Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
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
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
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
vii
viii Contents
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
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.
„ 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
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.
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.
7
8 Quasiconformal geometry
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
i pθ` π
2q
e
|a|p1´|μ|q
S R
ei θ
|a|p1`|μ|q
θ
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.
10 Quasiconformal geometry
1
L´1 pwq “ paw ´ bwq.
|a|2 ´ |b|2
It follows that
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
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.
J pz q
L
“ L´1 pi Lpzqq
i Lpzq
J J pzq “ ´z
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
12 Quasiconformal geometry
Figure 1.5 Some similar concentric ellipses. They are equal up to scaling
by a real constant.
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 “ ´i , Bz f “ `i .
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
1.2 Almost complex structures and pullbacks 15
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
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
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
16 Quasiconformal geometry
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
f ˚ μg “ f ˚ pg ˚ μ0 q “ pg ˝ f q˚ μ0 “ μg ˝f .
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
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.
G˚ pf ˚ μq “ pf ˝ Gq˚ μ “ pF ˝ f q˚ μ “ f ˚ pF ˚ μq “ f ˚ μ.
Kf ´1 “ Kf .
18 Quasiconformal geometry
In particular
f f μ0 :“ μf and Kf :“ Kf .
μpzq “ μpzq,
f f μ0 :“ μf and Kf :“ Kf .
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.
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.
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.
22 Quasiconformal geometry
ordinary sense.
We are ready for the first definition of a quasiconformal mapping.
(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 .
1.3 Quasiconformal mappings 23
Remarks 1.6
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.
1.3 Quasiconformal mappings 25
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
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
26 Quasiconformal geometry
x1 x2 xn
f pxq “ .x 1 x 2 ¨ ¨ ¨ x n ¨ ¨ ¨ “ ` 2 ` ¨¨¨ ` n ` ¨¨¨
2 2 2
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.
z1 z4 ϕ pz4 q ϕ pz3 q
ϕ
b
z2
z3 ϕ pz1 q a ϕ pz2 q
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.
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.
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.
30 Quasiconformal geometry
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
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
The next statement comes directly from the first analytic definition, although
it is less obvious when we think about it in geometric terms:
Finally, the following area distortion estimate due to Astala will be useful
later on, see [As].
AreapφpEqq ď MpAreapEqq1{K ,
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].
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
1.3 Quasiconformal mappings 35
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.
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:
1.3 Quasiconformal mappings 37
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)
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.
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
1.4 The Integrability Theorem 39
}μ}8 “ k ă 1.
}μ}8 “ k ă 1.
1.4 The Integrability Theorem 41
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 .
42 Quasiconformal geometry
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 .
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.
(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).
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).
1.4 The Integrability Theorem 45
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)
}T h}p ď Cp }h}p
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
1.4 The Integrability Theorem 47
φ ˚ μ0 “ pI ´1 q˚ φr˚ I ˚ μ0 “ pI ´1 q˚ φr˚ μ0 “ pI ´1 q˚ μ
r “ I˚ μ
r “ μ,
48 Quasiconformal geometry
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
r :“ pφ2 q˚ μ,
μ
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
φ2
pC, μq / pC, μ rq
HH
HH
HH
HH φr
φ H$
pC, μ0 q
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˚ .
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
hence
ν ` ν0 ν ´ ν0
a“ , b“ ,
ν0 ` ν0 ν0 ` ν0
ζ ÞÑζ `ν
H ÝÝÝÝÝÑ
0
H
§ §
Lν đ
§ §L
đ ν
ζ ÞÑζ `ν
H ÝÝÝÝÝÑ H
1.5 An elementary example 51
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 .
#
pexp ˝Lν ˝ logqpzq if z P D˚
φν pzq :“
0 if z “ 0
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
1 1 1 1
mod A0 “ log and mod Aν “ log ν .
2π |λ0 | 2π |e |
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.
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
54 Quasiconformal geometry
and
#
μν pzq if z P D,
r ν pzq :“
μ
pM0n q˚ μν pzq if z P An .
