preserving corners when extended to the boundaries (in particular mapping
ζ1 to ζ2 ), extends as near translations on the boundaries close to infinity.
On the other hand, if two strips have different heights and we are given
boundary maps which are near translations, then they extend to the interior as
a quasiconformal map (see Chapter 2). In what follows, we apply these results
to our present setup.
Lemma 8.4 (Near translation boundary maps) Let fj and Sj be as above,
and let R : S1 Ñ S2 be the unique Riemann mapping, whose extension to the
boundary maps corners to corners, in particular ζ1 to ζ2 . Denote the boundary
maps by BL{R R : BL{R S1 Ñ BL{R S2 and Bout R : Bout S1 Ñ Bout S2 . Suppose
the opening moduli of the sectors Sj relative to fj at ζj are equal. Then
the boundary mappings BL{R R when expressed in normalized log-linearizing
coordinates are near translations.
Proof We may assume that RpS1 X 1 q Ă S2 X 2 . If not, we shrink 1
and redefine ϕ1 . Consider the restriction of R to S1 X 1 . The standard half
strips obtained by the normalized log-linearizing coordinates composed with
conformal mappings have the same height. Hence the boundary maps between
the standard half strips are near translations (see Proposition 2.15).
We now formulate the General Quasiconformal Interpolation Lemma.

Notice, that Lemma 8.2 above is a special case of the lemma below and could
have been skipped. It is included since it fits well with the common idea of
spreading by the dynamics.
Lemma 8.5 (General Quasiconformal Interpolation Lemma) Let fj , Sj
and ϕj be as above. Let gL{R : BL{R S1 Ñ BL{R S2 be analytic maps and
gout : Bout S1 Ñ Bout S2 a C 2 -diffeomorphism. Assume that the gL{R are near
translations when expressed in normalized log-linearizing coordinates close
to ζj . Then there exists an extension g : S 1 Ñ S 2 , which is quasiconformal in
the interior.
Proof We may assume that gL{R pBL{R S1 X 1 q Ă BL{R S2 X 2 . Consider
the restriction of gL{R to BL{R S1 X 1 . The standard half strips obtained by
the normalized log-linearizing coordinates composed with conformal map-
pings have different heights in general. But the induced boundary maps are
near translations by hypothesis. Hence we may extend them to the interior
quasiconformally (see Lemma 2.26).
8.1.3 Limbs of the Mandelbrot set

In this chapter the surgeries start from quadratic polynomials in a limb of the
Mandelbrot set or quadratic-like maps hybrid equivalent to such. We recall
the characteristics of quadratic polynomials in limbs. Let 0 denote the main

hyperbolic component of the Mandelbrot set, consisting of all c-values for
which Qc pzq “ z2 ` c has an attracting fixed point. The boundary B0 is
the main cardioid, parametrized by γ0 : t P T ÞÑ pe2π i t {2q ´ pe2π i t {2q2 . For
any 0 ă p ă q with gcdpp, qq “ 1 the p{q-limb, Lp{q , of the Mandelbrot
set is the connected component of Mz0 whose closure is attached to 0 at
γ0 pe2π i p{q q (see Figure 8.8).
L1

3
L1
L1
4
2 γ0 p 13 q
?
γ0 p 12 q 0
γ0 p 23 q

L3
5
Figure 8.8 The limbs of the Mandelbrot set are the connected components of Mz0 .
The polynomials in the p{q-limb Lp{q share the following features. The
inner fixed point αc is repelling and the landing point of a cycle of q external
rays mapped among themselves by combinatorial rotation number p{q. We
denote by Rαc the q rays landing at αc plus the fixed point itself. The only other
preimage of αc , apart from itself, is ´αc and it is the landing point of q other
external rays which are mapped onto the q-cycle of rays. We denote by R´αc
this set of rays plus the pre-fixed point ´αc . The complement of Rαc Y R´αc
consists of 2q ´ 1 domains, as shown in Figure 8.9 for p{q “ 1{3. We denote
these domains by tVci u0ďi ăq where Vc0 contains the critical point 0, and by
Vrci “ ´Vci for 0 ă i ă q, all numbered in cyclic order relative to ˘αc . The
quadratic polynomial acts on these domains as
2:1 p
Vc0 Ñ Vrc ,
1:1 i `ppmod q q
Vci , Vrci Ñ Vrc for i ‰ q ´ p,
q ´p r q ´p 1:1 0
Ťq ´1
Vc , Vc Ñ R´αc Y Vc Y i “1 Vci .
Rc p 17 q 1 q
Rc p 14
Vrc1 Vc2
Rc p 27 q αc Vc0 ´αc Rc p 11
14 q
Vrc2 Vc1
Rc p 47 q 9 q
Rc p 14
Figure 8.9 The partition of the plane induced by the set of rays Rαc Y R´αc
for p{q “ 1{3.
The set Kc tαc , ´αc u has 2q ´ 1 connected components, one in each set Vci
for i “ 0, 1, . . . , q ´ 1, and Vrci for i “ 1, . . . , q ´ 1. We denote the compo-
nents of Kc in Vci and Vrci by Kci and K r i respectively.
c
8.1.4 Controlling the opening modulus of sectors

In the surgeries to come we need to map certain sectors conformally onto
others, in such a way that the boundary maps have nice properties. In particular,
we require the opening moduli of these sectors to be equal.
In Section 8.3 we shall use polynomials Qc with c P Lp{q zp{q , where
p{q is the hyperbolic component with the same root point as the limb Lp{q .
In what follows we show how it is possible to control the opening moduli if
the equipotential η ą 0 is chosen sufficiently small so that the slope s can be
chosen sufficiently large (see Lemma 8.6). Later, in Section 8.4, Epstein and
Yampolsky show how one can control the opening moduli by working in the
hybrid class of a quadratic polynomial Qc with c P Lp{q .
Working with quadratic polynomials directly For any Qc with c P Lp{q , con-
sider the quadratic-like restriction Qc : W 1 Ñ W , where W 1 , W are bounded
by equipotentials of potential η{2 and η for an arbitrary choice of η ą 0.
Consider the forward invariant set of bounded sectors Scs pθi q of slope s around
the q rays Rc pθi q, i “ 1, . . . , q, landing at αc . Let be a linearizing domain
q
around αc and let Tαc “ ztαc u{Qc denote the quotient torus. Observe that
the polynomial Qc induces a conformal map Q p c : Tα Ñ Tα . The q sectors
c c
of slope s project to a q-cycle under the map Qc of annuli in Tαc , of common
p
modulus mpsq, given by the formula (8.1) in Lemma 8.1. Observe that mpsq
does not depend on c and η. The annuli in between form another q-cycle under
the map Q p c . Denote the common modulus of these ‘in between’ annuli by
m pc, sq. These moduli are independent of η.
Lemma 8.6 The modulus m pc, sq is a continuous function of pc, sq P Lp{q ˆ

R` . For fixed c P Lp{q zp{q the modulus m pc, sq tends to 0 as s tends to 8.
A proof can be found in [BD, Chap. III, Sect. 13].

Remark 8.7 For any c0 P Lp{q zp{q we can choose a slope s (increasing η
if necessary) such that m pc, sq ă mpsq holds for all c in a neighbourhood of
q
c0 in Lp{q zp{q . We can then split sectors of slope s into three Qc -invariant
subsectors SL , SM , SR so that the middle sector SM has opening modulus equal
to m pc, sq.
Working in the hybrid class of a quadratic polynomial Let f : W 1 Ñ W be

a quadratic-like map, which is hybrid equivalent to a quadratic polynomial Qc
with c P Lp{q . Let ζf denote the inner fixed point of f , and let K r i pf q for
i “ 0, 1, . . . , q ´ 1, denote the q components of Kf ztζf u, numbered cyclicly
with the critical point ωf of f contained in K r 0 pf q. Let Tζ “ pf ztζf uq{f q
f
be the associated quotient torus, and fp : Tζf Ñ Tζf the quotient map induced
by f . Let Kp i pf q denote the projection of K
r i pf q to Tζ . Following Epstein and
f
Yampolsky (compare with Subsection 8.4.3) we define
modζf K
p i pf q “ inftmod A | K
p i pf q Ă Au,
where the infimum is taken over open annuli A in Tζf . The modulus
modζf K p i pf q is the same for all i and does not depend on the representative
f in the hybrid class. It follows from Lemma 8.6 that the modulus is equal to
zero for all c P Lp{q zp{q , since K p i pf q is included in the ‘n between’ annuli
for all i.
The following lemma is due to Epstein and Yampolsky (cf. [EY]) and is
applied in Section 8.4.
Lemma 8.8 (Sectors with prescribed opening modulus) Choose c P Lp{q .

For each m ą 0 there exists a quadratic-like mapping f : W 1 Ñ W , hybrid
equivalent to Qc , such that f admits a q-cycle of disjoint f q -invariant sectors
with vertex at ζf , bounded by analytic curves at the sides and with opening
modulus relative to f q equal to m.
Proof Let as above Qc : V 1 Ñ V be a quadratic-like restriction where V

is bounded by an equipotential of potential η ą 0. Then consider the q
log-Böttcher sectors Scs pθi q of slope s around the rays Rc pθi q, i “ 1, . . . , q,
landing at αc . Let A pq denote the corresponding annuli in the quotient

p1 , . . . , A
torus Tαc . Choose ą 0 and a quasiconformal homeomorphism ψ : A p1 Ñ
Ae´2π pm`q ,1 , where Ae´2π pm`q ,1 “ te ´ 2π p m` q ă |z| ă 1u is the standard
annulus of modulus m ` . Change the complex structure in Scs pθ1 q to the one
induced by ψ ˚ pσ0 q in A p1 , and then in the other sectors S s pθi q and all their
c
preimages by successive pullbacks by Qc . Keep σ0 elsewhere. Through inte-
gration we obtain a quadratic-like mapping f : W 1 Ñ W , hybrid equivalent to
Qc , with a q-cycle of annuli in the quotient torus Tζf of modulus m ` . Inside
this q-cycle we can choose a q-cycle of annuli of modulus m with analytic
boundaries at the sides.
8.1.1 Consider the linear map z ÞÑ d q z and the unbounded sector S s p0q of
slope s centred at Rp0q “ R` . The normalized log-linearizing coordi-
nate map z ÞÑ pLog zq{pLog d q q maps the sector S s p0q onto the standard
strip with upper boundary equal to the horizontal line of imaginary part
Arg p1 ` i 2π q{ Log d q and lower boundary equal to the horizontal line
of imaginary part Arg p1 ´ i 2π q{ Log d q . Show that the height of the
strip is equal to the opening modulus and hence
1 1`
mod0 pS s p0q, pMd qq q “
˘
Logp1 ` i 2π sq ´ Logp1 ´ i 2π sq
Log d q i
2
“ Arctan p2π sq.
q Log d
You may recall the formula
1
Arctan z “ pLogp1 ´ i zq ´ Logp1 ` i zqq.
i
8.2 Creating new critical points

The cut and paste surgery in this section involves constructing a truncated
complex plane, an abstract Riemann surface. The Integrability Theorem allows
for Riemann surfaces conformally equivalent to D, C or C.p However, when the
construction depends on a parameter we need to be extra careful.
Previously we discussed several examples of surgery, depending on a param-
eter and inducing well-defined maps between parameter spaces. In all of them
we constructed a model map by pasting different maps on one copy of the
complex plane or the Riemann sphere.
The new feature of the surgery is the creation of an extra critical point,
obtained by cutting out part of the original dynamical plane and applying the
first return map on the remaining region. This kind of construction was origi-
nally done by Branner and Douady (see [BD, Part II]) to explain an observation
by Yoccoz, concerning combinatorial similarities between quadratic polynomi-
als in the 1{2-limb L1{2 of the Mandelbrot set and certain cubic polynomials.
The construction was later generalized by the authors (see [BF1]), relating
quadratic polynomials in any p{q-limb Lp{q of the Mandelbrot set to certain
polynomials of degree pq ` 1q. We explain the original surgery and sketch in
Section 8.2.4 that the same result can be obtained by pasting different maps on
one complex plane, as observed by Shishikura.
First we state the content of the main theorem in [BF1]. Then we explain
the surgery in the simplest case (q “ 3). Finally we list some properties of the
resulting maps between the parameter spaces; we point out the different steps
in proving them, but refer to details in the original paper.
For any q ě 3 consider the one-parameter family of polynomials of degree
pq ` 1q
z q
ˆ ˙
Pq,λ pzq “ λz 1 ` where λ P “ C˚ .
q
These polynomials of degree pq ` 1q are characterized by having a fixed point

at the origin of multiplier λ, a critical point of multiplicity pq ´ 1q at ´q,
q
mapped onto that fixed point, and a simple free critical point ω “ ´ q ` 1 . Let
Cq denote the connectedness locus of this family,
Cq “ tλ P C˚ | KPq,λ is connectedu.
Notice that for any λ P Dzt0u the two critical points belong to KPq,λ , which is
therefore connected. Hence Dzt0u Ă Cq . The unit circle is the analogue of the
main cardioid in the Mandelbrot set, and analogously the r{s-limb Lq,r {s of
Cq is defined as the connected component of Cq zD attached to the unit circle at
e2π i r {s , where either r “ 0 or 0 ă r ă s and gcdpr, sq “ 1. The 0-limb Lq,0
is the focus in the following.
Theorem 8.9 (Comparing limbs in the two families) For any q ě 3 and any
p ă q such that gcdpp, qq “ 1 there exists a homeomorphism
“ p,q : Lp{q ÝÑ Lq,0 ,
which is holomorphic in the interior of Lp{q .

Observe that the image p,q pLp{q q is independent of p. Hence we have the
following corollary.
Corollary 8.10 (Limbs of M of denominator q are homeomorphic) Given

p{q and p1 {q with p, p1 ă q such that gcdpp, qq “ gcdpp1 , qq “ 1, the map
r p,p1 ,q “ ´11 ˝ p,q : Lp{q ÝÑ Lp1 {q

r “
p ,q
is a homeomorphism, which is holomorphic in the interior of Lp{q .
One can actually prove this corollary without going through the higher
degree families (as observed by Dierk Schleicher at the time), using a cut and
paste surgery without leaving the complex plane. This approach is carried out
in [BF2] in order to prove that the homeomorphisms r p,p1 ,q between the limbs
in the Mandelbrot set preserve their embeddings in the plane.
The limbs L1{3 and L3,0 are shown in Figures 8.10 and 8.11 respectively.
Figure 8.10 The 1{3-limb L1{3 of the Mandelbrot set and the filled Julia set of z2 ` c
for c “ ´0.127997 ` i 0.871489. The value of c is marked in the limb and the
attracting 6-cycle is marked in the filled Julia set.
We organize the section as follows. First we give the characterizations of

the polynomials Pq,λ in the 0-limb Lq,0 . Then we explain the surgery for
p{q “ 1{3 turning a quadratic polynomial Qc in the 1{3-limb into a quartic
polynomial P3,λ in the 0-limb L3,0 . We explain why this gives a well-defined
Figure 8.11 The 0-limb L3,0 of C3 , the connectedness locus of the family P3,λ , and
the filled Julia set of P3,λ with λ “ 1,3 pcq “ 5.64 ` i 0.54 and c as in Figure 8.10.
The value of λ is marked in the limb and the attracting 2-cycle is marked
in the filled Julia set.
map “ 1,3 : L1{3 Ñ L3,0 . Finally we sketch how the construction varies
with the parameter and what properties the map has. The last part is the most
delicate.
8.2.1 The family of polynomials of degree pq ` 1q

The 0-limb Lq,0 of the connectedness locus Cq of the family Pq,λ is the
connected component of Cq zD whose closure is attached to the unit disc at
the point λ “ 1 (see Figure 8.11, where q “ 3). Polynomials Pq,λ in Lq,0
share the following features. The repelling fixed point at z “ 0 is the landing
point of a unique invariant ray. As a consequence, the multiple critical point
at ´q is the landing point of q rays, which are preimages of the invariant ray.
Let ψλ : CzD Ñ CzKPq,λ be the inverse of a choice of Böttcher coordinate,
conjugating z ÞÑ zq `1 to Pq,λ and so that the invariant ray has argument 0.
We shall denote dynamic rays of argument t by Rq,λ ptq or Rλ ptq if it does
not lead to confusion. The preimages of Rq,λ p0q are Rq,λ pj {pq ` 1qq for j “
1, . . . , q.
The q rays landing at ´q partition the plane into q domains, which
we denote by Vq,λ0 “ V 0 (the one containing the critical point ω), V 1 “
λ q,λ
q ´1 q ´1
Vλ1 , . . . , Vq,λ “ Vλ (see Figure 8.12, where q “ 3). The dynamics on
these domains are as follows:
2:1
Vλ0 ÞÝÑ CzR q,λ p0q,
(8.2)
1:1
Vλi ÞÝÑ CzR q,λ p0q for i “ 1, . . . , q ´ 1.
Rλ p 34 q
Vλ2
Rλ p0q 0 Vλ0 ´3 Rλ p 12 q
Vλ1
Rλ p 14 q
Figure 8.12 Sketch of the partition in the dynamical plane of Pq,λ for q “ 3. The
plane is rotated by π to show the resemblance with the partition in the dynamical
plane of a quadratic polynomial.
8.2.2 The surgery

Starting from a quadratic polynomial Qc with c P L1{3 , we construct a map
which has the topological properties of a polynomial Pλ “ P3,λ with λ “ λpcq
in L0 “ L3,0 . We show that this determines a well defined map : L1{3 Ñ L0
with pcq “ λpcq. The rays landing at αc and ´αc have external arguments
1 2 4 1 9 11
7 , 7 , 7 and 14 , 14 , 14 respectively.
Step 1: Cut and paste to define an abstract Riemann surface CTc For any Qc
with c P L1{3 we start by cutting out the domains Vrc1 , Vrc2 and the external ray
Rc p 27 q in between, and identifying the rays Rc p 17 q and Rc p 47 q equipotentially.
This means that z1 „ z2 if z1 P Rc p 17 q and z2 P Rc p 47 q both have the same
potential. Let Rrc denote the set of equivalence classes on the two identified
rays, the gluing curve. We obtain the truncated plane
CTc “ Vc0 Y Vc1 Y Vc2 {„ .
The space CTc has a natural structure as a Riemann surface isomorphic to C

(see Exercise 8.2.1).
Step 2: Define the first return map on CTc In the truncated complex plane CTc
p1q
we let fc denote the first return map, that is, we iterate each point in CTc by
Qc until it returns to CTc . On the wedges tVci ui “0,1,2 the map takes the form
$
3 0
&Qc in Vc ,
’
’
p1q
fc :“ Q2c in Vc1 ,
’
in V 2 .
’
%Q
c c
We observe that this map shares many features with the polynomial Pλ . Indeed
it is holomorphic in Vc0 Y Vc1 Y Vc2 and maps these sets in the same fashion as
(8.2). Roughly speaking it has degree four and the point at ´αc is now topo-
logically a double critical point, since any neighbourhood of ´αc is mapped
3:1 to a neighbourhood of αc . However, the first return map is discontinuous
along the three rays landing at ´αc . The discontinuity is of shift type. As
1
an example, if we approach a point of potential ρ on Rc p 14 q from Vc2 , the
images tend to the point of potential 2ρ on the gluing curve Rc . However, if
r
we approach the same point from Vc0 , the images tend to the point of potential
23 ρ on Rrc (see Figure 8.13). As a consequence, the preimage under the first
return map of an equipotential curve in CTc of potential ρ will consist of three
pieces of equipotential curves of potential 2´3 ρ, 2´2 ρ, 2´1 ρ in Vc0 , Vc1 , Vc2
respectively.
1 q
Rc p 14
p1q
fc Q3c Qc
Vc2
Vc0
Rrc Rc p0q
αc ´αc
Vc1 Q2c
Q3c
9 q
Rc p 14
Figure 8.13 Sketch of the first return map, showing how the preimage of an
equipotential curve in CTc consists of three disconnected pieces of equipotential
curves. There is a shift discontinuity along the rays landing at ´αc .
p1q p2q
Step 3: Modify fc to a quasiregular map fc on a restricted region For an
arbitrary but fixed choice of potential η ą 0 let Xc denote the restricted region
in CTc of points of potential less than or equal to η. On a neighbourhood of
each ray of discontinuity (a sector) we modify the first return map to make it
continuous. The modifications will be quasiconformal maps, keeping the map
unchanged on the truncated KcT :“ Kc X CTc .
We use bounded log-Böttcher sectors as neighbourhoods of these rays. As
explained in Section 8.1 they can be defined in the right half plane Hr , once
and for all, independently of c, with a choice of slope s compatible with the
Rc p 27 q
11
14
Rc p 17 q 9
14
ψc ˝ exp 4
αc
7 Sp 47 q
Rc p 47 q 1 q
Rc p 14
2
7 Sp 27 q
´αc
9 q
Rc p 14 1
7 Sp 17 q
1
14
Rc p 11
14 q η {4 η {2 η
Figure 8.14 Sectors overlap, but not if we restrict to the inside of a given
equipotential curve and an appropriate slope. The right figure shows the sectors in the
right half plane. The map ψc ˝ exp brings them to the dynamical plane of Qc .
potential η so none of the sectors and their iterated preimages overlap (see
Figure 8.14).
Let Sc pθq denote the sectors for the six external arguments corresponding
to the rays landing at αc and ´αc . The sectors Sc pθ q, θ “ 17 , 47 , merge into
a single one, which we denote by Src . It consists of half of Sc p 17 q and half of
Sc p 47 q, joined along the gluing curve Rrc (see Figure 8.15). This sector Src is
‘invariant’ under the first return map, in the sense that points in Src remain in
p1q
Src until they leave the region Xc . Moreover, fc is holomorphic in Src , where
it equals Q3c .
Consider the sectors around the rays landing at ´αc . Each of them is
mapped, under the first return map, to Src , with a shift discontinuity along the
rays. We will modify the first return map inside these sectors.
First choose a C 2 -Jordan curve γc , which coincides with the three pieces of
equipotentials of potential 2´3 η, 2´2 η, 2´1 η in Xc outside the three sectors
landing at ´αc , and connects them with C 2 -Jordan arcs inside these three
sectors (see Figure 8.15). Let Xc1 Ă Xc denote the domain bounded by γc .
Note that the choice of curve can be made once and for all, independently of c,
in Hr , and then transported to the dynamical plane by ψc ˝ exp. Let Sc1 denote
the part of the sectors landing at ´αc , which is inside Xc1 , that is
´ ¯
Sc1 :“ Sc p 14
1 9
q Y Sc p 14 q Y Sc p 11
14 q X Xc1 .
Xc
γc “ BXc1 1 q
Sc1 p 14
?
Xc1
Src R
rc
αc ´αc Sc1 p 11
14 q
9 q
Sc1 p 14
Figure 8.15 Sketch of the domain Xc and its preimage Xc1 under the modified map
p2q
fc . Shadowed regions show the sector Src and its three preimages, disjoint from Src ,
under the same map.
We now modify the first return map on the sectors Sc1 landing at ´αc . We
use the following extension lemma.
p2q
Lemma 8.11 For some K ą 1 there exists a K-quasiregular map fc : Xc1 Ñ
Xc which satisfies the following conditions:
p2q p1q
(a) fc pzq “ fc pzq if z P Xc1 zSc1 ;
p2q
(b) for θ “ 141 9 11
, 14 , 14 , the map fc : pSc pθ q X Xc1 q Ñ Src is a
K-quasiconformal homeomorphism.
The proof of this lemma is left as an exercise (see Lemma 8.2 and Exer-
p2q
cise 8.2.2). By working on the right half plane, fc can actually be defined so
that it depends holomorphically on the parameter c (see Exercise 8.2.3).
p3q
Step 4: Changing the complex structure to obtain a holomorphic map fc
We now define a Beltrami form μc on Xc as follows. First define μc “ μ0 on
p2q
Src . Since fc is holomorphic on´Src , it ¯
follows that μ0 is invariant.
p2q ˚
We proceed by defining μc “ fc μ0 on Sc1 . The dilatation is K, from
Lemma 8.11. Observe that orbits pass through these three sectors at most once,
since they are all mapped to Src which is invariant. Hence we can spread μc
p2q
recursively by the dynamics of fc , that is
$
μ on Src ,
’´ 0 ¯ ˚
’
’
& fcp2q μ0 on Sc1 ,
’
’
μc :“ ´ p2q ¯˚ p2q
’
’
’ pfc qn μc on pfc qń pSc1 q, for n ě 1,
’
’ p2q
on Xc1 z n pfc qń pSc1 q.
Ť
μ0
%
p2q
In particular, μc “ μ0 on KcT . Observe that μc is invariant under fc by
construction, and it has bounded dilatation (precisely K), since all pullbacks,
except the first one, are done by holomorphic maps.
Hence, applying the Integrability Theorem we obtain quasiconformal home-
˝
omorphisms φc : Xc Ñ D such that φc˚ μ0 “ μc . Observe that, since μc “ μ0
on KcT , it follows that Bφc “ 0 on KcT .
˝
p3q p3q p2q
Set Dc1 :“ φc pXc1 q and define fc : Dc1 Ñ D as fc :“ φc ˝ fc ˝ φc´1 .
Then the following diagram commutes:
˝ p2q ˝
f c
pXc1 , μc q ÝÝÝ ÝÑ pXc , μc q
§ §
φc đ
§ §φ
đ c
p3q
f c
pDc1 , μ0 q ÝÝÝ ÝÑ pD, μ0 q
p3q
and the map fc is holomorphic (see Key Lemma for surgery (Lemma 1.39)).
p3q
Step 5: Obtaining a polynomial in L0 Since fc is quasiconformally con-
p2q p3q
jugate to the map fc , the properties of the latter are preserved. Hence, fc
is a ramified holomorphic covering of degree four, with two critical points (at
z “ 0 and at φc p´αc qq. Moreover, Dc1 is relatively compact in D. It follows
p3q
that fc is polynomial-like of degree four.
By the Straightening Theorem (Theorem 7.4) there exists a hybrid equiva-
p3q
lence χc , conjugating fc to an actual polynomial fc of degree four. Due to
the conjugacy, we conclude for fc that:
• zpcq :“ χc pφc pαc qq is a repelling fixed point,

• z2 pcq :“ χc pφc p0qq is a simple critical point,
• z3 pcq :“ χc pφc p´αc qq is a double critical point,
• z1 pcq :“ fc pz2 pcqq, and
• z1 pcq is the landing point of a unique fixed ray.
Conjugating by the affine map which sends z1 pcq to 0 and z3 pcq to ´3, we
obtain a unique polynomial in the 0-limb of C3 , that is
fc „ Pλpcq , with λpcq P L0 .

affine
See Exercise 8.2.4. Finally we need to check that λpcq is independent of the
many choices made during the construction.
Proposition 8.12 ( is well defined) The map : L1{3 Ñ L0 defined by c ÞÑ

λpcq is well defined.
Proof Suppose we made different choices: an equipotential ηr, a slope r s,

a Jordan curve γrc bounding X r 1 , quasiconformal extensions on the sectors
c
r 1 landing at ´αc , resulting in the quasiregular map frc p2q , the induced
in X c
p3q
r c and, after integration, the holomorphic map frc , which is
Beltrami form μ
p2q p2q
hybrid conjugate to a unique polynomial Pr . Then fc and frc are hybrid
λpcq
equivalent. To see this, observe that the choices only affect the shape of the
sectors and the modification of the first return map on those sectors landing
at ´αc . Construct recursively a quasiconformal conjugacy between the maps
on the sectors (restricting to an outer level corresponding to the smallest of the
equipotentials η and ηr, and enlarging to the largest of the slopes s and r
s ), so that
these conjugating maps equal the identity on the side boundaries of the sectors.
p2q p2q
Outside the sectors the maps fc and frc coincide with the first return map.
Let hc denote the constructed quasiconformal conjugacy on sectors, extended
by the identity outside the sectors. Then hc is a hybrid equivalence.
p3q p3q
It follows that the holomorphic maps fc and frc are also hybrid equiv-
´1
alent by the conjugating map φrc ˝ hc ˝ φc , and so are the polynomials Pλ
and Prλ obtained by straightening. Since the filled Julia sets, KPλpcq and KPrλpcq ,
are connected, the polynomials Pλpcq and Prλpcq are affine conjugate (Proposi-
tion 3.53). The affine conjugacy must fix 0 and the double critical point ´3
and is therefore the identity. Hence λpcq “ r
λpcq.
We end this part with a series of remarks, which are consequences of the
fact that, on the filled Julia set, the new polynomial acts as the first return map
of the quadratic polynomial we started with.
Remarks 8.13 (Properties of corresponding polynomials) Let c P L1{3 .
(1) If the critical point 0 of Qc is attracted to an attracting cycle, so is the

free critical point ω of Pλpcq . The periods of the cycles are not the same,
since the first return map skips the first q ´ 1 points in the critical orbit of
Qc (and maybe others). However, the multipliers are the same, since all
conjugacies are holomorphic in the interior of the filled Julia set.
(2) For the same reason, if the critical point of Qc is attracted to a parabolic
cycle, so is the free critical point of Pλpcq . Also if Qc has a periodic orbit
of Siegel discs, so does Pλpcq .
(3) If the critical point of Qc is eventually mapped onto a repelling periodic
cycle, so is the free critical point of Pλpcq .
8.2.3 Properties of : L1{3 Ñ L0 – dependence of the parameter

We proceed to sketch how to study the map in the parameter plane. All details
can be found in [BF1]. We explain how to prove continuity and bijectivity of
. Since L0 is compact, the continuity of ´1 follows.
Surjectivity To prove that is surjective one needs to perform an inverse

surgery. That is, start with a parameter λ P L0 and end up with a quadratic
polynomial Qcpλq with cpλq P L1{3 , such that pcpλqq “ λ. This surgery
requires to cut the dynamical plane of Pλ along the invariant ray Rλ p0q, open
it up and add pieces in order to glue back what we previously removed.
p2q
To prove injectivity of , we observe that if pcq “ pc1 q, then fc and
p2q
fc1 are hybrid equivalent. This means that the first return maps are confor-
mally conjugate in the interior of the truncated filled Julia sets. One can prove
that this implies c “ c1 .
Continuity The proof of continuity of is the most delicate part, and it makes
a difference to be working on an abstract Riemann surface, which varies with
the parameter. A naı̈ve approach would be to start by examining the following
diagram:
˝ p2q ˝
f
c
Xc1 ÝÝÝ ÝÑ Xc
§ §
φc đ
§ §φ
đ c
p3q
f
Dc1 c
ÝÝÝ ÝÑ D
§ §
χc đ
§ §χ
đ c
Pλpcq
χc pDc1 q ÝÝÝÝÑ χc pDq
If we consider for a moment Xc and Xc1 as subsets of the complex plane, we
p2q
can argue that the Beltrami form μc and the map fc vary holomorphically
with the parameter c, since both were constructed once and for all in the
complex plane and then transported by the Böttcher coordinates, which vary
holomorphically with the parameter. However, this works on the basin of
infinity, which itself is moving with c. So we can not make the same statement
for all points in the dynamical plane, since a given point z might belong to
Kc for a given c, but might be in the basin of infinity for arbitrarily nearby
parameters.
To make things worse, once the identifications are made, the Riemann
surface Xc varies with c. If we ever wanted to apply the Integrability Theorem
with parameters, to conclude that φc depends continuously on c, we would
need to uniformize Xc to the unit disc. But then we should take into account
the dependence of the uniformizing map with respect to c.
All these observations indicate that this is not the best approach.
In the interior of L1{3 we can prove not only continuity of but also
conformality. This is proved by using Remark 8.13 above. Let M denote a
hyperbolic component of L1{3 . It is mapped into some hyperbolic component
in L0 , say Cq . Let M : D Ñ M and Cq : D Ñ Cq be the respective
multiplier maps. They are both conformal isomorphisms. (For Cq this fol-
lows by applying the surgery in Section 4.1 to Pλ with λ P Cq , noticing that
the free critical point ω is simple.) Therefore
1
|M “ Cq ˝ ´
M
is a conformal isomorphism.
It is conjectured that the interior of M is the union of hyperbolic com-
ponents. As long as the conjecture is not proved, we have to consider the
possibility of non-hyperbolic components, also known as queer. In this case
a similar argument would show that is continuous and holomorphic in such
components. One would use the continuous parametrization obtained when
deforming the invariant line field supported on the Julia set (in fact, another
soft surgery construction).
The remaining case is the continuity on the boundary of L1{3 . It is proved
via sequences. Suppose c P BL1{3 , and cn Ñ c, with cn P L1{3 . Set λ “ pcq.
Then λ must belong to BL0 , because the inverse map ´1 is mapping the
interior of L0 to the interior of L1{3 . If we set λn “ pcn q, we must prove that
all convergent subsequences of tλn uně0 have λ as limit. Suppose we have a
convergent subsequence (which we denote by λn again), such that λn Ñ r λ.
The key is the following rigidity property: if Prλ and Pλ are quasiconformally
conjugate, and λ P BL0 , then λ “ r λ. We leave the proof of this rigidity fact as
an exercise (see Exercise 8.2.5 or [BF1, Lem. 5.22]). It remains to show that
Prλ and Pλ are quasiconformally conjugate.
We claim there exists a convergent subsequence
Pλnk ÝÑ Pλ˚ ,
k Ñ8
such that Pλ˚ is quasiconformally conjugate to Pλ .

To see the claim, let φn :“ φcn and χn :“ χcn denote the corresponding
integrating maps for each of the parameters. Observe that the dilatation of φn is
p2q
uniformly bounded by K for all n, since fcn was constructed once and for all
in the right half plane. To find convergent subsequences of these maps, we first
need to fix a common domain for all of them. We use the following lemma.
Lemma 8.14 Given c P L1{3 , there exists r “ rpcq ą 0 such that for each t P
Dr one can find a quasiconformal homeomorphism
Ht : Xc ÝÑ Xc`t ,
with dilatation bounded by

r ` |t|
Kt “
r ´ |t|
and H0 “ id|Xc , making

p2q p2q
Ht´1 ˝ fc`t ˝ Ht ÝÑ fc
uniformly on compact subsets of Xc as t Ñ 0.
We refer to [BF1, Lem. 5.22] for the proof, which uses holomorphic motions
and the λ-lemma.
Picking the right tn values so that c ` tn “ cn (with n large enough) we have
quasiconformal homeomorphisms Hn :“ Htn : Xc ÝÑ Xcn with dilatation
Kn :“ Ktn . Then the maps tφn ˝ Hn : Xc ÝÑ Dun form a K 1 -quasiconformal
family, with K 1 ă KKn for all n. We normalize by setting φn Hn pαc q “ 0.
Then, the maps tφn Hn un form an equicontinuous family and, by Arzelà–
Ascoli’s Theorem, there exists a convergent subsequence
φnk ˝ Hnk Ñ φ ˚ .
The limit map is a K 1 -quasiconformal homeomorphism. The map

p3q p2q
f˚ :“ φ ˚ ˝ fc ˝ pφ ˚ q´1 ,
being the uniform limit of

´ ¯
p3q p2q
fnk “ φnk ˝ Hnk ˝ Hn´k 1 ˝ fnk ˝ Hnk ˝ Hn´k 1 ˝ φn´k1 ,
is holomorphic and
p3q φ ˚ p2q φc p3q
f˚ „ qc fc „hb fc .
p3q p3q
Hence fc „qc f˚ .
p3q
Abusing notation, we have a sequence of polynomial-like mappings fn
p3q
converging to f˚ uniformly on compact subsets of D, and a sequence of
hybrid equivalences χn , with uniform bounded dilatation, since the moduli
of the exterior annuli are uniformly bounded. We may assume χn is defined
in all of D and normalized so that χn p0q “ 0 and χn pϕn pHn p´αc qqq “ ´3.
Then, we can find a subsequence χnk converging to a quasiconformal homeo-
morphism χ ˚ .
The polynomials Pλnk tend to the polynomial
p3q
P ˚ :“ χ ˚ ˝ f˚ ˝ pχ ˚ q´1 .
Therefore
χ˚ p3q p3q χc
P ˚ „ qc f˚ „qc fc „hb Pλ ,
so it follows that Pλ „qc P ˚ .

Finally, note that P ˚ must be of the form Pλ˚ . Hence, Pλn Ñ P ˚ implies
that λn Ñ λ˚ , which proves the claim. It follows that λ˚ “ r λ, and therefore
we have proved Pλ „qc Prλ .
This concludes the proof of continuity at boundary points.
8.2.4 Alternative surgery construction without truncating the plane

We may obtain the homeomorphism : Lp{q Ñ Lq,0 by an alternative
surgery construction. The idea is due to Shishikura. One may do the con-
struction in the original dynamical plane. As before, starting from any Qc
with c P Lp{q the construction should lead to a polynomial-like map of degree
pq ` 1q, which is hybrid equivalent to Pλ where λ “ p{q pcq P Lq,0 .
In this section we sketch the main steps in this alternative surgery, in the
simplest case q “ 3.
For c P L1{3 , choose a potential η ą 0 and a slope s so that the bounded sec-
tors Sc pθ q “ Scs pθ q of slope s and argument θ “ 17 , 27 , 47 are pairwise disjoint.
Let Yc denote the domain in the dynamical plane of Qc of potential less than η.
Spc
αc
1 q
Sc1 p 14
´αc
Yc
9 q
Sc1 p 14
Sc1 p 11
14 q
Figure 8.16 Sketch of the domain Yc and its preimage Yc1 under the modified map Fc ,
and the sector Spc and its three preimages, disjoint from Spc , under the same map.
Let Yc1 Ă Yc be the simply connected domain inside Yc , which is bounded by

the equipotential of potential η{23 everywhere except in Vc1 Y Vc2 Y Sc1 where
the bounding curve is chosen similar to the bounding curve of Xc1 (compare
Figures 8.16 and 8.15). Set
´ ´ ¯ ´ ¯ ´ ¯¯
Spc “ Sc 17 Y Vrc1 Y Rc 27 Y Vrc2 Y Sc 47 X Yc
and Spc1 “ Spc X Yc1 . The sector Spc plays a role similar to the sector Src in the
truncated case.
The goal is to define a modified map Fc : Yc1 Ñ Yc whose dynamics resem-
bles the dynamics of a polynomial in L3,0 . The map is defined so that it is
quasiregular and there exists an Fc -invariant Beltrami form μc of bounded
dilatation, and so that after integrating we obtain a polynomial-like map hybrid
equivalent to a unique polynomial Pλ , where λ “ 1{3 pcq P L3,0 .
We proceed to define the modified map Fc from each of the sectors Spc1 pθ q
to Spc . The sector Spc contains part of the filled Julia set Kc and is cer-
tainly not invariant under Q3c . However, the sector is locally invariant under
Q3c . Let ϕc : c Ñ D be a linearizing coordinate around αc conjugating Qc
to z ÞÑ λc z, where λc “ Q1c pαc q is the multiplier of Qc at αc . Note that
Spc X c “ Q3c pSpc q X c . We use this fact and construct the map Fc so that
it coincides with Q3c in a neighbourhood of αc .
For notation compare with Figure 8.17. Let A0L denote the intersection point
j j ´1 3
in BYc X BL Spc , and define AL recursively as the preimage of AL under Q´ c
j
belonging to BL Spc . Define the points AR belonging to BR Spc similarly. Choose
N `1
N so that AN L{R
and AL {R
belong to c . Choose γ N as a C 2 -curve inside c
connecting AN N
L and AR , and let γ
N `1 denote the preimage of γ N under Q´3
c
belonging to c . Set Fc :“ Qc on the domain in Spc inside the curve γ N `1 .
3
Let γ 0 and γ 1 be the pieces of equipotentials of potential η and η{23

respectively belonging to the closure of Spc and choose C 2 -curves for j “
2, . . . , N ´ 1, also belonging to the closure of Spc , pairwise disjoint, and num-
bered as indicated in Figure 8.17. Denote the closed quadrilateral bounded
by γ j , γ j `1 and pieces of BL{R Spc by Q p j . For j “ 1, . . . , N choose C 2 -
diffeomorphisms from γ j to γ j ´1 .
γ0
γ1
γ2
γ3
γ4
p0 Q
Q p1 Q p3 Q
p2 Q p4 γ5 A0L
1
2 AL
Fc “ Q3c αc 5
A A4L A3L AL
A5R L
A4R c
A3R
A2R
A1R
0
AR
Figure 8.17 Sketch of the subdivision of the sector Spc into pN ` 1q “ 5

p j , j “ 1, . . . , N ` 1, the bounding curves γ j , j “ 0, 1, . . . , N ` 1,
quadrilaterals Q
j j j
connecting the points AL and AR respectively. The points AL{R belong to BL{R Spc
j j ´1
and satisfy Q3c pAL{R q “ AL{R for j “ 1, . . . , N ` 1. The curves γ0 and γ1 are
equipotentials. The curves γ2 and γ3 are arbitrary, while γ5 is the preimage of
γ4 under Q3c .
Let Fc : Qpj Ñ Q p j ´1 be a quasiconformal map, which extends the given

boundary maps, and suppose the map is Kj -quasiconformal for some Kj . This
defines Fc : Spc1 Ñ Spc .
To complete the definition of the modified map Fc we need to define it on the
remaining part of Yc1 . This is done similarly to the original surgery explained
above. The map Fc is defined as before on Yc1 zpSpc1 Y Sc1 q, i.e.
$
3 in Vc0 zpSpc1 Y Sc1 q,
&Qc
’
’
Fc :“ Q2c in Vc1 zSc1 ,
’
in Vc2 zSc1 .
’
%Q
c
1 9 11
Each of the sectors Sc pθ q with θ “ 14 , 14 , 14 are invariant under ´Q3c . Using
this and the local invariance of Spc under Q3c it follows that there exist K 1 -
quasiconformal maps Fc : Sc1 pθ q Ñ Spc for θ “ 14 1 9 11
, 14 , 14 and some K 1 . (Com-
pare with Lemma 8.2, Lemma 8.5 and Exercise 8.2.2.)
ŤN
Set K :“ K 1 ˆ N j “1 Kj , and set Z :“ Sc Y j “1 Q , the subset of Yc on
1 pj 1
which Fc is quasiconformal; everywhere else Fc is holomorphic. Observe that

the orbits of Fc pass through Z at most pN ` 1q times. We can therefore define
an Fc -invariant Beltrami form μ p c by spreading μ0 on Q p 0 by the dynamics to all
of n“0 Fc pSpc q and leave it as μ0 on the complement Yc1 z 8
Ť8 ´ n
Ť ń p
n“0 Fc pSc q.
The map Fc is K-quasiregular.
After integrating the Fc -invariant Beltrami form μ p c we obtain a polynomial-
like map, which is hybrid equivalent to a unique polynomial Pλ with λ P L3,0 .
It remains to justify that λ “ 1{3 pcq. It can be checked that Fc is hybrid
p2q
equivalent to fc from the original construction, using the identity map on
p2q
Xc zSpc and a quasiconformal map from Spc to Src conjugating Fc to fc . By the
p2q
same arguments as in Proposition 8.12, the λ-values obtained from fc and
from Fc are the same.
8.2.1 Prove that CT is a Riemann surface isomorphic to C.

8.2.2 Prove Lemma 8.11 by constructing quasiconformal extensions as in
Lemma 8.2.
8.2.3 Show that the proof of Lemma 8.11 can be done independently of the
parameter c. More precisely, define a map f p2q in Hr and conjugate by
p2q
ψc ˝ exp to obtain fc in Xc zKcT , varying holomorphically with c.
8.2.4 Let q ě 3. Show that any polynomial P of degree pq ` 1q with a fixed
point at z “ 0, one simple critical point, and a second critical point at
´q of multiplicity q ´ 1 such that P p´qq “ 0, must be of the form
P pzq “ λzp1 ` z{qqq , for some λ P C˚ .
8.2.5 Fix q ě 3 and λ P C˚ . Let Pλ pzq “ λzp1 ` z{qqq . Let Cq denote the
connectedness locus of tPλ uλPC˚ . Suppose Pλ and Pλ1 are quasiconfor-
mally conjugate and that λ P BCq . Show that λ “ λ1 .
Hint: Let μ denote the Beltrami form induced by the quasiconformal

conjugacy in Kλ . Extend it to equal 0 outside Kλ . Consider μt “ tμ for
appropriate values of t. Use μt to construct a family Pλpt q of polynomi-
als, all quasiconformally conjugate to Pλ with λptq P Cq . Show that λptq
is holomorphic and conclude that it is constant. Compare λp0q and λp1q
to deduce, using Theorem 3.53, that λ “ λ1 .
8.2.6 Consider the two one-parameter families of cubic polynomials
´ z ¯2
PrA pzq “ Apz3 ´ 3zq, A P C˚ , P2,λ pzq “ λz 1 ` , λ P C˚ .
2
Note that for PrA the critical points are ˘1 and the critical orbits
are symmetric w.r.t. 0. For P2,λ the critical points are ´q “ ´2 and
´q{pq ` 1q “ ´2{3, and the critical point ´2 is mapped to the fixed
point 0.
Show that PrA and P2,λ are semiconjugate by a quadratic polynomial,
which sends both critical points of PrA to the critical point ´2{3 of P2,λ
and the preimages of the fixed point 0 of PrA (that are different from 0)
to the prefixed critical point ´2 of P2,λ .
Conclude that the connectedness locus of the family PrA in the A-plane
is the same as the connectedness locus of P2,λ in the λ-plane under the
relation λ “ p3Aq2 .
Figure 8.23 in Section 8.4 illustrates the connectedness locus of the
family PrA , and hence also the connectedness locus of the family P2,λ in
the λ-plane.
8.3 Embedding limbs of M into other limbs

As in the preceeding section, we present a surgery where the topological part
involves abstract Riemann surfaces.
The basic construction was done by Branner and Douady (see [BD, part III])
in order to obtain an embedding of the 1/2-limb L1{2 into the 1/3-limb L1{3 of
M, so that the image of the interval r´2, ´3{4s Ă L1{2 is a topological arc
connecting the landing point of the external ray RM p1{4q to the root point of
the limb L1{3 . The interval and the image arc are called the principal veins of
L1{2 and L1{3 respectively.
Recall from Chapter 3 that the Mandelbrot set is known to be connected and
is conjectured to be locally connected. If the MLC Conjecture is proven, it fol-
lows immediately that M is arcwise connected, since a closed connected and
locally connected set is always arcwise connected. Douady’s main motivation
for the surgery was to provide support to the conjecture of arcwise connectivity.
The construction was later generalized in [R] showing the exixtence of other
veins in M as topological arcs.
In the following we concentrate on the original embedding of L1{2 into L1{3 ,
although it can easily be generalized to an embedding of L1{2 into an arbitrary
p{q-limb, and to to an embedding of L1{q into L1{q `1 .
The steps in the surgery are similar to those in Section 8.2. However, in this
case, one needs to control the opening modulus of certain sectors.
The common features for polynomials in Lp{q are given in Section 8.1.3.
To distinguish c-values in L1{pq `1q from those in L1{q we shall use c P L1{q
and λ P L1{pq `1q . Figure 8.18 shows the domains Vci , Vrci and Vλi , Vrλi and the
external rays landing at ˘αc and ˘αλ respectively.
Rc p 13 q Rc p 16 q Rλ p 17 q 1 q
Rλ p 14
Vrλ1 Vλ2
Rλ p 27 q Rλ p 11
14 q
Vrc1 Vc0 Vc1 Vλ0
Vrλ2 Vλ1
Rc p 23 q Rc p 56 q Rλ p 47 q 9 q
Rλ p 14
Figure 8.18 Sketch of the partition of dynamical plane for c P L1{2 and λ P L1{3 .
Theorem 8.15 (Embedding of limbs) There exists a map
q : L1{q Ñ L1{pq `1q ,
which is a homeomorphism onto its image and holomorphic in the interior of

L1{q . The image N1{pq `1q :“ q pL1{q q consists of parameters λ P L1{pq `1q
for which Qnλ p0q R Kλ X Vλ1 for all n P N.
8.3.1 The surgery

Recall that the multiplier map : Ñ D from any hyperbolic compo-
nent Ă M extends continuously to the boundaries. Set q : 1{q Ñ
1
1{pq `1q equal to ´
1{pq `1q
˝ 1{q . It remains to define q : L1{q z1{q Ñ
L1{pq `1q z1{pq `1q . In the following we restrict to q “ 2.
Starting from a quadratic polynomial Qc with c P L1{2 z1{2 we construct
a map which has the topological properties of a quadratic-like map, hybrid
equivalent to a polynomial Qλ with λ “ λpcq P L1{3 .
Step 0: Restricting the polynomial to a quadratic-like map We restrict Qc

to a quadratic-like map Qc : Wc1 Ñ Wc by a choice of potential η ą 0 so that
Wc is bounded by the chosen equipotential. As explained in Section 8.1 we
can choose sectors of an arbitrary slope s around the rays Rc p 13 q and Rc p 23 q in
Wc , satisfying ηs ă 16 so that the bounded sectors Scs p 13 q and Scs p 23 q and their
successive preimages are all disjoint. The sectors are outside the filled Julia
set Kc .
Applying Lemma 8.6 and Remark 8.7 we may choose η and s so
that m pc, sq ă mpsq, where m pc, sq is the opening modulus of cs “
Vrc1 zpScs p 13 q Y Scs p 23 qq and mpsq the opening modulus of the two sectors of slope
s, with vertex at αc and relative to Q2c . Note that ´αc is a fixed point of ´Q2c
of the same multiplier as αc under Q2c and that the sectors of slope s around
Rc p 16 q and Rc p 56 q are ´Q2c -invariant and of opening modulus mpsq. Hence,
we can choose a middle sector SM around Rc p 56 q of slope sM ă s so that
the opening modulus mod´αc pSM , ´Q2c q “ m pc, sq. We have subdivided the
sector around Rc p 56 q into three subsectors SL , SM , SR so that SL Y SM Y SR “
Scs p 56 q X Wc1 (compare with Figure 8.19).
Rc p 13 q Rc p 16 q
Scs p 13 q Scs p 16 q
cs 0 ´αc
αc
SR
SL
Scs p 23 q
SM
Rc p 23 q Rc p 56 q
Figure 8.19 The restriction Qc : Wc1 Ñ Wc with sectors and wedge

as explained in Step 0.
Step 1: Cut and paste to obtain an abstract Riemann surface CE c We define

the space CE c by extending C as explained below (see Figure 8.20).
Slit along Rc p 23 q and open the slit up creating two copies of the ray which we
denote by Rc p 23 qp1q Ă B Vrc1 and Rc p 23 q Ă BVc0 . Insert a copy of Vrc1 in between
these two rays. The original set is denoted by pVrc1 qp1q and the copy by pVrc1 qp2q .
The latter is bounded by a copy of each of the original rays, which we denote
by Rc p 13 qp2q and Rc p 23 qp2q . We paste by identifying zp2q on Rc p 23 qp2q to the
Rc p 13 q Rc p 16 q
pVrc1 qp1q
Vc0
R
pc αc ´αc Vc1
Rc p 23 qp1q „ Rc p 13 qp2q
pVrc1 qp2q
Rc p 23 qp2q „ Rc p 23 q Rc p 56 q
Figure 8.20 The extended plane CE

c as explained in Step 1. The arrows show the
p1q
discontinuity of the map fc , as defined in Step 2.
point z on Rc p 23 q, which is on the same equipotential curve as zp2q , and by

identifying zp2q on Rc p 13 qp2q to Qc pzqp1q on Rc p 23 qp1q . The identification of
Rc p 23 q and Rc p 23 qp2q is still be denoted Rc p 23 q, while the gluing of Rc p 23 qp1q
with Rc p 13 qp2q is denoted by Rpc . The sectors around Rc p 13 q and Rc p 23 q are still
denoted by Scs p 13 q and Scs p 23 q, while the sector around Rpc is denoted by Spc . The
extended space CE c has a natural structure as a Riemann surface isomorphic
to C.
p1q
Step 2: Define the map fc on the extended plane We define the map
p1q
fc by
$
Q pzq z P Vc1 ,
’ c
’
’
&Q pzqp1q
’
z P Vc0 ,
p1q c
fc pzq :“
’zp2q
’ z “ zp1q P pVrc1 qp1q ,
’
z “ zp2q P pVr 1 qp2q .
’
Qc pzq
%
c
It is holomorphic on its domain of definition. By continuous extension, Rc p 16 q

is mapped to Rc p 13 q, and Rc p 13 q to Rpc , while Rc p 56 q is a line of discontinuity.
p1q
Note that fc pzq approaches Rc p 23 q and Rpc respectively, as z approaches
Rc p 5 q within V1 pcq and within V 0 respectively. Points in pVr 1 qp2q only have
6 c c
p1q
one preimage under fc .
p2q p2q
Set Kc :“ pKc X pVrc1 qqp2q , and let KcE :“ Kc Y Kc denote the extended
p1q
filled Julia set. Note that KcE is forward invariant under fc .
p1q p2q
Step 3: Modify fc to a quasiregular map fc on a restricted domain We
start by choosing the domain of definition Xc1 and the range Xc of the modified
p2q
map fc (compare with Figure 8.21). The boundary of Xc is chosen as a
C 2 -curve, which coincides with the boundary of WcE except in the interior of
E
pVrc1 qp2q Y Scs p 23 q, where it is chosen to lie in WcE zWc1 . The boundary of Xc1
coincides with the boundary of W 1E except in the interior of pVr 1 qp1q Y Spc ,
c c
p1q
where it is determined by fc mapping Xc1 onto Xc . We shall modify the map
p1q
fc in the sector Scs p 56 q X Wc1 .
Rc p 13 q
Traces of
equipot. η{2
Scs p 13 q
pcs qp1q
Rpc Spc αc ´αc
SL SR
pcs qp2q
Traces of
Scs p 23 q SM
equipot. η{2
Traces of
equipot. η
Rc p 23 q Rc p 56 q
Figure 8.21 The figure illustrates where Xc1 and Xc differ from Wc1E and WcE and
p1q p2q
shows the sectors that are important when modifying fc to fc .
Let R : SM Ñ pcs qp2q X Xc denote the unique Riemann mapping, which

when extended continuously to the boundaries is mapping corners to corners
and so that the vertex ´αc of SM is mapped to the vertex αc of pcs qp2q . Since
we arranged the opening moduli to be the same, the boundary maps on the
sides are near translations when expressed in log-linearizing coordinates(recall
Lemma 8.4). By Lemma 8.5 we can find quasiconformal maps
p2q p2q
fc : SL Ñ Spc X Xc and fc : SR Ñ Scs p 23 q X Xc
p1q
which extend the boundary maps given by BRL{R and fc . This finishes the
p2q
construction of fc : Xc1 Ñ Xc , a quasiregular map of degree 2. Note that
orbits pass through the side-sectors SL{R at most once, while orbits may pass
through SM infinitely often.
p3q
Step 4: Changing the complex structure to obtain a holomorphic map fc
We define a Beltrami form μc on Xc as follows. Set μc :“ μ0 on the three
sectors Scs p 13 q, Spc , Scs p 23 q in Xc . The union of these form a forward invariant set
p2q ˚
´ ¯
p2q
for fc whenever defined. Then set μc :“ fc μ0 on SL{R . Since orbits
pass at most once through SL{R we can spread μc by successive pullbacks
p2q p2q
by fc to ně1 pfc qń pSL{R q. Set μc :“ μ0 on the remaining part of Xc ,
Ť
p2q
i.e. on Xc z ně0 pfc qń pSL{R q. By construction μc is f p2q -invariant. The
Ť
dilatation of μc is bounded by the dilatation of μc in SL{R , since all pullbacks,

except the first one, are done by holomorphic maps.
By applying the Integrability Theorem we obtain a quasiconformal home-
˝ ˝
omorphism φc : Xc Ñ D such that φc˚ μ0 “ μc . Set Dc1 :“ φpXc1 q and define
p3q p3q p2qq p3q
fc : Dc1 Ñ D as fc :“ φc ˝ fc ˝ φc´1 . Then fc is holomorphic.
p2q
In particular, μc “ μ0 on ně0 pfc qń pKcE q.
Ť
p3q
Step 5: Obtaining a polynomial in L1{3 It follows from above that fc :
Dc1 Ñ D is a quadratic-like mapping with a critical point at φc p0q. The critical
p3q
orbit belongs to φc pKcE q. Therefore the filled Julia set of fc is connected. By
the Straightening Theorem 7.4 there exists a hybrid equivalence χc , conjugat-
p3q
ing fc to a unique quadratic polynomial Qλ with λ P L1{3 .
Proposition 8.16 (The map is well defined) The map : L1{2 Ñ L1{3
defined by c ÞÑ λpcq is well defined. Its image satisfies
pL1{2 q “ N1{3 :“ tλ P L1{3 | Qnλ p0q R Vλ1 for any n ě 0u.
Moreover, if Qc has a periodic orbit, which is attracting, parabolic or Siegel

respectively, then so does Qλpcq .
Proof (Sketch) Suppose we made other choices throughout the construc-

tion (a different equipotential level, slope, extensions, etc.), and obtained a
p2q p2q
map frc . Then there is a quasiconformal conjugacy hc between fc and
p2q
frc which is actually a hybrid equivalence. It follows that the holomor-
p3q p3q
phic maps fc and frc are also hybrid equivalent by the conjugating map
´1
φrc ˝ hc ˝ φc , and so are the polynomials Pλ and Pr obtained after straight-
λ
ening. Since the filled Julia sets, KPλpcq and KPrλpcq , are connected, the poly-
nomials Pλpcq and Prλpcq are affine conjugate (Proposition 3.53). Because of
the normalizations the affine conjugacy must be the identity and therefore
λpcq “ rλpcq.
8.4 Intertwining surgery 343
8.3.2 Concluding remarks

To end the proof of the Embedding Theorem 8.15 for q “ 2 one needs to
prove that the map : L1{2 Ñ N1{3 is a bijection and a homeomorphism.
To prove surjectivity one performs an inverse surgery, so that starting from
λ P N1{3 one ends up with Qcpλq P L1{2 satisfying pcpλqq “ λ. In the inverse
construction we either start by removing part of the plane corresponding to
what we added before, or we apply a construction similar to the alternative
surgery in Section 8.2.4. For details we refer to [BD].