A2
A1
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:
1.6 Quasiregular mappings 55
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
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 :
g “ f ˝ φ,
f ˚ pμ0 q “ pφ ´1 q˚ g ˚ pμ0 q “ pφ ´1 q˚ μ “ μ0 ,
g
pU, μq / pgpU q, μ0 q
q8
qqqqq
q
qqq f
φ
pφpU
p q, μ0 q
1.6 Quasiregular mappings 57
We prove that this definition is indeed equivalent to the two previous ones.
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.
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.
pC, μq pC, μq
φ φ
pC, μ0 q pC, μ0 q
fλ ˝ φλ ” φλ ˝ gλ
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
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 Application to holomorphic dynamics 63
64
2.1 Preliminaries: quasisymmetric maps and quasicircles 65
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.
Therefore G : T ˆ T Ñ R defined by
# |H px q´H py q|
if x ‰ y,
Gpx, yq :“ |x ´y |
| H 1 pxq | if x “ y,
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
68 Extensions and interpolations
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
f
D G
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
2.2 Extensions of mappings from their domains to their boundaries 71
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].
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
2.2 Extensions of mappings from their domains to their boundaries 73
1 h
σ
2
C
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).
: z ÞÑ eπ z : z ÞÑ eπ z
1 2
Dγ1 Dγ2
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
´eπ f px `i q ´F p´uq
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
´C 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).
2.2 Extensions of mappings from their domains to their boundaries 75
fp :“ gp ˝ φ : R
p Ñ R,
p
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.
Exercises Section 2.2
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.
2.3 Extensions of boundary maps 77
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.
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
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
80 Extensions and interpolations
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.
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 .
2.3 Extensions of boundary maps 81
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
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.
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 .
and
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
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.
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]).
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.
ż1
1
Bf px ` i yq :“ tf px ` tyq ` f px ´ tyqudt
2 0
ż1
`i tf px ` tyq ´ f px ´ tyqudt,
0
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
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.
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 `i 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 .
R
p1 R1 R2 R
p2
D gp D
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.
fo
A1
γ1i γ2i
f
γ1o γ2o
A2
fi
g
r1 1 r2 1
gi
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.
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
and
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
ψ8 : Ar,1{r Ñ Ar ˚ ,1{r ˚ .
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.
2.3 Extensions of boundary maps 91
(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
92
Preliminaries on dynamical systems and actions of Kleinian groups 93
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.
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.
specified. Most of the facts which follow can easily be checked, and those
which are not obvious can be found for example in [Ro].
pf p q1 px0 q “ pg p q1 py0 q.
96 Preliminaries on dynamical systems and actions of Kleinian groups
We shall see that many properties of dynamically defined sets are preserved
under topological conjugacy. As an example, we mention the following.
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.
f
X ÝÝÝÝÑ X
§ §
h1 đ
§ §h
đ 2
g
Y ÝÝÝÝÑ Y
Equivalences other than topological ones are defined in the obvious way.
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.
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 .
˝ F “ f ˝ .
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
(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.
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]).
3.2 Circle homeomorphisms and rotation numbers 101
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.
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).
102 Preliminaries on dynamical systems and actions of Kleinian groups
ρ pα q
1.0
Ò
0.8
0.6
0.4
0.2
Figure 3.1 A devil’s staircase, the graph of the rotation function for
fα pxq “ α ` x ` 2π 1 sinp2π xq pmod 1q.
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].
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
Jf “ SzFf .
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).
3.3 Holomorphic dynamics: the phase space 107
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
108 Preliminaries on dynamical systems and actions of Kleinian groups
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
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:
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.
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.
We define the exceptional set Epf q as the set of points with a finite grand orbit.
Theorem 3.27 (Properties of Julia sets and Fatou sets) Let the map
f : S Ñ S be in Rat, Ent or Ent˚ .