Adam Epstein and Michael Yampolsky
8.4.1 Cubic connectedness locus
Our discussion of the parameter space of cubic polynomials follows the
detailed presentation by Milnor in [Mi3].
Observe that every cubic polynomial is affine conjugate to a map of the form
Fa,b pzq “ z3 ´ 3a 2 z ` b, (8.3)
with critical points a and á. This normal form is unique up to conjugation by
z ÞÑ ´z, which interchanges Fa,b and Fa,´b . The ordered pair of complex
numbers A “ a 2 and B “ b2 parametrizes the space of cubic polynomials
modulo affine conjugacy.
The cubic connectedness locus is the set C Ă C2 of all ordered pairs pA, Bq
for which the corresponding polynomial Fa,b has connected Julia set. As in
the quadratic case, the connectedness locus is compact and connected with
connected complement. These results were obtained by Branner and Hubbard
[BH] who showed moreover that this set is cellular, the intersection of a
sequence of strictly nested closed discs. On the other hand, Lavaurs [Lav]
proved that C is not locally connected.
Milnor distinguishes four different types of hyperbolic components, accord-
ing to the behaviour of the critical points: adjacent, bitransitive, capture and
disjoint [Mi3]. We are exclusively interested in the last possibility: a compo-
˝
nent H Ă C is of disjoint type Dm,n if Fa,b has distinct attracting periodic
orbits with periods m and n for every pa 2 , b2 q P H. By definition, the Pern pλq-
curve consists of all parameter values for which the cubic polynomial Fa,b has
1 This section is a revised version of the paper [EY] and is reproduced with the permission of the
Annales Scientifiques de l’ENS.
a periodic point of period n and multiplier λ. The geography of Per1 p0q was
studied in [Mi4] and [Fau].
Notice that any cubic polynomial with real coefficients is affine conjugate
to a map of the form (8.3) with A, B P R, that is a, b P R Y i R. Thus we may
consider the connectedness locus of real cubic maps, the set of pairs pA, Bq P
R2 such that JFa,b is connected. This locus CR is also compact, connected and
cellular [Mi3]. We refer the reader to Figure 8.22 which was generated by a
computer program of Milnor. The real slices of various hyperbolic components
are rendered in different shades of grey. Certain disjoint type components are
indicated, as are the curves Per1 p1q X CR and Per2 p1q X CR .
1,1
2,2
Figure 8.22 Connectedness locus CR in the real cubic family.
To avoid ambiguities arising from the choice of normalization, we will

actually work in the family of cubics
PA,D pwq “ Apw 3 ´ 3wq ` D, A ‰ 0,
with marked critical points ´1 and `1. The reparametrization
C˚ ˆ C Q pA, Dq ÞÑ pA, AD 2 q “ pA, Bq P C˚ ˆ C
is branched over the symmetry locus B “ 0 consisting of normalized cubics

which commute with z ÞÑ ´z (see Figure 8.23). In particular,
C# “ tpA, Dq Ă C˚ ˆ C| JPA,D is connectedu
is a branched double cover of C X pC˚ ˆ Cq. The marking of critical points

allows us to label the attracting cycles of maps in disjoint type components
Figure 8.23 Symmetry locus in the family PA,D .
H Ă C# , and we denote the corresponding multipliers λ˘

H pA, Dq. It is shown
in [Mi5] that the maps H : H Ñ D ˆ D given by
H pA, Dq “ pλ´ `
H pA, Dq , λH pA, Dqq
are biholomorphisms. The omitted curve A “ 0, consisting of maps with a

single degenerate critical point, is irrelevant to the discussion of disjoint type
components.
This useful change of variable has the unfortunate side-effect that the val-
ues pA, Dq P R˚ ˆ R only account for the first and third quadrants of the
real pA, Bq-plane, the second and fourth quadrants being parameterized by
R˚ ˆ i R. We are therefore unable to furnish a faithful illustration of the entire
locus
C#R “ tpA, Dq| pA, AD 2 q P CR u.
8.4.2 Outline of the results

In the picture of the real cubic connectedness locus (Figure 8.22) one observes
several shapes reminiscent of the Mandelbrot set. We quote Milnor ([Mi3]):
these embedded copies tend to be discontinuously distorted at one particular point,
namely the period one saddle node point c “ 1{4, also known as the root of the
Mandelbrot set. The phenomenon is particularly evident in the lower right quadrant,
which exhibits a very fat copy of the Mandelbrot set with the root point stretched out
to cover a substantial segment of the saddle-node curve Per2 p1q. . . . As a result of
this stretching, the cubic connectedness locus fails to be locally connected along this
curve.
In this section, we present the following results of [EY] (see also Haı̈ssinsky
[Ha1]), which explain the appearance of these distorted copies of the Mandel-
brot set embedded in CR .
For p{q P Q{Z we consider the set Cp{q Ă C# consisting of cubic polyno-
mials for which 2q distinct external rays with combinatorial rotation number
p{q land at some fixed point ζ . As there can be at most one such point, the var-
ious Cp{q are disjoint. Each Cp{q is in turn the disjoint union of subsets Cp{q,m
indexed by an odd integer 1 ď m ď 2q ´ 1 specifying how many of these
rays are encountered in passing counterclockwise from the critical point ´1
to the critical point `1. In particular, C0 consists of those cubics in C# whose
fixed rays RA,D p0q and RA,D p 12 q land at the same fixed point.
Recall from Section 8.1.3 that the p{q-limb, Lp{q , of the Mandelbrot set
is the connected component of Mz0 whose closure is attached to 0 at
the point rootp{q :“ γ0 pe2π i p{q q, its root point, and that p{q denotes the
hyperbolic component in Lp{q with root point rootp{q .
Theorem 8.17 (Main Theorem on Quadratic Intertwining) Given p{q and

m as above, there exists a homeomorphic embedding
hp{q,m : Lp{q ˆ Lp{q ÝÑ Cp{q,m
which maps the boundary into the boundary. The product of hyperbolic com-
ponents p{q ˆ p{q is mapped by the embedding onto a component Hp{q,m
of type Dq,q .
We note that Hp{q,m is the unique Dq,q component contained in Cp{q,m as

will follow from Theorem 8.35. The restriction of hp{q,m to p{q ˆ p{q is
easily expressed in terms of the multiplier map
p{q : p{q Ñ D,
which associates the parameter with the multiplier of the attracting cycle
1
hp{q,m pc, crq “ ´
H p p{q pcq, p{q pr
cq q.
p{q,m
Discontinuity of hp{q,m at the corner point p1, 1q is a special case of a phe-

nomenon studied by one of the authors [Ep1].
Theorem 8.18 (Discontinuity at corner point) Each algebraic homeomor-

phism
Hp{q,m : Hp{q,m Ñ D ˆ D
extends to a continuous surjection H p{q,m Ñ D ˆ D. The fibre over p1, 1q is

the union of two closed discs whose boundaries are real-algebraic curves with
a single point in common, and all other fibres are points.
The following reasonable conjecture appears to be inaccessible by purely
quasiconformal techniques.
Conjecture 8.19 Each hp{q,m extends to a continuous embedding
Lp{q ˆ Lp{q z tprootp{q , rootp{q qu ÝÑ Cp{q,m .

We draw additional conclusions from the natural symmetries of our con-
struction. The central disc in Figure 8.23 is parametrized by the eigen-
value ´3A of the attracting fixed point at 0; this region corresponds to symmet-
ric cubics whose Julia sets are quasicircles. Each value A “ ´ 13 e2π i p{q yields
a map with a parabolic fixed point at 0. These parameters are evidently the
roots of small embedded copies of M, and our results confirm this observation
for odd-denominator rationals. More specifically, it will follow that the latter
copies are the images of h0 ˝ δ and hp{q,q ˝ δ for odd q ą 1, where
δ
C Q c ÞÑ pc, cq P C2
2 and P 2
is the diagonal embedding. As PA,0 Á,0 are conjugate by z ÞÑ ´z, our
construction also accounts for the copies with q ” 2 (mod 4q. Every map
in the symmetry locus is semiconjugate, via the quotient determined by the
involution, to a cubic polynomial with a fixed critical value. Such maps were
studied by Branner and Douady [BD] and are the subject of Section 8.3.
Similar considerations applied to the antidiagonal embedding yield results
for the real connectedness locus. In view of the fact that real polyno-
mials commute with complex conjugation, C#R X Cp{q “ H unless p{q ”
´p{q (mod 1), and it therefore suffices to consider the real slices of C1{2
and C0 .
Theorem 8.20 There exist homeomorphic embeddings
1{2,1 : L1{2 Ñ C#R X C1{2,1 ,
1{2,3 : L1{2 Ñ C#R X C1{2,3

and
0 : Mztrootu Ñ C#R X C0 .
It follows from work of Buff [Bu] that these maps are compatible with
the standard embeddings in the plane (see the discussion in the 8.4.4). Their
projections in CR are indicated in Figure 8.22. Notice that the two images of
L1{2 have been identified while the image of Mztrootu has been folded in
half.
8.4.3 Intertwining surgery

History The intertwining construction was described in the 1990 Conformal
Dynamics Problem List [Bi2]:
Let P1 be a monic polynomial with connected Julia set having a repelling fixed point
x0 which has a ray landing on it with rotation number p{q. Look at a cycle of q rays
which are the forward images of the first. Cut along these rays and get q disjoint
wedges. Now let P2 be a monic polynomial with a ray of the same rotation number
landing on a repelling periodic point of some period dividing q (such as 1 or q). Slit
this dynamical plane along the same rays making holes for the wedges. Fill the holes in
by the corresponding wedges above making a new sphere. The new map is given by P1
and P2 , except on a neighbourhood of the inverse images of the cut rays where it will
have to be adjusted to make it continuous.
A tool: surgery of invariant sectors Let f : W 1 Ñ W be a quadratic-like

map with connected Julia set, ζ a repelling fixed point with combinatorial
rotation number p{q and associated quotient torus Tζ “ pztζ uq{f q , where
is a fixed but otherwise arbitrary linearizing neighbourhood Q ζ . Given
B Ă W ztζ u with f q pB X q “ B X f q pq, we denote B p its projection to Tζ ;
q
in particular, Kf , . . . , Kf Ă Tζ are the quotients of the various components of
p 1 p
Kf ztζ u. As any two annuli A1 Ą K p i1 and A2 Ą Kp i2 are isotopic we may speak
f f
of a distinguished isotopy class of open annuli A Ă Tζ , namely A „ K p f if and
only if A is isotopic to an annulus containing some Kf . Moreover, it is easy to
p i
see if A Ă K p i does not separate Tζ then A „ K

f
p f ; it follows then that ζ is on
the boundary of an immediate basin of attraction. Consider
modζ K
p i :“ suptmod A| A Ă K
f
pi u
f
and
modζ K
p i :“ inftmod A| A Ą K
f
pi u
f
over open annuli A „ Kp f . Notice that these quantities are independent of i. In

view of the following we may simply write modζ K pf .
Lemma 8.21 In this setting modζ K

p i “ modζ K
f
pi .
f
Proof It is obvious that modζ K

p ď modζ K
p for K
p“K
p i . Conversely, given
f
Rn Œ R8 “ emodζ K there exist conformal embeddings hn : A1,Rn Ñ Tζ such
p
that hn pA1,Rn q Ą K
p i , where
f
A1,R “ tz|1 ă |z| ă Ru.
It follows from standard estimates in geometric function theory that the hn

form a normal family on A1,R8 ; moreover, as all of these embeddings are
isotopic, every limit h8 “ limk Ñ8 hnk is univalent. Clearly, h8 pA1,R8 q Ă K
p
and therefore modζ K ď modζ K.
p p
As mod Kp f is defined in terms of the interior of Kf , we observe the

following.
Remark 8.22 The quantity modζ K p f depends only on the hybrid equivalence
class; furthermore, modζ Kf ą 0 if and only if ζ lies on the boundary of an
p
immediate basin of attraction.
Recall from Section 8.1 that an invariant sector with vertex ζ is a simply
connected domain S Ă W bounded by an arc of BW and two additional arcs γ1
and γ2 with γj Ă f q pγj q and a common endpoint at ζ . We write S “ zγ1 , γ2 {
for the sector between γ1 and γ2 as listed in counterclockwise order. The
quotient Sp Ă Tζ is an open annulus whose modulus is referred to as the
opening modulus modζ S “ modζ pS, f q q of the sector S (see Figure 8.6).
Let λ “ f 1 pζ q denote the multiplier of the repelling fixed point ζ . Let ϕ :
Ñ C be a linearizing map conjugating the action of f in the neighbourhood
Q ζ to multiplication by λ in a neighbourhood of the origin. For an invariant
sector S Ă select a branch of the mapping
1
z ÞÑ w “ logpϕpzqq
q log λ
defined in S. In the log-linear coordinate w the map f q becomes the unit

translation τ : w Ñ w ` 1; the image of the sector S is a horizontal strip ă
with τ ´1 p ă q Ă ă (compare with Figure 8.5).
In what follows we shall need to construct quasiconformal maps between
sectors, or equivalently between half strips. In order to do so we will use
concepts and results from Chapter 2 and from Section 8.1.2.
Construction of a cubic polynomial Fix p{q written in lowest terms and an

odd integer m “ 2k ` 1 between 1 and 2q ´ 1. Our aim is to construct a map
hp{q,m : Lp{q ˆ Lp{q ÝÑ Cp{q,m .
Fixing parameter values c and cr in Lp{q , consider quadratic-like maps f :

W 1 Ñ W and fr : W r1 Ñ W r hybrid equivalent to Qc and Qrc respectively, with
the choice of the hybrid equivalences to be made below. In what follows we
will identify W and W r to obtain a new surface. The reader is invited to follow
the construction in the particular case p{q “ 1{2 with m “ 3, as illustrated in
Figure 8.24.
ψ
r
W Sr1 zM
r1
W
r
N0 zS0 R
r1
L
r1
L1 M0 M
r1
M1
S1 zM1 L0
L
r0
R1
Figure 8.24 Superimposing the regions W and W

r.
We lose no generality by assuming that 0 is the critical point for both f and
fr. Let ζ be the unique repelling fixed point of f with combinatorial rotation
number p{q, that is ζ “ βf for p{q “ 0 and ζ “ αf otherwise, and Si ”
zri , li {, i “ 0, . . . , q ´ 1 in W zKf a cycle of disjoint invariant sectors with
vertex ζ , indexed in counterclockwise order so that the critical point 0 lies in
the complementary region between Sq ´1 and S0 . We similarly specify ζr and a
cycle of invariant sectors Sri for fr. Let
´1
qď ´1
qď
φ: li Y r i Ñ li Y r
r ri ,
i “0 i “0
sending li to r
ri `q ´k and ri to r
li `q ´k ´1 , where k “ pm ´ 1q{2 and indices are
understood modulo q, be any smooth conjugacy which therefore satisfies
φpf pzqq “ frpφpzqq.
The sector Si should now correspond to the component of Kfrztζru con-

taining fri `q ´k p0q. Since Shishikura’s Principle warns against altering the
conformal structure on regions of uncontrolled recurrence, we will there-
fore employ invariant sectors Mi Ă Si Ă Ni Ă W zKf and M r i Ă Sri Ă N
ri Ă
Wr zK r to be determined below. For i “ 0, . . . , q ´ 1 denote Li the component
f
Ť q ´1
of W z j “0 Nj containing f i p0q, and L r i the corresponding component of
Ť q ´1
W j “0 Nj . Let
rz r
R i : Mi Ñ L
r i `q ´k
be the Riemann map which extends continuously to the boundary, mapping

BpMi X W q to BpL r q so that ζ maps to ζr, and
r i `q ´k X W
R
ri : M
r i Ñ Li ´ q ` k ` 1
the Riemann map sending BpM r q to BpLi ´q `k `1 X W q so that ζr maps to

ri X W
ζ . It remains to fill in the gaps.
Proposition 8.23 (Quasiconformal interpolation) For any pair Qc and Qrc

as above the hybrid equivalent quadratic-like maps f and fr and the invariant
sectors
Ni Ą Si Ą Mi and N
ri Ą Sri Ą M
ri
may be chosen so that there exist quasiconformal maps

´1
qď ´1
qď
ψ: pSi zMi q Ñ pN
ri zSri q,
i “0 i “0
´1
qď ´1
qď
r:
ψ pSri zM
ri q Ñ pNi zSi q,
i “0 i “0
with
ψ|BMi “ Ri |BMi and ψ|BSi “ φ|BSi ,
r r “R
ψ| r r “ φ ´1 | r .
r i | r and ψ|
B Mi B Mi B Si B Si
Let us complete the construction assuming the truth of Proposition 8.23. We

choose sectors Ni , Mi , N
ri , M r Ri and R
r i , and maps ψ, ψ, r i as specified above.
Ťq ´1
Consider the almost complex structure σ on W , given by ψ ˚ σ0 on i “0 Si zMi
and by σ0 elsewhere; similarly, let σ r be the almost complex structure on W r
˚ Ťq ´1
given by ψ r σ0 on
i “0 Si zMi , and σ0 elsewhere. In view of the Integra-
r r
bility Theorem (Theorem 1.28), there exist quasiconformal homeomorphisms
h : W Ñ X and h r:W r ÑX r such that σ “ h˚ σ0 and σr “ h r˚ σ0 . Consider the
Riemann surface obtained from W \ W with identifications ψ, ψ,
r r Ri , Rri ,
whose atlas is given by h and h. r It has the conformal type of a punctured disc,
and we obtain a conformal disc D by replacing the puncture with a point ‹.
Setting
qď´1 ´1
qď
˜ ¸ ˜ ¸
D :“ W z
1 1
Si Y W z r 1
Si Y t‹u Ă D,
r
i “0 i “0
we define a new map F : D1

Ñ D by
$
1 Ť q ´1
’ f pzq for z P W z i “0 Si ,
’
&
F pzq :“ frpzq for z P W r 1 z q ´1 Sri ,
Ť
’ i “0
’
‹ for z P t‹, ´ζ, ´ζru.
%
It is easily verified that F is a three-fold branched covering with critical points

0 P W and 0 P W r , and holomorphic except on the preimage of
´1
qď
S :“ pSi zMi q Y pSri zM
r i q.
i “0
Recalling that the sectors Si and Sri are invariant and disjoint, we consider
the almost complex structure on D defined as
#
pF n q˚ σ0 on F ń pSq,
σp :“
σ0 elsewhere.
By construction, the complex dilatation of σp has the same bound as that of
F ˚ σ0 , and moreover
F ˚ σp “ σ
p.
It follows from the Integrability Theorem (Theorem 1.28) that there is a qua-
siconformal homeomorphism ϕ : D Ñ V Ă C with σp “ ϕ ˚ σ0 . Setting U “
ϕpD 1 q, we obtain a cubic-like map
G :“ ϕ ˝ F ˝ ϕ ´1 : U Ñ V .
In view of the Straightening Theorem (Theorem 7.4), there is a unique

hybrid equivalent cubic polynomial PA,D whose critical points ´1 and `1
correspond to the critical point of f and fr respectively. The construction yields
extensions of the natural embeddings
π : Kc Ñ KPA,D and π
r : Krc Ñ KPA,D
to neighbourhoods of the filled Julia sets.

Remark 8.24 By construction, the projections π and π
r are infinitesimally
conformal on the respective filled Julia sets:
Bπ “ 0 a.e. on Kc and B π
r “ 0 a.e. on Krc .
We write PA,D « Qc O Qrc for any cubic polynomial so obtained. It is not

p{q,m
yet clear that this correspondence is well defined, let alone continuous. These
issues will be addressed in Section 8.4.4.
Proof of Proposition 8.23 Note first that were it not for the condition of
quasiconformality, the existence of the interpolating maps ψ and ψ r would
follow without any additional argument. Any smooth interpolations are qua-
siconformal away from the points of ζ and ζr: the issue is the compatibility of
the local behaviour of φ at ζ with that of Ri and R
ri .
For a proof of the following lemma see Lemma 8.8.
Lemma 8.25 Given any c P Lp{q and υ ą 0, there exists a quadratic-like map
f which is hybrid equivalent to Qc and admits disjoint invariant sectors Si as
above with modζ Si ą υ.
Given c, cr P Lp{q , we apply Lemma 8.25 to Qc and Qrc to obtain hybrid

equivalent quadratic-like maps f and fr admitting invariant sectors Si and Sri
with
modζ Si ą modrζ K
p rc and modr Sri ą modζ K
ζ
pc .
In view of Remark 8.22 we may then choose disjoint invariant sectors Ni Ą Si

and Nri Ą Sri so that
modζ Si ą modrζ L
r j and modr Sri ą modζ Lj
ζ
for the complementary invariant sectors Lj ,L

r j as above. Finally, we choose
Mi Ă Si and Mi Ă Si with
r r
modζ Mi “ modrζ L
r j and modr M
ζ
r i “ modζ Lj .
Note that the boundary maps
φ|BSi and φ ´1 |BSri Ri |BMi and R

ri | r .
B Mi
are all near translations in log-linearizing coordinates due to Lemma 8.3 and
Lemma 8.4 respectively. With this in mind the following lemma easily follows
from Lemma 8.5.
Lemma 8.26 With this choice of maps and invariant sectors there exist desired
quasiconformal interpolations
´1
qď ´1
qď
ψ: pSi zMi q Ñ pN
ri zSri q,
i “0 i “0
´1
qď ´1
qď
r:
ψ pSri zM
ri q Ñ pNi zSi q,
i “0 i “0
with
ψ|BMi “ Ri |BMi and ψ|BSi “ φ|BSi ,
r r “R
ψ| r r “ φ ´1 | r .
r i | r and ψ|
B Mi B Mi B Si B Si
8.4.4 Renormalization
Preliminary facts about external rays The following lemma is proved in
[GM].
Lemma 8.27 Let Pt be a continuous family of monic degree d polynomials

with continuously chosen repelling periodic points ζt . If the ray of argument θ
for Pt0 lands at ζt0 , then for all t close to t0 the ray of argument θ for Pt lands
at ζt .
The following useful Separation Principle is proven in [GM] for fixed points
and in [Kiw] for periodic points, and it directly illustrates why a degree d poly-
nomial admits at most d ´ 1 non-repelling cycles; the latter result was earlier
shown by Douady and Hubbard and appropriately generalized to rational maps
by Shishikura (see Section 7.7).
Theorem 8.28 (The Separation Principle for polynomials) Let P be a

polynomial with connected Julia set, N a common multiple of the periods of
non-repelling periodic points, the graph formed by the union of all external
rays fixed under P N together with their landing points, and let U1 , . . . , Um be
the connected components of Cz N ´j pq. Then:
Ť
j “0 P
• each component Ui contains at most one non-repelling periodic point;

• given any non-repelling cycle ζ1 , . . . , ζ passing through Ui1 , . . . , Ui , at
least one of the components Uik also contains some critical point.
We assume henceforth that KP is connected. Let R “ RP pθ q be a periodic
external ray landing at the n-periodic point ζ P KP , whose orbit we enumerate
ζ “ ζ0 ÞÑ ζ1 ÞÑ ¨ ¨ ¨ ÞÑ ζn “ ζ.
Denote by #i Ă Q{Z the set of arguments of the rays in the orbit of R landing
at ζi . The iterate P n fixes each point ζi , permuting the various rays landing
there while preserving their cyclic order. Equivalently, multiplication by d n
carries the set #i onto itself by an order-preserving bijection. For each i we
may label the arguments in #i as 0 ď θ 1 ă θ 2 ă ¨ ¨ ¨ ă θ q ă 1; then
d n θ i ” θ i `p pmod 1q
for some integer p, and we refer to the ratio p{q as the combinatorial rotation
number of R. The following theorem of Yoccoz (see [Hu1]) relates the combi-
natorial rotation number of a ray landing at a period n point ζ to the multiplier
λ “ pP n q1 pζ q.
Theorem 8.29 (Yoccoz’s Inequality) Let P be a monic polynomial with
connected Julia set, and ζ P KP a repelling fixed point with multiplier λ. If
ζ is the landing point of m distinct cycles of external rays with combinatorial
rotation number p{q then
Re ρ mq
2
ě , (8.4)
|ρ ´ 2π i p{q| 2 log d
where ρ is the suitable choice of log λ.
More geometrically, the inequality asserts that ρ lies in the closed disc of
radius log d{pmqq tangent to the imaginary axis at 2π i p{q.
Birenormalizable cubics Throughout this section we will work with fixed

values of p{q and m as specified above. Here we describe the construction
which will provide the inverse to the map hp{q,m .
We start with a cubic polynomial P “ PA,D with pA, Dq P Cp{q,m . Let ζ
be the landing point of the periodic rays with rotation number p{q. Denote
V0 , . . . , V2q ´1 the components of C with the point ζ and the 2q rays landing
at ζ removed. We enumerate them in counterclockwise order so that V0 Q ´1.
The polynomial P is renormalizable to the left if the forward orbit of the

Ťq ´ 1
critical point ´1 is contained in tζ u Y i “0 V2i . In this case the map P has
a quadratic-like restriction to a neighbourhood of the critical point ´1, as seen
from the thickening construction below.
Assuming that P is renormalizable to the left, let K´ i ĂV
1 2i for i “
0, . . . , q ´ 1 denote the connected component of KP ztζ u. Let RP pθ1i q and
RP pθ2i q be the two external rays landing at ζ which separate K´ i
1 from the
i i
other rays landing at the same point, where the values θ1 and θ2 are chosen so
that θ2i ă θ1i and the rays landing at K´ i i i
1 have arguments in rθ2 , θ1 s. Choose
a neighbourhood U Q ζ corresponding to a round disc in the local linearizing
coordinate. Fix an equipotential E and a small ą 0, and consider the seg-
ments of the rays RP pθ1i ` q and RP pθ2i ´ q connecting the boundary of U
q ´1 i
to E. Let Ą Yi “0 K´ 1 be the region bounded by these two ray segments and
the subtended arcs of E and BU , and consider the component 1 of P ´1 pq
1
with 1 Ă . In view of the fact that ζ is repelling, Ă provided that is
sufficiently small. Thus,
P : 1 Ñ
is a quadratic-like map which filled Julia set will be denoted KR . Since

q ´1 i
tP n p´1qu8 n“0 Ă Yi “0 K´1 , this set is connected, and we refer to the unique
hybrid conjugate quadratic polynomial Qc as the left renormalization RpP q.
By construction Qc has a fixed point with combinatorial rotation number p{q,
that is c P Lp{q .
Figure 8.25 illustrates this construction for a cubic polynomial in C0 . Notice
that ζ becomes the β-fixed point of the new quadratic polynomial.
Figure 8.25 Construction of the left renormalization for a cubic in C0 . The left
renormalization has the hybrid type of a rabbit. The beast on the right is more exotic.
The polynomial P is renormalizable to the right if the forward orbit of `1 is

Ťq ´1
contained in tζ u Y i “0 V2i `1 , and the set K and the right renormalization
pP q are correspondingly defined. It follows from general considerations
discussed in [McM2] that the left and right renormalizations do not depend
on the choice of thickened domains.
A cubic polynomial P is said to be birenormalizable if it is renormalizable
on both left and right, in which case
ωP p´1q X ωP p`1q Ă KR X K “ tζ u, (8.5)
and we set
8
ď
q P :“
K P í pKR Y K q.
i “0
The following is an easy consequence of the Goldberg–Milnor–Kiwi Separa-

tion Principle (Theorem 8.28) and the standard classification of Fatou compo-
nents.
Lemma 8.30 Let P be a birenormalizable cubic polynomial. Then K q P is

dense in KP , and every periodic orbit in KP zK
q P is repelling.
We denote by Bp{q,m the set of birenormalizable cubics in Cp{q,m , writing
R ˆ : Bp{q,m Ñ Lp{q ˆ Lp{q
for the map pA, Dq ÞÑ pc, crq, where Qc “ RpPA,D q and Qrc “ pPA,D q. In
view of Lemma 8.27 the thickening construction may be performed so that
the domains of the left and right quadratic-like restrictions vary continuously
for pA, Dq P Bp{q,m . Applying the Straightening Theorem, we obtain the
following.
Proposition 8.31 R ˆ : Bp{q,m Ñ Lp{q ˆ Lp{q is continuous.
We make use of the intertwining surgery in the following result.
Proposition 8.32 R ˆ : Bp{q,m Ñ Lp{q ˆ Lp{q is surjective.
Proof Fix c, cr P Lp{q . We saw above that
Qc O Qrc « P ,
p{q,m
for some cubic polynomial P “ PA,D , and we show here that R ˆ pA, Dq “
pc, crq; more precisely, we prove that RpP q “ Qc , the argument for right renor-
malization being completely parallel.
π
Let Kc Ă W 1 Ă W ÝÑC be as in Section 8.4.3. By construction, π is a
quasiconformal map conjugating Qc |Kc to P |KR and Bπ pzq “ 0 for almost
every z P Kc . Let ϕ0 : W Ñ be a quasiconformal homeomorphism with
ϕ0 ˝ Qc |BW 1 “ P ˝ ϕ0 |BW 1 ,
which agrees with π on a small neighbourhood of Kc . As ϕ0 maps the critical
value of Qc to the critical value of P |1 there is a unique lift ϕ1 : W 1 Ñ 1
such that
ϕ
W 1 ÝÝÝ1ÝÑ 1
§ §
Qc đ
§ §
đP
ϕ
W ÝÝÝ0ÝÑ
commutes and ϕ1 |BW 1 “ ϕ0 |BW 1 . Setting ϕ1 pzq “ ϕ0 pzq for z P W zW 1 , we
obtain a quasiconformal homeomorphism ϕ1 : W Ñ with the same dilata-
tion bound as ϕ0 ; moreover, ϕ1 |Kc “ π |Kc . Iteration of this procedure yields a
a sequence of quasiconformal homeomorphisms ϕn : W Ñ with uniformly
bounded dilatation. The ϕn stabilize pointwise on W , so there is a limiting
quasiconformal homeomorphism ϕ : W Ñ . By construction,
ϕ ˝ Qc |W 1 “ P ˝ ϕ|W 1
and furthermore ϕ|Kc “ π |Kc ; it follows from Rickman’s Lemma
(Lemma 1.20) that ϕ is a hybrid equivalence.
Properness Here we deduce the properness of birenormalization from the

Separation Theorem (Theorem 8.28).
Proposition 8.33 R ˆ : Bp{q,m Ñ Lp{q ˆ Lp{q is proper.
In view of the compactness of the connectedness loci, it suffices to prove
that if pAk , Dk q P Bp{q,m with
pAk , Dk q Ñ pA8 , D8 q P C# and R ˆ pAk , Dk q
Ñ pc8 , cr8 q P Lp{q ˆ Lp{q ,

then pA8 , D8 q P Bp{q,m if and only if c8 ‰ rootp{q ‰ cr8 . Let ζk be the
unique repelling fixed point of Pk “ PAk ,Dk where 2q external rays land, its
q ´1
multiplier is denoted μk . Denote by tζki ui “0 the periodic orbit of period q
q ´1
contained in KR with multiplier λk , and by tζrki ui “0 the similar orbit in K
with multiplier λk . These orbits renormalize to periodic orbits tγki ui Ă Kck
r
γki ui Ă Krck , whose multipliers will be denoted ρk and ρrk . Passing to a
and tr
subsequence if necessary, we may assume without loss of generality that the

ζk converge to a fixed point ζ8 of P8 “ PA8 ,D8 with multiplier μ8 , and ζki
and ζrki converge to periodic points ζ8
i and ζ
r8i with multipliers λ and r
8 λ8 .
Lemma 8.34 In this setting, if c8 ‰ rootp{q ‰ cr8 then ζ8 , tζ8 i u and tζ i u

i r8 i
belong to disjoint orbits. Moreover, the fixed point ζ8 is repelling.
Proof It follows from the Implicit Function Theorem that these orbits are
distinct unless one of tζ8 i u , tζ i u is parabolic with multiplier 1. Without
i r8 i
loss of generality, λ8 “ 1, and we may further assume that either |λk | ď 1
for every k or |λk | ą 1 for every k. In the first case, λk “ ρk as straightening
preserves the eigenvalues of attracting orbits, and λk Ñ 1 implies c “ rootp{q .
In the second case, it similarly follows that |ρk | ą 1 for every k; in view of
Yoccoz’s Inequality, the combinatorial rotation numbers of γki are pk {qk Ñ 0,
whence ρk Ñ 1 and again c “ rootp{q .
Because tζ8 i u , ζ and tζ i u lie in distinct orbits, it follows from Theo-
i 8 r8 i
i u is
rem 8.28 that at least one of these orbits is repelling. Suppose first that tζ8 i
repelling, and let pζk q P KPk be the points which renormalize to ´γk P Kfck .
i ˚ i
Then pζki q˚ Ñ pζ8 i q˚ , where P ppζ i q˚ q “ P pζ i q, and the rays landing at

8 8 8 8
i q˚ separate ζ from the critical point ´1. Similarly, if the orbit
the points pζ8 8
i u is repelling then the rays landing at the corresponding points pζ
tζr8 i q˚
i r8
separate ζ8 from `1. Applying Theorem 8.28 once again, we conclude that
ζ8 is repelling.
Continuing with the proof of Proposition 8.33, we observe by Lemma 8.27

that ζ8 is the common landing point of the same two cycles of rays with
rotation number p{q. The thickening procedure yields a pair of quadratic-like
restrictions
P8 : 18 Ñ 8 and P8 :
r 18 Ñ
r 8,
and we may arrange for 18 and r 18 to be the limits of thickened domains
1
k and r for the quadratic-like restrictions of Pk . As P n p´1q P 1 and
1
k k k
Pkn p`1q P r 1 , it follows that P8
k
n p´1q P 1 and P n p`1q P
8 8
r 18 . Thus, P8
is birenormalizable, that is, pA8 , D8 q P Bp{q,m . l
Injectivity The time has come to show that the intertwining operations
pf, frq ÞÑ f O fr
p{q,m
are well defined.

Proposition 8.35 R ˆ : Bp{q,m Ñ Lp{q ˆ Lp{q is injective.
The relevant pullback argument is formalized as follows.
Lemma 8.36 Let P “ PA,D and Pr “ PA,

rD r where pA, Dq, pA, Dq P
r r
Bp{q,m . If
R ˆ pA, Dq “ R ˆ pA,
r Dq,
r
then there exists a quasiconformal homeomorphism ϕ : C Ñ C conjugating P

to Pr with B̄ϕ “ 0 almost everywhere on K
qP .
Proof We begin by once again restricting P and Pr to domains G Ą KP and

Gr Ą K r bounded by equipotentials. Our first goal is the construction of a
P
quasiconformal homeomorphism ϕ0 , which is illustrated in Figure 8.26 for
p{q “ 1{2 and m “ 3. Let Rpθ1 q, . . . , Rpθ2q q be the rays landing at ζ , enu-
merated in counterclockwise order so that the connected component K´1 Q ´1
of KP ztζ u lies between Rpθ1 q and Rpθ2 q; the component K`1 Q `1 then
lies between Rpθm`1 q and Rpθm`2 q. We label the remaining components of
KP ztζ u as K˘ 1 , . . . , Kq ´1 , so that Ki
1 ˘1 ´1 lies between the rays Rpθ2i `1 q,
Rpθ2i `2 q and similarly for K . The corresponding objects associated to Pr
`1
i
are similarly denoted with an added tilde.

It will be convenient to introduce further notation. Let Si Ă G be dis-
joint invariant sectors centred at Rpθi q, and let Li˘1 be the component of
Ť2q
P ´1 pGqzp j “1 Sj q containing K˘i . The thickening procedure yields left and
1
right quadratic-like restrictions
P : 1R Ñ R and P : 1 Ñ
and
Pr : R
r R and Pr :
r1 Ñ r1 Ñ

r .
By assumption, there exist hybrid equivalences hR between P |1 and Pr| r1 ,

R R
and h between P |1 and Pr| r 1 . We now replace the domains G and G by
r

G1 “ P ´1 pGq and G r We define the map ϕ0 on q ´1 Li Ă G1 ,
r 1 “ Pr´1 pGq.
Ť
i “0 ˘1
setting it equal to hR pzq for z P Li´1 , and h pzq for z P Li`1 .
Let Bi be the quadrilateral of G1 zP ´1 pG1 q contained in Si , and B ri its
1 ´ 1 1
counterpart in G zP pG q. We smoothly extend ϕ0 to map Bi Ñ B
r r r ri in
i
agreement with the previously specified values of ϕ0 on BpL´1 Y L`1 q and i
G P
G1 S4
´1
h
`1
K`
0 ϕ0 Pr
L0´1 1
L0`1
S1
S3 Sr4 G
r
K1
´1
r1
G
B1
hR ´1 `1
K
r0
`1
r0
L r0
´1 L `1
h Sr1 Sr3
K
r1
´1
Figure 8.26 Construction of the map ϕ0 in the case p{q “ 1{2 with m “ 3.
ϕ0 ˝ P “ Pr ˝ ϕ0 on the inner boundary of Bi . We now extend ϕ0 to the entire

sector Si X X by setting
ϕ0 pzq “ Prń ˝ ϕ0 ˝ P n pzq when P n pzq P Bi .
The quasiconformal homeomorphism ϕ0 : G1 Ñ G r 1 so defined conjugates

P to Pr on KR Y K , with B̄ϕ0 “ 0 almost everywhere on this set, sending KR
to K
r R and K to Kr . We further extend ϕ0 to a quasiconformal homeomor-
phism from G to G so that
r
ϕ0 ˝ P |BG1 “ Pr ˝ ϕ0 |BG1 .
As ϕ0 is a conjugacy between postcritical sets, there is a unique lift ϕ1 : G1 Ñ

r 1 agreeing with ϕ0 on KR Y K such that the following diagram commutes:
G
ϕ
G1 ÝÝÝ1ÝÑ Gr1
§ §
§ §
Pđ Pđ
r
ϕ
G ÝÝÝ0ÝÑ G
r
As in the proof of Proposition 8.32, we set ϕ1 pzq “ ϕ0 pzq for z in the annulus
GzG1 , and iterate the lifting procedure to obtain a sequence of quasiconformal
maps ϕn with uniformly bounded dilatation. In view of the density of K qP
in KP , the limiting map ϕ : G Ñ G conjugates P to P . As ϕn stabilises
r r
pointwise on K q P with ϕ|K “ hR and ϕ|K “ h by construction, it fol-
R
lows from Rickman’s Lemma (Lemma 1.20) that B̄ϕ “ 0 almost everywhere
on K
qP .
To conclude the proof of Proposition 8.35, we show that the conjugacy

just obtained is actually a hybrid equivalence: that any measurable invariant
line field on KP zK q P has support in a set of Lebesgue measure zero. In
view of Lemma 8.36, it follows from the standard considerations of param-
eter dependence in the Integrability Theorem (Theorem 1.30) that F “ pR ˆ
q´1 pA, Dq is the injective complex-analytic image of a polydisc Dk for
some k P t0, 1, 2u; see [MSS] and [MS]. On the other hand, F is compact
by Proposition 8.33, whence k “ 0 and F is a single point. l
Conclusions Setting
hp{q,m pc, crq “ pR ˆ q´1 pc, crq,
so that hp{q,m pc, crq “ pA, Dq if and only if PA,D “ Qc O Qrc , we obtain
p{q,m
the embeddings
hp{q,m : Lp{q ˆ Lp{q ÝÑ Bp{q,m Ă Cp{q,m ,
whose existence was asserted in Theorem 8.17. As observed in Section 8.4.2,
if P “ PA,D is birenormalizable and pA, Dq P C#R then p{q “ 0 or 1{2. It
follows by symmetry that R ˆ pA, Dq “ pc, c̄q for some c P C; conversely,
if R ˆ pA, Dq “ pc, c̄q then pA, Dq P C#R . Writing δ̄ for the antidiagonal
embedding C Q c ÞÑ pc, c̄q P C2 , we define
1{2,1 “ h1{2,1 ˝ δ̄ : L1{2 Ñ C#R X C1{2,1 ,
1{2,3 “ h1{2,3 ˝ δ̄ : L1{2 Ñ C#R X C1{2,3

and
0 “ h0 ˝ δ̄ : Mztrootu Ñ C#R X C0 .
These are the embeddings whose existence was asserted in Theorem 8.20.
Compatibility with the standard planar embeddings is a consequence of the
following result of Buff [Bu] (see also [BF2]).
Theorem 8.37 Let K1 and K2 be compact, connected, cellular sets in the

plane, and ϕ : K1 Ñ K2 a homeomorphism. If ϕ admits a continuous exten-
sion to an open neighbourhood of K1 such that points outside K1 map to points
outside K2 , then ϕ extends to a homeomorphism between open neighbour-
hoods of K1 and K2 .
Let us sketch the argument for the map 0 . It is easily verified from the
explicit expressions in [Mi3, p. 22] that for each ρ P Czt1u there is a unique
pair pA, Bq P R2 such that the corresponding polynomial in the normal form
(8.3) has a pair of complex conjugate fixed points with multipliers ρ and ρ̄, the
remaining fixed point having multiplier
|ρ ´ 1|2
ν “1´ .
2 Repρ ´ 1q
We may continuously label these multipliers as ρpA, Dq, ρ̄pA, Dq and
νpA, Dq for parameter values pA, Dq P R ˆ i R in a neighbourhood of
0 pMztrootuq; in particular, pA, Dq ÞÑ ρpA, Dq is a homeomorphism on
such a neighbourhood. It follows from Yoccoz’s Inequality (8.4) that
νpA, Dq ą 1, and therefore ρpA, Dq R r1, 8q, for pA, Dq in 0 pMztrootuq.
Similarly, ρp0 pcqq Ñ 1 as c Ñ root, and thus c ÞÑ ρp0 pcqq extends to a
embedding
ϒ : M Ñ Czp1, 8q,
which clearly commutes with complex conjugation.
We claim that ϒ ´1 : ϒpMq Ñ M admits a continuous extension meeting
the condition of Theorem 8.37. The idea is to allow renormalizations with dis-
connected Julia sets. We note that the rays RA,D p0q and RA,D p1{2q continue to
land at the same fixed point for pA, Dq in a neighbourhood of 0 pMztrootuq.
As before, we may construct left and right quadratic-like restrictions with con-
tinuously varying domains 1A,D . It is emphasized in Douady and Hubbard’s
original presentation [DH3] that straightening, while no longer canonical for
maps with disconnected Julia set, may still be continuously defined: it is
only necessary to begin with continuously varying quasiconformal homeomor-
phisms from the fundamental annuli A,D z1A,D to the standard annulus. We
thereby obtain a continuous extension to a neighbourhood of ϒpMztrootuq;
it is easily arranged that this extension commutes with complex conjugation,
so that it is trivial to obtain a further extension to an open set containing the
point 1.
9
Trans-quasiconformal surgery
In this chapter we shall refer to the classical Integrability Theorem of Morrey,

Bojarski, Ahlfors and Bers (see Section 1.4) as the qc Integrability Theorem in
order to distinguish it from its generalization formulated in Theorem 9.7. We
shall refer to the latter as the trans-qc Integrability Theorem or the David Inte-
grability Theorem. The qc Integrability Theorem is the key element that makes
quasiconformal surgery such a successful technique in holomorphic dynam-
ics. Recall that this remarkable result establishes conditions under which a
Beltrami differential μ can be integrated by a quasiconformal homeomorphism
φ. This means that φ must satisfy the Beltrami equation
Bφ “ μ Bφ,
or equivalently, the pullback of μ0 ” 0 under φ, is equal to the starting
Beltrami differential μ almost everywhere. The qc Integrability Theorem states
that φ exists as long as
||μ||8 :“ kμ ă 1, (9.1)
or equivalently, when the dilatation Kμ ă 8, where Kμ is the essential supre-
mum of the dilatation over all points in the domain.
It is for this reason that quasiregular maps are used as models when doing
qc surgery, since those are the ones that may admit invariant almost complex
structures satisfying (9.1).
One may ask whether it is possible to relax the condition slightly, at
the expense of obtaining an integrating map, which is not quasiconformal.
This question has a positive answer: integration is possible as long as the
dilatation is unbounded in a set of small enough area. Indeed, Guy David
in [Da] introduced a new class of Beltrami differentials, the David–Beltrami
differentials, which are integrable by David homeomorphisms (he called them
364
μ-homeomorphisms), a wider class of homeomorphisms than the quasiconfor-

mal class.
As we shall see, David homeomorphisms share some properties with qua-
siconformal homeomorphisms, obviously not all, but enough to make them
useful for surgery.
The David Integrability Theorem is a marvellous generalization of the qc
Integrability Theorem, which opens a wide range of possibilities for new
surgeries to be performed. In its short life, it has already served to prove many
theorems in holomorphic dynamics. In this chapter we give two applications.
The first one (Section 9.2), is Petersen and Zakeri’s result (Theorem 9.8)
extending the class of rotation numbers for which the Siegel disc of a quadratic
polynomial is a Jordan domain containing the critical point on the boundary.
The second construction (Section 9.3) is Haı̈ssinsky’s result (Theorem 9.18),
which provides David conjugacies between hyperbolic polynomials and
parabolic ones on their Julia sets.
9.1 David maps and David–Beltrami differentials

The main hypothesis in the qc Integrability Theorem is that the given Beltrami
differential μ has essential supremum }μ}8 ă 1 or equivalently bounded
dilatation. A wider class of Beltrami differentials is defined by the following
condition.
Definition 9.1 (David–Beltrami differential) A measurable Beltrami differ-

ential
dz
μ “ μpzq on a domain U Ă C
dz
with }μ}8 “ 1 is called a David–Beltrami differential if there exist constants
M ą 0, α ą 0 and 0 ă 0 ă 1 such that
α
Areaptz P U | |μpzq| ą 1 ´ uq ă Me´ for ă 0 . (9.2)
This condition is often referred to as the area condition. David–Beltrami

differentials can be defined on any subset of the sphere C;
p to do so one needs
on the one hand to take into account that μ is really a p´1, 1q form and on the
other hand to replace the Euclidean area by spherical area in the area condition.
Recall that a homeomorphism ϕ : U Ñ C belongs to the Sobolev space

p1,1q
Wloc pU q if its distributional partial derivatives exist and are locally integrable
in U . It follows from Theorem 1.10 that such maps are ACL (absolutely con-
tinuous on lines, see Section 1.3.2). Therefore the ordinary partial derivatives
366 Trans-quasiconformal surgery
Bz ϕ, Bz ϕ exist almost everywhere, and even more, ϕ is differentiable almost

everywhere.
Definition 9.2 (David homeomorphism) An orientation preserving homeo-
p1,1q
morphism ϕ : U Ñ C in Wloc pU q is called a David homeomorphism if the
induced Beltrami differential
Bz ϕ dz
μϕ :“
Bz ϕ dz
satisfies the inequality (9.2) or, equivalently, if there exist constants M ą 0,
α ą 0 and K0 ą 1 such that
Areaptz P U | Kϕ pzq ą Kuq ă Me´αK for K ą K0 , (9.3)
1`|μ pzq|
where Kϕ pzq “ 1´|μϕ pzq| (see Exercise 9.11).
ϕ
Remark 9.3 (The Sobolev condition can be replaced by ACL) In the defi-
nition of a David homeomorphism one can replace the condition of ϕ being in
p1,1q
Wloc pU q by the a priori weaker requirement of ϕ being ACL.
This is a consequence of the following lemma.
Lemma 9.4 Suppose ϕ : U ÝÑ C is an orientation preserving homeomor-
phism onto its image with partial derivatives a.e. and with a Beltrami coef-
ficient μϕ satisfying (9.2). Then the Jacobian Jac ϕ as well as the partial
derivatives Bz ϕ and Bz ϕ belong to L1loc . In particular, if ϕ is ACL and μϕ
1,1
satisfies (9.2), then ϕ P Wloc .
Proof By [LV, Lem. 3.3, p. 131] we have for every compact subset
KĂU
ż
Jac ϕ dx dy ď AreapϕpKqq,
K
so that Jac ϕ “ p|Bz ϕ|2´ |Bz ϕ|2 q P L1loc . Since |Bz ϕ| “ |μϕ ||Bz ϕ| and ϕ is
orientation preserving so that |μϕ | ă 1 a.e. we have
Jac ϕ Jac ϕ
|Bz ϕ|2 “ ď ,
1 ´ |μϕ |2 1 ´ |μϕ |
and
1 ´1
|Bz ϕ| ď pJac ϕq 2 p1 ´ |μϕ |q 2 .
1
By the growth condition (9.2), the function p1 ´ |μϕ |q 2 is locally integrable
(see Exercise 9.1.3). It follows that Bz ϕ and Bz ϕ are in L1loc (see Exercise 9.1.4).
The same is true for Bz ϕ since |Bz ϕ| ď |Bz ϕ|.
If ϕ is ACL and μϕ satisfies (9.2), then the partial derivatives Bz ϕ and Bz ϕ

are in L1loc and hence coincide with the distributional derivatives of ϕ (see
1,1
Lemma 1.10). Hence ϕ P Wloc .
David homeomorphisms differ from quasiconformal maps in many respects.

For example, the inverse of a David homeomorphism is not necessarily David.
However, they share many convenient properties (such as compactness, see
[Tu2]), some of which are important when performing surgery.
Proposition 9.5 (David Removability of quasiarcs) 1 Suppose ϕ : U Ñ C is

an orientation preserving homeomorphism onto its image. Suppose there exists
1,1
a quasiarc Ă U such that the restriction ϕ|pU z q belongs to Wloc pU zq and
1,1
its partial derivatives satisfy (9.2) locally in U . Then ϕ belongs to Wloc pU q.
Proof Clearly the problem is local. Let z P be arbitrary. After pre-

composing ϕ with a quasiconformal homeomorphism if necessary, we may
assume that z P “ I Ă R and that U “ I 1 ˆ i pá, aq, where a ą 0 and
I 1 is an open interval with z P I Ă I 1 . By hypothesis, the ordinary partial
derivatives g “ Bz ϕ and h “ Bz ϕ exist a.e. in U , are L1loc functions in U by
Lemma 9.4 and are equal to the distributional partial derivatives Bϕ and Bϕ,
locally in U zI Ą U zI 1 . Thus we need only to show that g and h are also the
distributional partial derivatives of ϕ locally in U . In other words, we need to
show that
żż żż
ψ ¨ g dxdy “ ´ Bz ψ ¨ ϕ dxdy
U U
for any test function ψ P Cc8 (see equation (1.12)).

For any 0 ă ă a, set A :“ pá, ´q Y p, aq and Ac :“ p´, q. Let 1A
denote the characteristic function that takes the value 1 in A Ă R and 0 in RzA.
Then, for any 0 ă ă a, we split the integral by inserting the characteristic
functions 1A and 1Ac (as functions of y)
żż żż
ψ ¨ g dxdy “ ψ ¨ p1A ` 1Ac q ¨ g dxdy
U U
żż żż
“ ψ ¨ 1A ¨ g dxdy ` ψ ¨ 1Ac ¨ g dxdy.
U U
The products ψ ¨ 1A and ψ ¨ 1A are no longer test functions. However, since
the characteristic functions can be approximated by C 8 bump functions, the
products ψ ¨ 1A and ψ ¨ 1A can be approximated by test functions for which
1 We thank Carsten Lunde Petersen for providing this proposition and its proof.
the partial derivatives exist. With this in mind, and knowing that g “ Bϕ locally
in I 1 ˆ i A , the following equations can be justified:
żż żż
ψ ¨ 1A ¨ g dxdy “ ´ Bpψ ¨ 1A q ¨ ϕ dxdy
U U
żż żż
“´ Bz ψ ¨ 1A ¨ ϕ dxdy ` ψ ¨ B p1A q ¨ ϕ dxdy.
U U
The distributional derivative of the characteristic functions equal the (shifted)

Dirac delta şdistribution at the discontinuity point. The Dirac delta distribution
δα satisfies J f ptqδα ptqdt “ f pαq for any continuous function f and any α P
J Ă R. Hence we obtain
żż
ψ ¨ B p1A q ¨ ϕ dxdy
U
żż
“ ψ ¨ 12 i pδ´ ´ δ q ¨ ϕ dxdy
U
ż
“ 12 i pψpx, ´q ¨ ϕpx, ´q ´ ψpx, q ¨ ϕpx, qq dx ÝÑ 0.
I1 Ñ0
Putting everything together, and using that ψ is C 8 pU q with compact support,

ϕ is continuous and g is in L1loc pU q, we conclude
żż żż żż
ψ ¨ g dxdy “ ψ ¨ 1A ¨ g dxdy ` ψ ¨ 1Ac ¨ g dxdy ÝÑ
U U U Ñ0
żż
´ Bz ψ ¨ ϕ dxdy,
U
and hence g is the distributional B derivative of ϕ locally in U . Similarly, it can

be shown that h is the distributional B derivative of ϕ.
Proposition 9.6 (Absolute continuity) Every David homeomorphism ϕ and

its inverse are absolutely continuous, i.e. for a measurable set E Ă U
AreapEq “ 0 if and only if AreapϕpEqq “ 0.

The following is a generalization of the qc Integrability Theorem, which

asserts that David–Beltrami differentials can always be integrated by David
homeomorphisms [Da].
Theorem 9.7 (David Integrability Theorem) Let μ be a David–Beltrami

differential on a domain U Ă C. p Then there exists a David homeomorphism
ϕ : U Ñ V , whose complex dilatation μϕ coincides with μ almost everywhere.
The integrating map is unique up to post-composition by a conformal map,
i.e. if ϕr : U Ñ Vr is another David homeomorphism such that μϕr “ μ almost
everywhere, then ϕr ˝ ϕ ´1 : V Ñ Vr is conformal.
In the below sections we shall see two examples of trans-quasiconformal

surgery where the constructed Beltrami differential μ satisfies }μ}8 “ 1. Both
examples follow the same scheme, which we sketch below.
Let U Ă C be a domain bounded by a Jordan curve γ which is a quasicircle.
(In the first example, γ will actually be the unit circle, while in the second
example, γ is the boundary of an immediate basin of attraction for a poly-
nomial.) The model that will be constructed is a continuous map F : C Ñ C,
holomorphic on CzU and satisfying
F “ H ´1 ˝ R ˝ H on U ,
where R : H pU q Ñ H pU q is holomorphic and H is a David homeomorphism.

Let μ be the Beltrami form on C obtained by defining μ “ H ˚ μ0 on U and
then spreading it by the dynamics of F .
The first delicate step is to prove that such μ is a David–Beltrami differen-
tial, first showing that F has large dilatation only in regions of small area (in
the sense of (9.2) and (9.3)), and then proving, as we spread μ by the dynamics,
that further preimages of these small sets do not add too much area. This step
needs to be done in each special case, since it depends clearly on the maps F
and H .
Once this is proven, it follows from the David Integrability Theorem that μ
can be integrated by a David homeomorphism, say ϕ.
The second delicate step is then to show that the composition map
ϕ ˝ F ˝ ϕ ´1
is holomorphic. In qc surgery this step is immediate (see Lemma 1.39) from

Weyl’s Lemma. In the David case, however, another type of argument is
needed, since the hypothesis of quasiconformality is no longer satisfied. In
the next sections this problem is handled in Proposition 9.14 and Lemma 9.25.