We end this section defining a special type of conjugacy which is often used
in holomorphic dynamics.
h˝f “g˝h
f pzq “ λz ` a2 z2 ` a3 z3 ` ¨ ¨ ¨ ,
(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]).
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.
(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
112 Preliminaries on dynamical systems and actions of Kleinian groups
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
The following theorem is due to Leau [Lea], Fatou [Fat] and Julia [J] in
successive approximations.
P1 Y ¨ ¨ ¨ Y P2m Y t0u
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).
3.3 Holomorphic dynamics: the phase space 113
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).
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]).
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.
(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
3.3 Holomorphic dynamics: the phase space 115
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]).
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].
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.
3.3 Holomorphic dynamics: the phase space 117
The set of singular values, denoted by Singpf ´1 q, may contain three types
of points:
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.
(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 .
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 ´n 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.
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.
3.3 Holomorphic dynamics: the phase space 119
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.
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.
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).
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´ ´a
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.
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.
f pzq “ ad zd ` ad ´1 zd ´1 ` ¨ ¨ ¨ ` a1 z ` a0
`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.
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.
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.
3.3 Holomorphic dynamics: the phase space 123
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.
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
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
3.3 Holomorphic dynamics: the phase space 125
(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.
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.
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.
Exercises Section 3.3
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
tfλ : C
p Ñ C,
p λ P u Ă Ratd ,
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.
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.
The conjecture has been proven for parameters on the real line [GS, KSvS].
3.4 Families of holomorphic dynamics: parameter spaces 129
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.
130 Preliminaries on dynamical systems and actions of Kleinian groups
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).
The most important conjecture about the Mandelbrot set concerns a topo-
logical property.
Conjecture 3.62 (MLC) M is locally connected.
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.
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:
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 ´n ˝ η ˝ γ n Ñ Id as n Ñ 8 (or
n Ñ ´8), contradicting the discreteness of the group.
ˆC p ÝÑ C, p
pγ , zq ÞÝÑ γ pzq.
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.
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.
(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
138 Preliminaries on dynamical systems and actions of Kleinian groups
(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:
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.
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.
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.
140 Preliminaries on dynamical systems and actions of Kleinian groups
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.
Figure 3.16 Similar to the previous figure, illustrating the group in Example 3.76.
The ordinary set has infinitely many bounded components.
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.
Q1
Figure 3.17 The fundamental domain Q1 in Example 3.74 and the Riemann surface
which corresponds to 1 pq{.
In Section 4.3 it is explained how the corollary below follows from Ahlfors’
Finiteness Theorem.
series of papers Mahan Mj has claimed an extension of this result to all finitely
generated Kleinian groups.
147
148 Introduction to surgery and first occurrences
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
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.
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
152 Introduction to surgery and first occurrences
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,
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
p λ “ 0 in CzApαq.
μ p
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
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
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
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.
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.
˚ 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.
158 Introduction to surgery and first occurrences
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
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:
Set
k “ sup kpλq.
λPU0
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 .
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.
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
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|.
D D
Aλ0
Bλ
Aλ
r 1 r 1
hλ
Bλ0
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.
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.
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 .
˝
# 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
$
&μ0
’
’ on CzApαq,
η λ “ ηλ on A˝0 ,
’
%pQn q˚ η
’
on Q´ n ˝ ˝
c0 λ c0 pA0 qzA0 .
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.
168 Introduction to surgery and first occurrences
Remark: The exercises below are special cases of a more general theorem of
McMullen (see Section 7.5).
(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
170 Introduction to surgery and first occurrences
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
4.3 No wandering domains for rational maps 171
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
172 Introduction to surgery and first occurrences
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.
ψ : B ˆ D ÝÑ D,
pλ, zq ÞÝÑ ψλ pzq,
satisfying:
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
4.3 No wandering domains for rational maps 173
$
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
gλ :“ φλ ˝ f ˝ φλ´1 .
ϕt :“ φλ´p10q ˝ φλpt q ,
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.
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
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.
4.3 No wandering domains for rational maps 175
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.
which satisfies
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.