9.1.1 Check that the conditions on the decay of area expressed in the inequal-
ities (9.2) and (9.3) are equivalent.
9.1.2 By using Theorem 1.18, prove that the composition of a quasiconformal
map with a David map is a David map.
9.1.3 Show that if a Beltrami coefficient μ satisfies the area condition (9.2),
then the measurable function p1 ´ |μ|q´1 belongs to L1loc , i.e. it is
locally integrable.
9.1.4 Apply Hölder’s inequality (stated in Exercise 1.3.1) to show that if f
and g are in L1 pU q, then pf ¨ gq1{2 is in L1 pU q.
9.2 Siegel discs via trans-quasiconformal surgery

Carsten Lunde Petersen
This section is based on the results in [PZ]. To perform the Douady–Ghys qua-
siconformal surgery to Blaschke products (see Section 7.2 and Section 7.2.2),
the rotation numbers considered must be of bounded type. In this section we
extend the construction to a larger class of rotation numbers.
In order to do so we need to go beyond the class of quasiconformal maps and
enter the realm of David maps, i.e. maps which do not have uniformly bounded
dilatation, but whose dilatation is unbounded only in a very controlled manner.
This leads however to a new set of complications relative to those encountered
in the Douady–Ghys surgery.
To be more precise: given θ P r0, 1s, let λθ “ e2π i θ P S1 and Pθ pzq “
λθ z ` z2 . Denote by Aθ p8q the basin of attraction of 8, by Kθ the filled Julia
set, that is the compact full complement of Aθ p8q, by Jθ the Julia set, which
is the common boundary of the two, and when Pθ is linearizable around 0, let
θ be the maximal Siegel disc of Pθ .
The ultimate outcome of our labour will be the following theorem, which is
the main result in [PZ].
Theorem 9.8 (Main Theorem) There exists an explicit full-measure subset

E Ă r0, 1szQ such that for all θ P E, the Julia set Jθ is locally connected, has
planar area zero, and the boundary of θ is a Jordan curve passing through
the critical point. (The set E is defined in equation (9.6).)
The primary goal of this section will be a trans-quasiconformal surgery

result, which when combined with other theorems will yield Theorem 9.8.
Before we formulate this result it is however convenient to recall the Douady–
Ghys cut and paste surgery and introduce some auxiliary notation.
9.2 Siegel discs via trans-quasiconformal surgery 371
9.2.1 The Blaschke model

Given an irrational number 0 ă θ ă 1, consider the cubic Blaschke product
ˆ ˙
z´3
f :“ fθ : z ÞÑ e 2π i t pθ q 2
z , (9.4)
1 ´ 3z
which has superattracting fixed points at 0 and 8 and a double critical point
at z “ 1. Here 0 ă tpθ q ă 1 is the unique parameter for which the critical
circle map f |S1 : S1 Ñ S1 has rotation number θ . By a theorem of Poincaré
the restriction is semi-conjugate to the rigid rotation Rθ pzq “ λθ z by a unique
continuous map h “ hθ : S1 Ñ S1 fixing 1. And by a theorem of Yoccoz
[Yo2], fθ is minimal, i.e. has no wandering intervals, so that h is in fact a
conjugacy, i.e. a homeomorphism (see also Section 3.2.2).
Let H : D Ñ D be any homeomorphic extension of h and define
#
f pzq if |z| ě 1,
F pzq “ Fθ pzq :“ Fθ,H pzq :“ (9.5)
pH ´1 ˝ Rθ ˝ H qpzq if |z| ă 1.
It is easy to see that F is a degree 2 topological branched covering of

the sphere which is holomorphic outside of D and is topologically conju-
gate to a rigid rotation on D. By imitating the polynomial case, we define
AF “ AFθ p8q as the immediate basin of 8, the ‘filled Julia set’ KF “ KFθ,H
as the compact, connected and full complement of AF and the “Julia set” JF
of F as the common topological boundary of KF and AF :
JF “ BKF “ BAF .
Note that although the homeomorphism H is by no means canonical, neither

JF nor KF nor any of the definitions to follow depend on a particular choice of
H . This is simply because the constructions do not involve the values of F on
D. The main purpose of introducing F for the following constructions is to for-
get about the f -preimages of CzD in D. A particular choice of H is only used
in the final step of the proofs, where we need H to have some regularity. In the
quasiconformal surgery, H should be a quasiconformal homeomorphism and,
in our construction here, a David homeomorphism. The following theorem is
proved in [Pe1].
Theorem 9.9 (Local connectivity and zero measure) For every irrational
θ P r0, 1szQ the ‘Julia set’ J pFθ q is locally connected and of planar Lebesgue
measure zero.
There are by now several other similar constructions synthesizing other

maps by quasiconformal surgery, e.g [Z1] and [YZ] for variants in the case
of cubic polynomials and quadratic rational maps, [Ch] for horn maps and
[KZ] for sine maps.
Recall that the idea in the Douady–Ghys surgery is as follows. (1) Choose
H to be quasiconformal whenever it is possible. (2) When this is possible let μ
denote the unique F “ Fθ,H invariant Beltrami differential which coincides
with the standard one μ0 “ 0 dz dz on JF Y AF and with H μ0 on D. (3)
˚
Let φ “ φθ,H : C p ÝÑ C p be the unique integrating quasiconformal homeo-

morphism for μ with φp0q “ 0, φp8q “ 8 and φ ˝ F “ φ 2 ` Opzq near 8
(note that μ “ μ0 on a neighbourhood of 8 so that F and φ are holomorphic
near 8). (4) Let P “ φ ˝ F ˝ φ ´1 and note that P is holomorphic (by Weyl’s
Lemma (Theorem 1.14)) thus a quadratic rational map, which by the choice
of normalization equals Pθ . Note also that φpDq “ θ is a quasidisc, whose
boundary contains the critical point φp1q “ ´λθ {2.
However, H can be chosen to be quasiconformal only when θ is of bounded
type, as we shall see later in Section 9.2.3.
Consider the continued fraction expansion (see Section 3.2.1)
1
θ “ ra1 , a2 , a3 , . . .s “
1
a1 `
1
a2 `
a3 ` ¨ ¨ ¨
with an “ an pθ q P N. The nth convergent of θ is the irreducible fraction

pn {qn :“ ra1 , a2 , . . . , an s.
The number θ is said to be of bounded type if tan u is a bounded sequence.
The main goal of this section is to extend the surgery above to the class of
rotation numbers
?
E “ tra1 , a2 , . . .s| log an “ Op nq as n Ñ 8u (9.6)
at the expense of relaxing the boundedness condition on the F -invariant

Beltrami differential. This leads us to the David Integrability Theorem (Theo-
rem 9.7) for unbounded Beltrami differentials with exponential decay of areas
supporting large dilatation, which was explained in the previous section.
Theorem 9.10 (David conjugation) For every θ “ ra1 , a2 , . . . , an , . . .s P E

there exists a David extension H “ Hθ : D ÝÑ D of hθ : S1 ÝÑ S1 and a
David homeomorphism ϕ “ ϕθ : C p ÝÑ C
p conjugating Fθ,H to Pθ .
So why is this result so much better than the one obtained by the Douady–
Ghys surgery? The easy answer is that E is a full measure subset of r0, 1s,
where as the set of bounded type numbers is a subset of zero measure. Indeed,
the following theorem due to Khinchin [Kin] characterizes the asymptotic
growth of the sequence tan u for random irrational numbers.
Theorem 9.11 (Asymptotic growth of coefficients) Let ψ : N Ñ R be a

given positive function.
(i) If 8 1
ř
n“1 ψ pnq ă `8, then for almost every irrational 0 ă θ ă 1 there are
only finitely many n for which an pθ q ě ψpnq.
(ii) If 8 1
ř
n“1 ψ pnq “ `8, then for almost every irrational 0 ă θ ă 1 there are
infinitely many n for which an pθ q ě ψpnq.
?
By part (i), taking ψpnq “ eM n, we obtain the following.
Corollary 9.12 The set E has full measure in r0, 1s.
Thus it is a worthwhile extension. But where does the restriction θ P E come

into the picture? It turns out we only know how to construct a David extension
Hθ of hθ whenever θ P E (see Theorem 9.15 below). We even conjecture that
such an extension exists if and only if θ P E.
However, we can prove independently, that whenever such an extension
H exists, then the unique Fθ,H -invariant Beltrami differential μθ , which
coincides with μ0 on the closure of AF and with H ˚ pμ0 q on D is also a
David differential. To see this, let f “ fθ and let g0 “ g0,θ : D Ñ U0 denote
the unique branch of f ´1 on D with values outside D and image U0 (see
Figure 9.1).
Define
G “ tg : D Ñ CzD | g is a branch of f ń and f n´1 ˝ g “ g0 u,
D U0
g0
Figure 9.1 The set U0 and the branch g0 of f ´1 .
that is, G consists of all the inverse branches of iterates of f factoring through
g0 , i.e. g “ gp ˝ g0 , where gp is a branch of f ´pn´1q on U0 . Note that g0 extends
univalently to a neighbourhood of Dztf p1qu and that any of the maps gp extends
univalently to a neighbourhood of U0 .
Define a Borel measure ν supported on D by
ÿ
νpEq :“ AreapEq ` AreapgpEqq,
g PG
for any Borel set E Ă D. That is ν is the sum of the Lebesgue measure on D
and of all the pushforwards of Lebesgue measure on each of the components
of the interior of KF , i.e. on nPN F ń pU0 q. Then ν is a finite measure since
Ť
KF is compact and evidently ν is also absolutely continuous with respect to

Lebesgue measure. Observe that if E denotes the set where the dilatation of
H is larger than 1 ´ ε, then νpEq is the area of the region where μθ has norm
larger than 1 ´ ε, since μθ is the spreading of H ˚ μ0 under the dynamics of F ,
which is holomorphic outside D.
However, much more is true.
Theorem 9.13 (Power law) The Borel measure ν is dominated by a universal

power of Lebesgue measure. That is, there exists a universal constant 0 ă β
ă 1 and a constant C (depending on θ ) such that for any Borel set E Ă D,
νpEq ď CpAreapEqqβ .
Thus it follows that if H is David, then μθ is also David.

We shall motivate later why it is not obvious that μθ is a David–Beltrami
differential, when Hθ is a David map. Also we shall comment on the principal
idea underlying the proof above. For a comprehensive treatment and complete
proof of the theorem we refer the reader to the paper [PZ].
If we have obtained an F -invariant David–Beltrami differential μθ , then the
David Integrability Theorem provides a unique normalized integrating homeo-
morphism ϕ “ ϕθ : C p ÝÑ C, p ϕp0q “ 0, ϕp8q “ 8 and f ˝ ϕ “ ϕ 2 ` Opzq.
It would then be nice if P “ ϕ ˝ F ˝ ϕ ´1 was holomorphic. And indeed it is.
Proposition 9.14 (Straightening) In the notation above P “ ϕ ˝ F ˝ ϕ ´1 is

holomorphic and thus equals Pθ .
In contrast to the classical quasiconformal surgery, where no such consid-

eration occurs, the basic difficulties are that the inverse of a David homeo-
morphism need not be a David homeomorphism and the absence of a David
Gluing Lemma similar to the Quasisymmetric/Quasiconformal Gluing Lemma
as Theorem 1.19.
Proof P is a degree 2 branched covering of the sphere with P ´1 p8q “ 8,

so it will be a quadratic polynomial once we show that it is holomorphic.
We will heavily use the uniqueness statement in the David Integrability
Theorem. Observe that the hypothesis of the Proposition imply that the two
following diagrams commute:
θR F
pD, μ0 q ÝÝÝÝ Ñ pD, μ0 q pC, μq ÝÝÝÝÑ pC, μq
İ İ § §
ϕđ
§ § § §ϕ
H§ §H đ
F P
pD, μq ÝÝÝÝÑ pD, μq pC, μ0 q ÝÝÝÝÑ pC, μ0 q
Note that the map ϕ ˝ F also integrates the David–Beltrami differential μ.

The proof consists in showing that ϕ ˝ F is locally ACL (or that it belongs to
1,1
Wloc , see Remark 9.3) in every open set U Ă C where F is injective, and
hence is a David integrating map. Indeed, suppose this is true. Then, both
ϕ |U and pϕ ˝ F q |U are David maps integrating the same Beltrami differen-
tial μ. Hence, by the uniqueness statement in Theorem 9.7, the composition
P “ ϕ ˝ F ˝ ϕ ´1 must be conformal on ϕpU q. It follows that P is locally con-
formal around every point, except for a discrete number of points, and hence
it is holomorphic. This discrete set is the image of the set of critical points
of F .
1,1
Hence we need to show that ϕ ˝ F is locally ACL or belongs to Wloc in
every open set U Ă C, where F is injective. If U belongs to CzD, then F is
1,1
holomorphic and therefore ϕ ˝ F is the composition of a map in Wloc and a
1,1
holomorphic one, hence in Wloc (see [Ah5, Lem. 3]). If U belongs entirely to
D then we have ϕ ˝ F “ ϕ ˝ H ´1 ˝ Rθ ˝ H in U . Again by the uniqueness of
David integrating maps, ϕ ˝ H ´1 is conformal since both H and ϕ integrate μ
on D. Hence ϕ ˝ F is again the composition of a conformal map with a map in
1,1 1,1
Wloc , thus it is also in Wloc .
1,1
Since ϕ ˝ F belongs to Wloc pC˚ zS1 q with Beltrami differential satisfy-
ing (9.2) and S is a quasiarc, it follows from Proposition 9.5 that ϕ ˝ F P
1
1,1
Wloc pC˚ zt1uq.
Finally let us return to the issue of existence of David extensions Hθ of hθ .
Theorem 9.15 (David extension) Let f : S1 Ñ S1 be a critical circle map

whose rotation number θ “ ra1 , a2 , a3 , . . .s belongs to the arithmetical class
E. Then the normalized linearizing map h : S1 Ñ S1 , which satisfies h ˝ f “
Rθ ˝ h, admits a David extension H : D Ñ D so that
# ˇ ˇ +
ˇ ˇ BH pzq ˇ α
Area z P D ˇ ˇ ˇ ą 1 ´ ε ď M e´ ε for all 0 ă ε ă ε0 .
ˇ ˇ ˇ
ˇ BH pzq ˇ
Here M ą 0 is a universal constant, while in general the constant α ą 0

?
depends on lim supnÑ8 plog an q{ n and the constant 0 ă ε0 ă 1 depends
on f .
Summarizing the proof of Theorem 9.8 Let θ P E be arbitrary. Then by Theo-
rem 9.15 we can choose a David extension H : D Ñ D of the conjugacy hθ :
S1 Ñ S1 conjugating fθ |S1 to Rθ |S1 with H p0q “ 0. Let μ denote the unique
Fθ “ Fθ,H invariant Beltrami differential, which equals μ0 on AFθ p8q, and
which equals H ˚ pμ0 q on D, where μ0 “ 0 dz dz denote the standard Beltrami
differential. Then μ is a David–Beltrami differential by Theorem 9.13. Let
ϕθ : Cp ÑC p denote the integrating David homeomorphism for μ, normalized
by ϕθ p8q “ 8, ϕθ p0q “ 0, ϕθ p1q “ éi 2π θ {2. Then from Proposition 9.14, it
follows that P “ ϕθ ˝ Fθ ˝ ϕθ´1 “ Pθ and ϕθ pDq “ θ , and by Theorem 9.9,
Jθ “ ϕθ pJ pFθ qq is locally connected and of measure zero.
We refer the reader to the appendix of the paper [PZ] for a complete proof
of this theorem. The proof is based in part on an ingenious geometric con-
struction by Yoccoz in the unpublished manuscript [Yo2]. It is a reminiscence
of Carleson squares, but it is specially tailored to the particular geometry of
critical circle maps.
In the remainder of the section we shall survey this result.
9.2.2 The Yoccoz Conjugacy Theorem: idea of the Construction

The idea of the proof is to construct two dynamically defined graphs in the disc
linking the backward orbit of the critical point say 1 for f to the backward orbit
for 1 under Rθ , in such a way that the conjugacy hθ of f to Rθ extends to a
‘piecewise affine map’ on the graphs. In fact we shall lift the problem to R
and the upper half plane by the exponential map. In the upper half plane the
extension will be piecewise affine. The graphs will be so that the piecewise
affine extension extends quasiconformally to each cell of the associated cell
decompositions of the upper half plane. If θ is not of bounded type then the
supremum of the dilatations of these extensions necessarily diverge to infinity
when we consider cells closer and closer to R.
The ingeniousness comes into choosing the right graphs and the right qua-
siconformal extensions so that we may use as little quasiconformal distortion
(dilatation) as possible and also use it in a controlled manner.
To be slightly more precise the cells in these decompositions have bounded
geometry and are labelled by an integer, called their generation or level.
The union of the cells of any generation n is a belt separating the previous
generation from R and the following generations. The distance between the
nth belt and R tends to zero at least geometrically with n and thus (by bounded
geometry) the area of the nth-level belt tends to zero geometrically with n. By
making the right constructions the quasiconformal extension of the piecewise
affine boundary map on each level n cell can be chosen to have a dilatation
bounded by Cp1 ` plog an`1 q2 q for some universal constant C depending only
on the derivative f 1 .
9.2.3 Two cell decompositions for the upper half plane

Let f : S1 Ñ S1 be a critical circle map with a critical point at 1 and
irrational rotation number θ “ ra1 , a2 , a3 , . . .s with convergents pn {qn “
ra1 , a2 , . . . , an s, and let R “ Rθ denote the corresponding rigid rotation.
Using the canonical projection e2π i x : R Ñ S1 we identify S1 with the quo-
tient space T “ R{Z and denote the induced maps from T to T by f and R as
before.
We set xn :“ f ń p0q and x 1 n :“ R ń p0q for all n P Z.
The natural partitions n pf q and n pRq of the circle with dividing points
txj |0 ď j ă qn ` qn`1 u and tx 1 j |0 ď j ă qn ` qn`1 u respectively are for
instance used when studying renormalization of circle maps. However, writing
In “ r0, xqn s and I 1 n “ r0, x 1 qn s for n P N, then
1 |I 1 n | 1
ď 1 ď ,
1 ` an`1 |I n´1 | an`1
whereas by the Herman–Świa̧tek Theorem (see Theorem 3.17 or [Pe2]) there

exists some constant C “ Cpf 1 q such that
1 |In |
ď ď C.
C |In´1 |
If the sequence an is unbounded then this shows that h “ hθ can not

be quasi-symmetric. This is where the ‘only if’ conclusion of the Herman–
Świa̧tek Theorem comes from.
However, Yoccoz saw that if we instead consider the partitions with dividing
points
Qn :“ txj | 0 ď j ă qn u and Q1 n :“ tx 1 j | 0 ď j ă qn u
on T, so that Q0 “ Q1 0 “ t0u, then neighbouring intervals are always of alike

lengths for both partitions (see (9.9) below). To see this, first let us note that
¨ ˛ ¨ ˛
ď ď
T Qn “ ˝ sxj `qn´1 , xj r‚Y ˝ sxj , xj `qn ´qn´1 r‚.
0ďj ăqn ´qn´1 0ďj ăqn´1
Thus xj and xk , with j ă k, are adjacent in Qn if and only if either k “

j ` qn´1 and 0 ď j ă qn ´ qn´1 , or k “ j ` qn ´ qn´1 and 0 ď j ă qn´1 .
It follows that in the first case
rxk , xj s X Qn`1 “ txk , xk `qn , xk `2qn , . . . , xk `pan`1 ´1qqn “ xj `qn`1 ´qn , xj u,

(9.7)
and in the second case
rxj , xk s X Qn`1 “ txj , xj `qn , xj `2qn , . . . , xj àn`1 qn “ xk `qn`1 ´qn , xk u.

(9.8)
As a result, we see that xj and xk , with j ă k, are adjacent in both Qn and

Qn`1 if and only if an`1 “ 1, 0 ď j ă qn ´ qn´1 and k “ j ` qn´1 . Since f
and R are conjugate, the similar statement holds for Q1 n
We lift the sets Qn and Q1 n in T to the two sets in R, which are invariant
under the translation z ÞÑ z ` 1:
rn :“ Qn ` Z
Q and Q
Ă1 n :“ Q1 n ` Z.
In comparison with the partition n´1 pf q we have removed the intervals

which are short for R, when an is large by uniting it with its neighbouring
interval across the common end point xj , where qn ď j ă qn ` qn´1 . The
gain is that now as before any two adjacent intervals in R Q
rn have lengths
comparable up to a bound which is asymptotically universal [PZ, Lem. 5.1],
" *
|I | ˇˇ
max ˇ I, J are adjacent in R Qn — 1,
r (9.9)
|J |
and any two intervals I and J in R Q

Ă1 n (including neighbouring ones)
satisfy
1{2 ă |I |{|J | ă 2.
We now proceed to construct the graphs. For n ě 0 and x P Q

rn , let
1
Mn pxq :“ pxr ´ x q,
2
where xr and x are the points in Q

rn immediately to the right and left of x.
Evidently Mn pxq ě Mn`1 pxq. Observe that M0 pxq “ 1 for all x P Q r0 “ Z.
Define
zn pxq :“ x ` i Mn pxq n ě 0, x P Q
rn .
Using the sequence tzn u, we shall define an embedded graph in the upper
halfplane H as follows: The vertices of are the points tzn pxq | n ě 0 and x P
Qrn u. Note that zn pxq “ zn`1 pxq if and only if Mn pxq “ Mn`1 pxq, in which
case the corresponding vertex of is doubly labelled. The edges of are the
vertical segments
trzn pxq, zn`1 pxqs | n ě 0 and x P Q

rn with Mn pxq ‰ Mn`1 pxqu,
as well as the non-vertical segments
trzn pxq, zn pyqs | n ě 0 and x, y are adjacent in Q

rn u.
By a cell of we mean the closure of any bounded connected component of

H . Any cell γ of is uniquely determined by a pair of adjacent points
x ă y in Q rn with the property that either Mn pxq ‰ Mn`1 pxq or Mn pyq ‰
Mn`1 pyq. The integer n ě 0 will be called the level of γ , or we say that γ is an
n-cell. The top of the n-cell γ is formed by the non-vertical edge rzn pxq, zn pyqs
while its bottom is formed by the union of non-vertical edges
rzn`1 pt0 q, zn`1 pt1 qs Y rzn`1 pt1 q, zn`1 pt2 qs Y ¨ ¨ ¨ Y rzn`1 ptk ´1 q, zn`1 ptk qs,
where the points x “ t0 ă t1 ă ¨ ¨ ¨ ă tk “ y form the intersection rx, ys X

Qrn`1 . The sides of γ are formed by the vertical edge rzn pxq, zn`1 pxqs (which
collapses to a single point if Mn pxq “ Mn`1 pxq) as well as rzn pyq, zn`1 pyqs
(which similarly collapses to a single point if Mn pyq “ Mn`1 pyq). If k “ 1 so
that x, y are also adjacent in Qrn`1 , then γ is either a triangle or a trapezoid.
Otherwise k ě 2 and by (9.7) or (9.8), γ is a pk ` 3q-gon, where k is either
an`1 or an`1 ` 1.
The bounds (9.9) imply that the cells of have ‘bounded geometry’ in the
sense that there is a constant C ą 1 such that the top, bottom and sides of any
n-cell γ of have lengths comparable up to C. Moreover, the slopes of non-
vertical edges of γ are bounded by C. The constant C is even asymptotically
universal.
For every n P N the union of the n-cells, which is invariant under z ÞÑ z ` 1,
form a connected set Sn converging geometrically to R in the sense (for proofs
see [PZ, Lem. 2.7 and Lem. 5.2]) that there exists C ą 1 and σ1 , σ2 with 0 ă
σ1 ă σ2 ă 1 such that
" ˇ σn *
ˇ 1 n
Sn Ă z “ x ` i y ˇ ď y ď Cσ2 . (9.10)
C
Completely analogously we can construct a graph 1 Ă H for the rigid rotation
R with cells generically denoted γ 1 (compare with Figure 9.2). In analogy with
the critical circle map case above the cells of 1 have ‘bounded geometry’:
there is a universal constant C ą 1 such that the top, bottom, and sides of any
cell γ 1 of 1 have lengths comparable up to C. Moreover, the slopes of non-
vertical edges of γ 1 are bounded by 1{2.
n´1
n`1
xq1 1
x3q 1
x2q xq1 0 xq1 n xq1 ´q xq1
n´1 ´qn´2 n´1 n´1 n´1 n n´1 n´2
Figure 9.2 The embedded graph 1 for the rigid rotation with selected cells and points
in Q1n . In this picture an “ 3 and an`1 “ 4. The labels on cells denote their level.
9.2.4 Constructing the extension

Let h : R Ñ R be the lift to R of the conjugacy h normalized by hp0q
r r “ 0.
Note that h fixes the integer points and hpQn q “ Qn for all n ě 0. We shall
r r r r 1
extend hr to a homeomorphism Hr between the embedded graphs and 1

by mapping each vertex of to the corresponding vertex of 1 and each
edge of affinely to the corresponding edge of 1 . Strictly speaking, for each
n ě 0 and x P Q rn , we define Hr pzn pxqq :“ z1 phpxqq.
n
r Then rz, ws is an edge
of if and only if rHr pzq, Hr pwqs is an edge of 1 . Thus we can extend Hr
»
further to a homeomorphism ÝÑ 1 by mapping each such edge rz, ws
affinely to rHr pzq, Hr pwqs. Note that Hr defined this way is the identity on the
horizontal line R ` i so we can define Hr pzq “ z for all z P H with Impzq ě 1.
It is easy to check that for each cell γ of , the boundary Bγ is mapped
by Hr homeomorphically and edgewise affinely onto the boundary Bγ 1 of a
unique cell γ 1 of 1 . Moreover, as the angles of interior corners of cells are
(uniformly) bounded from below, the piecewise affine homeomorphism Hr
from the boundary of one cell γ of to the boundary of the corresponding cell
γ 1 of 1 is quasisymmetric and thus has a quasiconformal extension. The key
issue is how much quasiconformal distortion is needed. The following result

due to Yoccoz is the key estimate.
Theorem 9.16 (Key estimate) There exists a constant C ą 0 with the fol-
lowing property: for any n-cell γ of , the edgewise affine boundary home-
omorphism Hr : Bγ Ñ Bγ 1 extends to a quasiconformal homeomorphism Hr :
γ Ñ γ 1 whose dilatation is at most Cp1 ` plog an`1 q2 q. The constant C is
asymptotically universal.
Assuming this result for the moment, let us show how Theorem 9.10 follows.
Proof of Theorem 9.10 Consider the extension Hr : H ÝÑ H, obtained

by gluing the various extensions to cells given by Theorem 9.16. Clearly
Hr is ACL and is chosen so that Hr pz ` 1q “ Hr pzq ` 1 for all z P H. Since
?
log an “ Op nq by the assumption, there is a constant C1 ą 0 and an integer
N1 ě 1, both depending on θ , such that 1 ` plog an`1 q2 ď C1 n whenever
n ą N1 .
By Theorem 9.16, there is a universal constant C2 ą 0 and an integer
N2 ě 1 depending on f such that the dilatation KHr in the interior of any n-cell
of is at most C2 p1 ` plog an`1 q2 q whenever n ą N2 .
Finally, by Equation (9.10), there is a universal constant C3 ą 0, a constant
0 ă σ2 ă 1, and an integer N3 ě 1 depending on f such that if n ą N3 ,
8
ď
tγ | γ is an m -cell of u Ă tz P H | 0 ă Impzq ď C3 σ2n u.
m“n
Set N :“ maxtN1 , N2 , N3 u and define

K0 :“ maxtKHr pzq | z belongs to the interior of an m-cell of with m ď N u.
If KHr pzq ą K ą K0 , then either z P (which has Lebesgue measure zero),
or else z belongs to the interior of an n-cell of with n ě N , so that K ă
KHr pzq ď C1 C2 n, or equivalently n ą K{pC1 C2 q. It follows that
K log σ2
C C2 K
Areatz P H | 0 ď Repzq ď 1 and KHr pzq ą Ku ď C3 σ2 1 “ C3 e C1 C2 .
The exponential map z ÞÑ e2π i z does not change the dilatation and has norm
of the derivative bounded by 2π when restricted to the upper halfplane H.
Therefore, the induced ACL homeomorphism H : D Ñ D satisfies
log σ2
K
Areatz P D | KH pzq ą Ku ď 4π 2 C3 e C1 C2
whenever K ą K0 . It follows that H is a David homeomorphism as in
(9.3), with M “ 4π 2 C3 , α “ p´ log σ2 q{pC1 C2 q, and K0 defined as above.
Moreover, M is universal, α depends on C1 (which in turn depends on