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
178 Introduction to surgery and first occurrences
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
179
180 General principles of surgery
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 ´N pU q, for some N ě 0.
Then f is quasirational.
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.
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 ´n pU q Ď ¨ ¨ ¨ Ď f ´n pU q,
ně0
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
5.1 Shishikura principles 183
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.
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:
184 General principles of surgery
#
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.
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
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
186 General principles of surgery
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
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
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 .
Such families have been studied by Zakeri [Z1] (see also [BHe]).
First, fa p´zq “ ´f´a pzq, so that the affine map z ÞÑ ´z conjugates fa to
f´a . 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.
6.1 Deformation of rotation rings 191
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
Proof On the one hand, if a 2 “ 4ρ, the cubic polynomial fa has a critical
point at ´a{3 and fa p´a{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 .
φα pzq :“ z |z|α ´1 .
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
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.
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
6.1 Deformation of rotation rings 195
α : Hα Ñ Ar α
1
φα ˝ ϕ ˝ Q´
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.
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
Fα :“ Qα ˝ f ˝ Q´
α
1
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
´ ´a ¯¯
bα P p0, 1{9q and ρα “ exp 2π i Tθ bα .
as required.
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.
SX :“ tx ` i y | X ă x ă 0u with X “ logprq.
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 .
6.1 Deformation of rotation rings 199
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.
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.
exp exp
A|eτ0 | / A|eτ | .
φτ
ςτ “ pφτ ˝ φq˚ σ0 .
Fτ :“ Qτ ˝ f ˝ Q´
τ :CÑC
1 p p
τ : H Ñ A|eτ |
1
φτ ˝ ϕ ˝ Q ´
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 .
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 .
6.1 Deformation of rotation rings 203
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
h˚ ςτ “ ςτ `2π i .
h2
f ´1 pH q / f ´1 pH q .
f f
H / H
h1
hk
C
p / C
p
f f
C
p / C.
p
hk´1
204 Soft surgeries
A point z 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
gλ pzq “ ρz ` aλ z2 ` z3 ,
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 .
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 .
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:
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
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.
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
and spread to the remaining part of C by the dynamics of Ppa,bq . The Beltrami
coefficient is well defined on Cz ně0 Pp´a,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.
φω,pa,bq : pC,
p μω,pa,bq q Ñ pC,
p μ0 q,
normalized so that:
diagram:
6.2 Branner–Hubbard motion 213
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
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
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.
(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
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
Ñ pCzDr s 1 s , μ0 q
pa,bq
It follows that
Applying this to the fastest escaping critical value of P pa, bq, say
Ppa,bq paq, with Böttcher coordinate equal to
we obtain that Ppaω ,bω q paω q is the fastest escaping critical value of Ppaω .bω q
with Böttcher coordinate
3ps `i t q 2π i α
ϕpaω ,bω q pPpaω ,bω q paω qq “ rpa,bq e .
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
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].
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 | gp`aq “ log r ą
gp´aqu, Sr´ “ tpa, bq P Sr | gp´aq “ log r ą gp`aqu, and their common
boundary BSr` “ BSr´ “ tpa, bq P Sr | gp`aq “ log r “ gp´aqu.
The critical value P p`aq has three preimages when counted with multi-
plicity, the critical point `a counts for two and its co-critical point ´2a for
one. Analogously, the preimages of the critical value P p´aq are the critical
point´a 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.
6.2 Branner–Hubbard motion 217
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.
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.
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).
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 ´n 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 .
7.1 Polynomial-like mappings and the Straightening Theorem 221
The proof of the first part of the theorem is by surgery and is presented
below.
Proof
V
zd
f
ψ
U
ρ ρd
A0
P :“ φ ˝ F ˝ φ ´1
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.
V :“ tpλ, zq | z P Vλ u,
U :“ tpλ, zq | z P Uλ u,
f pλ, zq :“ pλ, fλ pzqq.
CF :“ tλ P | Kfλ is connected u.
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.