?
lim supnÑ8 plog an q{ n), and K0 depends on f and θ . l
The remainder of the section is dedicated to sketch the idea and the main
steps of the proof of Theorem 9.16. By bounded geometry, if k is small say
less than 5, then γ and γ 1 are uniformly quasiconformally equivalent with a
constant depending only on the real bounds (9.9).
9.2.5 Near parabolic

For critical circle mappings f : S1 ÝÑ S1 Yoccoz proved further the existence
of a bound C “ Cpf 1 q (the Yoccoz near parabolic bound) such that for any two
neighbouring points x, y P Q rn the intersection
rx, ys X Q
rn`1 “ tx “ t0 ă t1 ă ¨ ¨ ¨ ă tk “ yu,
as above, satisfies, for all 0 ă j ď k,
1 |tj ´ tj ´1 | C
2
ď ď (9.11)
Cpmintj, k ` 1 ´ j uq |tk ´ t0 | pmintj, k ` 1 ´ j uq2
(see [Yo2] and [dFdM]). This disposition of the orbit, at least in the real
analytic case, is the result of the orbit passing the narrow gate of two qn -
periodic points, which are close to merging into a parabolic point.
To see how this a priori bound can be used we consider as a model a
rigid rotation Rη pzq “ e2π i η z with pk ` 1qη ă 1 ă pk ` 2qη and k even. We
conjugate Rη by a real symmetric Möbius transformation
z`r
Mpzq “ ,
1 ` rz
and obtain T :“ M ˝ Rη ˝ M ´1 with fixed points r and 1r . Let us choose

1
r “ kk ´ π
`π . As k tends to infinity the fixed points r and r tend to 1. For k
large the conjugate map T is near parabolic. Consider the Rη -orbit through
1, and its segment wj1 “ e2π i pj ´k {2qη corresponding to 0 ď j ď k. (Note that
wk1 {2 “ 1.) The orbit is equidistantly separated, and the orbit points w01 “
e´π i kη and wk1 “ eπ i kη are close to ´1 for k large. The points ˘1 are fixed
points of the Möbius transformation M with derivatives M 1 p´1q “ 11` r
´r “ π
k
and M 1 p1q “ 11´ r π 1

`r “ k . Let wj “ Mpwj q, 0 ď j ď k, denote the correspond-
ing segment of the orbit through 1 “ Mp1q, of the conjugate map T (see
p. 46–47 in [PZ]). We invite the reader to check that there exists δ ą 0 so that
M
wk
wk1
´1 0 1
1 “ wk{2 r 1{r
w01 1
1 “ wk{2
w0
Figure 9.3 Changing the fixed points 0, 8 of a rigid rotation to the fixed points r and
1 of a near parabolic rotation.
r
´π ` δ ă arg w0 ă ´δ ă 0 ă δ ă arg wk ă π ´ δ,
and the arguments of the wj are spaced according to (9.11) (compare with
Figure 9.3).
Let φ : γ ÝÑ D and φ 1 : γ 1 ÝÑ D be homeomorphically extended Rie-
mann maps of the level n cells γ and γ 1 associated with t0 , . . . , tk and the
corresponding set (by Hr ) t01 , . . . , tk1 . Normalize φ by mapping zn`1 pt0 q to
í , zn`1 ptk q to i and zn`1 pj0 q to 1, where j0 is the integer part of k{2.
And similarly for γ 1 and φ 1 . Then the images wj of zn`1 ptj q will be spaced
according to (9.11) for some C 1 depending only on the derivative f 1 . And the
images wj1 of zn1 `1 ptj1 q will be approximately equidistant.
Let M be as above with r “ kk ´ π
`π , and let P denote the Möbius transforma-
tion, which fixes 1, maps í to ´1 and ´1 to i . Note that the inverse Möbius
transformation P ´1 satisfies P ´1 pzq “ P pzq. Define ψ : S1 ÝÑ S1 by
$
´1 if 0 ď argpzq ď π {2,
&P ˝ M ˝ P pzq
’
’
ψpzq “ z if π {2 ď argpzq ď 3π {2,
’
%P ´1 ˝ M ˝ P pzq
’
if 3π {2 ď argpzq ď 2π.
Note that ψ is a homeomorphism, which is piecewise Möbius with fixed

points at 1, ˘i as well as the points on the left halfcircle. Then there exist
quasiconformal self-homeomorphisms χ , χ 1 : D ÝÑ D fixing 1 and ˘i and
of uniformly bounded (in terms of f 1 ) dilatations such that
´1 ´1
ψpzq “ χ ˝ φ ˝ Hr ´1 ˝ pφ 1 q ˝ pχ 1 q pzq
(compare with Figures 9.4 and 9.5).

H̃ ´1
γ1 γ
t01 1
tk{2 tk1 t0 tk{2 tk
χ ˝φ
χ1 ˝ φ1 wk1 “ i wk “ i
1
1 “ wk{2
−1 1 “ wk{2
w01 “ ´1 w0 “ ´1
Figure 9.4 Uniformizing γ and γ 1 and post-adjusting by quasiconformal

homeomorphisms of uniformly bounded distortion.
i
P ˝ M ˝ P ´1
Id 1
P ´1 ˝ M ˝ P
í
Figure 9.5 The map ψ : S1 Ñ S1 , which is piecewise Möbius, to be

quasiconformally extended.
To prove the claims above about uniform control of the quasiconformal

dilations of χ and χ 1 , it is convenient, because of the ‘squareness’ of the
cells γ and γ 1 to map the cells quasiconformally to the upper half plane
‘by hand’ instead of factoring through the Riemann maps, see also [PZ, Sect.
6.3, pp. 46–50]. The problem with making estimates in the above equivalent
approach is the non-explicitness of the Riemann maps. To complete the proof
9.3 Turning hyperbolics into parabolics 385
of Theorem 9.16, it suffices to prove that the maps ψ have quasiconformal

extensions with dilatations bounded by 2p1 ` plog kq2 q. This is the content
of the last Proposition, which is due to K. Strebel.
Proposition 9.17 (Extensions to D of piecewise Möbius maps on S1 ) Let

ψ : S1 ÝÑ S1 be an orientation preserving homeomorphism, which is piece-
wise Möbius in the following sense: there exist n ě 2 fixed points x1 “
xn`1 , x2 , . . . , xn ordered counterclockwise and n Möbius transformations
ζ 1 , ζ 2 , . . . , ζ n preserving S1 such that ψ|rxj ,xj `1 s “ ζ j for 1 ď j ď n. Let
k ą 1 be the largest among the multipliers of the repelling fixed points of the
ζ j . Then ψ has a quasiconformal extension : D Ñ D whose dilatation is
bounded by 2p1 ` plog kq2 q.
Proof Let us first consider a related but easier problem on the horizontal strip
S :“ tz P C | 0 ď Impzq ď π {2u with ψpzq “ z on the bottom edge R and
ψpzq “ z ` log λ on the top edge R ` i π {2, where λ ą 1. In this case, we can
extend ψ to a quasiconformal self-homeomorphism of S by interpolating
linearly:
2
pzq “ z ` π Impzq log λ.
It is easy to verify that the dilatation of this is less than 2p1 ` plog λq2 q. (As
an exercise, the reader can show that this is the best possible extension.)
Back to the original situation, consider the hyperbolic convex set Dj
bounded by the interval rxj , xj `1 s Ă S1 and the hyperbolic geodesic ϒj in
D with endpoints xj and xj `1 . Each Dj is conformally isomorphic to the strip
S above, with rxj , xj `1 s mapping to R ` i π {2 and ϒj mapping to R. The
action of ψ on rxj , xj `1 s corresponds to z ÞÑ z ˘ log λj , where λj ą 1 is the
multiplier of the repelling fixed point of ζ j . Thus, by the initial construction,
ψ can be extended to a quasiconformal homeomorphism : Dj Ñ Dj which
interpolates between ψ|rxj ,xj `1 s and the identity on ϒj , with dilatation less
that 2p1 ` plog λj q2 q. On D nj“1 Dj , an ideal hyperbolic n-gon, extend
Ť
as the identity map. Evidently the dilatation of on D is less than

2p1 ` plogpmaxj λj qq2 q “ 2p1 ` plog kq2 q.
9.3 Turning hyperbolics into parabolics

Peter Haı̈ssinsky
Let P be a polynomial of degree d ě 2 and let KP and JP be its filled Julia set
and Julia set respectively. We denote by PP its postcritical set, that is the clo-
sure of its critical orbits (see Definition 3.38). If α is an attracting or parabolic
point, we set Apαq to be its basin of attraction and A˝ pαq its immediate basin.
The purpose of this section is to prove the following theorem, Theorem 9.18,
and to apply it to the construction of matings (cf. Theorem 9.32).
Theorem 9.18 (Main Theorem) Let P be a polynomial of degree d ě 2 with

an attracting fixed point α and a repelling fixed point β P BA˝ pαq, β R PP .
Suppose also that β is accessible from A˝ pαq.
Then there exist a polynomial Q of the same degree and a David map ϕ,
such that:
(i) ϕpKP q “ KQ ; ϕpβq is a parabolic fixed point with Q1 pϕpβqq “ 1, and

ϕpA˝ pαqq “ A˝ pϕpβqq;
(ii) for all z R Apαq, ϕ ˝ P pzq “ Q ˝ ϕpzq; in particular, ϕ : JP Ñ JQ is a
homeomorphism which conjugates P to Q.
Remark 9.19 By a slight and sensible change of the proof, this theorem can
be stated replacing the attracting fixed point by several cycles and the repelling
point by cycles such that their periods divide those of the attracting points
related to them, provided the repelling points that are to become parabolic are
not accumulated by the recurrent critical orbits of P .
The proof proceeds by a cut and paste surgery where we will only change the
map in the immediate basin of the attracting fixed point to create a parabolic
behaviour (see Lemma 9.21). The problem that we encounter is that we will
have to change drastically the geometry of the Julia set by creating cusps.
Hence we will not be able to use the qc Integrability Theorem and, instead,
we will rely on David’s generalization, the David Integrability Theorem
(Theorem 9.7).
There are several technical difficulties to overcome. The first is to check that
the assumptions of the theorem are indeed satisfied. This will come from a sim-
ple calculation made on a model map (Lemma 9.20), and spread to the whole
plane by bounded distortion. The second is that composing non-quasiregular
maps together does not imply at once that the resulting function retains enough
analytic regularity, e.g. the ACL property. The proof follows [Ha2].
9.3.1 Regularity of maps between local models

Consider the map f : z ÞÑ λz with λ ą 1 (repelling model). For n ą N where
N is large, we define the sector
" *
1
S :“ z P C | θ ď arg z ď 2π ´ θ and 0 ă |z| ă N ,
λ
f
where 0 ă θ ă π . We write Qn for the quadrilateral bounded by the segments
rp1{λn`1 qe˘i θ , p1{λn qe˘i θ s and the circle arcs of radii 1{λn`1 and 1{λn
f f
contained in S. Note that Qn “ f ń pQ0 q.
The map z ÞÑ wpzq “ Log λ{ Log z conjugates f on Dλń zS to
w
g : w ÞÑ (parabolic model)
1`w
on the cusp C “ wpDλń zSq with vertex at the origin (see Figure 9.6).
f pzq “ λz χ gpwq “ ww`1
S X Dλn
g
θ Qn
f λ´pn`1q ń C X Drn
Qn 0
λ
0
|w| — n1
1
|w| — n` 1
Figure 9.6 Definition of χn . The total of the shaded region on the left picture
corresponds to S X Dλn .
Lemma 9.20 There is a David extension χ of z ÞÑ wpzq to a neighbourhood

f
of the origin, with Kχ — n on Qn .
Proof We define rn :“ |wpλń e˘i θ q| and observe that rn — n1 . Define
˝
χ : S Ñ DrN zC
log λ
ρ ¨ ei t ÞÑ ¨ exp i aρ ptq “ |wpρei θ q| exp i aρ ptq ,
| log ρ ` i θ |
where aρ ptq is an affine map in t P rθ, 2π ´ θ s, for fixed ρ.

The map χ is a homeomorphism, which maps arcs of circles centred at the
origin to arcs of circles centred at the origin.
g f f
Write Qn “ χ pQn q (see Figure 9.6). Observe that in log coordinates, Qn
f
corresponds to a rectangle Rn of width ln λ and height 2pπ ´ θ q, both indepen-
g g 1
dent of n. Asymptotically, Qn corresponds to a rectangle Rn of width ln n` n
and height 2π . Also note that χ is asymptotically conjugate to an affine map
f g
from Rn to Rn . It follows that we can estimate Kχ by comparing the modulus
f g
of Rn with that of Rn . More precisely,
f
mod Rn 1
Kχ — g :“ Kn — — n.
mod Rn lnp1 ` n1 q
Hence, for n0 big enough,

ď f π ´θ
Areatz P D | Kχ pzq ě n0 u — Area Qn “ “ c e´2no ln λ
,
něn0
λ2n0
where c ą 0 is independent of n0 . Therefore χ is David.
9.3.2 Comparison of hyperbolic and parabolic basins

Consider the two Blaschke products
zd ` a
Bpar pzq “ , a “ pd ´ 1q{pd ` 1q
1 ` azd
and
zd ` b
Bpzq “ , 0 ă b ă a.
1 ` bzd
Note that both maps are a composition of z ÞÑ zd with a Möbius transformation

of the unit disc onto itself. Hence both leave D invariant as well as CzD and
thus S1 . They are branched coverings of D of degree d, with a unique critical
point of multiplicity d ´ 1 located at z “ 0.
The map B has an attracting fixed point α P p0, 1q, which attracts every point
in D (the Schwarz–Pick Lemma), while z “ 1 is a repelling fixed point with
real multiplier λ “ 11´ b
`b d. On the other hand, Bpar has a parabolic fixed point
at z “ 1, which also attracts every point in D. In both cases, observe that the
interval r0, 1s is invariant, and the critical orbit marches monotonically along
this segment towards z “ α, in the case of B, and towards z “ 1 in the case of
Bpar .
Observe that, locally, the dynamics of B around z “ 1 is conjugate via
linearizing coordinates to the local model f in the section above. Likewise,
those of Bpar is conjugate, via Fatou coordinates, to the local parabolic model
g. We will use these local models and the map χ constructed in the last section,
to build a partial conjugacy between the dynamics of B and those of Bpar in
the corresponding basins of attraction.
Lemma 9.21 There exist a piecewise C 1 homeomorphism φ : D Ñ D and a

sector SB Ă D with vertex at 1, which is a neighbourhood of α, such that:
(i) for all z P DzSB , φ ˝ Bpzq “ Bpar ˝ φpzq;

(ii) there is a set SB1 which is the intersection of SB with a neighbourhood of 1,
such that φ : Dz k B ´k pSB1 q Ñ φpDz k B ´k pSB1 qq is quasiconformal;
Ť Ť
1 f
(iii) on the quadrilaterals QB n in SB defined as Qn for Lemma 9.20, we have
Kφ — n, for all n ě n0 .
Remark 9.22 In (ii), the inverse branches are chosen so that no sub-branch
fixes 1.
Proof Let us first have a look at the dynamics of Bpar . We let : D Ñ C

be the Fatou coordinate such that ˝ Bpar pzq “ pzq ` 1 and p0q “ 0 (see
Section 3.3.2). Choose R ą 0 large enough so that ´1 t|Im z| “ Ru has two
components which bound invariant sepals at the parabolic point 1. Let
P “ Cztz P C | Re z ď 1, |Im z| ď Ru,
and let Spar be the connected component of ´1 pPq which contains a “
Bpar p0q and 1 in its boundary. Observe that since p0q “ 0, then paq “ 1.
We let Epar “ Spar zBpar pSpar q. It is naturally endowed with a quadrilat-
eral structure coming from its image by the Fatou coordinate (pEpar q “
r1, 2q ˆ r´R, Rs) (see Figure 9.7).
B φ w ÞÑ w ` 1
Bpar
SB Spar 1 q
EB 1
SB Epar φ pSB Ri
Sp
0
b
α
1
0
a 1 0 1 2 P
´Ri
pEpar q
Figure 9.7 Sketch of the construction of φ, for the degree d “ 2. The total of the
shaded regions correspond to SB , Spar and P respectively. The white region is the
1 , then extended to
first preimage of these sets. The map φ is first defined in SB
SB and finally extended to D using the dynamics.
Next we look at the dynamics of B. In the quotient torus associated to α,

we choose (in a way to be specified later) an annulus which is disjoint from
the critical orbit and with pα, 1q{pBq as its equator. We let Sp be the connected
component of its lift which contains pα, 1q. It is a simply connected forward
invariant set such that the action of B is conjugate to a translation on a strip.
Indeed, since the multiplier is real, Sp is conformally mapped to a sector by the
linearizing coordinates, which is then sent to an infinite strip by the principal

branch of the logarithm. Observe that close to z “ 1, the set Sp belongs to the
linearizing domain of this repelling fixed point. We assume that we chose the
annulus in such a way that, under the linearizing coordinates around z “ 1, Sp
is a straight sector. We then define SB1 Ă Sp and quadrilaterals QBn analogously
to what we did in the repelling model in the section above, for values of n large
enough.
Let be a linearizing domain for α such that b P B. Note that we can
choose the point b in B X B S. p We set SB “ Y S. p Therefore, BpSB q Ă SB ,
and we define EB “ SB zBpSB q.
We proceed now to the construction of φ. We first define φ in a neighbour-
hood of 1 as we did for the model maps (Lemma 9.20). Observe that φ maps
the set SB1 and the quadrilaterals QB 1
n in SB to their counterparts in Spar . (Note
that the cusp part, is outside Spar .) In particular, it maps the boundary of SB
close to 1, to the boundary of Spar also close to 1, conjugating the dynamics of
B to the dynamics of Bpar on this boundary. Since linearizing coordinates and
Fatou coordinates are conformal, the dilatation of φ is asymptotically n on the
quadrilaterals QB 1
n . We will only use φ wherever defined in SB (i.e. on SB ), that
is, we forget it on the rest of the neighbourhood of 1 where it was originally
defined.
We now extend φ to a homeomorphism φ : SB Ñ Spar such that, for all
z P BSB , φ ˝ Bpzq “ Bpar ˝ φpzq, and such that φpbq “ a. By the choices
above, we can adjust it so that φpEB q “ Epar , still keeping the conjugacy
on the boundary. We impose φ to be piecewise C 1 and quasiconformal on the
complement of any neighbourhood of 1.
We record the following facts:
(a) BpSB q Ă SB and Bpar pSpar q Ă Spar ;

(b) recall that B : D Ñ D and Bpar : D Ñ D are ramified coverings of the
same topological degree, and 0 is the unique critical point for B and Bpar
with the same multiplicity, and φpBp0qq “ φpbq “ a “ Bpar p0q;
(c) for all z P D, there are minimal iterates nB and npar such that B nB pzq P SB
npar
and Bpar pzq P Spar .
Therefore, we may inductively define φ : D Ñ D so that φ ˝ B “ Bpar ˝ φ

on DzSB . There are no topological obstructions for its definition thanks to (a)
and (b), and the map is well defined on D and onto thanks to (c).
Point (i) is satisfied by construction. Since compositions by holomorphic
maps do not change the dilatation, (ii) holds as well. Finally, (iii) follows from
the use of Lemma 9.20.
9.3.3 Parabolic surgery

Let P be as in Theorem 9.18, that is, a polynomial of degree d ě 2 with an
attracting fixed point α and a repelling fixed point β on BA˝ pαq, β R PP . It
follows from the surgery construction in Section 7.5, due to Curt McMullen,
that there is a quasiconformal map ψ, conformal outside Apαq, which con-
jugates P to another polynomial Pr except on Apαq, and such that Pr has the
following properties: there is only one critical grand orbit in ψpApαqq, with
at most one critical point in each component of ψpApαqq, and Pr1 pψpαqq ą 0.
(See Exercises 7.5.5 and 7.5.6, where the conjugacy is called φ.) Note that
in particular, ψpαq is an attracting fixed point of Pr, with immediate basin
A˝ pψpαqq “ ψpA˝ pαqq. Also ψpβq is a repelling fixed point on BA˝ pψpαqq,
not in the postcritical set of Pr, and ψpβq is accessible from the immedi-
ate basin. We also observe that ψpKP q “ KPr and ψpJP q “ JPr , where it
conjugates the dynamics of P and Pr. It follows that it is enough to prove
Theorem 9.18 for Pr. Therefore, we assume from now on that P “ Pr.
Topological surgery Let A˝ be the immediate basin of α. Consider the Rie-

mann map R “ RA˝ from A˝ onto the unit disk D, which conjugates P to
a Blaschke product BA˝ . Define the map φ : D Ñ D given by Lemma 9.21,
which partially conjugates B “ BA˝ to Bpar . Write B p the conjugate map of
Bpar by φ. Hence, the map B coincides with B except on the sector SB .
p
We can then define on A˝ the map F “ R´1 ˝ B p ˝ R which coincides with
P except on the preimage of the sector SB under R.
Finally set
#
F pzq if z P A˝ ,
Gpzq :“
P pzq if z R A˝ .
This map G is our topological model: a ramified covering of degree d, piece-

wise C 1 , and thus ACL.
We now need to go back to holomorphic dynamics.
Straightening of the almost complex structure and conclusion of the proof

of the Main Theorem (Theorem 9.18) Let us emphasize that the full strength
of the David Integrability Theorem is needed: the existence of the solution
in order to straighten the almost complex structure, and its uniqueness to
conclude that the resulting composition map is indeed a complex polynomial
(to be seen in Lemma 9.25).
We proceed to define a G-invariant almost complex structure in C. Let

p “ Bz φ{Bz φ. This Beltrami form is defined in D and invariant by B.
μ p We now
transport it to A˝ by defining μ “ R˚ μ
p . We have the following commutative
diagram:
Bpar
pD, μ0 q ÝÝÝÝÑ pD, μ0 q
İ İ
φ§
§ §φ
§
B
p
pD, μ
p q ÝÝÝÝÑ pD, μ
pq
İ İ
R§ §R
§ §
F
pA˝ , μq ÝÝÝÝÑ pA˝ , μq
We now spread it by the dynamics of F (which equals P outside D) to the

rest of the dynamical plane. As usual, we recursively define
#
pP n q˚ μ on P ń pA˝ q,
μ :“
on Cz n P ń pA˝ q.
Ť
μ0
The map G leaves μ invariant by definition.
Lemma 9.23 If β R PP then μ satisfies the hypotheses of the David Integra-

bility Theorem (Theorem 9.7).
Proof Let V be a simply connected linearizable neighbourhood of β disjoint

from PP . Let U be a connected component of P ´1 pV q compactly contained
in V . Let β be the trace of R´1 pSB1 q in U . Set ρ “ |P 1 pβq| and let Kμ be the
dilatation ratio of μ; by Koebe’s Distortion Theorem and Lemma 9.21,
Areatz P β | Kμ ą nu — p1{ρ 2n q Area β .
If y P C satisfies P n pyq “ β, then let y be the connected component of

P ń pβ q with vertex y.
The map P n : pP ń pV q, yq Ñ pV , βq is conformal, and since P : U Ñ V
fixes β, it lifts by P n as a repelling germ of same multiplier in a neighbourhood
of y. By bounded distortion, there exists a constant C ą 0 such that
Areatz P y , Kμ ą nu ď pC{ρ 2n q Area y ,
for all preimages y of β.

We thus deduce that for n large enough,

ÿ ÿ
Areatz P C, | Kμ ą nu “ Areatz P y , Kμ ą nu
p ě0 P p y “β
ÿ ÿ
ď pC{ρ 2n q Area y .
p ě0 P p y “β
Hence,
Areatz P C | Kμ ą nu ď pC{ρ 2n q Area X “ C 1 e´2n ln ρ ,

Ť
where X “ pě0, P p py q“β y . But Area X is finite because X is the union
of disjoint open sets, all contained in KP , which is a bounded domain. This
proves the estimate on Kμ .
Remark 9.24 This lemma is also true, with the same proof, if we only assume
that β is not accumulated by any recurrent critical point.
The David Integrability Theorem (Theorem 9.7) asserts the existence of a
map ϕ which integrates μ. However, to obtain a holomorphic function there
is still some work to be done. The following lemma is a standard trick (to be
compared with Proposition 9.14).
Lemma 9.25 (Straightening) If there exists an ACL homeomorphism ϕ :

C Ñ C, such that the equation Bz ψ “ μϕ Bz ψ has local ACL solutions, unique
up to post-composition by a conformal map, and such that G˚ μϕ “ μϕ a.e.,
then Q “ ϕ ˝ G ˝ ϕ ´1 is the polynomial of Theorem 9.18.
Proof Indeed, on every disk where G is injective, ϕ ˝ G is ACL (see the

proof of Proposition 9.14), and has the same Beltrami coefficient as R; hence,
there exists a conformal map, say Q, such that ϕ ˝ G “ Q ˝ ϕ (uniqueness of
the solution). The map Q is thus holomorphic. Since Q´1 pt8uq “ t8u, it is
a polynomial. The dynamical properties of Q are automatically satisfied by
construction.
To prove the local uniqueness of solutions of μ, it suffices to normalize

two solutions on the unit disk and to extend it to the plane by reflection
(see Section 1.2.1): the hypotheses of the David Integrability Theorem are
still fulfilled, hence these solutions are the same. By Lemma 9.25, the map
Q “ ϕ ˝ G ˝ ϕ ´1 is the polynomial we are looking for.
9.3.4 Application to matings

We will use the techniques above to prove the existence of certain matings of
geometrically finite polynomials. Details may be found in [HT].
The different steps involve constructing subhyperbolic and postcritically
finite polynomials from the given geometrically finite polynomials. Hence, we
start by recalling that a rational map R is said to be geometrically finite if its
postcritical set intersects the Julia set JR in a finite set, or equivalently if every
critical point is either preperiodic, or attracted to an attracting or parabolic
cycle. If R has no parabolic cycles it is called subhyperbolic. If the postcritical
set is finite R is said to be postcritically finite. Moreover, recall that if the Julia
set of a geometrically finite polynomial is connected then it is always locally
connected, see [DH2].
Definition, existence and uniqueness There are many equivalent ways to

define matings of polynomials. We have chosen the one presented by Milnor
in [Mi2].
Definition 9.26 (The sphere) Consider S2 as the subspace of C ˆ R defined
by S2 “ H` Y H´ Y tpz, 0q P C ˆ R | |z| “ 1u, where H` “ tpz, rq P C ˆ
R` | |z|2 ` r 2 “ 1u and H´ “ tpz, rq P C ˆ R´ | |z|2 ` r 2 “ 1u. Let τ˘ :
C Ñ H˘ be the gnomonic projections defined by:
b b
τ` pzq “ pz, 1q{ |z|2 ` 1, τ´ pzq “ pz̄, ´1q{ |z|2 ` 1 .
The hemispheres H˘ are equipped with conformal structures via τ˘ .
Definition 9.27 (Topological mating) Let F, G : C p ÑC p be two ramified
coverings of degree d ě 1. Assume that 8 is totally invariant for both F
and G, and that Fánd G are¯ holomorphic in a neighbourhood of 8 with
local expansion zd 1 ` Op 1z q , i.e. F, G have the same leading term in their
local expansion at 8. We define the topological mating F K K G : S2 Ñ S2 to
´1 ´1
be the unique extension on S of τ` ˝ F ˝ τ` |H` and τ´ ˝ G ˝ τ´
2 |H´ . By
´1
abuse of notation we will not distinguish F and τ` ˝ F ˝ τ` |H` (resp. G
´1
and τ´ ˝ G ˝ τ´ |H´ ). An easy calculation shows that the map F KK G is a
well defined branched covering of degree d. In particular, if d “ 1 we get a
homeomorphism of the sphere.
Assume now that f and g are two monic geometrically finite polynomials
of degree d ě 2 with connected Julia sets. The sphere S2 on which f K Kg
is defined is equipped with the ray equivalence relation, which is defined to
be the smallest equivalence relation generated by x „ y if x, y belong to the
closure in S2 of an external ray of f or an external ray of g. The map f K