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 fr´n pGγ q, whenever defined
’
’ Ť ´n
%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.
(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.
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 “ ´e2π 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
´e2π 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:
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,
230 Cut and paste surgeries
since Ha,t has no zeros. One can check that such a map must be of the form
Eθ pzq “ e2π i θ zez .
?
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
7.2 Gluing Siegel discs along invariant curves 231
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
r´n pS1 qq “ JP .
Ť
φθ p ně0 Bθ θ
Figure 7.7 Left: The set ně0 Bθ´n pS1 q. Right: The set ně0 B r´n pS1 q. The right
Ť Ť
θ
figure is quasiconformally homeomorphic to Figure 7.5, via the integrating map φθ .
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.
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
z´a
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
234 Cut and paste surgeries
(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
in relation to the semi-standard map Sθ pzq “ e2π i θ zez (see Exercise 7.2.4
below).
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
(a) |γ “ h;
(b) pGγ q “ CzG
p γr and pCzG
p γ q “ Gγr ;
7.3 Turning Siegel discs into Herman rings 237
f fr
r
h
Gγ Gγr
ϕ ϕr
Rθ R1´θ
r L r
1
r
1
0 0
Figure 7.11 The global definition of the gluing map , as the result of pasting four
maps together.
238 Cut and paste surgeries
Let
Ro : CzG
p γo Ñ G i
γr and Ri : Gγ i Ñ CzG
p γro
define
ho : Ao Ñ A
ri and hi : Ai Ñ A
ro
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.
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]).
Moreover,
ebz
F pzq “ az2 ,
z`1
for some parameters a, b P C˚ .
Proof We start by
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.
242 Cut and paste surgeries
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
7.3 Turning Siegel discs into Herman rings 243
$
&R on CzG
’
’ p γ1,
:“ h on A,
’
’
%ϕ on Gγ 1 .
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´
´a 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
´a
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 .
D
R
pU
γ
ψ
U
D
V
R
pV
QS{ „ ÐÑ QC ` { «,
7.4 Simultaneous uniformization of Blaschke products 247
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
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
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].)
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.
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].
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 .
fp :“ φpV´1 ˝ gp ˝ φpU : I pU q Ñ I pV q.
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.
g
g“f
U g “ φV´1 B φU V
φU φV
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
is quasirational.
7.5 Gluing along continua in the Julia set 255
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
Definition 7.32 (Rigid models) We define the three rigid models for proper
self-maps of the unit disk D as:
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
The main application is stated in the following theorem ([McM1, Prop. 6.9]).
(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.
7.5 Gluing along continua in the Julia set 257
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.
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
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
7.5 Gluing along continua in the Julia set 259
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 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
260 Cut and paste surgeries
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.
fp
RU RV
U V
φ1 φ2
U1
f
V1
ϕ1 ϕ2
ϕ2 ˝ f ˝ ϕ1´1
ϕ ϕ
z ÞÑ zd
In the proof of the above lemma we shall apply the following theorem of
Shub (cf. [Shu], [dMvS, Chapt. II, Sect. 2]).
The next proposition deals with the buried case where BU is not part of the
boundary of any Fatou component.
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.
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 .
264 Cut and paste surgeries
Wn
Wn ´ 1
An
f
f P2
γn`1 1
γn`1
An A1n A1n
An γn1
γn
Wn Wn1
1 1
An
An A1n A1n
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.
266 Cut and paste surgeries
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.
268 Cut and paste surgeries
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 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.
is quasirational.
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 .
7.6 Disc-annulus surgery on rational maps 271
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 .
272 Cut and paste surgeries
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
274 Cut and paste surgeries
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
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
components.
Next we lift φ0 to obtain φ1 . We call this the lifting step.
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.
V h V
H G F H G
f p pV q f p pV q
g
f pV q f pV q
• 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;
Then
$
&f on CzV
’
’ p
F :“ h on H
’
’
%g on V zH
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.
Exercises Section 7.6
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
282 Cut and paste surgeries
(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.
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
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.
7.7 Perturbation and counting of non-repelling cycles 285
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
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)
286 Cut and paste surgeries
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.
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.
7.7 Perturbation and counting of non-repelling cycles 287
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 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
288 Cut and paste surgeries
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 .
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.
7.7 Perturbation and counting of non-repelling cycles 289
pψ ´1 ˝ f p ˝ ψqpuq “ λu ` Opuk `3 q,
pψ ´1 ˝ gεp ˝ ψqpε q Ă ε .
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.
290 Cut and paste surgeries
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:
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.
aj z ` bj
γj pzq “ , 1 ď j ď n,
cj z ` dj
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.
ψc : CzD
p Ñ CzK
p c,
• 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` .
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
(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 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
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].
P
Q
R –1S U T 0 +1
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.
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.
7.8 Mating a group with a polynomial 303
• 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.
304 Cut and paste surgeries
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.
(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.
306 Cut and paste surgeries
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.
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.
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
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 ´n pζ q and all rays landing at these points. Their arguments
Ť
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.
0 ρ
η{d η
Figure 8.3 Disjoint bounded sectors of slope s “ smax in Hr .
f
S B L S “ γ1
B R S “ γ2
W
W1
ζ
1
z ÞÑ w “ plog ˝ϕqpzq.
log λq
2
modζ pS, P q q “ Arctanp2π sq.
q log d
8.1 Preliminaries: sectors and opening modulus 313
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.
314 Cut and paste surgeries with sectors
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 qzfj´n 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.
8.1 Preliminaries: sectors and opening modulus 315
ψ ˝ f q “ frq ˝ ψ.
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).
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 .
318 Cut and paste surgeries with sectors
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
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
8.1 Preliminaries: sectors and opening modulus 319
the map Q p c . Denote the common modulus of these ‘in between’ annuli by
m pc, sq. These moduli are independent of η.
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.
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
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
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
Observe that the image p,q pLp{q q is independent of p. Hence we have the
following corollary.
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.
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.
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.
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
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
8.2 Creating new critical points 325
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
326 Cut and paste surgeries with sectors
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 .
8.2 Creating new critical points 327
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
328 Cut and paste surgeries with sectors
$
μ on Src ,
’´ 0 ¯ ˚
’
’
& fcp2q μ0 on Sc1 ,
’
’
μc :“ ´ p2q ¯˚ p2q
’
’
’ pfc qn μc on pfc q´n pSc1 q, for n ě 1,
’
’ p2q
on Xc1 z n pfc q´n 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:
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
See Exercise 8.2.4. Finally we need to check that λpcq is independent of the
many choices made during the construction.
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 .
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 .
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
8.2 Creating new critical points 331
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
332 Cut and paste surgeries with sectors
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
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 ,
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 ˚ .
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 ,
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.
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
8.2 Creating new critical points 335
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
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
$
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
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 .
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
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
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
8.3 Embedding limbs of M into other limbs 341
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
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 .
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.
342 Cut and paste surgeries with sectors
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 q´n pSL{R q. Set μc :“ μ0 on the remaining part of Xc ,
Ť
p2q
i.e. on Xc z ně0 pfc q´n pSL{R q. By construction μc is f p2q -invariant. The
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
with critical points a and ´a. 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.
344 Cut and paste surgeries with sectors
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
H pA, Dq “ pλ´ `
H pA, Dq , λH pA, Dqq
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 .
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 .
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
Hp{q,m : Hp{q,m Ñ D ˆ D
8.4 Intertwining surgery 347
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.
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
that hn pA1,Rn q Ą K
p i , where
f
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
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
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
8.4 Intertwining surgery 351
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
R i : Mi Ñ L
r i `q ´k
R
ri : M
r i Ñ Li ´ q ` k ` 1
Ni Ą Si Ą Mi and N
ri Ą Sri Ą M
ri
´1
qď ´1
qď
r:
ψ pSri zM
ri q Ñ pNi zSi q,
i “0 i “0
with
r r “R
ψ| r r “ φ ´1 | r .
r i | r and ψ|
B Mi B Mi B Si B Si
352 Cut and paste surgeries with sectors
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 ´n 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 .