Kg
preserves this relation.
Definition 9.28 (Marked mating and matable polynomials) We say that

pf, g, q, Rq is a marked mating, if f, g are monic polynomials of degree d,
R is a degree d rational map, and q : S2 Ñ Cp is a continuous map such that:
(1) (semi-conjugacy) q ˝ pf KK gq “ R ˝ q;
(2) (identification) qpxq “ qpyq if and only if x and y are ray-equivalent;
(3) (maximal conformality) q is conformal in intpKf q Y intpKg q,
qpintpKf q Y intpKg qq “ C p JR , and q ´1 pC p JR q “ intpKf q Y
intpKg q.
We will say that two monic polynomials f and g are matable if there exist
a continuous map q and a rational map R such that pf, g, q, Rq is a marked
mating.
The basic result of matability is the following, due to Rees, Shishikura and
Tan Lei (cf. [Sh3] and [Ta1]).
Theorem 9.29 (Matability of postcritically finite polynomials) Let f and g

be two monic postcritically finite polynomials of the same degree. Then either
fK K g is equivalent to a Lattès example, or f and g are matable if and only
if f K
K g has no Thurston obstructions. In particular, two postcritically finite
quadratic polynomials fc pzq “ z2 ` c and fc1 pzq “ z2 ` c1 are matable if and
only if c and c1 do not belong to the same limb of the Mandelbrot set.
Uniqueness of R and q Pilgrim has provided examples of non-uniqueness of

the conformal conjugacy class of R in a marked mating pf, g, q, Rq in a case
where R is a Lattès example. For details, see [Mi2, Appendix B.5]. However,
we have the following uniqueness result.
Proposition 9.30 (Uniqueness) Let pf, g, q, Rq be a marked mating. If

f, g, R are geometrically finite and R is not a Lattès example, then R is unique
up to conformal conjugacy, and for a given R, the map q is unique up to post-
composition by an automorphism of C. p
Postcritically finite polynomials associated to geometrically finite

polynomials Let f be a monic and centred geometrically finite polynomial of
degree d ě 2 with connected Julia set. Recall that there is a unique conformal
map (the inverse of Böttcher coordinates) ψf : CzD Ñ CzKf , which is
tangent to the identity at infinity and satisfies ψf pzd q “ f ˝ ψf pzq. Since the
Julia set is locally connected this map extends continuously to the closure
and induces the Carathéodory loop γf : S1 Ñ Jf (see Theorem 3.50 and the
remark afterwards).
Suppose f has one or more parabolic cycles. We wish to associate to f a
canonical monic postcritically finite polynomial T pf q. We will proceed in two
steps.
The first step follows from [Ha3]. If f is a monic geometrically finite
polynomial with connected Julia set, then there exist a monic subhyperbolic
polynomial Spf q and an orientation preserving homeomorphism h : Jf Ñ
JSpf q such that h ˝ f “ Spf q ˝ h on Jf . The map h is obtained as the
continuous extension of the composition ψSpf q ˝ ψf´1 : CzKf Ñ CzKSpf q
of the Böttcher coordinates of f and Spf q. So formally h “ γSpf q ˝ γf´1 .
Moreover, it follows from the construction in [Ha3] that Spf q satisfies the
assumptions for P in Theorem 9.18, giving rise to Q “ f and a David home-
omorphism ϕ1 : C Ñ C, which in particular is a homeomorphism on the Julia
sets, conjugating Spf q to f .
The second step is obtained by a surgery construction that associates
a canonical monic postcritically finite polynomial T pf q to Spf q, see for
instance Section 4.2 and Exercise 4.2.2, or Section 7.5. From the surgery
construction it follows that there is a quasiconformal map ϕ2 : C Ñ C such
that ϕ2 ˝ T pf q “ Spf q ˝ ϕ2 is satisfied on some neighbourhoods of their
respective Julia sets JT pf q and JSpf q .
The subhyperbolic polynomial in the first step in not uniquely determined.
However, the different choices give rise to the same T pf q, which is therefore
canonically determined.
Consider the diagram
T pf q
pC, JT pf q q ÝÝÝÝÑ pC, JT pf q q
§ §
ϕ2 đ
§ §ϕ
đ 2
Spf q
pC, JSpf q q ÝÝÝÝÑ pC, JSpf q q
§ §
ϕ1 đ
§ §ϕ
đ 1
f
pC, Jf q ÝÝÝÝÑ pC, Jf q
Observe that ϕ :“ ϕ1 ˝ ϕ2 : C Ñ C is a David homeomorphism since it

is a composition of a quasiconformal map by a David homeomorphism
(see Exercise 9.1.2). Moreover, the diagram commutes on the Julia sets. There-
fore we have the following result.
Proposition 9.31 (Existence of a David conjugacy on the Julia sets) Let f

be a monic geometrically finite polynomial with connected Julia set and let
T pf q denote the canonical associated monic postcritically finite polynomial.
Then there is a David homeomorphism ϕ : C Ñ C such that ϕ ˝ T pf q “ f ˝
ϕ on JT pf q .
We are now ready to state our main application, which will be proved in the
next section.
Theorem 9.32 (Matability of geometrically finite polynomials) Let f, g be

two monic and centred geometrically finite polynomials with connected Julia
sets. If T pf q and T pgq are matable, then f and g are also matable.
Corollary 9.33 (Precise condition for matability) Two geometrically finite

quadratic polynomials fc and fc1 are matable if and only if c̄ and c1 do not
belong to the same limb of the Mandelbrot set.
Surgery for matings The existence of marked matings of certain postcritically

infinite polynomials can now be established through surgery. We state this
in the next proposition, which in particular provides us with an extension of
Theorem 9.29 to hyperbolic polynomials with infinite critical orbits.
Proposition 9.34 (Matability of postcritically infinite polynomials) Let

pf, g, q, Rq be a marked mating of polynomials f and g. Let us assume
that fp (resp. gp) is a quasiregular map which coincides with f (resp. g)
on its basin of infinity, and that μ (resp. ν) is a fp-invariant Beltrami form
(resp. gp-invariant) supported on its filled Julia set. If we let ϕ (resp. ψ) be a
quasiconformal homeomorphism which integrates μ (resp. ν) normalized at
infinity to be tangent to the identity, then the polynomials f1 “ ϕ ˝ fp ˝ ϕ ´1
and g1 “ ψ ˝ gp ˝ ψ ´1 are matable.
Proof Define
$
´1
&τ` ˝ ϕ ˝ τ`
’
’ on H` ,
K ψ : S2 Ñ S2 by: ϕK
ϕK K ψ :“ Id on the equator,
’
%τ ˝ ψ ˝ τ ´ 1
’
on H´ .
´ ´
The map ϕ restricted to the basin of 8 of f realizes a conformal conjugacy

from f to f1 . The normalization of ϕ guarantees that ϕ maps the external ray
of f of argument θ onto the external ray of f1 of the same argument. Therefore
the map ϕKK ψ : S2 Ñ S2 is a homeomorphism, conjugating fpK K gp to f1 K

K g1
and preserving the external rays of the same argument.
We define the Beltrami form ξ on C p as the pushforward by q as follows:
$
&q˚ μ on qpintpKf qq,
’
’
ξ :“ q˚ ν on qpintpKg qq,
’
’
%μ
0 elsewhere.
Since q is conformal on intpKf q Y intpKg q, the Beltrami form ξ is R-

invariant. Therefore, the Integrability Theorem (Theorem 1.28) provides us
with a quasiconformal homeomorphism χ : pC, p a, b, cq Ñ pC,
p a, b, cq which
solves the Beltrami equation associated to ξ .
Set R1 “ χ ˝ R ˝ χ ´1 and q1 “ χ ˝ q ˝ pϕK K ψq´1 : S2 Ñ C.
p
We can then check that pf1 , g1 , q1 , R1 q is a marked mating.
Remark 9.35 The proof of Theorem 9.32 will follow the same lines as
Proposition 9.34, but instead of using the classical Integrability Theorem
(Theorem 1.28), we shall use the David Integrability Theorem (Theorem 9.7)
from Section 9.1.
Proof of Theorem 9.32 We proceed as in Proposition 9.34. Let us apply

Proposition 9.31 to f and g. Let T pf q and Spf q be the associated monic
postcritically finite polynomials, and ϕ and ψ the David homeomorphisms that
are given by the proposition.
By assumption T pf q and T pgq are matable. Let pT pf q, T pgq, q, Rq denote
a marked mating. We let μ and ν be the Beltrami forms associated to ϕ and ψ.
We push them forward by q on C p and denote the resulting Beltrami form by ξ .
Let us define
´ ¯
H :“ q ˝ pϕ ´1 ˝ f ˝ ϕqK K pψ ´1 ˝ g ˝ ψq ˝ q ´1 .
This map is well defined and we can check that it is ACL. Moreover, ξ is
H -invariant, and we may complete the proof analogously to the proof of
Proposition 9.34 using the David Integrability Theorem (Theorem 9.7). The
proofs of Lemma 9.23 and Lemma 9.25 apply to this setting as well.
Proof of Corollary 9.33 Let fc and fc1 be geometrically finite polynomials.

If c and c1 belong to conjugate limbs of the Mandelbrot set their respective
α-fixed point have opposite external arguments and therefore S2 { „ray is not
a sphere and the marked mating can not exist. See [HT] for further details.
On the other hand, if c and c1 are not in conjugate limbs, then it is the same
for the parameter values of T pfc q and T pfc1 q, the centre of the hyperbolic
component having c, resp. c1 , as its root. Therefore the marked mating of fc
and fc1 exists by Theorem 9.29 and Theorem 9.32.
References
[Ah1] L. Ahlfors (1964) Finitely generated Kleinian groups, Amer. J. Math. 86, 413–429.
[Ah2] L. Ahlfors (1966) Lectures on Quasiconformal Mappings. Van Nostrand.
[Ah3] L. Ahlfors (1973) Conformal Invariants: Topics in Geometric Function Theory.
McGraw-Hill.
[Ah4] L. Ahlfors (1978) Complex Analysis. An Introduction to the Theory of Analytic Func-
tions of One Complex Variable, 3rd edition. McGraw-Hill.
[Ah5] L. Ahlfors (2006) Lectures on Quasiconformal Mappings, 2nd edition. Vol. 38, Ameri-
can Mathematical Society.
[AB] L. Ahlfors and L. Bers (1960) Riemann mapping’s theorem for variable metrics, Annals
of Math. 72, 385–404.
[An] J. W. Anderson (1998) A brief survey of the deformation theory of Kleinian groups,
in The Epstein Birthday Schrift, eds I. Rivin, C. Rourke and C. Series. Geometry and
Topology Monographs 1, pp. 23–49. Mathematics Institute, University of Warwick.
[AMa] J. W. Anderson and B. Maskit (1996) On the local connectivity of limit sets of Kleinian
groups, Complex Variables Theory Appl. 31, 177–183.
[AIM] K. Astala, T. Iwaniec and G. Marten (2009) Elliptic Partial Differential Equations and
Quasiconformal Mappings in the Plane. Princeton University Press.
[AM] K. Astala and G. J. Martin (2001) Holomorphic motions, in Papers on Analysis,
pp. 27–40. Rep. Univ. Jyväskylä Dep. Math. Stat., 83.
[Ar1] V. I. Arnol’d (1965) Small denominators. I: Mappings of the circumference onto itself,
Am. Math. Soc., Transl., II. Ser. 46, 213–284; translated from Izv. Akad. Nauk SSSR, Ser.
Mat. 25 (1961), 21–86.
[Ar2] V. I. Arnol’d (1983) Geometrical Methods in the Theory of Ordinary Differential Equa-
tions. Springer.
[As] K. Astala (1994) Area distortion of quasiconformal mappings Acta Math. 173, 37–60.
[BA] A. Beurling and L. Ahlfors (1956) The boundary correspondence under quasiconformal
mappings, Acta Math. 96, 125–142.
[BC] X. Buff and A. Cheritat (2004) Upper bound for the size of quadratic Siegel discs, Invent.
Math. 156, 1–24.
[BD] B. Branner and A. Douady (1986) Surgery on Complex Polynomials, Proceedings of the
Second International Colloquium on Dynamical Systems, Mexico, pp. 11–72. Springer-
Verlag.
[BDH] C. Bodelón, R. Devaney, M. Hayes, G. Gareth, L. Goldberg and J. H. Hubbard (2000)
Dynamical convergence of polynomials to the exponential, J. Differ. Eq. Appl. 6,
275–307.
400
References 401
[Be1] L. Bers (1960) Simultaneous uniformization, Bull. Amer. Math. Soc. 66, 95–97.
[Be2] L. Bers (1972) Uniformization, moduli, and Kleinian groups, Bull. Lon. Math. Soc. 4,
257–300.
[Be3] L. Bers (1985) On Sullivan’s proof of the finiteness theorem and the eventual periodicity
theorem, Amer. J. Math. 109, 833–852.
[Bea] A. Beardon (1991) Iteration of Rational Functions. Springer Verlag.
[Bel] E. Beltrami (1868) Saggio di interpretazione della geometria non euclidia, Giorn Math.
6, 285–315.
[Ber] W. Bergweiler (1993) Iteration of meromorphic functions, Bull. Amer. Math. Soc. 29,
151–188.
[BerE] W. Bergweiler and A. Eremenko (1995) On the singularities of the inverse of a mero-
morphic function of finite order, Rev. Mat. Iberoamericana 11, 355–373.
[BF1] B. Branner and N. Fagella (1999) Homeomorphisms between limbs of the Mandelbrot
set, J. Geom. Anal. 9, 327–390.
[BF2] B. Branner and N. Fagella (2001) Extension of homeomorphisms between limbs of the
Mandelbrot set, Conf. Geom. and Dyn. 5, 100–139.
[BFGH] X. Buff, N. Fagella, L. Geyer and C. Henriksen (2005) Herman rings and Arnold disks,
J. Lon. Math. Soc. 72, 689–716.
[BH] B. Branner and J. H. Hubbard (1988) The iteration of cubic polynomials. Part I: the
global topology of parameter space, Acta Math. 160, 143–206.
[BHai] S. Bullett and P. Haı̈ssinsky (2007) Pinching holomorphic correspondences, Conf.
Geom. Dyn. 11, 65–89.
[BHar] S. Bullett and W. Harvey (2000) Mating quadratic maps with Kleinian groups via
quasiconformal surgery, Electronic Research Announcements of the AMS, 6, 21–30.
[BHe] X. Buff and C. Henriksen (2001) Julia sets in parameter space, Comm. Math. Phys. 220,
333–375.
[Bi1] B. Bielefeld (1989) Changing the order of critical points of polynomials using quasi-
conformal surgery, PhD thesis, Cornell University.
[Bi2] B. Bielefeld (1990) Questions in quasiconformal surgery, 2–8, in Conformal Dynamics
Problem List, ed. B. Bielefeld. Stony Brook IMS Preprint 1990/1. Also published in
part 2 of Linear and Complex Analysis Problem Book 3 (1994), eds. V. P. Havin and
N. Nikolskii. Springer-Verlag.
[Bl] P. Blanchard (1984) Complex analytic dynamics on the Riemann sphere, Bull. Amer.
Math. Soc. 11, 85–141.
[Bo1] B. Bojarski (1955) Homeomorphic solutions of Beltrami systems, Dokl. Akad. Nauk.
SSSR 102, 661–664.
[Bo2] B. Bojarski (2010) On the Beltrami equation, once again: 54 years later, Ann. Acad. Sci.
Fenn. Math. 35, 5973.
[Bö] L. E. Böttcher (1904) The principal law of convergence of iterates and their applications
to analysis (in Russian), Izv. Kazan. Fiz.-Mat. Bsch. 14, 155–234.
[BP1] S. Bullett and C. Penrose (1994) Mating quadratic maps with the modular group, Invent.
Math. 115, 483–511.
[BP2] S. Bullett and C. Penrose (2001) Regular and limit sets for holomorphic correspon-
dences, Fund. Math. 167, 111–171.
[BPer] A. Bridy and R. Pérez (2005) A count of maximal small copies in multibrot sets,
Nonlinearity 18, 1945–1953.
[BR] L. Bers and H. L. Royden (1986) Holomorphic families of injections, Acta Math. 157,
259–286
[Bra1] B. Branner (1989) The Mandelbrot set, Proc. Symp. Applied Math. 39, 75–105.
[Bra2] B. Branner (1993) Cubic polynomials: turning around the connectedness locus, in
Topological Methods in Modern Mathematics, eds. L. R. Goldberg and A. V. Philips,
pp. 391–427. Publish or Perish, Inc.
402 References
[BrHa] M. R. Bridson and A. Haefliger (1999) Metric Spaces of Non-positive Curvature.

Springer.
[Bry1] A. D. Bryuno (1965) Convergence of transformations of differential equations to normal
forms, Dokl. Akad. Nauk USSR 165, 987–989 (Soviet Math. Dokl., 1536–1538).
[Bu] X. Buff, Personal communication.
[C] S.-S. Chern (1955) An elementary proof of the existence of isothermal parameters on a
surface, Proc. Amer. Math. Soc. 6, 771–782.
[CG] L. Carleson and T. Gamelin (1993) Complex Dynamics. Springer Verlag.
[Ch] A. Cheritat (2006) Ghys-like models for Lavaurs and simple entire maps, Conform.
Geom. Dyn. 10, 227–256.
[CL] E. F. Collingwood and A. J. Lohwater (1966) The Theory of Cluster Sets. Cambridge
University Press.
[Co] J. B. Conway (1995) Functions of One Complex Variable II. Springer Verlag.
[Com] M. Comerford (2013) The Carathéodory topology for multiply connected domains I,
Cent. Eur. J. Math. 11, 322–340.
[Cr] H. Cremer (1932) Über die Schrödersche Funktionalgleichung und das Schwarzsche
Eckenabbildungsproblem, Ber. Verh. Sächs. Akad. Wiss. Leipzig, Math.-Phys. Kl. 84,
291–324.
[Da] G. David (1988) Solutions de l’équation de Beltrami avec }μ}8 “ 1, Ann. Acad. Sci.
Fenn. Ser. A 1 Math. 13, 25–70.
[DB] A. Douady and X. Buff (2000) Le théorème d’intégrabilité des structures presque
complexes, in The Mandelbrot Set, Theme and Variations, ed. Tan Lei, pp. 307–324.
Cambridge University Press.
[Den] A. Denjoy (1932) Sur les courbes définies par les équations differentieles á la surface
du tore, J. Math. Pure et Appl. 11, série 9, 333–375.
[Dev1] R. L. Devaney (ed.) (1994) Complex Dynamical Systems: The Mathematics Behind the
Mandelbrot and Julia sets, Proceedings of Symposia in Applied Matehmatics, Vol. 49.
American Mathematical Society.
[Dev2] R. L. Devaney (1989) An Introduction to Chaotic Dynamical Systems, 2nd edition.
Addison-Wesley.
[DE] A. Douady and C. Earle (1986) Conformally natural extension of homeomorphisms of
the circle, Acta Math. 157, 23–48.
[DF] P. Domı́nguez and N. Fagella (2004) Existence of Herman rings for transcendental
meromorphic functions, Comp. Var. Theory and Appl. 49, 851–870.
[DH1] A. Douady and J. H. Hubbard (1982) Iterations des Polynomes Quadratiques Com-
plexes, C.R. Acad. Sci., Paris, Ser. I 29, 123–126.
[DH2] A. Douady and J. H. Hubbard (1984–1985) Étude dynamique des polynômes com-
plexes, I, II, Publ. Math. Orsay. English version at www.math.cornell.edu/„hubbard/
OrsayEnglish.pdf.
[DH3] A. Douady and J. H. Hubbard (1985) On the dynamics of polynomial-like mappings,
Ann. Scient., Ec. Norm. Sup. 4e series, 18, 287–343.
[dFdM] E. de Faria and W. de Melo (1999) Rigidity of critical circle mappings I, J. Eur. Math.
Soc. (JEMS) 1, 339–392.
[dMvS] W. de Melo and S. van Strien (1991) One-Dimensional Dynamics. Springer-Verlag.
[Do1] A. Douady (1986) Chirugie sur les applications holomorphes, in Proc. Int. Congr. Math.,
Berkeley, pp. 724–738.
[Do2] A. Douady (1983) Systèmes dynamiques holomorphes, Astérisque, 105/106, 39–63.
[Do3] A. Douady (1987) Disques de Siegel et anneaux de Herman, Astérisque, 152/153,
151–172.
[DT] R. L. Devaney and F. Tangerman (1986) Dynamics of entire functions near the essential
singularity, Erg. Th. Dyn. Syst. 6, 489–503.
References 403
[EL1] E. L. Eremenko and M. Lyubich (1990) The dynamics of analytic transformations,

Leningrad Math. J. 1, 563–634.
[EL2] E. L. Eremenko and M. Lyubich (1992) Dynamical properties of some classes of entire
functions, Ann. Inst. Fourier 42, 989–1020.
[Ep] A. L. Epstein (1993) Towers of finite type complex analytic maps, PhD thesis, CUNY.
Available at http://homepages.warwick.ac.uk/„mases/Thesis.pdf.
[Ep1] A. Epstein, Counterexamples to the quadratic mating conjecture (unpublished
manuscript).
[Ep2] A. L. Epstein (1999) Infinitesimal Thurston rigidity and the Fatou–Shishikura inequal-
ity, Stony Brook I.M.S. Preprint 1999#1 or arXiv:math./9902158 at arXiv.org.
[Ep3] A. L. Epstein (2013) Algebraic dynamics I: Contraction and finiteness principles (in
preparation).
[EY] A. Epstein and M. Yampolsky (1999) Geography of the cubic connectedness locus I:
Intertwining surgery, Ann. Scient. Ec. Norm. Sup. 32, 151–185.
[Fat] P. Fatou, Sur les équations fonctionnelles, Bull. Soc. Math. Fr. 47 (1919), 161–271; 48
(1920), 33–94, 208–304.
[Fau] D. Faught (1992) Local connectivity in a family of cubic polynomials, PhD thesis,
Cornell University.
[FK] H. Farkas and I. Kra (1980) Riemann Surfaces. Springer.
[FG] N. Fagella and L. Geyer (2003) Surgery on Herman rings of the complex standard
family, Erg. Th. and Dyn. Sys. 23, 493–508.
[FSY] M. Flexor, P. Sentenac and J. C. Yoccoz (eds.) (2000) Géométrie complexe et systémes
dynamiques, Astérisque 261.
[Ga] C. F. Gauss (1973) Werke. Band IV, 193–216. Reprint of the 1863 original. Georg Olms
Verlag, Hildesheim.
[Ge1] L. Geyer (1995) Iteration of quasiregular mappings in two dimensions (preprint).
[Ge2] L. Geyer (2001) Siegel disks, Herman rings and the Arnold family, Trans. Amer. Math.
Soc. 353, 3661–3683.
[Gh] E. Ghys (1984) Transformations holomorphes au voisinage d’une courbe de Jordan,
C.R. Acad. Sc. Paris 289, 383–388
[GK] L. R. Goldberg and L. Keen (1986) A finiteness theorem for a dynamical class of entire
functions, Erg. Th. and Dyn. Syst. 6, 183–192.
[GM] L. Goldberg and J. Milnor (1993) Fixed points of polynomial maps II, Ann. Scient. Éc.
Norm. Sup., 4e série, 26, 51–98.
[Go] G. M. Goluzin (1969) Geometric Theory of Functions of a Complex Variable, Vol. 26.
American Mathematical Society.
[GRSY] V. Gutlyanskii, V. Ryazanov, U. Srebro and E. Yakubov (2012) The Beltrami Equation,
A Geometric Approach, Developments in Mathematics, Vol. 26, Springer.
[GS] J. Graczyk and G. Świa̧tek (1997) Generic hyperbolicity in the logistic family, Ann.
Math. 146, 1–52.
[Ha1] P. Haı̈ssinsky (2000) Chirurgie croisée, Bull. Soc. Math. de France 128, 599–654
[Ha2] P. Haı̈ssinsky (1998) Chirurgie parabolique, C.R. Acad. Sci. Paris Sér. I Math. 327,
195–198.
[Ha3] P. Haı̈ssinsky (2000) Déformation J -équivalente de polynômes géométriquement finis,
Fund. Math. 163, (2), 131–141.
[He1] M. Herman (1979) Sur la conjugaison différentiable des difféomorphismes du cercle à
des rotations, Publ. Math. I. H. E. S. 49, 5–234.
[He2] M. Herman (1986) Conjugaison quasi-symétrique des difféphismes du cercels et appli-
cations aux disques singuliers de Siegel (unpublished manuscript).
[He3] M. Herman (1987) Conjugaison quasi-symétrique des homeomorphismes analytiques
du cercels à des rotations (unpublished manuscript).
404 References
[HP] P. G. Hjorth and C. L. Petersen (eds.) (2006) Dynamics on the Riemann Sphere. Euro-
pean Mathematical Society.
[HT] P. Haı̈ssinsky and Tan Lei (2004) Convergence of pinching deformations and matings
of geometrically finite polynomials, Fund. Math. 181(2), 143–188.
[Hu1] J. H. Hubbard (1992) Local connectivity of Julia sets and bifurcation loci: three the-
orems of J.-C. Yoccoz, in Topological Methods in Modern Mathematics, pp. 467–511
and 375–378. Publish or Perish, Inc.
[Hu2] J. H. Hubbard (2006) Teichmüller Theory and Applications to Geometry, Topology and
Dynamics, Vol. 1. Matrix Editions.
[HY] Xin-Hou Hua and Chung-Chun Yang (1998) Dynamics of Transcendental Functions.
Gordon and Breach Science Publishers.
[J] G. Julia (1918) Memoire sur l’itération des fonctions rationnelles, J. Math. Pure Appl.
8, 47–245.
[K] L. Keen (ed.) (2010) Special issue dedicated to Robert L. Devaney on the occasion of
his 60th Birthday, J. Diff. Eq. Appl. 16.
[Ke] J. L. Kelley (1975) General Topology. Springer-Verlag.
[Kin] A. Ya. Khinchin (1997) Continued Fractions. Dover Publications.
[Kiw] J. Kiwi (2000) Non-accessible critical points of Cremer polynomials, Erg. Th. Dyn. Syst.
20, 1391–1403.
[Ko] A. Korn (1914) Zwei Anwendungen der Methode der sukzessiven Annäherungen, in
Schwarz-Festschr. pp. 215–229.
[Kœ] G. Kœnigs (1884) Recherches sur les intégrales de certaines équations fonctionelles,
Ann. Sci. École Norm. Sup. Paris 1, supplém., 1–41.
[Kre] E. Kreyszig (1978) Inroductory Functional Analysis with Applications. Wiley.
[Kri] H. Kriete (ed.) (1998) Progress in Holomorphic Dynamics. Longman.
[KSvS] O. Kozlovski, W. Shen and S. van Strien (2007) Density of hyperbolicity in dimension
one, Ann. Mat. 166, 145–182.
[KZ] L. Keen and G. Zhang (2009) Bounded-type Siegel disks of a one-dimensional family
of entire functions, Erg. Th. and Dyn. Syst. 29, 137–164.
[Lan] S. Lang (1995) Introduction to Diophantine Approximations. Springer-Verlag.
[Lav] P. Lavaurs (1989) Systèmes dynamiques holomorphes: explosion de points périodiques,
Thèse, Université de Paris-Sud.
[Lavr] M. A. Lavrentiev (1935) Sur une classe de représentations continues, Rec. Math. 48,
407–423.
[Lea] L. Leau (1897) Étude sur les equations fonctionelles à une ou plusiérs variables, Ann.
Fac. Sci. Toulouse 11, E1–E110.
[Le] O. Lehto (1986) Univalent Functions and Teichmüller Spaces. Springer-Verlag.
[Li] L. Lichtenstein (1916) Zur Theorie der konformen Abbildung nichtanalytischer singu-
laritätenfreier Flächenstücke auf ebene Gebiete, Kram. Anz., 192–217.
[LV] O. Lehto and K. I. Virtanen (1973) Quasiconformal Mappings in the Plane. Springer-
Verlag.
[Ly] M. Lyubich, Conformal geometry and dynamics of quadratic polynomials (in prepara-
tion).
[Mar] A. Marden (2007) Outer Circles. An Introduction to Hyperbolic 3-Manifolds. Cam-
bridge University Press.
[Mas] B. Maskit (1967) A characterization of Schottky groups, J. d’Analyse Mat. 19, 227–230.
[McM] C. T. McMullen, Simultaneous uniformization of Blaschke products (unpublished
manuscript).
[McM1] C. T. McMullen (1986) Automorphisms of rational maps, in Holomorphic Functions
and Moduli, Vol. I (Berkeley, CA, 1986), pp. 31–60. Math. Sci. Res. Inst. Publ., 10,
Springer, 1988.
References 405
[McM2] C. T. McMullen (1994) Complex Dynamics and Renormalization. Princeton University

Press.
[McM3] C. McMullen (1996) Renormalization and 3-Manifolds which Fiber Over the Circle.
Princeton University Press.
[McM4] C. T. McMullen (1998) Self-similarity of Siegel discs and Hausdorff dimension of Julia
sets, Acta. Math. 180, 247–292.
[Mi1] J. Milnor (2006) Dynamics in One Complex Variable: Introductory Lectures, 3rd ed.,
Princeton University Press.
[Mi2] J. Milnor (2004) Pasting together Julia sets, Exp. Math. 13(1), 55–92.
[Mi3] J. Milnor (1992) Remarks on iterated cubic maps, Exp. Math. 1, 5–24.
[Mi4] J. Milnor (2009) On cubic polynomials with periodic critical point, in Complex Dynam-
ics: Families and Friends, p. 257. A.K. Peters.
[Mi5] J. Milnor (2012) Hyperbolic components, in Conformal Dynamics and Hyperbolic
Geometry, pp. 183–232. American Mathematical Society.
[MNTU] S. Morosawa, Y. Nishimura, M. Taniguchi and T. Ueda (1999) Holomorphic Dynamics.
[Mo] C. B. Morrey (1938) On the solutions of quasi-linear elliptic partial differential quations,
Trans. Amer. Math. Soc. 43, 126–166.
[MS] C. McMullen and D. Sullivan (1998) Quasiconformal homeomorphisms and dynamics
III: the Teichmüller space of a rational map. Adv. Math. 135, 351–395.
[MSS] R. Mañé, P. Sad and D. Sullivan (1983) On the dynamics of rational maps, Ann. Scie.
École Norm. Sup.(4) 16, 51–98 and 193–217.
[MSW] D. Mumford, C. Series and D. Wright (2002) Indira’s Pearls: The Vision of Felix Klein.
[MT] K. Matsuzaki and M. Taniguchi (1998) Hyperbolic Manifolds and Kleinian Groups.
Clarendon Press.
[N] V. I. Năishul’ (1983) Topological invariants of analytic and area preserving mappings
and their application to analytic differential equations in C2 and CP2 , Trans. Moscow
Math. Soc. 42, 239–250.
[NZM] I. Niven, H. S. Zuckerman and H. L. Montgomery (1991) An Introduction to the Theory
of Numbers, 5th ed. Wiley.
[Pe1] C. L. Petersen (1996) Local connectivity of some Julia sets containing a circle with an
irrational rotation, Acta Math. 177, 163–224.
[Pe2] C. L. Petersen The Herman–Świa̧tek Theorems with applications, in The Mandelbrot
Set, Theme and Variations, ed. Tan Lei, pp. 211–225. Cambridge University Press.
[Pe3] C. L. Petersen (2004) On holomorphic critical quasi-circle maps, Erg. Th. Dyn. Syst. 24,
1739–1751.
[PeT] C. L. Petersen and Tan Lei (2006) Branner–Hubbard motions and attracting dynamics,
in Dynamics on the Riemann Sphere, eds. P. G. Hjorth and C. L. Petersen, pp. 45–70.
European Mathematical Society.
[Pi] G. Piranian (1958) The boundary of a simply connected domain, Bull. Amer. Math. Soc.
64, 45–55.
[PM] R. Perez-Marco (1997) Fixed points and circle maps, Acta Math. 179, 273–294.
[Poi] H. Poincaré (1885) Sur les courbes défines par des équations différentielles, J. Math.
Pures et Appl., 4e série, 1, 167–244.
[Pom] Ch. Pommerenke (1992) Boundary Behavior of Conformal Maps. Springer.
[PT1] K. Pilgrim and Tan Lei (1999) On disc-annulus surgery of rational maps, in Proceedings
of the International Conference in Honor of Professor Shantao Liao, eds. Y. Jiang and
L. Wen, pp. 237–250. World Scientific.
[PT2] K. Pilgrim and Tan Lei (2000) Rational maps with disconnected Julia set, Astérisque
261, 349–384.
406 References
[PZ] C. L. Petersen and S. Zakeri (2004) On the Julia set of a typical quadratic polynomial
with a Siegel disk, Ann. Math. (2) 159, 1–52.
[R] J. Riedl (2000) Arcs in multibrot sets, locally connected Julia sets and their construc-
tion by quasiconformal surgery, PhD thesis, Technische Universität München Zentrum
Mathematik.
[Ro] R. C. Robinson (2004) An Introduction to Dynamical Systems. Northwestern University.
[Roh] S. Rohde (1997) Bilipschitz maps and the modulus of rings, Ann. Acad. Sci. Fen. Math.
22, 465–474.
[Rü] H. Rüssman (1967) Über die Iteration analytischer Funcionen (in German), J. Math.
Mech 17, 523–532.
[RS] P. J. Rippon and G. M. Stallard (eds.) (2008) Transcendental Dynamics and Complex
Analysis. Cambridge University Press.
[Sc] D. Schleicher (ed.) (2009) Complex Dynamics: Families and Friends. A.K. Peters.
[Sh1] M. Shishikura (1987) On the quasiconformal surgery of rational functions, Ann. Sci. Ec.
Norm. Sup. 20, 1–29.
[Sh2] M. Shishikura (1998) The Hausdorff dimension of the boundary of the Mandelbrot set
and Julia sets. Ann. of Math. 147, 225–267.
[Sh3] M. Shishikura (2000) On a theorem of M. Rees for matings of polynomials, in The
Mandelbrot Set, Theme and Variations, pp. 289–305. Cambridge University Press.
[Shu] M. Shub (1970) Expanding maps, in Global Analysis, AMS Proc. of Symp. XIV,
pp. 273–276. American Mathematical Society.
[Si] C. L. Siegel (1942) Iteration of analytic functions, Ann. of Math. 43, 607–612.
[Sl] Z. Słodkowski (1991) Holomorphic motions and polynomial hulls, Proc. Amer. Math.
Soc. 111, 347–355.
[St] N. Steinmetz (1993) Rational Iteration: Complex Analytic Dynamical Systems. Walter
de Gruyter.
[Su1] D. Sullivan (1983) Conformal dynamical systems, in Geometric Dynamics (Rio de
Janeiro, 1981), pp. 725–752. Springer.
[Su2] D. Sullivan (1985) Quasiconformal homeomorphisms and dynamics. I. Solution of the
Fatou–Julia problem on wandering domains. Ann. of Math. 122, 401–418.
[ST] D. Sullivan and W. P. Thurston (1986) Extending holomorphic motions, Acta Math. 157,
243–257.
[Sw1] G. Świa̧tek (1988) Rational rotation numbers for maps of the circle, Comm. Math. Phys.
119, 109–128.
[Sw2] G. Świa̧tek (1998) On critical circle homeomorphisms, Bol. Soc. Bras. Mat. (N. S.), 29,
329–351.
[Ta1] Tan L. (1992) Matings of quadratic polynomials, Erg. Th. Dyn. Syst. 12, 589–620.
[Ta2] Tan L. (ed.) (2000) The Mandelbrot Set, Theme and Variations. Cambridge University
Press.
[Tu1] P. Tukia (1986) On quasiconformal groups, J. Analyse Math. 46, 318–346.
[Tu2] P. Tukia (1991) Compactness properties of μ-homeomorphisms, Ann. Acad. Sci. Fenn.
Ser. A I Math. 16, 47–69.
[TY] Tan L. and Yin Y. (1996) Local connectivity of the Julia set for geometrically finite
rational maps, Sci. China Ser. A 39, 39–47
[Why] S. Whyburn (1942) Analytic Topology, AMS Coll. Publ. 28.
[Yo1] J.-C. Yoccoz (1988) Linéarisation des germes de difféomorphismes holomorphes de
pC, 0q, C.R. Acad. Sci. Paris 306, 55–58.
[Yo2] J.-C. Yoccoz (1989) Structure des orbites des homéomorphismes analytiques posedant
un point critique (unpublished manuscript).
[Yo3] J.-C. Yoccoz, Conjugaison des diffeomorphismes analytiques du cercle (unpublished
manuscript).
References 407
[Yo4] J.-C. Yoccoz (1992) An introduction to small divisors problems, in From Number
Theory to Physics, pp. 659–679. Springer-Verlag.
[YZ] M. Yampolsky and S. Zakeri (2001) Mating Siegel quadratic polynomials, J. Amer.
Math. Soc. 14, 25–78.
[Z1] S. Zakeri (1999) Dynamics of cubic Siegel polynomials, Comm. Math. Phys. 206,
185–233.
[Z2] S. Zakeri (2010) On Siegel disks of a class of entire maps, Duke Math. J. 152, 481–532.
[Zh] G. Zhang (2011) All bounded type Siegel disks of rational maps are quasidisks, Ivent.
Math. 185, 421–466.
Index
Page numbers in bold indicate where a concept is defined.
(´1, 1)-differential, see Beltrami form symmetric, 18, 20

differential
absolutely continuous, 15 see Beltrami form 364
on an interval, 23 David, 364
on lines, see ACL equation, 40
access, 177 form, 35, 42
ACL, 23, 24, 366 Beltrami, E., 44
affine conjugacy, 95 Bers, L., 40, 44, 127, 145, 244, 364
Ahlfors, L., 21, 40, 44, 134, 364 Sewing Lemma, 34
Conjecture, 145 Beurling–Ahlfors extension, 83
Finiteness Theorem, 143 Bielefeld, B., 308
almost complex structure, 14, 15 bilipschitz, 31
on a Riemann surface, 36 Blaschke product, 120, 162, 190, 228, 230,
annulus, 28 245, 371, 388
area condition, 365 generalized, 121
area distortion, 32 Bojarski, B., 40, 44, 364
Arnol’d, V., 99, 103 boundary
family, 229, 234 connect sum, 267
Arzelà–Ascoli Theorem, 105 correspondence, 245
Astala, K., 21, 127 ideal, 249, 250
attracting topological, 250
p-cycle, 108 value problem, 77
fixed point, 110 bounded type number, 100, 230, 372
branched covering, 268
Böttcher coordinates, 111, 123, 155 Bryuno, A.D., 115
Baker, I. N., 170, 273 number, 101, 190, 240
domain, 116 Buff, X., 189, 347, 362
basin of attraction Bullett, S., 291, 298, 299
hyperbolic, 388
immediate, 116 Canary, D., 145
of an attracting cycle, 108 Cantor function, 25
parabolic, 113, 388 capture, 190
Beardon, A., 283 component, 192
Beltrami Carathéodory, C.
coefficient, 8, 14 semi-conjugacy or loop, 125
of a Beltrami form, 36 Theorem, 69
408
Index 409
Carleson, L., 376 David, G., 364

chordal metric, 94 conjugacy, 372
circle homeomorphism, 97 Beltrami differential, 364, 365
critical, 101 conjugacy, 365, 397
lift of, 98 homeomorphism, 364, 366
circle map Integrability Theorem, 364, 369,
analytic linearization, 101 386
critical, 101, 126 degenerate group, 145
expanding, 262 degree, 93
linearizable, 99 Denjoy, A., 99, 190
real analytic, 101, 190, 230 counterexample, 99
circumcentre, 186 Denjoy’s Theorem, 99
compactly contained, 23 Density Conjecture, 145
complex analytic manifold, 34 devil staircase, 25, 29, 32, 104
complex dilatation, 9, 13 diameter, 118
complex structure dilatation, 9, 14, 15
almost, see almost complex maximal, 27
structure of a quadrilateral, 27
analytic, 34, 42 Diophantine number, 100, 115
invariant, 17 bounded type, see bounded type number
standard, 15 constant type, see bounded type number
stretching, 208, 215 order of, 100
turning, 208, 215 distribution, 21
wringing, 208, 213 distributional derivatives, 22
conformal conjugacy, 95 Douady, A., 44, 115, 150, 219, 225, 230, 282,
conformal welding, 245 321, 337, 347, 354
conjecture rabbit, 124
hyperbolicity, 132 Douady–Earle extension, 83
MLC, 131, 132 dynamical space, 105
rigidity, 132
conjugacy, 60, 95
J -conjugacy, 167 ellipse field, see field of ellipses
David, see David conjugacy elliptic
invariant, 95, 104 Möbius transformation, 134
quasiconformal, 131 model, 255
for groups, 135 Epstein, A.L., 308, 319, 343
conjugate diameter, 11 equipotential, 123
connectedness locus, 128, 132, 191, 211, equivalence, 96
321, 343 hybrid, 125, 131, 221
of polynomial-like mappings, escape locus, 211
223 foliation, 216
continued fraction, 100, 372 essential singularity, 116
continuum, 250 exceptional set, 108
Cremer, H., 116 expanding map, 119
fixed point, 114, 115 extensions
critical point, 56, 97, 108 Beurling–Ahlfors, 83
free, 321 David, 375
marked, 133, 344 disc-annulus, 270
curve, 64 Douady–Earle, 83
closed, 64 of boundary maps, 77, 83
differentiable, 64 of conformal maps, 69, 71
exterior of, 225 of coverings, 71
interior of, 225 of quasiconformal maps, 75
Jordan, see Jordan curve to the unit disc, 81
piecewise regular, 71 external
smooth, 64 argument, 124
cycle, see p-cycle ray, 124
410 Index
Fatou component of disjoint type, 343

classification of, 116 root of, 130
periodic, 115 hyperbolic model, 255
preperiodic, 115 hyperbolicity, 119
wandering, see wandering domain conjecture, 128, 132
Fatou, P., 112, 118, 282
component, see Fatou component ideal boundary, 276
coordinates, 113 compatibility, 252
set, 107, 109 indifferent p-cycle, 108
Fatou–Shishikura inequality, 118 Integrability Theorem, 40
fibre, 177, 216 David, 364
field of ellipses, 14, 15 dependence on parameters, 42
fixed point qc, 364
attracting, 110 trans-qc, 364
Cremer, see Cremer fixed point integrating map, 40
irrationally indifferent, 114 normalization of, 42
parabolic, see parabolic fixed point interpolation
repelling, 110 on annuli, 81
Siegel, 114 on half strips, 82
superattracting, 111 on quadrilaterals, 80
Flower Theorem, 112 on standard annuli, 78
Fuchsian group, 141, 160, 244 quasiconformal, see quasiconformal
fundamental domain, 142 interpolation
invariant set, 96
Gauss, C. F., 44 totally, 107, 109
Gehring, F., 23 under a group, 135
genus, 244
geometrically finite, 120, 144, 394 J -stability, 127
Geyer, L., 232 Jacobi’s Theorem, 98, 104
Ghys, E., 102, 224, 228, 230 Jacobian, 24, 32
surgery, 225, 370 J -conjugacy, 109
Gleason, A., 150, 159 Jordan
Gluing Lemma, 253, 254 annulus, 71
Goldberg, L., 357 arc, 64
Grötzsch, H., 21 curve, 64, 119
grand orbit, 108, 135, 183 Theorem, 64
Green’s function, 123 domain, 64
Julia set, 107
Haı̈ssinsky, P., 184, 346, 365, 385 filled, 122
Harvey, W., 299 of a polynomial-like mapping, 220
hedgehog, 115 truncated, 325
Henriksen, C., 189 of a polynomial-like mapping, 220
Herman ring, 116, 122, 126, 189, 190, 232, properties, 109
233, 235 Julia, G., 107, 112
Herman, M., 101, 103, 198, 225, 228, 230
number, 101, 102, 228 Key Lemma for surgery, 60
ring, see Herman ring dependence on parameters, 62
Herman–Świa̧tek Theorem, 102, 225 Khinchin, A.Y., 373
Hölder condition, 31 Kiwi, J., 357
Hölder inequality, 38 Kleinian group, 135, 245, 267, 291
holomorphic correspondence, 293 elementary, 137
holomorphic map with respect to μ, 17, Korn, A., 44
192 Kœnigs’ Theorem, 110
holomorphic motion, 126, 208, 332
Hubbard, J.H., 21, 150, 219, 282, 343, 354 λ-lemma, 127, 332
hyperbolic component, 128, 131, 151 Lavaurs, P., 343
centre of, 130, 150 Lavrentiev, M. A., 44
Index 411
Leau, L., 112 Gluing Lemma, 253

Lehto, O., 21, 23 Measurable Riemann Mapping Theorem,
Lichtenstein, L., 44 see Integrability Theorem
lift, 98 Milnor, J., 105, 343–345, 357, 394
limb, 316, 321, 346 Misiurewicz, M.
embedding, 337, 347 map, 117
linear conjugacy, 95 point, 131
linearization, 99, 110 polynomial, 131
analytic, 101 model map, 147, 369, 386
conformal, 114 rigid, 255
parabolic, 113 modular group, 292, 296
quasisymmetric, 104 modulus, 30, 192
linearizing coordinate, 110, 151, 154 continuity of, 88
log, 312, 341, 349 of a quadrilateral, 27
Liouville number, 100 of a torus, 161
local connectivity, see set locally connected of an annulus, 28
loxodromic transformation, 134 opening, 308, 310, 315, 319, 339
Lyubich, M., 21 Montel’s Theorem, 105, 136
Morrey, C. B., 40, 44, 364
Möbius transformation, 134 Multibrot set, 133, 156
classification of, 134 multiplicity of a parabolic fixed point, 112
of finite order, 134 multiplier, 108
Mañé, R., 126 map, 130, 150, 151, 331
Mandelbrot set, 128, 224 Mumford, D., 138
generalized, 133
limb of, 316, 346 Năishul’, V. I., 114
Mandelbrot-like family, 224 near translation, 72, 308, 315, 341
map neutral p-cycle, 108
R-linear, 8 Newton’s method, 220
entire transcendental, 93, 187, 239 No Wandering Domains Theorem, 169
expanding, 119 for Kleinian groups, 143
first return, 324 for rational maps, 116, 150
geometrically finite, see geometrically normal family, 105
finite normal form, 110
hyperbolic, 119
integrating, see integrating map opening modulus, 349
linearizing, see linearizing coordinate orbit, 60, 92
meromorphic transcendental, 93, 187, 240 periodic, see p-cycle
Minkowski question mark, 295 singular, 116
Misiurewiz, see Misiurewicz map
multiplier, see multiplier map p-cycle, 92
orientation reversing, 18, 19 classification of, 108
proper, 120 multiplier of, 108
quasiconformal, see quasiconformal parabolic
mapping p-cycle, 108
quasirational, 180, 251 basin, 388
quasiregular, see quasiregular mapping fixed point, 111
quasisymmetric, see quasisymmetric multiplicity of, 112
mapping Flower Theorem, 112
rational, 93 linearization, 113
degree of, 93 Möbius transformation, 134
subhyperbolic, 119, 394 model, 256
Martin, G., 127 surgery, 391
mating, 292, 294, 296, 394 parameter
marked, 395 hyperbolic, 128
topological, 394 plane, 127
McMullen, C., 168, 225, 244, 248 space, 126
412 Index
parametrization, 64 quasimeromorphic, 180

conformal, 70 quasirational, 268
quasisymmetric, 68 quasiregular mapping, 55–57
partial derivatives, 23 properties, 58
Penrose, C., 298 pullback by, 59
petal, 112 quasisymmetric mapping, 65, 66, 85,
Petersen, C.L., 102, 365, 370 246
phase space, 105 composition, 67
Pilgrim, K., 120, 267 quotient torus, 312, 348
point
accessible, 177 radial limit, 177
critical, see critical point recurrent, 92
fixed, see fixed point Rees, M., 395
periodic, see p-cycle reflection, 18, 20, 242
pre-essential, 240 renormalization, 354, 356
preperiodic, 92 bi, 355
recurrent, 92 left, 356
regular, 116 right, 357
pole, 240 repelling
polynomial, 93 p-cycle, 108
centred, 97 fixed point, 110
monic, 97 Rickman’s Lemma, 34
unicritical, 133, 156 Riemann surface, 34
polynomial-like mapping, 219, 291 abstract, 320, 324, 339
holomorphic family of, 223 atlas, 34
postcritically finite, 394 chart, 34
prime end, 249 Riesz, F., 172, 177
principal vein, 337 Riesz, M., 172, 177
proper discontinuity, 136 rigid rotation, 97
pullback, 15, 16, 16, 55 gluing, 226
by quasiregular mappings, 59 rigidity, 331
notation, 16, 60 conjecture, 132
of Beltrami forms, 37 rotation function, 103
pushforward, 17 rotation number, 98, 103, 104, 228
combinatorial, 113, 125, 308, 355
quadrilateral, 26, 80 of a fixed point, 114
modulus of, 27 on invariant curves, 102
quasi-Fuchsian group, 141, 162, 244 rotation ring, 189, 191
quasiannulus, 71, 85 twisting, 198
quasiarc, 33, 68 Royden, H. L., 127
quasicircle, 68, 245 Rüssman, H., 115
quasiconformal
conjugacy, 60, 95 Sad, P., 126
deformation, 144, 148, 201, 245 Schleicher, D., 322
interpolation, 87, 314, 316, 351 Schottky
removability, 33 circles, 139
rigidity, 149 group, 142
surgery, see surgery sector, 308
quasiconformal mapping log-Böttcher, 309, 325
analytic definition, 22, 24 opening modulus, 308
between Riemann surfaces, 35 slope, 309
composition, 31 semi-conjugacy, 96
families of, 34, 38 Series, C., 138
geometric definition, 28 set
properties, 31 cellular, 343
quasidisc, 68, 85 exceptional, 177
quasientire, 180 full, 122
Index 413
limit, 136, 244, 291 cut and paste, 149, 162, 219, 235, 244, 248,
properties, 137 267, 282, 307, 320, 370, 386
locally connected, 120, 124, 131, 177, disc-annulus, 267
371 inverse of, 277
ordinary, 136, 291 for matings, 397
properties, 137 Ghys, see Ghys, E.
postcritical, 117 intertwining, 343
postsingular, 117 parabolic, 391
recurrent, 92 soft, 148, 151, 160, 169, 198, 207
regular, 136 topological, 147
Sewing Theorem, 245 trans-quasiconformal, 148
Shishikura, M., 118, 179, 225, 232, 235, 240, Świa̧tek, G., 230
282, 321, 333, 354, 395 symmetry, 18, 20
First Principle, 180, 182, 222 locus, 344
for transcendental maps, 186
Perturbation Theorem, 283 Tameness Conjecture, 145
Second Principle, 183, 268 Tan Lei, 120, 267, 395
Shub, M., 262 Teichmüller space, 149
Siegel disc, 114, 116, 119, 189, 225, 228, 232, test function, 21
233, 235, 370 thickening construction, 356
Siegel, C.L., 114 Thurston, W., 145
capture, 190 topological conjugacy, 95
disc, see Siegel disc Tukia, P., 184
point, 114 twisting, 198
singular
orbit, 116 Uniformization Theorem, 35
value, 116 Simultaneous, 244
singularity, 116
Słodkowski, Z., 127 value
Sobolev space, 22, 24, 365, 366 asymptotic, 116
spherical metric, 94 critical, 116
standard singular, 116
annulus, 28, 71 Virtanen, K. I., 21
half strip, 72
Steinmetz, N., 170 wandering domain, 115, 170
Straightening Theorem, 221, 374, 393 Weyl Lemma, 32, 61
Sullivan, see Sullivan, D. Wright, D., 138
Strebel, K., 385
subhyperbolic, see map subhyperbolic Yampolsky, M., 308, 319, 343
Sullivan, D., 116, 126, 133, 143, 145, 150, Yin, Y., 120
169 Yoccoz, J.-C., 101, 115, 321, 355, 371, 376
dictionary, 146 inequality, 355
Straightening Theorem, 184, 223 near parabolic bound, 382
for transcendental maps, 186 Young centre, 186
superattracting
p-cycle, 108 Zakeri, S., 190, 232, 365
fixed point, 111 Zhang, G., 232
surgery, 147 Zoretti’s Theorem, 262

Branner B., Fagella N.-Quasiconformal Surgery in Holomorphic Dynamics-CUP (2014)

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Branner B., Fagella N.-Quasiconformal Surgery in Holomorphic Dynamics-CUP (2014)

Caricato da

Copyright:

Formati disponibili

C A M B R I D G E S T U D I E S I N A DVA N C E D M AT H E M AT I C S 1 4 1

QUASICONFORMAL SURGERY IN HOLOMORPHIC

Bodil Branner is Professor Emerita at the Technical University of Denmark, Lyngby.

Núria Fagella is currently Associate Professor at Universitat de Barcelona. Her

Cambridge University Press is part of the University of Cambridge.

List of contributors page ix

4 Introduction to surgery and first occurrences 147

Carsten Lunde Petersen

Bodil Branner and Núria Fagella

A Siegel disc or a linearizing domain around an attracting

Mer8 The set of transcendental maps which are meromorphic outside

The theory of one-dimensional complex dynamical systems, understood as the

About the book structure

In Chapter 4, Introduction to surgery and first occurrences, we present

Chapter 9, Trans-quasiconformal surgery, is dedicated to the contributions

On how to read this book

1.1 The linear case: Beltrami coefficients and ellipses

Any R-linear map L : CR Ñ CR can be written, using the coordinates

Lpzq “ az ` bz, with a, b, z P C.

Figure 1.1 The unit square is mapped to the parallelogram spanned by

The absolute value of the determinant of L is the area of the parallelogram,

Lpzq “ ei α |a| z ` |μ| ei 2θ z .

Hence, L is the R-linear map Spzq “ |a| z ` |μ| ei 2θ z post-composed with

the rotation Rpzq “ ei α z. We have split L into the composition of a self-

Figure 1.2 The ellipse EpSq “ EpLq.

It is easy to check that ei θ and ei pθ `π {2q are eigenvectors of S corresponding

We shall denote by σ0 the standard conformal structure of CR , that is to

μpL´1 q “ ´μpLqei p2 arg a q , (1.1)

and hence |μpL´1 q| “ |μpLq|, which implies that

KpL´1 q “ KpLq. (1.2)

Now suppose j P t1, 2u and we have two R-linear maps Lj pzq “ aj z `

KpL1 ˝ L2 q ď KpL1 qKpL2 q,

pL1 ˝ L2 qpzq “ pa1 a2 ` b1 b2 qz ` pa1 b2 ` b1 a2 qz,

1.1.1 Geometric interpretation

Figure 1.4 E Ă CR so that LpEq is a circle in CR . A pair of conjugate diameters in

family determines a conformal structure σ in CR . For any z ‰ 0, choose the

Exercises Section 1.1

(ii) Justify that the determinant of the matrix is |a|2 ´ |b|2 .

(iii) Consider in particular Spzq “ |a| z ` |μ| ei 2θ z , and confirm that

the matrix of S in the standard basis is symmetric.

Note that LpS1 q “ EpL´1 q. Show that

and let E be the ellipse satisfying LpEq “ S1 .

(i) Show that points px, yq P E satisfy the equation

1.1.4 (Complex dilatation of an orientation reversing linear map) Given

the complex conjugate ellipse E1 equals μ1 . The formula states the

1.2 Almost complex structures and pullbacks

μf puq “ f ˚ μ0 puq or σf puq “ f ˚ σ0 puq.

Figure 1.6 The almost complex structure σf in U is the pullback of σ0 in V , under f .

T V corresponding to an almost complex structure σ in V . Let Ev denote the

we mean that f : U Ñ V and f ˚ μ2 “ μ1 , in the sense explained above.

From the linear discussion in Section 1.1 it follows that

Bz f puq ` μpf puqqBz f puq

μf puq ` μg pf puqqe´i 2 argpBz f puqq

Furthermore, note that if f is holomorphic then (1.9) reduces to

Definition 1.1 (f-invariant almost complex structure) Let U be an open

Almost complex structures that are f -invariant will be of crucial importance

We observe that if the inverse map f ´1 is also absolutely continuous then

As a final remark, note that we even may allow f to be non-invertible as

1.2.1 Almost complex structures and symmetries

Definition 1.2 (Pullback under orientation reversing maps) In the setup

Observe that f is orientation preserving and therefore the definitions make

Definition 1.3 (Symmetries) A Beltrami coefficient μ is symmetric with

or equivalently if cf μ “ μ, where cpzq “ z.

Define f as the unique non-decreasing function, which on x P as above

Theorem 1.19 (Quasiconformal removability of quasiarcs) If is a