8.4 Intertwining surgery 353
π : Kc Ñ KPA,D and π
r : Krc Ñ KPA,D
Bπ “ 0 a.e. on Kc and B π
r “ 0 a.e. on Krc .
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.
modζ Si ą modrζ K
p rc and modr Sri ą modζ K
ζ
pc .
modζ Si ą modrζ L
r j and modr Sri ą modζ Lj
ζ
modζ Mi “ modrζ L
r j and modr M
ζ
r i “ modζ Lj .
354 Cut and paste surgeries with sectors
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
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].
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).
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
P : 1 Ñ
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.
8.4 Intertwining surgery 357
and we set
8
ď
q P :“
K P ´i pKR Y K q.
i “0
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.
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.
358 Cut and paste surgeries with sectors
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.
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
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
R ˆ pA, Dq “ R ˆ pA,
r Dq,
r
and
Pr : R
r R and Pr :
r1 Ñ r1 Ñ
r .
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 |BG1 “ Pr ˝ ϕ0 |BG1 .
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 .
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 ,
364
9.1 David maps and David–Beltrami differentials 365
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|.
9.1 David maps and David–Beltrami differentials 367
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.
368 Trans-quasiconformal surgery
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
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
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
F “ H ´1 ˝ R ˝ H on U ,
ϕ ˝ F ˝ ϕ ´1
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.
JF “ BKF “ BAF .
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.
372 Trans-quasiconformal surgery
1
θ “ ra1 , a2 , a3 , . . .s “
1
a1 `
1
a2 `
a3 ` ¨ ¨ ¨
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.
(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.
D U0
g0
Figure 9.1 The set U0 and the branch g0 of f ´1 .
374 Trans-quasiconformal surgery
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 ´n pU0 q. Then ν is a finite measure since
Ť
νpEq ď CpAreapEqqβ .
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
# 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
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 .
1 |I 1 n | 1
ď 1 ď ,
1 ` an`1 |I n´1 | an`1
1 |In |
ď ď C.
C |In´1 |
Qn :“ txj | 0 ď j ă qn u and Q1 n :“ tx 1 j | 0 ď j ă qn u
378 Trans-quasiconformal surgery
1{2 ă |I |{|J | ă 2.
1
Mn pxq :“ pxr ´ x q,
2
9.2 Siegel discs via trans-quasiconformal surgery 379
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
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,
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.
rx, ys X Q
rn`1 “ tx “ t0 ă t1 ă ¨ ¨ ¨ ă tk “ yu,
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
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
´i , 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 ´i 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.
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
i
P ˝ M ˝ P ´1
Id 1
P ´1 ˝ M ˝ P
´i
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
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).
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].
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 ´n pQ0 q.
The map z ÞÑ wpzq “ Log λ{ Log z conjugates f on Dλ´n zS to
w
g : w ÞÑ (parabolic model)
1`w
on the cusp C “ wpDλ´n zSq with vertex at the origin (see Figure 9.6).
S X Dλn
g
θ Qn
f λ´pn`1q ´n 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 .
˝
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 |
f
mod Rn 1
Kχ — g :“ Kn — — n.
mod Rn lnp1 ` n1 q
zd ` a
Bpar pzq “ , a “ pd ´ 1q{pd ` 1q
1 ` azd
and
zd ` b
Bpzq “ , 0 ă b ă a.
1 ` bzd
Remark 9.22 In (ii), the inverse branches are chosen so that no sub-branch
fixes 1.
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.
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
Hence,
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).
(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]).
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
(see Exercise 9.1.2). Moreover, the diagram commutes on the Julia sets. There-
fore we have the following result.
We are now ready to state our main application, which will be proved in the
next section.
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´ .
´ ´
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.
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.
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
[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
[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
408
Index 409
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